Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
The small dollar political donor: why regular folks give money to politics
(USC Thesis Other)
The small dollar political donor: why regular folks give money to politics
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
The Small Dollar Political Donor
Why regular folks give money to politics
by
Joshua Timm
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulllment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
POLITICAL SCIENCE AND INTERNATIONAL RELATIONS
December 2021
Copyright 2021 Joshua Timm
Contents
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
1 Why Do People Contribute? 1
1.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Introduction and Contemporary Motivations for Research . . . . . . . . . . . 3
1.2.1 Candidates Refuse PAC Money . . . . . . . . . . . . . . . . . . . . . 4
1.2.2 Popular Press Coverage of Small Donors . . . . . . . . . . . . . . . . 6
1.2.3 Disclosure of Small Donor Donations increased in 2018 . . . . . . . . 7
1.2.4 Demonstrated commitment from the Democratic party to expand the
role of small donors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2.5 Money is Here to Stay . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.6 A Networked Democracy . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 Literature Review: Existing Theories of Individual Political Contributions . 16
1.3.1 Small Contributions are Substantively Dierent From Large Contribu-
tions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.2 Material, Solidary, and Purposive Incentives . . . . . . . . . . . . . . 18
ii
1.3.3 Civic Gratications and \Warm Glow" Giving . . . . . . . . . . . . . 20
1.3.4 Resource-Based Models of Giving . . . . . . . . . . . . . . . . . . . . 22
1.3.5 Income and Solicitation Eects . . . . . . . . . . . . . . . . . . . . . 23
1.3.6 Ideology and Strategic Contributors . . . . . . . . . . . . . . . . . . . 23
1.3.7 Rational Choice Theory and Political Contributions . . . . . . . . . . 25
1.4 The Two-part Small-Donor Behavior Framework: Traits x Solicitation . . . . 29
1.5 The Anxious Avoidance Theory of Contributor Behavior . . . . . . . . . . . 30
1.5.1 Political Anxiety: A Major Feature of Society . . . . . . . . . . . . . 37
1.5.2 Anxiety . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
1.5.3 Why Use Anxiety to Fundraise? The Theory of Anxious Avoidance-
Based Contributing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
1.5.4 Measurement of Anxiety . . . . . . . . . . . . . . . . . . . . . . . . . 43
1.6 Outline of this Dissertation . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2 Who Contributes? 48
2.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.3 Literature Review: The Traits that Predict Political Contributions . . . . . . 51
2.4 Research Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.4.1 Research Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.4.2 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.4.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
iii
2.6 Building a Predictive Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
2.7 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3 Beyond the Resource Model 86
3.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.2 The Importance of Democratic Partisanship, Liberal Ideology, and Contact . 87
3.2.1 Asymmetrical Partisan and Ideological Contact Rates . . . . . . . . . 90
3.2.2 Contact by Email or Text . . . . . . . . . . . . . . . . . . . . . . . . 93
3.2.3 Eect of party-matched solicitation . . . . . . . . . . . . . . . . . . . 95
3.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4 The Anxious Avoidance Model of Contributor Behavior 100
4.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.3 Modern Fundraising Emails: Anxious, Apocalpytic, and Frantic . . . . . . . 103
4.4 Theory: Fundraising Through Avoidance of Negative Emotions Leads to
Avoidance of Contributing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.5 Research Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.5.1 Research Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.5.2 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.6 Data and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.6.1 Collecting the Data: Political Emails . . . . . . . . . . . . . . . . . . 113
4.6.2 Classifying Anxiety . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.6.3 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
iv
4.7 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
5 Political Contribution Behavior In Anxious Individuals 140
5.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
5.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
5.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
5.3.1 The Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
5.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
5.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
5.6 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
6 Conclusion 166
6.1 Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
6.1.1 Study 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
6.1.2 Study 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.1.3 Study 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
6.2 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
6.3 Limitations and Suggestions for Future Research . . . . . . . . . . . . . . . . 174
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
v
List of Tables
2.1 CCES Multiple Regression Results: Eect of Political/Social Characteristics
on Donating to a candidate . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.2 CCES Regression Results: Eect of Income on Donating to a Candidate . . 69
2.3 CCES Regression Results: Eect of Education on Donating to a Candidate . 70
2.4 CCES Regression Results: Eect of Race on Donating to a Candidate . . . . 71
2.5 ANES Logistic Regression Results: Eect of Background Traits on Donating
to Any Political Entity, 2012 - 2016 . . . . . . . . . . . . . . . . . . . . . . . 74
2.6 ANES Regression Results: Eect of Background Traits on Donating to Vari-
ous Political Entities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.7 Logistic Regression: Contact Method . . . . . . . . . . . . . . . . . . . . . . 77
2.8 Logistic Regression: Simple Contact Count Model . . . . . . . . . . . . . . . 79
3.1 Contact DV Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.2 Eect of Being Contacted by Democrat vs Republican Party on Donating . . 96
3.3 Eect of Being Contacted by Democrat vs Republican Party on Donating to
parties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.1 OLS: Eect of Anxious Emails on Small Donor Fundraising . . . . . . . . . . 123
4.2 Log-Linear Model: Eect of Emails and Time on Fundraising Below $200 . . 127
vi
4.3 OLS: Eect of Emails and Time on Fundraising Below $200 . . . . . . . . . 130
4.4 OLS: Eect of Emails and Time on Fundraising Below $200 . . . . . . . . . 133
5.1 Logistic Regression: Eect of anxiety on propensity to contribute . . . . . . 152
5.2 OLS: Eect of anxiety on contribution frequency . . . . . . . . . . . . . . . . 154
5.3 Logistic Rgression: Eect of Anxiety on Propensity to Avoid Campaign Contact157
6.1 Low Income Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
6.2 High Income Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
6.3 Log-Linear Model: Eect of Emails and Time on Fundraising Below $200 . . 210
6.4 Logistic Regression: The eect of GAD-7 and IUS-12 items on contribution
likelihood . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
vii
List of Figures
1.1 Histogram displaying the number of articles on Proquest per day about small
donors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 CRP Individual Contributions Breakdown by Source . . . . . . . . . . . . . 9
1.3 Line graph displaying the percent of US Adults that at least occasionally use
the internet. Source: Pew Research Center (2019) . . . . . . . . . . . . . . . 13
1.4 Line graph displaying the percent of US Adults that at use at least one social
media site. Source: Pew Research Center (2019) . . . . . . . . . . . . . . . . 14
2.1 Line graph displaying the percentage of individuals in the CCES who con-
tributed after being solicited and without being solicited (spontaneously). . . 59
2.2 Line graph displaying the percentage of individuals in the CCES who either
donated or did not donate. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.3 Line graph displaying the percentage of individuals in the CCES who were
solicited and either contributed or did not contribute. . . . . . . . . . . . . . 62
2.4 Model Performance, CCES Data Predicting Donations . . . . . . . . . . . . 82
3.1 Line graph depicting the distribution of contributors by party ID between
2006 - 2018 according to CCES data. . . . . . . . . . . . . . . . . . . . . . . 88
viii
3.2 Line graph depicting the distribution of contributors by ideology between 2006
- 2018 according to CCES data. . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.3 Distribution of contacted individuals by party ID, 2006 - 2016. . . . . . . . . 91
3.4 Distribution of Individuals Contacted via Email/Text by Party ID, 2010 - 2016 94
3.5 Distribution of Individuals Contacted via Email/Text by Party ID, 2010 - 2016 95
4.1 This is an image sent from the Kamala Harris campaign for President, sent
on May 4, 2019 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.2 An email notication from Joe Biden, sent by the DNC HQ . . . . . . . . . 106
4.3 An example of an actual job MTurk workers were asked to do . . . . . . . . 115
4.4 A histogram displaying the distribution of anxiety prediction scores on a scale
from 0-1. 38% of the emails were classied as having an anxiety likelihood of
0.1, which is the threshold for classifying an email as anxiety-inducing. . . . 117
4.5 Graph displaying the mean number of emails and anxious emails sent by
campaigns over the period of study, July 2019 - November 2020. Generally,
campaigns send 3-4 emails per week, 2-3 of which were classied as anxiety-
inducing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.6 Graph displaying the mean number of anxious emails sent by campaigns over
time, grouped by candidates from the Democratic party and Republican party.
Democrats typically send slightly fewer anxious emails per week. . . . . . . . 120
4.7 Graph displaying the number of anxious emails sent by Donald Trump and
Joe Biden per week. Joe Biden sent slightly fewer anxiety-inducing emails per
week than Donald Trump did. . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.8 Graph displaying the mean weekly contribution total for all small contribu-
tions below $200, July 2019 - November . . . . . . . . . . . . . . . . . . . . 122
ix
4.9 Graph displaying the mean weekly contribution total for all small contribu-
tions below $200, June 2020 - November 2020. Grouped by non-Presidential
candidates in the top 10% of spending and bottom 90% of spending. . . . . . 125
4.10 Fundraising predictions for a candidate at various levels of anxiety over time,
holding all other features at their means. Includes all candidates. Black lines
are marked between Week 63 and 65 of an entire campaign cycle, correspond-
ing to between September 7 - 21. . . . . . . . . . . . . . . . . . . . . . . . . 128
4.11 Fundraising predictions for a candidate at various levels of anxiety over time,
holding all other features at their means. Includes all candidates. . . . . . . 135
4.12 Graph displaying the mean number of emails and anxious emails sent by
campaigns over the period of study, July 2019 - November 2020. Generally,
campaigns send 3-4 emails per week, 2-3 of which were classied as anxiety-
inducing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.13 A fundraising text sent at 6:31pm on May 26, 2021 . . . . . . . . . . . . . . 139
5.1 Primary DV Instrument: Has the subject ever made a political contribution 143
5.2 Secondary DV Instrument: Has the subject made a political contribution in
the past 12 months. Answers `Denitely yes' and `Probably yes' were coded
as a 1 for contributing, and other answers were coded as a 0. . . . . . . . . . 143
5.3 Tertiary DV Instrument: How frequently does the subject makes political
contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
5.4 Dependent Variable 2 Survey Instrument: A list of campaign avoidance be-
haviors subjects have engaged in . . . . . . . . . . . . . . . . . . . . . . . . . 144
x
5.5 The Instrumentation for the GAD-7, a scale to measure general anxiety disor-
der. The GAD-7 Avg. variable used throughout this chapter is geneated from
the rst of these two survey questionss. Answering `Not at all' was coded as
a 1, `Several days' = 2, `More than half the days' = 3, and `Nearly every day'
= 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
5.6 The instrumentation for the IUS-12, an index to measure intolerance of un-
certainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
5.7 Dot-whisker Plot Visualizing the Eect of Anxiety on Propensity to Contribute.150
5.8 Eect of anxiety on contribution frequency . . . . . . . . . . . . . . . . . . . 153
5.9 Visualizations of the regression model describing the eects of anxiety on
propensity to avoid campaign contact. . . . . . . . . . . . . . . . . . . . . . 156
5.10 Bar chart showing the most popular reasons survey respondents listed for
avoiding campaign contact. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
5.11 Bar chart showing the most popular reasons survey respondents listed for not
contributing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
6.1 Flowchart Describing the Relationship Between Variables that Lead to a Po-
litical Contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
6.2 Several gmail email openers on an iPhone XS Max . . . . . . . . . . . . . . . 204
6.3 Several Apple mail email openers on an iPhone XS Max . . . . . . . . . . . 205
6.4 Gmail client email opener for a shorter window: note fewer words appear as
preview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
6.5 Gmail client email opener for a full-size window . . . . . . . . . . . . . . . . 205
6.6 Gmail client email opener for a very narrow window: note that the text is
changed to resemble the mobile Gmail app user interface . . . . . . . . . . . 206
xi
6.7 Apple desktop mail opener for a large window . . . . . . . . . . . . . . . . . 206
6.8 Apple desktop mail opener for a large window with the settings customization
to view 5 lines of preview text . . . . . . . . . . . . . . . . . . . . . . . . . . 207
6.9 Fundraising predictions for a candidate at various levels of anxiety over time,
holding all other features at their means. Includes all candidates. Black lines
are marked between Week 63 and 65 of an entire campaign cycle, correspond-
ing to between September 7 - 21. . . . . . . . . . . . . . . . . . . . . . . . . 209
6.10 The begining of the survey deployed in chapter 5. . . . . . . . . . . . . . . . 211
xii
Abstract
\There are two things that are important in politics. The rst is money and I
can't remember what the second one is." - Mark Hanna, American Industrialist
and campaign adviser to President William McKinley (Birnbaum, 2000, 29-30).
As represented by this oft-used quote of Senator Mark Hanna, money in American politics
has been vital to its functioning since time immemorial, and its ongoing eminence in political
contests in large part motivates this project. In this dissertation, I seek to answer the
question: why do ordinary people contribute to political campaigns? The question of why
small donors give is theoretically important to the literature on political behavior, and
answering it will enhance our understanding of an increasingly important form of political
participation. Existing research on individual giving is characterized by two shortcomings:
(1) a focus on large donor contributors, whose motivations, capacities, and policy preferences
dier signicantly from those of small donors; and (2) research on small donors is often narrow
in scope.
In three studies, I contribute to the literature on this topic by placing a focus on the
impact of two understudied variables: modern emotional campaign email solicitation and
psychological anxiety. The rst study replicates and expands upon previous work into why
people contribute using the CCES and ANES survey data as far back as 2000. The second
and third studies argue that psychological phenomena such as anxiety and worry at times
encourage people to contribute, but at other times encourage people to avoid campaign
xiii
contact. This research is founded in the theory of anxious avoidance (Borkovec et al., 2004),
arguing that while anxious people are more likely to contribute as a result of anxiety, anxious
individuals are also more likely to engage in avoidance behaviors to avoid anxiety-inducing
campaign contact. Logically, this suggests diminishing returns for campaigns using anxiety,
which is explored in chapter 4 and chapter 5.
1. The rst study uses survey data from the CCES (Cooperative Congressional Election
Study) and the ANES (American National Election Study) to identify the demographic,
political, and social characteristics that are most predictive of who makes a political
contribution. The goals of the study are to both reexamine past research on the most
impactful variables on contribution decisions and measure the importance of campaign
contact/solicitation
1
on contribution decisions. Later chapters in this dissertation fo-
cus on the eects of anxiety-inducing campaign contact on contribution behavior, so
this rst study serves as a foundation to re-examine the ndings of past research such
as Schlozman et al. (2012) and highlight the important of campaign contact. Chap-
ters 2 and 3 use OLS and logistic regression to analyze survey data from the CCES
(Cooperative Congressional Election Study) and the ANES (American National Elec-
tion Study) to identify the demographic, political, and social characteristics that are
most predictive of who makes a political contribution. Results show that campaign
contact/solicitation (when an individual has been contacted by a political campaign),
income, education, political interest, frequency of news consumption, Democratic par-
tisanship and liberal ideology are the most predictive traits in determining who con-
tributes. This chapter's contribution to the literature is the nding that Democratic
partisanship, liberal ideology, and contact/solicitation are three of the most predictive
variables in determining who is likely to make a political contribution. Chapters 2-3
describe the full process and results of this study.
1
throughout this work, \contact" and \solicitation" are used interchangeably. Additionally, \donor" and
\contributor" are used interchangeably.
xiv
2. The second study is the rst of two projects researching the role of worry and anxiety
on contribution likelihood. It is well known that campaigns send many frantic, anxiety-
inducing messages to generate campaign contributions. Seeking to understand if these
tactics are successful or not, chapter 4 uses political emails
2
sent by campaigns during
the 2019-2020 election cycle in conjunction with FEC contribution data to determine
the relationship between campaign invocation of worry/anxiety and their fundraising
returns. I hypothesize that overuse of worry-inducing messages will have an overall
negative relationship with fundraising returns because anxious recipients will come to
view campaign communication as a source of anxiety and avoid campaign contact,
controlling for competing variables. OLS regression is used to analyze the data over
time and nds that while anxiety is shown to be an eective fundraising strategy at
the start of a campaign, over time anxiety-inducing emails become less eective at
fundraising than sending 0 anxiety-inducing emails. This eect held true for all groups
except Democrats, among whom anxiety actually increased in eectiveness towards
the end of a campaign. A striking statistic from this study is that 61% of all campaign
emails were classied to be anxiety-inducing, with the average campaign sending 2-3
anxiety-inducing emails per week throughout the entirety of July 2019 - November
2020.
3. The third and nal study uses survey data from a subject pool recruited through
MTurk to measure the relationship between two measures of psychological anxiety and
campaign contribution behaviors. Specically, the relationship between anxiety
3
and
(1) contributing to politics and (2) avoiding campaign communication is measured us-
ing descriptive statistics and regression. I hypothesize that higher levels of anxiety
are positively related with making a political contribution and campaign avoidance
behaviors, two seemingly contradictory behaviors. This contradiction is resolved by
the theory of anxious avoidance, considering a contribution as a behavior to avoid the
2
collected from https://politicalemails.org
3
measured via the GAD-7 and IUS-12, two measures explained in depth later
xv
anxiety of uncertain political events, and unsubscribing from a campaign as a behav-
ior to avoid the anxiety of constant reminders of uncertain negative political events.
Results show that anxiety has both a strong positive relationship with the likelihood
of having contributed in the past year, and a weaker but still positive relationship
with likelihood of avoiding campaign contact through unsubscribing or blocking email
communication.
Overall, the conclusion drawn is that anxiety-inducing fundraising emails suer from
diminishing returns. Because the mass distribution of anxiety-inducing emails from political
campaigns is likely to cause psychological harm to many individuals, the suggestion made
by this research is for campaigns to signicantly reduce the number of anxiety-inducing
fundraising emails sent and search for less harmful and more eective ways to conduct email
fundraising.
xvi
Chapter 1
Why Do People Contribute?
This dissertation is centrally about one thing: understanding why regular people contribute
to political candidates, campaigns, or causes.
1.1 Abstract
In this chapter, I discuss the motivations for researching small dollar political contributors,
justify their importance to political behavior research, and describe the primary research
question of this research. Then, I discuss the existing literature around why people con-
tribute, oer a framework for explaining political contributions, and outline the research
questions generated from the contribution framework.
The brief version explaining the motivations to research small donors is that they seem to
have become more important to campaign fundraising, they are understudied, and there are
interesting behavioral questions to be asked of their unique behavior. Small donors are very
unlikely to receive any tangible benets from contributing, and their individual contributions
are extremely unlikely make a dierence in an election outcome, so their behavior is curious.
This dissertation contains three major studies into small contributor behavior. First, survey
data is used to determine the background traits that are most predictive of making a political
1
contribution. Second, the relationship between anxiety-inducing campaign messages and
fundraising results are investigated using campaign emails and FEC data. Third, through a
survey the relationship between anxiety and contributing is investigated.
I provide an overall framework that explains small-dollar contributing: campaign solici-
tation activates an individual with complementary background traits or characteristics that
make them likely to contribute. Even though many individuals do contribute, some individ-
uals will never contribute because of their demographic or political characteristics, such as a
lack of interest in politics, lack of educational resources, or lack of nancial resources. On the
other hand, it is useful to consider some individuals \dormant" or un-activated donors who
simply haven't contributed yet because they haven't been asked at the right time. It is within
this framework that the anxious avoidance theory of small donor behavior is oered. This
theory views individual political behavior as a result of anxiety-avoidance behavior. While
campaign contributions are studied in this dissertation, this work argues that avoidance of
anxiety is a major driver of a variety of political attitudes and behaviors, such as forming
policy opinions, voting, and contributing. The theory functions by real-world political events
serving as elements in an individual's life that induce worry or anxiety that individuals then
seek to avoid.
A common adaptation strategy to deal with anxiety is to avoid it. Making a political
contribution serves as a way to avoid anxiety by decreasing the likelihood that a perceived
negative future event will happen. For example, if the election of a candidate called \Dr.
Evil" would cause an individual anxiety for various reasons, contributing to their rival can-
didate, \Dr. GoodPolicy" would alleviate their anxiety in the moment because they know
they have done what they can to decrease the likelihood of a future negative event and
gain a modicum of control. The theory also includes the caveat that if campaign solici-
tation becomes too anxiety-inducing, more anxious individuals would avoid the campaign
communication due to the same avoidance mechanism.
2
1.2 Introduction and Contemporary Motivations for
Research
The core relevance of small dollar donations to the larger body of political science research is
that contributing is an important form of political participation (Jones, 2019). Verba et al.
(1995) open their major work on political participation by asserting that citizen participation
is central to democracy, and that voice and equality are central to democratic participation.
They dene political participation as \activity that is intended to or has the consequence
of aecting, either directly or indirectly, government action[9]," and political contributions
certainly fall under this umbrella. They argue that for a democracy to be meaningful, \the
people's voice must be clear and loud so policymakers understand the citizens' concerns."
An important point Verba et al. (1995) make about political participation is that demo-
cratic participation must be equal for a functioning Democracy, but that large degrees of
participatory inequality exist. People who participate beyond simply voting are unrepresen-
tative of the population as a whole, tending to be from more socioeconomically advantaged
groups. And beyond that, there are clear hierarchies among these participants, especially
for contributors. Simply put, the impact of a garden variety American contributing $20 to
a campaign is far less than the impact of a wealthy individual contributing thousands of
dollars to a super PAC in support of a candidate.
A popular idea in support of campaign nance reform is that if more people were to
make political contributions, the political donor pool would become more representative
of the country, and so too would politics become more representative as the people funding
campaigns begin to look more like average citizens and less like the extremely wealthy. Equal
political participation is vital to a fair democracy, and money is necessary to the political
process. And yet, the campaign nance system has long been dominated by a small number
3
of very wealthy donors.
1
However, with the proliferation of the internet, political fundraising
from a large number of average people has become much more feasible. Unfortunately, not
much research has been conducted to explain why regular people give money to political
campaigns. This dissertation seeks to answer one central puzzle: why and under
what circumstances do ordinary people contribute money to political candidates
or causes?
This project is motivated by a number of contemporary factors. Primary motivations are
(1) candidates in the 2018 midterms publicly refusing PAC money; (2) popular press coverage
of candidates utilizing grassroots fundraising for their campaigns;
2
(3) signicantly more
money from small donors disclosed in 2018 than previous years, especially for a midterm; (4)
Demonstrated commitment from the Democratic party to expand the role of small donors
in politics and campaign nance. I will discuss each motivating factor individually.
1.2.1 Candidates Refuse PAC Money
During the 2018 midterm elections, many prominent Congressional candidates made claims
about not taking money from PACs, corporate PACs, or Super PACs.
3
These pledges about
not taking money from PACs or `monied interests' came primarily from Democrats. Beto
O'Rourke (D-TX) made a claim in his unsuccessful 2018 Texas Senate bid that his campaign
would be run by \all people, no PACs", which he carried on to his presidential bid (O'Rourke,
2019). Elizabeth Warren made a similar claim, and in fact many Democrats claimed to not
take corporate PAC money. In total, 44 Democrats who won their House elections took
the pledge to not accept corporate PAC money (Evers-Hillstrom, 2018) in 2018. It should
be noted that the candidates refusing corporate PAC money generally accept money from
1
\Donor" and \contributor" will be used interchangeably throughout this dissertation
2
Notably, this coverage was centered around a number of high-prole candidates. Examples include Beto
O'Rourke and Alexandria Ocasio-Cortez
3
this comment is notable by itself, as Super PACs have never been legally allowed to give money directly
to candidates
4
other types of PACs, such as union PACs or ideological PACs. It should also be noted that
most of the candidates who ran without corporate PAC money were challengers, and most
corporate PACs almost exclusively support incumbents, not challengers (Mutch, 2016, 70).
For progressive challengers to boast that they do not take corporate PAC money is a bit like
a college professor announcing that they refuse to play for the Patriots in the NFL; it may
be true, but it likely wasn't an option in the rst place. Particularly if one accepts the idea
that corporations make political contributions to gain access and in
uence over members of
congress, corporate PACs are not strongly incentivized to contribute to challengers who do
not yet hold oce, and especially not anti-corporate progressive candidates.
Despite the very public rejection of PACs in the late 2010s, it was not the rst time candi-
dates have publicly rejected PAC money to capitalize on negative public sentiment. After the
rst initial explosion of PACs from corporations in the 1980s, public sentiment was strongly
against PACs. Mutch (2016, 66) writes, \...most Democratic presidential hopefuls publicly
announced that they would turn down PAC money...publicly refusing to take PAC money
came to be something that was expected of Democratic presidential candidates. Arkansas
governor Bill Clinton and most of his rivals for the 1992 Democratic nomination made such
pledges. And when Senator Barack Obama announced his campaign for the presidency in
2007, he too said he would take no contributions from PACs or lobbyists. By 1992, however,
it was obvious{... that not taking PAC money was not the same thing as not taking money
from people in the organizations that sponsored PACs. Refusing contributions from a cor-
poration's PAC, for example, did not mean refusing contributions from that corporation's
executives...". Certainly some candidates still utilize PACs and wealthy donors, while others
fundraise largely from small donors, with some even raising a majority of their funds this
way.
In summary, fundraising from individual wealthy donors remains a vital component of
political fundraising despite surges in the importance of small-donor fundraising. As of
2021, there is no indication that wealthy donors have or will cease playing an enormous
5
role in campaign fundraising. However, that does not preclude small donors increasing in
importance.
1.2.2 Popular Press Coverage of Small Donors
Another primary motivating factor for researching small donors was the perception that
there was a great deal of media coverage about small donors during and especially after the
2016 election. To verify this perception, data was gathered from Proquest to observe the
number of articles containing the word "small donors" over time. To do this, a record of
every article containing the keyword "Small Donors" (case insensitive) was downloaded from
the Proquest U.S. Newsstream, a database containing articles from 1109 unique publications.
I aggregated the number of articles published on each day and created a simple frequency
histogram of the total number of articles published per day (across the U.S. Newsstream)
containing the keyword `Small Donors'. The Proquest database had entries from 1980 to
July 24, 2019. Results are visualized in Figure 1.1, which shows there is a signicant increase
in the number of articles about small donors since 2008, but particularly after 2012. There
has been a clear upward trend in the number of articles about small donors in popular press
publications across the United States, particularly since the rst large spike in 2008. Notice
that there were already more articles about small donors by July 2019 than the entirety of
2018, likely as a result of the 2020 Democratic presidential primaries. Because mentions of
small donors in the popular press have increased in recent years, scholarly research about
small donors is warranted, as it will enhance our understanding of a clear topic of public
interest.
6
Number of Articles per day About Small Donors
Date
No. of Articles per day
0 10 20 30 40
1980 1984 1988 1991 1995 1999 2003 2007 2011 2014 2018
Figure 1.1: Histogram displaying the number of articles on Proquest per day about small
donors
1.2.3 Disclosure of Small Donor Donations increased in 2018
More newspaper coverage about small donors does not necessarily mean that small donors are
actually giving signicantly more money than they have in the past, however. An important
question to answer is: just how much money are small donors giving?
7
The immediate problem with answering such a question is one of data availability arising
out of disclosure laws. The major data availability problem when studying small-dollar
donations is that campaigns are not required to disclose information about donors giving
less than $200 per cycle. While this is problematic for research, it is not to say the entire
environment of small donors is a mystery. Starting in 2010, the major donation processing
platform for Democratic candidates, ActBlue, has required complete disclosure from all
donors regardless of the amount given. Further, candidates are incentivized to disclose
their contributions to the FEC to give the appearance of raising a lot of money, a proxy
for campaign support/enthusiasm and a deterrent to challengers. Data was collected from
from Opensecrets.org
4
to determine if disclosures from small donors had increased over time.
While the data is not a perfectly accurate representation of the truth, results suggest that
small dollar giving has indeed increased.
The Center for Responsive Politics (2018) show that between the two midterm years
2010 and 2014, small donor fundraising increased from $1.138B to $1.140B, a 0.21% in-
crease (adjusted for in
ation). Between the two midterm years 2014 and 2018, small donor
fundraising increased from $1.140B to $1.58B, a 38.55% increase (adjusted for in
ation).
That is a striking increase in donations that is resistant to concerns about disclosure laws
and data accuracy. For further comparison, most midterm years see less money in small
dollar donations than presidential years. Small donors gave $1.567B in 2016 and $1.58B in
2018, a midterm year. Between 2012 and 2014, small donor funds dropped from $1.55B to
$1.14B, a decrease of -26.6%. Again, to contrast that, between 2016 and 2018, small donor
funds increased from $1.567B to $1.58B, an increase of 0.83%. While the increase is quite
small between the two, it is quite notable that small donor fundraising increased even slightly
between a presidential election year and a midterm election year.
4
a website run by the Center for Responsive Politics
8
Figure 1.2: CRP Individual Contributions Breakdown by Source
1.2.4 Demonstrated commitment from the Democratic party to
expand the role of small donors
Two major signals from the Democratic party have demonstrated the party as a whole
is taking small donors seriously. A Democratic National Committee rule required 2020
presidential candidates to either poll at 1% or have raised money from 65,000 unique donors
in 20 states with at least 200 donors per state in order to join the presidential debates (Zychal,
2019). Additionally, in the 116th Congress, House Democrats proposed House Resolution 1,
a major bill to overhaul voting rights and campaign nance in the United States. This bill
9
included a policy to match small donor contributions at a 6:1 rate by adding a 2.75% fee to
criminal and civil penalties committed by banks and corporations engaging in malfeasance
(Sarbanes, John P., 2019). The bill also included a number of campaign nance reforms
such as a constitutional amendment to end Citizens United, disclosure laws for `dark money'
contributions, mandatory disclosure of entities funding Facebook and Twitter ads, among
many others. While these bills did not pass, it clearly demonstrated campaign nance reform
was a public goal for the Democrats, and expanding the role of small donor fundraising is
part of that eort. As a result, studying why small donors give is important to understand
their role in campaign nance.
1.2.5 Money is Here to Stay
Despite negative perceptions of money in politics and the appearance of being a priority
for Democrats, campaign nance reform has not historically been a priority for Americans
(Primo, 2002), nor was it listed as a priority in recent polls (Confessore and Thee-Brenan,
2015). Additionally, campaign nance reform through legal challenges is exceedingly unlikely
for the foreseeable future. The decisions of two major campaign nance-deregulating court
cases, Citizens United v. FEC (2010 USCC) and Speechnow.org v. FEC (2010 USDC-DC)
and the conservative shift of the Supreme Court mean campaign nance reform is unlikely
for many years. In a book chapter titled \The Last Great Hope for Reform," Richard Hasen
outlines a persuasive argument that the most viable pathway to campaign nance reform
rested with the possiblity for the Supreme Court to shift to a direction more favorable
towards regulation Hasen (2016). The book was written in 2016, before justices Gorsuch,
Kavanaugh, and Barrett joined the court, eectively removing Hasen's last great hope. In
another 2016 book, Mutch (2016, 137) writes, \... the future looks bright for opponents of
reform." Reform bills tend to die in committee and the FEC is not eective at enforcing
existing law. Only at a local level has reform seen any progress as small-scale public funding
10
has been proposed and occasionally implemented. The use of public funding for presidential
elections died in 2008 when President Obama became the rst President to decline the use
public funding for the general election campaign (Milkis and Rhodes, 2009; Luo and Zeleny,
2008), which made John McCain in 2008 the last presidential candidate to use public funding.
Because campaign nance regulations have diminished and may even diminish further
under the new makeup of the Supreme Court, even more money is likely to become involved
in politics. Some question if it is even possible for regulations to keep money out of poli-
tics, as determined and clever attorneys will gure out workarounds and loopholes to keep
funneling money into campaigns. From the perspective of a prominent former Democratic
fundraiser, Lindsay Mark Lewis
5
argues that money will always nd its way into politics
because fundraisers and campaigns will gure out how to circumvent any legal regulations
(Lewis and Arkedis, 2014). Based on all these factors, campaigin nance reform seems im-
possible for the foreseeable future. It seems more likely that campaigns who do not have
access to megadonors will turn their eorts to increased small fundraising.
1.2.6 A Networked Democracy
Another major component that makes research about small donor fundraising necessary is
the expansion and maturation of the internet and related technologies including social media
platforms, digital advertising, news apps, and the general political data economy.
A core aspect of this dissertation is modernity - the widespread adoption of internet-
related technologies by the end of the 2010s has changed the way people and campaigns may
communicate with each other. Fundraising via email lists, websites, and social media allows
for cheaper fundraising, and thus more fundraising (Wilcox, 2008). The expansion of the
5
Director of the Progressive Policy Institute and former Democratic Fundraiser responsible for raising
$150 million for the Democratic Party
11
internet and social media in the past decade has been astonishing, and highlights the need
for research using more contemporary data.
Only 52% of American adults reported using the internet in 2000 (Pew Research Center,
2019), ve years after Verba et al. (1995) wrote their book on political participation. 74%
of American adults said they used the internet at all in 2008
6
, the year up to which the
authors updated their work with a second book (Schlozman et al., 2012). However, as shown
in Figure 1.3, in 2019 a massive 90% of American adult residents reported using the internet.
When isolating that number to older Americans, the slowest age cohort to adopt the internet,
only 14% of Americans over 65 used the internet in 2000. 38% of Americans over 65 used
the internet in 2008, and that number has doubled since: 73% of Americans over 65 use the
internet in 2019. To dismiss the passage of time as a compelling factor in the changing nature
of campaign fundraising would be to dismiss the rapid pace of technological change and usage
that are part and parcel to the work of campaigns. Examples of the internet's growth eect
on campaign fundraising perhaps most notably begin with the Obama campaign's famous
use of the internet to fundraise (Kenski et al., 2010). Because of the importance of the
internet on small dollar fundraising since 2008, much of this project is based on increasing
our understanding of the ever-changing dynamics of the campaign nance system. On a
basic level, more people being online means more fundraisers will nd them there to ask for
money.
6
The specic question wording has changed over time, but all variations ask if the respondent `ever' or
`occasionally' `goes online' or `checks email'.
12
52
55
59
61
63
68
71
74 74
76 76
79
83
84 84
86
88
89
90
20
40
60
80
100
2000 2005 2010 2015 2020
Y ear
Percent of US Adults that Use the Internet
Percent of U.S. Adults that Use the Internet, 2000 − 2019
Figure 1.3: Line graph displaying the percent of US Adults that at least occasionally use
the internet. Source: Pew Research Center (2019)
Beyond simple internet use, social media use was quite low among American adults during
2008, when Schlozman et al. (2012) wrote their famous book on participation and equality,
including data from the 2008 Pew Internet and American Life Project.
13
Schlozman et al. (2012, 502-503) cite Pew research, noting that 6% of respondents made
an online contribution, while 15% made an oine contribution. This nding was before
Obama's 2008 fundraising blitz, and is certainly not representative of modern contributions.
5%
8%
11%
21%
42%
50%
62%
65%
69% 69%
72%
20
40
60
2005 2006 2008 2010 2012 2014 2015 2016 2018 2019
Date
Percent of Adults on Social Media
Percent of US Adults Who Use at Least One Social Media Site
Figure 1.4: Line graph displaying the percent of US Adults that at use at least one social
media site. Source: Pew Research Center (2019)
14
Because some of the most recent landmark research on small donor fundraising relies on
data from 2008, updating the research is crucial. America has gone through a great deal of
technological and political change in the past decade, and updating the research about small
donor fundraising has never been more important. I turn now to examine the past scholarly
research on political giving and fundraising. After synthesizing extant research, I forward
the idea of emotional solicitation-based contributing.
15
1.3 Literature Review: Existing Theories of Individual
Political Contributions
Contribution decisions are an important part of political behavior, but a large amount of
the research on contribution decisions do not focus on small donors. A great deal of re-
search exists on PAC contributions, for example (Biersack et al., 1994; Box-Steensmeier
and Grant, 1999; Clawson et al., 1992; Cox and Magar, 1999; Endersby and Munger, 1992;
Evans, 1988; Sorauf, 1988; Lawton et al., 2013; Jenkins, 2021). PAC research generally sees
their subjects as strategic actors that factor in logical, power-maximizing elements such as
incumbency, seniority, and committee assignments. While there is a body of research on
individual contributors, much of the work follows in the tradition of viewing contributors as
strategic, power-maximizing actors and studies large donors primarily (Sorauf, 1992; Joul-
faian and Marlow, 1991; Francia et al., 1999; Milyo et al., 2000; Grant and Rudloph, 2002;
Apollonio and Raja, 2004; Gordon et al., 2007; Claessens et al., 2008; Ovtchinnikov and
Pantaleoni, 2012; Hansen and Rocca, 2019). Other, more general theories oer frameworks
for all contributions but de facto focus on large donors (Rhodes et al., 2018; Bonica, 2014;
Verba et al., 1995) or use data that is no longer representative of the campaign contribution
environment in 2021 (Schlozman et al., 2012).
Some research does account for and even focus on small donors (Ansolabehere et al.,
2003; La Raja, 2014; Overton, 2004; Corrado et al., 2010; Malbin et al., 2012; Malbin,
2013; Culberson et al., 2018; Frost, 2013; Magleby et al., 2018; Bouton et al., 2018). The
goals of most small donor research, however, does not study why small donors give, but
explores questions such as the relationship between polarization/ideological extremity and
small donors (La Raja, 2014; Keena and Knight-Finley, 2019; Malbin, 2013), reforms to
encourage small dollar donations (Overton, 2004; Ramsden and Donnay, 2001), the eects
of increased small donor importance in elections (Culberson et al., 2019), the role of social
16
media in contributions (Petrova et al., 2017), and other questions not related to why people
contribute. One of the only major works that examines small donors and specically studies
why they give is Magleby et al. (2018), which in addition to surveys such as the CCES and
ANES, uses samples of small donor data from 2008 and 2012 directly given by the Obama,
McCain, and Romney campaigns.
Magleby et al. (2018) studies candidate appeal as a motivator for donors to contribute,
and like this dissertation, they note the striking lack of emphasis on the role of cam-
paigns/candidates and message, even in major studies in donor behavior such as Clark and
Wilson (1961); Brown et al. (1995); Verba et al. (1995); Rosenstone et al. (1993). A major
argument of Magleby et al. (2018) is the importance of the candidate in donor contribution
decisions, citing that in a survey, being \inspired by the campaign" was mentioned by 64% of
donors in their 2008 survey as a reason for contributing, but dropped to 27% in 2012. Cam-
paigns became more negative, commonly using the fear and anxiety induced by opposing
candidates as fundraising strategies. In 2012, 89% of respondents in the survey by Magleby
et al. (2018) listed a candidate being \a bad choice for the country" as \very important" in
their participation decision. While Magleby et al. (2018) advocate for more research that
focuses on candidate appeal, much of their work in dierentiating small donors as separate
from large donors uses relatively standard demographic, political, and social characteristics
found in the ANES and CCES. This work is an excellent contribution to the literature around
small contributions as political behavior, and this dissertation aims to expand upon it by
researching the role of anxiety in small donor decision-making.
1.3.1 Small Contributions are Substantively Dierent From Large
Contributions
It is necessary to separate small donors from large donors, as the dierences between them
in terms of eects and motivations are vast. Large donors are typically older white male
17
business owners that give large sums, often at private fundraisers, in order to gain access
to politicians and improve regulations that impact them and their income streams (Francia
et al., 2003). Small donors are more demographically diverse, are not motivated by in
uence
or access because they are not capable of contributing large amounts, and they are more
vulnerable to political outcomes than the wealthy. Because of these dierences between large
and small contributors, further study on small donors is important to expand understanding
of political behavior.
The psychological processes of an individual contributing $20 to a candidate because
of genuine support are considerably dierent than those of a wealthy donor navigating the
campaign nance regulatory environment to funnel tens of thousands of dollars to a candidate
through direct contributions, PAC contributions, and super PAC contributions. Regular
donors can only signal their support to a candidate through small contributions that are
virtually guaranteed to have no reward, while wealthy donors are more likely to expect
social rewards in the form of invitations to events, political rewards in the form of meetings
with a candidate, and even individual ecacy rewards in the form of knowing that a $20,000
contribution may impact a campaign outcome, while a $20 contribution will not.
1.3.2 Material, Solidary, and Purposive Incentives
Francia et al. (2003) illustrate the small/large donor disconnect well: they created a con-
vincing framework to explain individual giving, forwarding three major reasons individuals
make political contributions: material incentives, solidary incentives, and purposive incen-
tives (Clark and Wilson, 1961). Material incentives for contributing are expectations of
personal gain as a result of making a contribution. Material incentives may exist in the form
of contributing with the expectation of access or in
uence over a member of Congress (Hall
and Wayman, 1990), patronage, or even explicit bribery/corruption (Overacker, 1974). Pub-
lic opinion polls show that many Americans believe wealthy donors receive unfair benets
18
as a result of their contributions. In a 2015 New York Times/CBS phone survey of 1,022
Americans, 55% believed public oce winners promote policies that directly help the people
and groups who donated money to their campaigns most of the time, and 30% answered
they did so sometimes (Confessore and Thee-Brenan, 2015). Material incentives such as in-
uencing a member of congress are not viable reasons for small donors to contribute because
contributions below $200 are generally not large enough to grant access or in
uence to an
elected ocial.
Francia et al. (2003, 43) write that \Solidary or social motives involve the psycholog-
ical benets that stem from the social side of political participation," and \Purposive or
ideological motives are mostly concerned with enacting general public policies or providing
public goods that apply equally to large classes of society." Solidary incentives are described
to be the social benets received from participating in activities such as attending political
fundraisers and \rubbing shoulders" with politicians and celebrities. Generally speaking,
the social benets received from making a large contribution to a campaign are substantially
dierent than making a small contribution.
Small donors are generally not capable of making large enough contributions to earn them
a spot at exclusive fundraisers, private meet-and-greets, or receive similar social rewards.
Further, many small donations take place online (Schlozman et al., 2012), where the social
benets of people witnessing one contribute are not generally applicable. Solidary incentives
for small donors do exist in the form of sharing one's act of contribution with their social
network (e.g. Facebook, Twitter), which is found to in
uence the political decisions of one's
peers, including contributions (Sinclair, 2012; Saxton and Wang, 2014).
Still, one should be careful in entirely dismissing solidary incentives for small donors,
as they may still receive social pressure to contribute through other means, such as being
asked to contribute by a peer. Additionally, campaigns occasionally capitalize on solidary
19
incentives themselves by emailing their supporters with chances to meet them, and people
can share that they contributed on social media.
I expect that solidary incentives are not particularly strong motivators for most small
donors, while material incentives are not applicable at all to small donors. Existing research
suggests that it is purposive incentives that are the most prevalent motivators for small
donors, encompassing ideology, partisanship, and issue alignment with a candidate (Johnson,
2013; Malbin, 2013). The dominance of purposive incentives as driving forces for small dollar
donors makes logical sense, as material incentives are not normally applicable to small donors,
and purposive incentives are essentially a pre-requisite to solidary incentives. If an individual
doesn't have a purposive draw to a candidate, it is unlikely they would be drawn to the social
benets of being associated with a candidate.
At the core of a small donor's decision to make a political contribution are several over-
lapping options: people want to make a dierence in an election; they want to support a
candidate that they believe in; they believe the policies of a candidate will make their lives
better; they believe electing candidate A will be better than electing candidate B. All these
options are encompassed by the concept of expressive or consumption benets, dierentiated
from contributing as a social action or a way to enhance their own well being or material
worth.
1.3.3 Civic Gratications and \Warm Glow" Giving
Verba et al. (1995, 115)'s research shows most people report civic gratications (80%) and
policy gratications (46%) for reasons they contributed to a candidate, followed by social
gratications (22%) and material benets (18%). When reporting why individuals gave to
non-candidates, such as party organizations, Work-Related PACs, or Issue Organizations, the
distribution changes only slightly. Civic gratications and policy gratications are always
20
the top-two reasons for contributing, while only contributions to work-related PACs were
driven in any large way by material benets (46%).
According to Ansolabehere et al. (2003, 117-118), \...individuals give because they are
ideologically motivated, because they are excited by the politics of particular elections, be-
cause they are asked by their friends or colleagues and because they have the resources
necessary to engage in this particular form of participation. In short, people give to politics
because of the consumption value associated with politics, rather than because they receive
direct private benets." Individuals may feel expressive benets from contributing, seeing
their contribution as a 'consumption good' that produces a `warm glow', in the language of
Ansolabehere et al. (2003) and Andreoni (1990).
In this view, small donors give because they are excited about a candidate and get an
intangible satisfaction from participating in the process that outweigh the costs of giving $20
or so to a campaign.
7
Giving for consumption benets ts into Francia et al. (2003)'s notion
of contributing for purposive benets. Ansolabehere et al. (2003) actually advocate for more
research on small donors, writing \On the subject of individual campaign contributions,
the idea of a campaign contribution as a form of consumption needs more empirical and
theoretical development. As with other forms of voluntary public-spirited activities such as
giving to charities or voting, the theoretical underpinnings of small campaign donations are
not well understood. It is unclear what specic empirical predictions distinguish consumption
from rent seeking or what evidence will prove compelling." I argue that for small donors, rent
seeking is not possible if we consider rent seeking to be manipulation of policy to increase
personal prot. When considering dierences in material benets for low-to-middle income
individuals vs wealthy individuals, it is immediately apparent that they are worlds apart.
A millionaire contributing the maximum legal amount to multiple candidates who favor
lowering a certain tax cannot be considered the same motivation as a fast-food worker who
contributes $10 to a candidate who supports raising the minimum wage. The motivations
7
note that they also dismiss material incentives as reasons for small donor giving
21
behind the fast-food worker are closer to survival or protection-motivated behaviors, rather
than wealth-increasing behaviors.
1.3.4 Resource-Based Models of Giving
Verba et al. (1995) developed the Civic Voluntarism model, considering three main forces
of participation: motivation, capacity, and networks of recruitment. Motivation exists as
people's desire to participate; capacity to participate is determined by factors such as people's
availability/free time to volunteer or money to donate; and networks of recruitment generally
refer to whether or not one was asked or solicited to participate, or saw that their peers
contributed (Saxton and Wang, 2014). Conversely, explanations for why people do not
participate are because they do not want to; they can't; or because nobody asked.
The civic voluntarism model concludes that the development of civic skills through edu-
cation and participation in organizations at work or in school are important predictors for
political participation, as does SES (level of education, income, and occupation). However,
they admittedly focus their attention on a resource-based model of participation. Particu-
larly when it comes to making a political contribution/donation, the argument is that income
matters most, especially because campaigns spend a lot of time and resources making sure
to target people who are likely to contribute. Brady et al. (1995) formulate a resource-based
model of political participation, of which contributions are only a part. The essential part
of that work is that time, money, and civic skills best explain participation. They nd quite
clearly, \The results are unambiguous: the major determinant of giving money is having
money." Survey results in Verba et al. (1995) back up the claim that a major determinant of
contributing is simply having money. Resource-based models are of less theoretical interest
to me because they do not truly explain why people give, but merely explain who is eligible or
capable to give. It should be expected that the less capable one is to contribute due to having
a lower income, the less likely they are to make a contribution. What is interesting about
22
small donors is not their income, however, but why they would give $20 to a campaign with
no real expectation of material gain and knowing that such a small contribution is unlikely
to change an election outcome. To explain this choice is to explain an interesting element of
modern political behavior. Many people with modest resources contribute to politics, while
their richer counterparts do not contribute; thus, the simplicity of resource-based theories is
incomplete. Income may be a major factor in contribution decisions, but it does not oer a
compelling or complete behavioral explanation for why individuals give.
1.3.5 Income and Solicitation Eects
While having money is an important predictor of contributing, it is also a strong determinant
of being solicited, which creates problems for researchers because being solicited aects
theoretical explanations about why people contribute. By and large, people tend not to
contribute unless they are explicitly asked by a campaign or political group to give (Grant
and Rudloph, 2002; Brady et al., 1999; Rosenstone et al., 1993; Cummings and Cummings,
2004; Burton et al., 2015).
Brady et al. (1999) researched how fundraisers \rationally prospect" for contributors by
only asking those people who are known beforehand to be likely to contribute, largely as a
result of income. This supports the idea that solicitation is an important factor in explaining
small dollar donations, although it makes analysis more dicult because it blurs the lines
between the eects of being solicited and the targeting/selection eects that arise when
fundraisers solicit people that are already more likely to give.
1.3.6 Ideology and Strategic Contributors
One school of thought regarding small donors is that they are motivated by strategic electoral
concerns. This argument stipulates that individuals contribute because they want to increase
23
their preferred candidate's chances of winning a race by contributing to their campaign. A
body of literature shows compelling evidence that electoral motives play a signicant role in
small donors' decisions to contribute. Bouton et al. (2018) argue convincingly that electoral
incentives are a major factor in the decision for small donors to contribute. They take this
theory to empirical testing, predicting that as election closeness increases, individual contri-
butions increase in response. They write, \The logic of this empirical prediction is similar
to that in the literature on voter turnout (see e.g. Cox and Magar (1999), or Herrera et al.
(2014, 2016)): like voters, electorally motivated donors should only `turn out' when they
think their eort will aect the election outcome. This is more likely when the election is
close. Empirically, this eect of (perceived) election closeness appears quantitatively impor-
tant: combining survey data on US donors with FEC data, Barber et al. (2017, 17) write, \a
standard deviation increase [in competitiveness] raises the likelihood a donor gives to that
campaign by 43%." This body of research suggests the closer a race is, the more people
perceive their contribution as needed or important and thus are more likely to contribute in
closer races or races perceived as close. This idea of contributing to close races is related to
the P term in Riker and Ordeshook (1968)'s rational choice theory of turnout, although this
will be discussed more in detail later.
One major determinant of small donor giving is ideology, which also raises concerns
about ideological extremity and polarization as small donors or \grassroots" donors may
be seen as ideologically extreme (Lewis and Arkedis, 2014). Con
icting research exists on
whether or not small donors increase the level of polarization among elected ocials or if
politicians strategically mobilize existing ideologues (La Raja and Wiltse, 2012; Keena and
Knight-Finley, 2019; Culberson et al., 2018). Theory suggests that ideology should per-
haps be an even stronger determinant of contributions for small donors than large donors.
Johnson (2013) writes, \If larger contributors expect material rewards for their substantial
contributions, smaller contributors have less reason to expect to be rewarded for pitching in
$10 or even $100 to a political campaign. Under those circumstances, smaller donors would
24
be mostly devoid of material concerns, driven more by other incentives. These may include
intense ideological commitment, identity-based issues, issue agreement, social pressure, or
the desire to in
uence an outcome." I agree with this statement, and argue factors such as
ideology, issue agreement, and electoral incentives are stronger determinants of contribut-
ing to a political campaign among small donors than large donors because the decision to
contribute is more costly to small donors.
Other research supports the notion that ideology and issue preferences are strong factors
in the decision to make a contribution. Culberson et al. (2018) nd that more extreme
candidates are better at attracting funds from small donors. Malbin (2013) nd the top 5%
of incumbents raising money from small donors are no more ideological on balance than the
rest of the incumbents raising from small donors. While Malbin nds that candidates who
raise from small donors are not on average more extreme than candidates who are less reliant
on small donors for fundraising, their research does not suggest that ideological extremism
was not used to raise money. In fact, Malbin explicitly acknowledges that raising money via
ideological extremism is a distinct fundraising strategy and was used in 2012 by candidates
such as Michele Bachman, Alan West, and Mark Grayson. Rather than argue broadly that
all candidates who raise money from small donors are ideologically extreme, it should simply
be acknowledged that ideology is but one mechanism by which candidates may attempt to
raise funds from small donors.
1.3.7 Rational Choice Theory and Political Contributions
One of the most obvious ways to view the small dollar donation question is through a
rational choice lens founded by Riker and Ordeshook (1968), who grounded the rational
choice calculation of turnout from Downs (1957), which was later updated by Aldrich (1993),
among others. The basic idea of rational choice theory is that individuals engage in behaviors
25
if they believe it is in their self-interest to do so (Downs, 1957, 271-272). Formally stated,
the calculus of rational choice voting is reiterated below:
R =PB +DC
Where
• R = the reward, in utiles, that an individual receives from their act of voting
• P = the probability that the citizen will, by voting, bring about the benet, B; where
P is greater than 0 but less than 1
• B = the dierential benet, in utiles, that an individual voter receives from the success
of their more preferred candidate over their less preferred one
• D = the expressive benets and personal satisfactions one receives from the act of
voting
• C = the cost to the individual of the act of voting
This theory seeks to explain the decision to turnout to vote or not. When applied to
contributing instead of voting, the terms change; some changes are signicant, while others
are more rhetorical:
R =PB +DC
• R = the reward, in utiles, that an individual receives from their act of contributing
• P = the probability that a contribution made by an individual will, by contributing,
bring about the benet, B
• B = the dierential benet, in utiles, that an individual voter receives from the success
of their more preferred candidate over their less preferred one
• D = the expressive benets and personal satisfactions one receives from the act of
contributing
26
• C = the cost to the individual of the act of contributing
In Riker and Ordeshook (1968), the theory stipulates that if the P, B, and D terms are
large enough and the C term is small enough, the R term will be greater than 0. If R
is greater than 0 for an individual, they will vote because the reward from voting will be
higher than the cost. The same idea holds for contributing. If R is greater than 0, individuals
will contribute. That is, if PB +D > C, an individual will contribute. One of the major
dierences is that in contributing, there is always a real monetary cost measured not in time
or eort, but in dollars.
The model suers from the classic problem that the P term would predict very few
people would contribute money to campaigns because the P term is innitesimally small
when considering small contributions because no contribution under $200 can be thought
to have made the dierence in an election outcome. And while it is true that people are
more likely to contribute when they believe their contribution will have an impact on the
election and are more likely to donate in a close race, it would be an exaggeration to argue
contributors believe their small individual contribution will
ip an election outcome. Michael
Whitney, chief fundraiser for Bernie Sanders, writes, \The act of donating is not generally a
calculated one for grassroots supporters. Very few people sit down with their credit card, look
at polling averages and turnout models, and then make donations to candidates based on the
likely impact their contribution could have at winning the race (Whitney, 2018)." That said,
in an earlier piece in Politico, Whitney (2017) argues voters do want their contributions to
make a dierence, writing \People are motivated to act when they feel like part of something
larger than themselves|and when they understand that their participation in that larger
something makes a real dierence".
Research shows that people are more likely to make contributions in close races, sug-
gesting people want their donations to make a dierence, indeed. Klandermans (1984)
applies expectancy-value theory to movement participation and mobilization. He assumes
27
that willingess to participate in a social movement, in this case by contributing to a po-
litical campaign, is a function of the perceived costs and benets of participation. This
includes potential contributors factoring in expectations about others' behavior, in this case
considerations about if other people will contribute.
Specically regarding the P term, people tend to contribute more when they think their
contribution will make a dierence, supporting Riker and Ordeshook (1968). In addition
to the work on competitive races above, multiple experiments conducted via M-Turk by
Rogers and Moore (2014) nd that among supporters of a candidate, support and willingness
to volunteer for/donate to a candidate increases in races that are close, but in which the
preferred candidate is losing. In another experiment by Rogers et al. (2017b), subjects who
were treated to believe that their preferred candidate in a Florida Governor's race would
win his race were 5% less likely to click on a donation link than subjects who were treated
to believe that their preferred candidate was down in the polls by 1 point.
In essence, people are more likely to donate if they believe their contribution will make
a dierence, but other factors such as civic gratications (which map onto the D term
of the equation) are also independently sucient motivators for donating. To reiterate,
Verba et al. (1995, 115)'s research found most people report civic gratications (80%) and
policy gratications (46%) for reasons they contributed to a candidate, followed by social
gratications (22%) and material benets (18%).
In sum, the rational choice equation is useful for thinking about certain situations in
which people are more likely to contribute:
1. P : they believe the race is close and their candidate is behind
2. B: they strongly favor one candidate winning
3. D: they feel expressive benets or satisfaction from the act of contributing
4. C: the cost of contribution is low.
28
However, while the rational choice equation provides a decent framework for thinking
about contribution decisions, it leaves out a good deal of explanatory power for why specic
people contribute at specic moments. For example, various psychological variables are
left out of the rational choice equation, as are solicitation eects. Aldrich (1993) argues
that strategic politicians in
uence the calculus through campaign strategy and activity. I
agree and argue that while a rational choice framework may explain much of contributor
behavior, the decision to contribute is inseparable from the strategic actions of campaigns
(and other political groups such as PACs). Because the rationality of individuals is limited,
their considerations may be in
uenced by fundraising solicitation messaging.
It is clear that the decision to contribute is frequently an emotional one, driven by political
convictions and strong feelings about desired political outcomes. Additionally, solicitation
must be considered a major factor in determining who contributes. Informed by the existing
theories and research discussed above, I develop a two-part framework of contributor behavior
that argues contribution behavior is explained by the interaction between solicitation and
background traits/psychological variables. While the framework is fairly generic, it is useful
to form specic research questions and test more specic theories regarding small donor
behavior.
1.4 The Two-part Small-Donor Behavior Framework:
Traits x Solicitation
Most people do not make political contributions, and most people who do make contributions
are asked to do so. Campaigns must work hard to identify people who are more likely to
give than the average member of the public. In essence, an individual must have the right
mix of traits that make them susceptible for solicitations, and campaigns must solicit that
individual in an eective way to produce a contribution. The rst part of this contribution
29
framework is the identication of variables or traits in respondents that make them more
likely to contribute. The second part of the framework consists of the various solicitation
techniques and strategies used by campaigns (or other political entities) to raise funds.
Existing work convincingly demonstrates that background traits such as political al-
iation, partisan strength, income, and interest in politics are highly predictive of making
a contribution. Additionally, existing research demonstrates that solicitation has great ex-
planatory power in determining who contributes. To learn more about small donor behavior,
this dissertation researches small contributors using demographic, political, and social char-
acteristics such as those found in the ANES and CCES, as well as the role of anxiety. While
other psychological traits are likely helpful in explaining small contributor behavior, anxiety
has a particularly important role in political behavior. Chapters 2-3 of this dissertation
study common demographic features using ANES and CCES survey data, and chapters 4
and 5 study the eect of anxiety on small contributor behavior.
1.5 The Anxious Avoidance Theory of Contributor Be-
havior
This work builds upon the small contributor framework set out by Frost (2013), predicting
contributions using background traits such as psychological characteristics, political alia-
tion, strength of political aliation, income, and solicitation. While background traits cover
most of the variance regarding factors that determine whether or not somebody is likely to
contribute to a political campaign, I argue that political background traits generally deter-
mine who is in the universe of cases of individuals that would even consider contributing.
And while these traits are predictive of who contributes and who does not, background traits
alone do not adequately explain why some politically active and nancially capable people
contribute and others do not. To truly better understand why some people contribute and
30
others do not, I turn to the literature on emotion and anxiety to oer the anxious avoidance
theory of contributor behavior.
In the past hundred years, much research has been done that investigates the role of
emotion in media and communication, as well as the importance of emotion in political
reasoning and behavior. As Kinder (1994) writes, emotion is an inevitable, natural force
that can not be wholly negative. It is unreasonable and unhelpful to consider emotions to
be a completely or primarily negative force, so I argue it is best to consider emotions as
a known, but volatile, aspect of human nature that can be manipulated by sophisticated
and well-nanced actors to produce action in society. I believe this research is valuable and
important to help understand the ways in which a variety of political actors attempt to
manipulate the public into believing and doing certain things.
A primary consideration regarding emotion in politics is the fact that there exists a
strong negativity bias in media coverage. Iyengar (1991) nds the media disproportionately
covers threatening issues like crime and terrorism. Baum and Groeling (2009) nds news
media tends to favor negativity and con
ict in coverage. Nacos et al. (2011) nds newscasts
focus on the most threatening parts of stories. It is generally accepted that people have
a bias towards negative information and use negative information to make decisions more
than positive information (Baumeister et al., 2001; Rozin and Royzman, 2001; Tversky and
Kahneman, 1991). Simply put, individuals tend to focus on negative information and facts
because they believe it better prepares them to avert bad outcomes.
The theory of aective intelligence (Marcus and MacKuen, 1993; Marcus et al., 2000;
Crigler, 2007) forwards two systems that help people organize their political environment:
dispositional and surveillance. Most relevant to this work is the surveillance system, which
helps warn people about novel situations and threats, signalling a response may be necessary.
When people detect a threat, they pay closer attention and seek more information to help
counter the threat. A relevant implication of this discussed by others is that once people are
31
made anxious, they are more likely to seek out more news on an issue and may base their
opinions more on information rather than partisanship (Brader, 2006; Marcus et al., 2000).
A problem arises when considering that strong partisanship may be a source of manipulating
the surveillance system itself, convincing people that certain things are threats when they
are not. Additionally, fabrication and misinformation may be rapidly spread and complete
lies may be considered fact by a sizable percentage of the public.
A large body of research and work argues that emotion is integral to information process-
ing, learning, attitude formation, and reasoning (Westen, 2008; Marcus et al., 2000; Redlawsk
et al., 2007; Valentino et al., 2008; Brader et al., 2008; Zajonc, 1980; Damasio, 1994; Lodge
and Taber, 2005; Osgood et al., 1957). Emotions are used to determine which information
to focus on - events are appraised as relevant when they favor or harm an individual's main
concerns (Frijda, 1988, 213). This large-scale importance of emotion in people's learning,
attitude formation, and reasoning is so important because people do not simply let emotions
happen to them - they actively try to increase positive/pleasant emotions and decrease nega-
tive/unpleasant emotions (Gross, 2013; Gross and Thompson, 2007). Because people have a
negativity bias, manipulating people to feel negative emotions is perhaps more eective than
using positive emotions to force people to regulate their emotional state. Using game theory,
Lupia and Menning (2009) predict politicians can send unwarranted fear signals when peo-
ple can't determine the nature of real threats, allowing politicians to manipulate the public
into supporting candidates or policies based on bad faith exaggerations or even completely
fabricated misinformation.
The implication of this is that by inciting a certain emotion within someone, one can
predict what many recipients will do in response to try to regulate that feeling. I argue that in
a fundraising setting, it is the goal of many campaigns to make people feel negative emotions
(fear, anxiety, anger) and present the act of contributing money to their campaign as a
solution to decrease those negative feelings. Albertson and Gadarian (2015) show through
multiple experiments that people who are made anxious by an appeal are more likely to
32
support protective policies that should safeguard against the anxiety-inducing events. They
write, \[Threatening appeals] may not be successful at making all people feel anxious, but
if they are, these appeals are particularly successful at persuading them to support a range
of policies meant to protect them from harm." If we extend their results to a campaign
fundraising scenario, emails written to induce anxiety may not work on most people to
become anxious, but may be eective at producing a contribution among people who are
made anxious, or are already anxious. From this, I reason that individuals more prone
to worry and anxiety are also more likely to contribute after processing anxiety-inducing
messages.
A caveat here is to note that generic inducement of worry or anxiety shouldn't be expected
to work, as political anxiety depends on party, ideology, and candidate/policy preferences.
Rather, in-party or in-group inducement of worry or anxiety is the precise phenomenon to
be observed. Albertson and Gadarian (2015) show that anxiety-inducing messages from
an opposing party are less eective at producing anxiety among out-partisans; that is, a
republican message meant to induce anxiety regarding immigration is less likely to produce
anxiety in a strong progressive with positive opinions of immigrants. Additionally, Westen
(2008) shows that partisans are able to rationalize inconsistencies in the actions of in-party
candidates, but punish out-party candidates for similar inconsistencies. At a baseline level,
it should be expected that while out-party anxiety-inducing fundraising solicitations will
virtually never result in a contribution, in-party inducement of anxiety should frequently be
successful at producing campaign contributions.
An email sends information about terrible events directly to people's inboxes and then
provides a low-cost, yet ephemeral, solution that's just a click away. Donors feel better emo-
tionally to have actually taken political action and done their part for democracy or a noble
cause. The catch, of course, is that donors will soon receive another email requesting a con-
tribution. And in fact, as soon as somebody gives money once, sophisticated campaigns will
track donor behavior and ask them for more money at an even higher rate. By contributing
33
once at the request of a campaign, the likelihood of being requested more increases. Still,
contributing or opting in to campaigns is not the only way to become targeted, as campaigns
use detailed individual data in sophisticated ways to nd, target, and track individuals via
methods such as email, phone, internet metadata and even social media activity (Hersh,
2015; Bakir, 2020; Bartlett et al., 2018).
People generally contribute only when asked, and typically people with background traits
such as political interest and money are likely to both give and be targeted by campaigns
through rational prospecting (Brady et al., 1999; Grant and Rudolph, 2002). Because so-
licitation is so central to small-dollar contributing and functions as a major form of direct
communication between campaigns and individuals, the nature of those fundraising mes-
sages occupies an important role in understanding why people give. From a normative
perspective, American elections being increasingly funded through the mass distribution of
anxiety-inducing messages is problematic. In addition to the very low cost of sending mes-
sages in 2021, contemporary data science allows campaigns to cheaply nd and target people
directly through text and email. As such, campaign messages warrant further study from a
variety of perspectives, including eectiveness for campaigns, harm done to individuals, and
long-term attitude changes in individuals. This dissertation only studies the eectiveness
of using anxiety as a fundraising tactic, but encourages follow-up research on the harm to
individuals from repeated exposure to anxious messages, as well as attitude changes.black
Fundraising messages are designed to produce emotions in people so strong that they
compel them to make a contribution to a campaign, despite receiving no tangible benet
in return.
8
The emotions invoked by campaigns run the spectrum, but commonly include
fear, worry, anxiety, and anger. I focus on the worry and anxiety because they are similar
concepts, and one of the biggest features of campaign fundraising is the truly apocalyptic
nature of many fundraising emails. The messages are designed to make people feel like the
world will end unless they give $20.
8
With the exception of merchandise and lottery-style events
34
Michael Whitney, a former fundraising manager for Bernie Sanders argues that organiza-
tions like the DCCC use fundraising appeals that appeal to anxiety, fear, shock, and shame
to incite people to action (Whitney, 2018, 2017). The strategy appears to be one that in-
duces whiplash about the outcome of an election, constantly putting in doubt a candidate's
ability to win, emphasizing how bad that would be, and asking for money to avoid it.
Whitney (2017) argues that the DCCC has used a variety of emails that invoke panic,
writing,
\Other DCCC emails cause messaging whiplash. In the week before the June special
election in Georgia featuring Democratic candidate Jon Osso, the DCCC sent a fundraising
email with the subject line `JON OSSOFF LOSES!' ... followed by another email four hours
later with the subject line `Osso DOMINATES!' Osso's campaign used nearly identical
tactics itself. Just two days before the election, it sent a fundraising email to supporters with
the subject line, `Accept defeat. Jon Osso lost. It's over.' The eect of these tactics is
to shock, depress or shame people into action. It's the opposite of empowering. They use
it because it works, at least initially-the DCCC is beating its own fundraising records, and
Osso raised an unprecedented amount of money for a House race, with two-thirds coming
from small-dollar donors. But while this strategy raises money in the short term, a party
can't build a future by telling supporters hour after hour that the sky is falling. It's bad for the
candidates, bad for the party, and, most vitally, bad for the voters whose support they need
to build up. It's also harmful to the Democratic Party's allies. A 2014 experiment conducted
between the DCCC and a left-leaning issue advocacy group sought to measure what eect the
DCCC's messaging would have on supporters of the advocacy group. What they found should
deeply concern Democrats: Email addresses exposed to the DCCC's program preformed 15
percent worse in both money raised and contributions made as compared to email addresses
that weren't on the DCCC's list."
9
This critique is not new, as Democratic fundraising
strategists and journalists have been noting and raising questions about DCCC fundraising
9
I was unable to nd the experiment Whitney refers to, so this claim should only be taken at face value.
35
strategies since at least 2014 (Huey-Burns, 2014; Roller and National Journal, 2014; O'Keefe,
2014). Huey-Burns (2014) writes of DCCC emails, \The subject lines on the emails read
like they could have been sent by exes, stalkers { or cultists awaiting the Apocalypse....
`Throw in the towel,' says one. `Kiss all hope good bye,' reads another. A third simply says,
`Doomed."' In a Washington Post piece, O'Keefe (2014) notes an email from the Democratic
Governors Association that read, \We're on the verge of the Dem-pocalypse".
While Whitney provides a contemporary articulation of an overarching fear-based emo-
tional framework used by campaigns, campaigns have been using emotional manipulation
for many years. Brader (2006) argues that political advertising is replete with emotional
appeals and in fact has been since the 1980s (Kern, 1989). Brader (2006) forwards the idea
that a useful bipolar dimension on which ads may be viewed is a spectrum from fear to
enthusiasm. It is well known that emotions contribute to public opinion (Marcus et al.,
2000), especially sound, images, and speech on television Kinder (2003); Brader (2006). One
dierence between Brader's work on television ads and digital ads Brader (2006, 4) is that
\...political ads, like product commercials, usually contain more than words. They are full of
pictures, sounds, and music. Why are simple messages packaged with this nonverbal fanfare?
It could be that these features make ads more entertaining and thus encourage an otherwise
disinterested public to watch. Even so, it is doubtful that people who are not already po-
litical junkies nd much entertainment in political ads. The music and images accomplish
something more than merely enhancing the pleasure of the viewing experience. They make
the ad compelling by eliciting specic emotions and, in doing so, change the way viewers
respond to the ad." Brader's work goes on to argue that music and images do not substitute
for the verbal or written component of the emotionally appealing ad, but rather sharpen
the written message. While Brader gives special attention to the nonverbal elements the
emotional appeals in ads are intended to make, I focus on the text itself of political emails
for sake of methodological parsimony.
36
No argument is made that stressing potential negative consequences as a result of an
election loss is always normatively negative or harmful to democratic norms. The reason so
many stress-inducing emails are sent is because they tap into real emotions. Kelly Ward,
DCCC executive director for the DCCC in 2014 wrote, \The level of outrage that our list feels
about the Republican Congress is very real... They continue to take action and continue to
set priorities that are out of step with the country, and our lists re
ect that and [DCCC email
subscribers] respond with contributions. They are very energized." The director's comments
re
ect a larger, important point about fundraising: while inducing strong negative emotions
in the public can be considered normatively harmful, it is not necessarily so. However, while
anxiety-inducing emails may be healthy on occasion, the potential for abuse is clear, and
they are used all throughout the year because campaigns think they work.
1.5.1 Political Anxiety: A Major Feature of Society
In 2015, Albertson and Gadarian (2015) published \Anxious Politics," a foundational text
for exploring anxiety in politics. They investigate the question: \What are the consequences
for an American public besieged with emotional appeals, evocative imagery, and threatening
news?"
Albertson and Gadarian (2015, xix) write, \whether the issue is an infectious disease
like polio or new immigrants, American life is often threatening. Crisis, threat, and worry
are relatively common phenomena in the American landscape...Candidates running for oce
and elected leaders warn the public of risks to health, safety, and security. It is dicult
to turn on a news channel without hearing about the latest virus, food, political regime,
or ideology that is an imminent threat to one's life and the security of the nation." This
passage emphasizes how emotion and anxiety have long been a central feature of American
politics. Especially in a political contribution setting, anxiety and worry have a particular
capacity to in
uence behavior.
37
When discussing future research, Albertson and Gadarian (2015, 222) write, \One set
of open questions from our research is about the duration of anxiety's eects. How long
do the eects of immigration anxiety or worries over a public health crisis last? what if
threatening stimuli are ever-present? If citizens face daily reminders of economic crisis,
increasing crime, or the threat of terrorism when they
ip on the television or talk to a
neighbor, do people remain vigilant and trusting, or do they tune out?" This dissertation
seeks to directly respond to that question by studying anxiety induced by political campaigns
for the purposes of fundraising. In particular, the use and overuse of of anxiety is studied,
as well as the eectiveness of anxiety over time.
Campaign fundraising serves as a sort of propaganda that attempts to in
uence minds
and generate actions among supporters of political causes. Scholars have long been wary of
the ways that mass communication and propaganda can have negative eects on the public
at large (Lasswell, 1927; Lippmann et al., 1943). While in
uencing attitudes through mass
communication is a core component of society, scholars like Edelman (1985) have shown that
emotionally manipulating a public can lead to harmful consequences. Even Hamilton et al.
(2008) in the federalist papers have expressed concern that anxiety created by politics hurts
people's abilities to make rational political decisions. Of course, that is not to say that all
emotion in politics are bad or serve as a barrier to reasonable and important discussions
(Fishkin, 1997; Ackerman and Fishkin, 2004; Habermas and Habermas, 1991), but emotions
such as anxiety are important, but have not always been major considerations for research
by political scientists (Albertson and Gadarian, 2015, 45).
1.5.2 Anxiety
Anxiety in particular has great potential for manipulation. Anxiety is an \unpleasant and
aversive state" that occurs in individuals when they appraise a situation as being unpleasant,
highly threatening, uncertain, or out of people's control (Eysenck, 1992; Lerner and Keltner,
38
2001, 2000; Roseman and Evdokas, 2004; Smith and Ellsworth, 1985; Steenbergen and Ellis,
2006). Spielberger (1972, 30) denes anxiety as \the complex emotional reactions that are
evoked in individuals who interpret specic situations as personally threatening."
It motivates individuals to avoid danger, seek protection, and create a safer environment
(Jarymowicz and Bar-Tal, 2006; Nabi, 1999; Roseman and Evdokas, 2004). Political anxiety
is created by a variety of sources, from advertising (Brader, 2006), candidates (Kinder et al.,
1979), policy changes (MacKuen et al., 2010), and real-world events (Boyle et al., 2004). As
stated before, people experiencing anxiety that search for more information will focus on
negative information, as it is more relevant to avoiding harm and thus more able to reduce
anxiety (Cacioppo and Berntson, 1994; Lau, 1982).
\Initiation of worry may also be in
uenced by appraisals regarding one's ability to per-
sonally in
uence, or cope with, threatening outcomes (Tallis and Eysenck, 1994)." I argue
that in political settings, taking action to cope with or mitigate threats is a likely behavioral
response.
Albertson and Gadarian (2015) write, \When anxiety originates from political life, we
expect that individuals will employ more conscious processes connected to politics (i.e.,
calling a Member of Congress) more often than interpersonal methods (i.e., calling a friend)."
In the context of supportive members subscribing to a campaigns' political email list, the
conscious process to cope with political anxiety is contributing money. This is generally
supported by existing research: an experimental study by Brader (2006) found that anxiety-
inducing campaign ads increased learning and the potential for persuasion. They also found
that political challengers were more likely to use fear-based appeals than front-runners, who
are more prone to using enthusiasm-based appeals.
Fear-inducing campaign communication is quite common: Ridout and Searles (2011)
found that 23% of ads for senate races in 2004 used fear appeals. Within this, Ridout
and Searles (2011) nd temporal eects: losing candidates tend to use fear-based appeals
39
toward the end of a race in an eort to shock people out of partisan habits. Witte (1992)
and Witte and Allen (2000) nd that urging people to pay attention to threats/risks is an
eective strategy for campaigns, so long as those anxiety appeals do not appear manipulative
(Dillard and Shen, 2005). The a priori expectation is that some subset of the population
will nd fundraising emails to be too annoying, manipulative, or anxiety-inducing at some
time horizon or email threshold.
1.5.3 Why Use Anxiety to Fundraise? The Theory of Anxious
Avoidance-Based Contributing
An important question to consider in a study of political fundraising and anxiety is why
anxiety would make people more likely to contribute. One explanation lies in the concept
of control. Thompson (1981, 89) denes control as \the belief that one has at one's disposal
a response that can in
uence the aversiveness of an event." Grupe and Nitschke (2013)
interpret this denition to mean that \when one is unable to prevent negative events from
occurring, increased certainty about the future provides one with control over adaptive antic-
ipatory responses that can mitigate these events' negative impact. Uncertainty, conversely,
precludes one from exercising this form of control and leads to preparations that are \diuse,
psychologically expensive, and of questionable eectiveness." I argue that making a political
contribution is the obvious behavior for people looking to relieve their political anxiety and
lack of control. Grupe and Nitschke (2013)'s point of preparations being diuse, psychologi-
cally expensive, and of questionable eectiveness map perfectly onto why people would make
a political contribution despite it not being pivotal to the outcome of the election, as outlined
in the discussion of rational choice theory. Because pivotality-based rational choice theories
would predict almost nobody contributes small sums of money, the emotional component of
contribution decisions provides a more convincing explanation for small-donor behavior.
40
Grupe and Nitschke (2013, 489-490) write that \Control can also be thought of as the
belief that one has at one's disposal a response that can in
uence the aversiveness of an event
(Thompson, 1981). Thus, even when one is unable to prevent negative events from occurring,
increased certainty about the future provides one with control over adaptive anticipatory
responses that can mitigate these events' negative impact." Their work directly informs
the reasoning for why people with greater-than-average levels of anxiety are expected to
contribute at higher rates than their less anxious counterparts. Extended to contribution
decisions, the theory of anxious avoidance predicts that people with higher levels of anxiety
seek to ease their political anxiety directly by making contributions to political groups to
gain a greater sense of control over potential negative future outcomes. The caveat, however,
is that small contributors have a nancial limit at which point they are unable to make any
more contributions. Repeatedly sending anxiety-inducing fundraising messages to donors
both annoys contributors and is anxiety inducing, both of which discourage small donors
from contributing again.
I theorize that many fundraising emails are being sent to individuals in an attempt to
produce worry in people that increases their psychological distress over political outcomes,
all with the end goal being to produce a contribution from the individual. Making a small
political contribution provides no tangible rewards and by itself is incapable of making a
dierence in a campaign. Because contributing does not provide tangible rewards, the job of
campaign fundraising solicitation is to produce emotion so strong that people are compelled
to make a contribution to a campaign. I theorize that in many cases, these decisions are
made as a coping mechanism to deal with psychological distress over political uncertainty
that is exacerbated by political emails. Of course, fundraising in response to dangerous
real-world events is not normatively harmful, per se, but the potential for campaigns to
inappropriately generate mass worry and anxiety is clear and dangerous. This channel of
worry, anxiety, and control to produce a political contribution is but one of many emotional
channels that campaigns can and do use to manipulate individuals into making political
41
contributions. Other common paths to producing emotion so strong in people that they
contribute might include overwhelming sense of civic duty, anger, fear, etc. Future research
into these paths is encouraged.
Stapinski et al. (2010) write that \Generalised anxiety disorder (GAD) is characterised
by persistent and uncontrollable worry. According to the cognitive avoidance theory of GAD,
worry may function as an aective dampening strategy motivated by intolerance of negative
emotional states... Worry is a ruminative process xated on possible future catastrophes."
Rumination on potential future catastrophes commonly leads people to take action to avoid
those catastrophes. In a political fundraising scenario, the election of a rival candidate and
the policies they support are the perceived catastrophes for potential contributors.
Craske et al. (1989) write, \Perceptions of diminished control over environmental out-
comes, including both external events and internal events such as emotions, are thought to
exacerbate the experience of anxious aect associated with any potential future threat." I
theorize that the less sense of control someone has, the more likely they are to contribute
to gain a sense of control. Merolla et al. (2012) describe how collective crises, including
political crises, have solutions that are simply beyond the control of any individual or house-
hold. While people can engage in behaviors such as stocking their home with supplies like
gas masks in preparation for a terrorist attack, they cannot stop a terrorist attack them-
selves. For any number of political issues from terrorism to immigration to job automation,
individual eorts to stop a large-scale event are not possible, but individuals still prepare
and try to mitigate negative outcomes. In a political environment, one of the most obvious
ways to participate (beyond voting) to gain control over an uncertain and threatening future
is to make a contribution to a cause larger than oneself. Because individual action is not
sucient to address collective and political crisis, individuals recognize their powerlessness
and contribute, quite literally, to a collective eort.
42
To summarise, I theorize people contribute due to an interaction between anxiety, sense
of loss-of-control, and the receipt of fundraising emails. Having justied the motivation to
study small donors, outlined existing theories of why people contribute, and forwarded the
theory of anxious-avoidance, I specify the measurement strategy used in this research.
1.5.4 Measurement of Anxiety
Anxiety is measured in two ways for this dissertation. In chapter 4, the intent to induce
worry/anxiety in email text is classied via text analysis. In chapter 5, anxiety is measured
via the survey instruments which assess participants' general anxiety and intolerance of
uncertainty.
The chapter 4 classication of worry/anxiety-inducement is more dicult because it is a
far less common task and is inherently subjective. The type of text that will induce worry
or anxiety in some people will not induce worry or anxiety in others. To work around this
problem, crowdsourced workers were used to build an automated classier to be applied
86,209 emails, according to the following denition of worry from Buhr and Dugas (2002).
\Worry can be dened as concern about future events in which there is uncertainty about
the outcome and where the individual experiences feelings of anxiety."
Each of the emails in the training data was classied by only one crowdsourced worker
as an attempt to maneuver around the inherent subjective nature of anxiety-inducement in
text. However in practice, this led to a poorly separated class. Because of this, the results
from a supervised machine learning classier were only about 12 - 25% better than random
depending on the metric used. While this is adequate for research, improvements on the
accuracy, prediction, and recall of the classier should be improved upon in any future work.
Chapter 5 uses two well-validated survey instruments to assess individual anxiety. The
rst instrument used is the GAD-7, a 7-item survey instrument to assess generalised anxiety
disorder, as a general measure of anxiety as a trait (Spitzer et al., 2006; Swinson, 2006;
43
L owe et al., 2008; Ruiz et al., 2011). Individuals with higher levels of GAD have lower
tolerance of negative emotional states. The second instrument used is the IUS-12, a measure
of intolerance of uncertainty (Freeston et al., 1994; Buhr and Dugas, 2002). Both the GAD-
7 and the IUS-12 are used because in addition to a general measure of anxiety (GAD-7),
intolerance of uncertainty was expected to be highly relevant to political anxiety. The GAD-7
measures the frequency of 7 anxiety-related problems over the last 2 weeks:
1. feeling nervous, anxious, or on edge
2. not being able to stop or control worrying
3. worrying too much about dierent things
4. trouble relaxing
5. being so restless that it is hard to sit still
6. becoming easily annoyed or irritable
7. feeling afraid as if something awful might happen
Then, it measures the severity of those problems by asking this question: \If you checked
o any problems, how dicult have these problems made it for you to do your work, take
care of things at home, or get along with other people?" with a 4-point answer scale ranging
from `not dicult at all` to `extremely dicult'.
10
This provides a general sense of how
anxious a person is. It is important to acknowledge that this question was elded to subject
participants in the summer of 2021 when the COVID-19 pandemic was still dominant in
the world, although a vaccine was available and many people were being vaccinated in the
United States, so general anxiety wouldn't have been as high as a period like October or
November 2020, when pandemic deaths were at an all-time high and the presidential election
was looming.
The IUS-12 is a 12-item panel of statements about how well an individual can tolerate the
feeling of uncertainty. Participants are asked on a 1-5 scale how characteristic each of these 12
statements are of themselves. The statements include `Unforeseen events upset me greatly;'
10
The specic implementation of this question will be included in chapter 5
44
A small, unforseen event can spoil everything, even with the best of planning;' `Uncertainty
keeps me from living a full life;' `I must get away from all uncertain situations;' and more.
11
The IUS-12 was used because it t nicely into the concept of control and the role it plays in
political anxiety. Individuals who have low tolerance of uncertainty feel more psychological
distress over the potential of future negative events/threats, and as the name implies, cannot
tolerate the uncertainty surrounding negative possibilities (Carleton et al., 2007). Lower
tolerance of uncertainty increases the amount of worry an individual experiences (Laugesen
et al., 2003) and the anxiety they feel (Greco and Roger, 2003). Uncertainty itself is a
stressor (Monat et al., 1972), so much so that study participants prefer immediate electric
shocks to unpredictable ones, feel unpredictable shocks are stronger than predictable ones
(Badia et al., 1966), and feel more anxious when being treated to unpredictable shocks (vs
predictable shocks) (Badia et al., 1966; Pervin, 1963; Lanzetta and Driscoll, 1966).
Generalised anxiety disorder is a very common medical disorder prevalent in 1.6 - 5.0%
of the population (Wittchen et al., 1994; Kessler et al., 2005, 2001). Uncertainty, stress,
and anxiety are not solely mental phenomena, but have physiological responses. Greco and
Roger (2003) write, \Other studies have shown that the anticipation of a future, possible
threat of unknown intensity constitutes a potent psychological stimulus that has an eect on
the pituitary-adrenocortical system and the sympathetic-adrenal medullary system (Mason
et al., 1973; Voigt et al., 1990; Zakowski, 1995)." Further, anxiety may also be related to
the physiological eects from the fear of pain (Asmundson, 1999; Barlow, 2004; Asmundson
et al., 2004). While the physiological eects of anxiety and uncertainty on the body are not
explored further, these references are included to make the point that the eects of anxiety
and producing anxiety in others is not so intangible or ephemeral, but has real eects on the
body and mind that can impact mood and behavior.
Politics centers around elections, an inherently uncertain and volatile process to make
large-scale changes to the country. Politically interested anxious individuals and individuals
11
The full and specic implementation of the IUS-12 is included in chapter 5
45
with lower tolerance of uncertainty are likely to feel such distress from the uncertainty of
political events like elections that they will be more likely to make political contributions,
as even a small contribution may help to make an outcome slightly more certain. At the
very least, making a contribution may ease the mind as even though people have limited
resources, spending $10 on a campaign is doing what one is capable of to try to thwart a
possible negative outcome where a `bad' candidate wins.
1.6 Outline of this Dissertation
Throughout this project, the question of why regular people make political contributions is
explored. In chapter 2, past research into donor motivations is replicated and expanded upon.
This is done by analysing time-series survey data from the CCES (Cooperative Congressional
Election Study) and the ANES (American National Election Study). Basic descriptive statis-
tics and regression analyses are conducted and presented to identify factors that are most
predictive of making a political contribution. The primary independent variable investi-
gated is the binomial variable of whether or not one has made a political contribution to
a political entity. This chapter mainly answers the question of what background traits are
most predictive of making a political contribution, updated to include more contemporary
developments to campaign fundraising after the 2016 presidential election. In chapter 3, this
analysis is expanded upon by directly comparing contribution behaviors between party ID
and ideological lines.
In chapter 4, over 80,000 emails sent by political campaigns between July 2019 - Novem-
ber 2020 were collected and classied as being anxiety/worry-inducing or not. Additionally,
all individual contributions to those political campaigns were collected. The eect of send-
ing worry-inducing emails on fundraising returns is analyzed, testing the hypothesis that
inducing anxiety is an ineective strategy in the long-term. In chapter 5, a survey is con-
ducted to understand the eect of anxiety on the likelihood that individuals contribute to
46
campaigns and avoid campaign contact. The prediction made is that greater individual
anxiety is associated with a higher likelihood of both making campaign contributions and
avoiding campaigns. Theoretically, this is explained by individuals using contributions as a
way to avoid worry and feel a greater sense of control over their world. Finally, chapter 6
provides a brief review of the results and discusses some of the implications of the ndings.
Additionally, limitations of this research are acknowledged and lines of future research are
suggested.
47
Chapter 2
Who Contributes?
2.1 Abstract
Using survey data from the Cooperative Congressional Election Study (CCES) and the
American National Election Study (ANES), this chapter explores the background traits
most strongly associated with making a political contribution. Evidence shows that that
Democratic partisanship, liberal ideology, and being contacted by a campaign are all highly
in
uential traits that have been less prominent in previous research. In addition, results
from past works such as Brady et al. (1995); Verba et al. (1995); Schlozman et al. (2012)
are conrmed, nding the basic facts that an individual's income, political aliation and
strength of partisanship, degree of interest in politics, and education are very explanatory
and predictive of who will make a political contribution. However, the data shows that after
2012, Democratic partisanship, liberal ideology, and being contacted have become much
more closely associated with making a political contribution over time. Additionally, survey
data suggests that text/email contact is the method of contact most strongly associated with
making a contribution.
48
2.2 Introduction
There are many variables that impact the decision of an individual to contribute. Most com-
monly studied are the demographic, political, and social characteristics of individuals, such
as party identication and the strength of partisanship, gender, race, age, income, education,
news consumption, political interest, and to a lesser degree the Big-Five personality traits
(Mondak and D Halperin, 2008; Gerber et al., 2012). Solicitation has been demonstrated
to be an incredibly important factor in contribution decisions, with over 3/4 of contribu-
tors having been solicited. However, not much research focuses on the relationship between
campaigns and their small dollar contributors, and almost no modern research studies why
small donors give while specically examining the role of campaign solicitation, aside from
Magleby et al. (2018)'s study of presidential campaigns. This chapter and chapter 3 expand
past research on demographic, political, and social characteristics that most increase an in-
dividual's contribution likelihood. This work examines Congressional, Gubernatorial, and
Presidential elections from 2000-2018, placing emphasis on both solicitation and changes in
the eects of variables over time. The widespread adoption of the internet between 2000-2020
as well as campaigns' use of readily available personal data have changed the political con-
tribution landscape (Kosinski et al., 2013; Hersh, 2015; Bartlett et al., 2018), necessitating
re-examination of the variables that impact political contribution decisions.
The ndings in this chapter provide an argument for two updates to general scholarly
understanding of why people make political contributions: First, research of modern cam-
paign fundraising/contributing must include contact and/or solicitation as central variables
in predicting contributions. Second, Democratic partisanship and liberal ideology should be
acknowledged as major factors in predicting contributions. In a surprising nding, Democrats
and liberals generally contribute more than Republicans and conservatives, with the trend
appearing after 2010. Particularly, though, strong Democrats and strong liberals contribute
at much higher rates than strong Republicans and strong conservatives, indicating that
49
stronger Democratic partisanship and liberal ideology increase the likelihood one is a polit-
ical contributor.
This chapter is divided into ve main parts. First, the extant political contribution litera-
ture is reviewed and theoretical reasons for the importance of certain variables in contribution
decisions are outlined. Second, research questions and hypotheses are made regarding which
political, demographic, and social characteristics are most associated with making political
contributions. Third, hypotheses are tested by running logistic regression models. Fourth,
logistic regression models are conducted using one half of the data as a training set to \pre-
dict" contributions from the other half of the data as a test set. This is done to test how
well a model can predict who will make a contribution using only observable traits recorded
in public observational surveys that entities such as campaigns or non-prots may already
know. The results are reported (accuracy, precision, recall, and F-1) to understand how
much explanatory power these variables have regarding contribution decisions. Fifth, in
the discussion two key ndings are highlighted: The rst is a surprising nding that
Democratic partisanship and liberal ideology are closely associated with making
political contributions, in contrast to Republican partisanship and conservative
ideology. The second key point is that contact is an extremely important factor
for determining who contributes and who doesn't. The nding that campaign
contact is a very strong predictor of who contributes serves to highlight the
necessity of studying campaign contact in research on political behavior. These
ndings are notable because(1) Democratic partisanship and liberal ideology have typically
not been considered variables that increase the likelihood of contributing more than Republi-
can partisanship and conservative ideology, so demonstrating and exploring that connection
is novel and important work. (2) Solicitation has not been adequately highlighted and ex-
plored as a central determining factor for why people contribute money to political entities,
so more research is warranted on campaign fundraising solicitation.
50
2.3 Literature Review: The Traits that Predict Polit-
ical Contributions
To begin, the theoretical connection between some of the known variables and propensity to
make a political contribution is outlined. Variables that have been argued to have a connec-
tion in academic literature include (1) income (2) partisanship (3) strength of partisanship
(4) ideology (5) strength of ideology (6) interest in politics (7) interest in the news/current
aairs/government aairs
1
(8) education (9) gender (10) race and (11) contact
2
In the appendix there is a
owchart in gure 6.1 describing some of the concepts and
variables that lead one to make a political contribution, providing a visual explanation of
how contribution decisions work. The general pathway that explains contributions begins
with the circumstances of life out of one's control { namely the income level, political beliefs,
and political activity of one's parents. Closely linked to one's circumstances of birth is family
income, which has a feedback loop with education: the higher family income one has, the
more likely they are to be educated, and the more educated one is the more likely they are
to have a higher income. Education aects one's issue knowledge and how likely one is to
follow news and politics, which in turn aect one's political identity and ideology. At the
intersection of one's issue knowledge and political identities is how intensely one feels about
various issues, which is a vital link in the pathway to contributing. How strongly one feels
about political issues impacts how gratied one will feel from political participation, which
increases the likelihood of being asked to give and saying `yes'. Caring more about issues
means one is more willing to give, at which point the only obstacles are whether one has the
capacity to give (which is essentially a function of income and/or wealth) and whether one
1
Of course, interest in politics and frequency with which one follows the news/current aairs are highly
correlated variables.
2
One could argue that whether or not one has been contacted is a \background trait" like income or
partisanship, but I include it in this section because being solicited has been shown to be a very important
variable to contribution decisions. Further, being contacted is included in some CCES and ANES surveys,
so it is included here to increase the explanatory power of the regression model.
51
is asked to give, which lowers the structural barriers to giving and can be seen as `activating'
dormant political contributors.
In addition, income directly aects both one's capacity to give and whether one is asked
to give, as rational prospecting makes campaigns and political groups more likely to request
donations from people with money who have donated in the past. Additionally, people
may opt-in to being solicited as a function of caring about political issues and engaging in
activities like signing up for a candidate's or political group's email list, which will almost
certainly email solicitations for money. But while all of these variables certainly can and do
in
uence one's propensity or likelihood to make a political contribution, some are stronger
than others.
Income
It is well researched that income has a large eect on one's propensity to make a contribution
(Verba et al., 1995; Brady et al., 1995, 1999). Income works to increase one's likelihood
of making a contribution via two main interconnected pathways. To start, one's income
determines one's relative capacity to give. If somebody has zero expendable income, it is
extremely unlikely one will make political contributions because with the limited resources
one has, it makes more sense to spend that money on essentials such as housing, food,
healthcare, and other tangible goods. The more money one has, the more capable one is
to make a contribution. The more capable one is of making a contribution, the more likely
they are to contribute because the greater one's income, the lower the real-world cost of
contributing is. In essence, giving a candidate $20 on a $35,000 income is a greater cost than
giving a candidate $20 on a $350,000 income.
The second pathway by which income increases one's likelihood of contributing is through
\rational prospecting" (Brady et al., 1999). Political recruiters and solicitors often have
information about the income of many Americans stored in proprietary databases. Because
it costs money, time, and resources for political entities to send solicitation materials to an
52
individual, political fundraisers generally attempt to target people who are known to be
likely to give. These two pathways of income both work to increase the likelihood of an
individual making a contribution.
Modern Fundraising Solicitation
Higher income levels mean one is more likely to be asked to give through rational prospect-
ing and strategic recruitment by campaign fundraisers, which increases one's likelihood to
contribute (Brady et al., 1999). As technology has improved, it has made large datasets
containing useful information about individuals more accessible to campaigns. Campaigns
are able to use massive, detailed datasets to more eciently discover and reach individuals
wherever they are, whether by mail, text, email, social media, or advertisement (Nickerson
and Rogers, 2014). Over time, and especially as the cost of computing power decreases,
campaigns hire sta and consultants trained in data science to better allow them to acquire
and use detailed individual-level data. Use of detailed individual datasets is not limited
to tech-savvy or resource-rich campaigns, as the major national parties often provide their
candidates with well-maintained national voter databases (McAulie and Kettmann, 2007).
Beyond national party datasets, some particularly tech-savvy campaigns make use of
private datasets and commercial tools, including the 2016 Trump campaign's controversial
use of the psychographic proling rm Cambridge Analytica as well as their use of Facebook
tools Bakir (2020); Bartlett et al. (2018). Companies and campaigns track individuals across
various online platforms and services using data including email addresses, usernames, cook-
ies, unique hardware/device identiers, IP addresses, browsing data, location data, and even
personality surveys cross referenced with the user's social media data (Kosinski et al., 2013;
Hersh, 2015; Bartlett et al., 2018). While there is much more to be said about this topic,
the scope of this project is limited to emails.
3
3
There is more discussion on this topic in chapter 4, where I use political emails as a data source. Because
campaigns may run experiments and send dierent messages to dierent users, there is no way to capture
all emails sent by a campaign with absolute certainty.
53
Partisanship, Ideology, and the Strength of Both
It is well demonstrated and theorized the myriad ways in which party ID and strength of
partisanship impact one's political attitudes and behaviors. Partisanship is a core component
of Campbell et al. (1960)'s funnel of causality, in which party ID is the most important
factor in predicting one's vote. Party ID determines which party one prefers, and thus which
types of candidates one supports. Because making a political contribution exists as the
act of giving money to support a cause or candidate one believes in, party identication
plays an enormous factor in determining who or what one will give money to. Beyond the
determination of where a contribution will go, partisanship and strength of partisanship
impact one's decision to make a contribution at all. Most donors cite purposive motivations
for contributing (Francia et al., 2003; Webster et al., 2001), as people give money to causes
they believe in for an intangible good such as feeling better about themselves, or feeling
a `warm glow'. Similarly, the stronger one feels about an issue the more likely they are
to contribute money towards that issue because the more strongly one holds a political
ideal, the more likely that person is to seek out candidates and organizations that share and
support that view (Schuman et al., 1981). Further, strength of beliefs can serve to lower the
contribution costs, as intensity of feeling about an issue will increase the satisfaction from
giving. Once people begin to seek out organizations, they begin to nd places where they
can engage politically with an issue they care about by donating money in support of that
issue.
Ideology and strength of ideology function in very similar ways as party ID and par-
tisanship, especially given the tremendous degree of partisan ideological sorting in the last
decade (Fiorina, 2016; Mason, 2015; Abramowitz and Saunders, 2008). Because ideology and
partisanship are so strongly correlated and Democrats are almost exclusively liberals and Re-
publicans are almost exclusively conservatives, the explanatory pathways for the impact of
ideology and partisanship on propensity to contribute are basically identical. Ideology and
partisanship may weight one's decision to contribute dierently, but the basic function of
54
both variables is that they change what one cares about and how much they care about it.
The more strongly one feels about an issue, the more likely one is to give money towards
advancing that issue.
Interest in Politics and Interest in News
Political interest is an essential component in contribution decisions. people who care about
politics and are concerned with the country are more politically active that those who do
not show much political interest (Verba et al., 1995). In fact, political interest is so vital to
participation that Milbrath and Goel (1977) nd \...many authors do not bother to report
it." Rather than exclude the variable, it is included here as an important causal factor that
leads to caring about political candidates, parties, and groups, which is a necessary step to
giving money to them.
Education
People who tend to be more educated are more knowledgeable on political issues and current
events. Knowledge about current events is a precursor to having opinions on current events,
and having opinions on political issues is an essential component in the pathway that leads
one to give money to a political candidate or cause. It is important to note that education
is strongly linked to income, in that income sharply increases as education does Verba et al.
(1995). Education is simply a vital component of any and all SES models that indicate
higher SES levels cause higher levels of political participation Verba et al. (1995); Brady
et al. (1995), of which contributing is a part.
Race and gender
Race and gender have a variety of consequences for one's socioeconomic status, namely that
white people and men, and white men in particular, are more likely to have higher incomes
and thus more likely to make political contributions Verba et al. (1995). Race and gender
aect socioeconomic status, which has an impact on propensity to contribute through issue
knowledge, capacity to give, and thus chances of being solicited. Higher SES levels generally
55
cause individuals to have higher levels of education, be more civically engaged, and thus
be more knowledgeable and informed of political issues, which increases one's willingness
to give (Brady et al., 1995; Grant and Rudolph, 2002; Rosenstone et al., 1993). However,
there does not appear to be a consensus on whether or not race has a consistent impact on
contribution decisions (Frost, 2013).
Contextual Variables
When considering election-specic contextual variables, Individuals are more likely to vote
when they believe their vote might impact the outcome of the election (Riker and Ordeshook,
1968). In competitive elections, the chances of a vote impacting the election increase, and
many voters are aware of this. As a result, close races are likely to increase the probability of
voter participation, because political participation is more likely to be perceived as in one's
self-interest (Blais, 2000, 57).
Many survey respondents provide the state or zip code they reside in, which allows for
two types of variable operationalization. Some states are classied as battleground states,
which can be considered competitive in a general sense, across time. Individuals living in
battleground states are more likely to turnout to vote (McDonald, 2004; Lipsitz, 2009). The
eect of being in a deeply embattled battleground state on voter participation was measured
by Cebula et al. (2013) to be 7.8% in presidential elections from 1964-2008. However, when
extending the battleground state hypothesis to political donations, empirical results fall short
of support. Lipsitz (2009) and Frost (2013) both nd respondents in battleground states are
not more likely to contribute money than respondents in safe states. I postulate this may
be due to the fact that donors across the country can donate anywhere, and donors located
in very safe districts (such as urban centers like Los Angeles or New York) may donate to
battleground states to increase their internal sense of ecacy, knowing their vote in a very
safe district will not have much impact on political outcomes.
56
Importantly, Frost (2013) does not nd evidence of presidential ad spending on propensity
to donate, which does not support the notion that people living in battleground states are
more likely to contribute, as more ads are run in battleground states. However, Frost does
nd strong evidence that direct contact by campaigns increases propensity to donate. Using
panel survey data from CCAP 2008, she crucially nds that most people who donated were
contacted by a campaign prior to donating, supporting an important temporal element to
the theory that solicitation largely impacts propensity to contribute.
While other research shows evidence that when an individual's favored candidate is down
in the polls (Rogers and Moore, 2014; Rogers et al., 2017b), they are more likely to make a
contribution to that candidate - however, that result was demonstrated using experimental
evidence. It is possible that electoral competition can increase one's propensity to contribute,
but that this is a latent eect that requires activation via solicitation or campaign contact
to function properly.
Contact and Solicitation
Basic empirical evidence shows that individuals are more likely to make political con-
tributions if they are asked to do so. This CCES data shows that the vast majority of
donors have been contacted by a campaign within the past year. About 16% of people give
spontanelously to a campaign without being solicited. This provides additional empirical
evidence to a large body of work showing that solicitation increases the likelihood of making
a donation (Grant and Rudolph, 2002; Graf, 2006b,a; Brady et al., 1999; Lewis and Arkedis,
2014; Cummings and Cummings, 2004; Burton et al., 2015). Political fundraisers may in-
crease the salience of political issues for individuals, thus increasing the degree to which one
cares about an issue, the satisfaction they might receive from acting positively towards that
issue, and thus increasing the chances that they will make a political contribution to advance
that issue. On a basic level, being asked to give money generates a binary opportunity for
contributing where there was none before. Without being asked to give, people will generally
57
not willingly make contributions because there is a cost without a tangible reward. Making
a political contribution is similar to making a charitable contribution in that it provides a
sense of gratication (Verba et al., 1995) or a positive feeling (Andreoni, 1990). By being
asked to give money to support a political cause, the costs of giving money are reduced
through greater issue salience, forcing an action (to give or not to give), and social pressure.
People generally want to help causes they believe in, but do not do so because of the cost
and intangible rewards from giving. Solicitation reduces those costs and may increase the
rewards from giving.
In Figure 2.1, a line graph depicts the percentage of people in the CCES (2006 - 2016)
that replied they have made a political contribution spontaneously or after being solicited.
The vast majority of contributors give after being solicited, not spontaneously. The percent-
age range for contributing survey respondents that were solicited ranges from 83% to 91%,
depending on the year.
58
0.25
0.50
0.75
2006 2008 2010 2012 2014 2016
colour Solicited Spontaneous
Percentage of Spontaneous or Solicited Donors, CCES 2006 − 2016
Figure 2.1: Line graph displaying the percentage of individuals in the CCES who contributed
after being solicited and without being solicited (spontaneously).
While the vast majority of donors have been solicited, the majority of those solicited do
not donate. This is representative of the fact that not many people contribute at all. In
2016, 23% of individuals surveyed in the CCES made a political contribution. Among those
59
solicited, 36.3% contributed. Taken together, it is evident that solicitation is a very strong
factor in individual contributions, signicantly increasing contribution likelihood.
0.2
0.4
0.6
0.8
2006 2008 2010 2012 2014 2016
colour donor nondonor
Percentage of Donors vs Non−Donors, CCES 2006 − 2016
Figure 2.2: Line graph displaying the percentage of individuals in the CCES who either
donated or did not donate.
Figure 2.3 shows the percentage of donors among solicited individuals, which is a help-
ful comparison to Figure 2.1. Together, these graphs show that even though most people
60
don't contribute, the vast majority of people who do contribute were solicited. However,
most people who are solicited do not contribute. Specically, among all respondents, 23%
reported contributing in 2016, while among all respondents who were solicited, 36% reported
contributing. While solicitation does not equate to contributing, it signicantly increases the
likelihood that one will contribute.
61
0.3
0.4
0.5
0.6
0.7
2006 2008 2010 2012 2014 2016
colour donor nondonor
Percentage of Donors Among Solicited Individuals, CCES 2006−2016
Figure 2.3: Line graph displaying the percentage of individuals in the CCES who were
solicited and either contributed or did not contribute.
62
2.4 Research Design
While there are many research questions that can and should be asked regarding small
donor contribution decisions, ve questions are isolated that may be adequately answered
using publicly available survey data. Answering these questions provides general insights
into who is more likely to make political contributions, and why.
2.4.1 Research Questions
• RQ 1) Which background traits or variables are the most associated with making a
political contribution?
• RQ 1a) Do dierent variables impact various types of contributions dierently? That
is, do the same variables associated with making a contribution to a candidate explain
making a contribution to a party or other political group?
• RQ 2) How important is solicitation/contact to the decision to contribute?
• RQ 3) Do any of the attributes that explain contributions become more or less salient
over time?
• RQ 4) Which type of contact is most eective at producing a donation?
2.4.2 Hypotheses
• H1) The traits that are associated with contributions are divided into three categories:
(1) high impact factors, (2) moderate impact factors, and (3) low-and-no impact fac-
tors. Hypothesis 1 predicts that high impact factors are (1) income, (2) being con-
tacted by a campaign, (3) interest in politics, (4) frequency with which one follows
the news/current aairs, and (5) strength of partisanship/ideology. Moderate impact
63
factors are gender and ideology. Low-to-no impact factors are race, education, and
many, many more variables not discussed.
• H1a) Strength of ideology has a stronger eect on contributions to political groups,
and strength of partisanship has a stronger eect on contributions to political parties.
• H2) Solicitation/contact will serve as an excellent predictor of contribution decisions
and enhance the accuracy of any model it is added to.
• H3) No hypothesis is made regarding RQ 3. Rather, this is an exploratory question, but
the numbering of hypotheses is preserved so that hypothesis 4 corresponds to research
question 4.
• H4) In-person contact will have the largest eect on respondents' likelihood of con-
tributing.
2.4.3 Data
To test these hypotheses, observational survey data from the Cooperative Congressional
Election Study (CCES) and the American National Election Study (ANES) data from 2000
- 2018 are used. Specically, data included are the CCES from 2006, 2008, 2010, 2012, 2014,
2016, 2017, and 2018. From the ANES, 2012 and 2016 data are used.
The unfortunate part of conducting analyses across dierent datasets that span over 12
years is that the variables measured vary by year and survey, and some questions asked
in 2006 were not asked in 2008, or 2016. This makes using a uniform model of analy-
sis unfeasible. Not every year of CCES/ANES have asked the same questions, so some
regression models show blank results denoting that that variable was not recorded that
year. For example, the CCES did not ask respondents if they were contacted by candi-
dates/campaigns/parties/groups in 2008, 2017, and 2018. As a result, there are blanks in
the data that show in the regression tables.
64
2.5 Results
To test these hypotheses, multivariate logistic regression was conducted for each year of the
CCES data to examine the varying degrees to which each variable increases the probability
of an individual making a political contribution. The following variables are included in
the logit models: income, strength of aliation with the Democratic party (3-point scale),
strength of aliation with the Republican party (3-point scale), strength of liberal ideology
(2-point scale), strength of conservative ideology (2-point scale), the degree to which one is
interested in politics (only present for some years), the frequency with which one follows the
news (only present for some years), level of education, race, and gender. Complete regression
results are reported in Table 2.1.
Results show that in each model, income is an important explanatory factor for making
a political donation, with the eect being generally linear. The income variable is measured
as a scale, because survey respondents report their income as a range (as an example, peo-
ple may list their income as between $10,000 - $14,999). Each year does not have exactly
the same income brackets, but the eect is consistent: as income bracket increases, so too
does the likelihood to make a political donation.
4
Regarding gender, men are signicantly
more likely than women to make a donation, which other work has shown to be true. The
partisanship variables and ideology variables work similarly on 7-point and 5-point scales,
respectively. Lower values mean strong Democrats/liberals and higher values mean strong
Republicans/conservatives. Because of the nature of the scale, regression output would re-
ally only show that the relationship is statistically signicant, while any eects of strong
Republican or strong Democratic partisanship would be obscured in a simple partisanship
regression. To avoid this, variables denoting the strength of Democratic and Republican
partisanship were created on a 3-point scale
5
, and the strength of liberal and conservative
4
full results of income regression found in Table 2.2
5
1 = lean Democrat/Republican, 2 = Democrat/Republican, 3 = Strong Democrat/Republican
65
Table 2.1: CCES Multiple Regression Results: Eect of Political/Social Characteristics on
Donating to a candidate
Dependent variable:
donated
2006 2008 2010 2012 2014 2016 2017 2018
(1) (2) (3) (4) (5) (6) (7) (8)
Income 0.124
0.125
0.100
0.113
0.099
0.068
0.100
0.106
(0.007) (0.005) (0.004) (0.004) (0.005) (0.004) (0.008) (0.004)
Contact 0.913
1.190
1.368
1.116
1.411
(0.053) (0.039) (0.036) (0.033) (0.030)
Education 0.126
0.183
0.125
0.154
0.149
0.134
0.196
0.154
(0.015) (0.011) (0.009) (0.010) (0.010) (0.009) (0.017) (0.009)
Whiteness 0.301
0.024 0.103
0.129
0.111
0.140
0.278
0.352
(0.051) (0.041) (0.037) (0.035) (0.037) (0.033) (0.060) (0.032)
Female 0.392
0.108
0.127
0.130
0.332
0.191
0.051 0.216
(0.042) (0.034) (0.028) (0.027) (0.029) (0.026) (0.050) (0.026)
Political Interest 1.374
1.482
(0.053) (0.039)
News Consumption 1.161
1.076
0.784
0.970
1.056
0.680
(0.029) (0.025) (0.027) (0.026) (0.051) (0.018)
Democrat Strength 0.238
0.209
0.162
0.257
0.208
0.105
0.189
0.189
(0.024) (0.019) (0.016) (0.016) (0.016) (0.014) (0.027) (0.015)
Republican Strength 0.166
0.093
0.105
0.109
0.099
0.045
0.028 0.073
(0.026) (0.021) (0.017) (0.017) (0.018) (0.016) (0.032) (0.017)
Liberal Strength 0.221
0.318
0.307
0.335
0.311
0.516
0.421
0.358
(0.038) (0.030) (0.031) (0.027) (0.028) (0.024) (0.045) (0.022)
Conservative Strength 0.047 0.044 0.359
0.181
0.123
0.083
0.170
0.105
(0.041) (0.031) (0.028) (0.029) (0.029) (0.027) (0.052) (0.025)
Constant 7.653
6.678
7.973
7.622
6.491
6.995
7.504
5.421
(0.180) (0.128) (0.120) (0.106) (0.110) (0.109) (0.205) (0.091)
Observations 16,595 23,402 38,167 35,627 38,669 43,421 14,030 43,498
Log Likelihood 7,560.862 14,272.010 14,436.060 16,567.360 15,748.790 19,197.590 5,212.316 19,762.810
Akaike Inf. Crit. 15,143.72028,564.030 28,894.120 33,156.730 31,519.580 38,417.18010,444.63039,545.630
Note:
p<0.1;
p<0.05;
p<0.01
66
ideology on a 2-point scale.
6
Strength of Democratic partisanship consistently has a stronger
eect on propensity to make a political contribution than strength of Republican partisan-
ship. Additionally, having a liberal ideology is consistently associated with stronger eects
on contributing than having a conservative ideology. Interest in politics/government aairs
and the frequency with which people follow the news are also closely associated with mak-
ing a contribution. Education has a similar eect as income, although respondents at the
upper levels of education are not as likely to contribute as respondents at the upper levels
of income. In fact, respondents with a post-graduate degree were less likely to contribute
than those with a 4-year degree. Regarding demographic variables, being white is generally
associated with a signicantly positive eect on contributing, and being female is generally
associated with a negative eect on contributing.
When considering the specic eect sizes for each variable, it is important to discuss how
to interpret these results. Logistic regression results are reported in log odds, which can
be dicult to interpret because of the necessary conversion to probability. The most basic
interpretation is that higher values means higher probability of making a contribution. The
eect size for income is 0.124, meaning for each increase in income, the log odds of making
a contribution increase by 0.124, a probability which by itself corresponds to a probability
of 0.53. Being contacted/solicited has a coecient indicating a log odds increase of +1.411
above the constant, which would correspond to a probability of 0.8 by itself. However, it
is crucial to keep in mind that making a political contribution is a relatively uncommon
action, and the log odds for the constant/intercept is -6.995, which corresponds to a base
contribution probability of 0.00092. The coecients add together the sum of the log odds
and then convert the total into a probability.
For example, consider the year 2016. Someone in the fth income group (0.068*5), who
was contacted (1.411), is in the third education category (0.134*3), who is white (0.14),
and female (-0.191), with the second level of news consumption (0.970*2), who is a fairly
6
1 = liberal/conservative, 2 = very liberal/conservative
67
strong Democrat (0.105), would have a log odds of 4.013 + the constant (-6.995) = -2.848.
Converting this log odd into a probability, we get 0.054, which indicates a 5.4% probability
of making a political contribution. Generally this shows that even for people who have fairly
ripe conditions for contributing, contributing is still a relatively uncommon activity.
These results generally conrm hypothesis 1, with an exception. As predicted, the factors
that are associated with the largest eect sizes are income, contact, education, and interest
in news/politics. In a surprising nding, strength of Democratic partisanship and strength of
liberal ideology were found to have considerably stronger eects than strength of Republican
partisanship and strength of conservative ideology. This is explored more in depth in chapter
3.
Gender does have a sizable impact on contribution decisions, as women are considerably
less likely to make political contributions than men. Regarding expected low-impact factors,
race generally has small eect sizes on balance in that white respondents are more likely to
have made contributions. Similarly, education shows somewhat low eect sizes, although it
is similar to income in that it has multiple levels of education. Further regression analysis
of income and education are necessary to conrm any ndings in this multivariate model.
Because some variables have a large number of factors, several of them were isolated for
more in-depth analysis of eects and eect sizes. First, the eect of income was isolated by
conducting logistic regression and using each income level as the independent variable and
donation/contribution as the dependent variable.
68
Table 2.2: CCES Regression Results: Eect of Income on Donating to a Candidate
Dependent variable:
donated
2006 2008 2012 2016 2017 2018
(1) (2) (3) (4) (5) (6)
Income Rank 2 0.198 0.044 0.539
0.467
0.410
0.845
(0.169) (0.131) (0.091) (0.087) (0.166) (0.102)
Income Rank 3 0.284
0.280
0.624
0.558
0.458
0.926
(0.162) (0.123) (0.089) (0.083) (0.157) (0.099)
Income Rank 4 0.333
0.501
0.839
0.738
0.947
1.147
(0.148) (0.114) (0.087) (0.082) (0.154) (0.097)
Income Rank 5 0.597
0.563
1.152
0.947
1.250
1.362
(0.144) (0.110) (0.087) (0.082) (0.152) (0.097)
Income Rank 6 0.694
0.772
1.241
1.105
1.194
1.572
(0.133) (0.100) (0.086) (0.081) (0.153) (0.097)
Income Rank 7 0.952
0.896
1.322
1.156
1.343
1.562
(0.132) (0.100) (0.090) (0.083) (0.157) (0.099)
Income Rank 8 0.991
1.220
1.560
1.317
1.497
1.741
(0.133) (0.100) (0.088) (0.083) (0.155) (0.097)
Income Rank 9 1.201
1.281
1.650
1.372
1.616
1.838
(0.133) (0.102) (0.087) (0.081) (0.151) (0.096)
Income Rank 10 1.363
1.484
1.915
1.530
1.806
2.108
(0.133) (0.102) (0.089) (0.083) (0.156) (0.098)
Income Rank 11 1.597
1.665
2.083
1.722
2.077
2.199
(0.131) (0.100) (0.092) (0.084) (0.157) (0.099)
Income Rank 12 1.764
1.916
2.294
1.924
2.289
2.372
(0.132) (0.103) (0.098) (0.090) (0.164) (0.102)
Income Rank 13 1.916
2.193
2.537
1.900
2.352
2.584
(0.136) (0.107) (0.121) (0.111) (0.203) (0.118)
Income Rank 14 2.396
2.436
3.028
2.330
2.379
2.719
(0.134) (0.105) (0.149) (0.131) (0.249) (0.139)
Income Rank 15 2.754
2.347
1.865
2.791
(0.202) (0.177) (0.309) (0.190)
Income Rank 16 3.184
2.823
2.587
2.538
(0.248) (0.213) (0.345) (0.183)
Constant 2.306
1.994
2.440
2.389
2.971
2.845
(0.124) (0.090) (0.079) (0.073) (0.135) (0.089)
Observations 26,077 25,425 39,467 47,395 15,780 46,700
Log Likelihood 13,788.390 17,903.240 20,991.300 23,849.070 6,083.705 22,497.150
Akaike Inf. Crit. 27,604.780 35,834.480 42,014.600 47,730.140 12,199.410 45,026.290
Note:
p<0.1;
p<0.05;
p<0.01
69
The regression results for income show that as income increases, so too does its eect size
on contributing. At the highest levels of income, the eect size of income on contributing is
greater than 3, which means for the the richest category of individuals, the base probability
of contributing is 0.0179 (1.8%). This maximum eect size is more than double the eect
of any single variable in the multivariate regression model. These results conrm the major
eect income has on contributing.
In Table 2.3, the eects of education on contributing are isolated.
Table 2.3: CCES Regression Results: Eect of Education on Donating to a Candidate
Dependent variable:
donated
2006 2008 2012 2016 2017 2018
(1) (2) (3) (4) (5) (6)
No HS Degree 0.261
0.072 0.443
0.011 0.111 0.299
(0.085) (0.054) (0.061) (0.053) (0.103) (0.062)
HS Graduate 0.686
0.759
1.276
0.690
0.454
0.890
(0.087) (0.053) (0.060) (0.052) (0.102) (0.060)
Some College 0.838
0.762
1.334
0.602
0.819
0.842
(0.093) (0.067) (0.066) (0.057) (0.108) (0.065)
2-Yr. College Degree 0.936
1.278
1.646
1.011
1.121
1.340
(0.087) (0.055) (0.060) (0.052) (0.099) (0.059)
4-Yr. College Degree 1.453
1.895
2.101
1.436
1.635
1.764
(0.091) (0.062) (0.063) (0.054) (0.102) (0.061)
Post-Grad Degree 1.747
1.472
2.326
1.895
2.382
2.245
(0.082) (0.046) (0.056) (0.047) (0.090) (0.055)
Observations 30,160 27,021 45,018 52,899 18,200 51,808
Log Likelihood 16,244.430 19,164.490 23,742.090 26,758.160 7,039.374 25,296.170
Akaike Inf. Crit. 32,500.870 38,340.970 47,496.190 53,528.310 14,090.750 50,604.330
Note:
p<0.1;
p<0.05;
p<0.01
When isolated, higher levels of education also show considerably large eects, regularly
above 1.2, which is very signicant.
70
Factored regression was performed on various racial categories to determine if whiteness
does have a signicant eect on the likelihood of contributing, but there is no consistent
eect for racial groups. Results are displayed in Table 2.4.
Table 2.4: CCES Regression Results: Eect of Race on Donating to a Candidate
Dependent variable:
donated
2006 2008 2012 2016 2017 2018
(1) (2) (3) (4) (5) (6)
Black 0.230 0.118 0.549
0.217
0.028 0.048
(0.177) (0.121) (0.092) (0.079) (0.171) (0.072)
Hispanic 0.680
0.259
0.057 0.022
(0.176) (0.126) (0.100) (0.085)
Latino 0.210 0.006
(0.179) (0.076)
Middle Eastern 0.883
1.303
0.227 0.800
0.423 0.040
(0.321) (0.457) (0.280) (0.223) (0.508) (0.312)
Mixed 0.317 0.215 0.922
0.697
0.505
0.153
(0.212) (0.167) (0.112) (0.098) (0.199) (0.106)
Native American 0.323 0.439
0.715
0.804
0.075 0.339
(0.243) (0.179) (0.149) (0.132) (0.294) (0.148)
Other 0.549
0.669
1.373
1.394
1.055
0.654
(0.197) (0.151) (0.112) (0.108) (0.209) (0.130)
White 0.466
0.246
0.680
0.663
0.634
0.514
(0.172) (0.115) (0.085) (0.070) (0.158) (0.063)
Constant 1.556
0.998
1.783
1.850
2.212
1.683
(0.171) (0.114) (0.084) (0.069) (0.156) (0.062)
Observations 30,213 27,021 45,018 52,899 18,200 51,808
Log Likelihood 16,561.850 20,454.310 24,873.990 27,615.090 7,378.879 26,292.560
Akaike Inf. Crit. 33,139.710 40,924.630 49,763.990 55,246.190 14,773.760 52,601.120
Note:
p<0.1;
p<0.05;
p<0.01
These regression results are for the entire respondent pool for CCES. Because this study
seeks to explain why small donors give as separate from large donors, additional regression
models were run to delineate between respondents with high incomes and low incomes. To
71
delineate lower-income from higher-income respondents, the datasets were divided between
respondents with family income above the mean for all respondents. The logistic regression
results tell similar stories for respondents with income below the mean and respondents with
income above the mean. The main dierences seem to be that partisanship is less important
for the lower-income respondents, while ideology is more important. The coecients for
being a strong Democrat are weaker for higher-income respondents, while being a strong
liberal has a larger coecient for lower-income respondents. Being a strong conservative
was not a statistically signicant coecient in either model, but the coecient for being a
strong Republican was larger for lower-income respondents than higher-income respondents.
Other eects do exist, such as gender appearing to have a smaller coecient among higher-
income respondents, but eect is inconsistent across the years analyzed so general conclusions
shouldn't be drawn. These results are not reported here, but can be found in Tables 6.2 and
6.1 in the appendix.
This suggests that beyond the material and social benets extremely wealthy donors
may receive from their massive contributions, similar demographic, social, and political
characteristics in
uence both groups: interest in politics, partisanship, being contacted by
a campaign, and having money. However, it would be unwise to extrapolate too far from
survey results, as there is a massive substantive dierence between a donor making $250,000
per year and a donor making $25,000,000 a year.
In addition to analyzing data from the CCES, analysis of the ANES is included for 2016
and 2012. Not all of the variables in the CCES data are available in the ANES data, but the
features selected were as similar as possible. For the ANES, results are generally replicated
with a few surprises. Income is still an important variable for the 2012 results, although for
2016 income had a very small increase in the likelihood of making a contribution in and was
not statistically signicant. This is a curious nding, and because every model shows income
being highly related to contributing, no explanation is oered for this eect. Being contacted
by a party had a large eect size, as well as political interest, campaign interest, and liberal
72
ideology (for 2016 only, however). Whiteness had a negative eect in the ANES data, which
is a reversal from the CCES data. Strength of Democratic partisanship and Republican
partisanship had roughly similar eects, taking into account both 2012 and 2016. Liberal
ideology had a large and strong positive impact on contribution likelihood but only for 2016.
Conservative ideology had a large negative eect in 2012 and a small positive eect in 2016.
Overall, the ANES generally replicates ndings in the CCES data, which suggests the core
demographic, political and social characteristics related to contribution likelihood are robust.
One benet of the ANES data is that 2016 has more ne-grained data about who subjects
contributed to: candidates, parties, or groups. The regression models in Figure 2.5 are
replicated on four models corresponding to four dependent variables: donating to anyone,
donating to a political party, donating to a candidate, and donating to a political group.
The model for donating to anyone is the same as the model reported above, but is repeated
here for easy comparison across all dependent variables.
The main multivariate regression results for ANES 2016 data reported in Figure 2.6
generally conrm hypotheses supported by the CCES data, with some exceptions. In this
regression, income for 2016 is not generally shown to have that signicant of an eect on
any category except contributing to political groups, as the direction of eect is small and
not signicant for most ranks of income. In fact, only at the nal 5 ranks of income is the
variable associated with a positive eect on contributing to any political group, and only
the very highest income rank has both statistical signicance and a large eect size. This
casts some doubt on just how important income is to contribution decisions. Rather, the
largest eect sizes from the ANES data are associated with being contacted by a party,
strong Democrat/Republican aliation, liberal ideology, political interest, and following
campaigns/the news regularly. Perhaps surprisingly, the 2016 ANES data support the no-
tion of a solicitation-driven model of political contributions, as being solicited was one of
the largest factors in every contribution decision, from candidates to parties (although not
groups, which is unsurprising). It is important to note that strength of Democratic par-
73
Table 2.5: ANES Logistic Regression Results: Eect of Background Traits on Donating to
Any Political Entity, 2012 - 2016
Dependent variable:
Donated
2012 2016
(1) (2)
income 0.078
(0.014) 0.004 (0.007)
contacted by party 0.877
(0.110) 0.648
(0.115)
education 0.051
(0.025) 0.108
(0.027)
whiteness 0.304
(0.125) 0.098 (0.140)
female 0.142 (0.107) 0.224
(0.116)
political interest 0.466
(0.073) 0.408
(0.091)
campaign interest 0.621
(0.109) 0.733
(0.133)
news followship 0.046
(0.022) 0.210
(0.050)
Democratic strength 0.219
(0.058) 0.153
(0.068)
Republican strength 0.113
(0.062) 0.182
(0.080)
liberal 0.027 (0.062) 0.399
(0.075)
conservative 0.366
(0.147) 0.028 (0.090)
Constant 5.671
(0.343) 7.732
(0.490)
Observations 3,461 2,807
Log Likelihood 1,117.237 1,025.783
Akaike Inf. Crit. 2,260.475 2,077.566
Note:
p<0.1;
p<0.05;
p<0.01
74
Table 2.6: ANES Regression Results: Eect of Background Traits on Donating to Various
Political Entities
Dependent variable:
Contrib. to Anyone Contrib. to Party Contrib. to Candidate Contrib. to Group
(1) (2) (3) (4)
Income 0.004 0.003 0.001 0.037
(0.007) (0.009) (0.008) (0.014)
Contacted by Party 0.648
0.534
0.694
0.252
(0.115) (0.143) (0.123) (0.196)
Education 0.108
0.053 0.087
0.107
(0.027) (0.033) (0.029) (0.048)
White 0.098 0.172 0.075 0.191
(0.140) (0.173) (0.151) (0.251)
Female 0.224
0.185 0.089 0.494
(0.116) (0.145) (0.125) (0.203)
Political Interest 0.408
0.298
0.443
0.643
(0.091) (0.113) (0.099) (0.170)
Campaign Interest 0.733
0.703
0.756
0.154
(0.133) (0.171) (0.148) (0.202)
Wekly News Consumption 0.210
0.146
0.214
0.187
(0.050) (0.062) (0.056) (0.086)
Democratic Strength 0.153
0.376
0.147
0.243
(0.068) (0.090) (0.073) (0.115)
Republican Strength 0.182
0.365
0.191
0.130
(0.080) (0.103) (0.087) (0.153)
Liberal Strength 0.399
0.138 0.351
0.431
(0.075) (0.096) (0.080) (0.124)
Conservative Strength 0.028 0.071 0.006 0.057
(0.090) (0.111) (0.098) (0.175)
Constant 7.732
7.213
7.943
8.055
(0.490) (0.601) (0.542) (0.779)
Observations 2,807 2,808 2,809 2,810
Log Likelihood 1,025.783 747.852 913.630 436.738
Akaike Inf. Crit. 2,077.566 1,521.704 1,853.260 899.476
Note:
p<0.1;
p<0.05;
p<0.01
75
tisanship and strength of Republican partisanship are near equal in the ANES, somewhat
weakening the trend suggested by the CCES data. In any case, liberal ideology holds up as
a much greater predictor of contributing than conservative ideology.
The results from the ANES appear to support hypothesis 1a, that partisanship is more
important for contributing to political parties, and ideology is more important for contribut-
ing to political groups. Democratic and Republican partisanship have similar but signicant
eects on propensity to contribute across entities: parties, candidates, and groups. This is
notable as it does not conrm ndings in the CCES data that show Democratic partisanship
impacts propensity to contribute more than Republican partisanship. Regarding contribut-
ing to a political group, Democratic strength, Republican strength, and liberal strength are
all signicant variables that increase likelihood to contribute, while conservative ideology
has no eect, substantive or statistical.
Because a core component of my theory of small dollar donations is solicitation, I use
a proxy variable of campaign contact to measure exposure to fundraising solicitations.
I note a strong caveat here that the form of contact is being approached by any cam-
paign/party/group, and that the questions asked in the CCES surveys do not pertain ex-
plicitly to fundraising solicitations, but rather any contact from a candidate/campaign. That
said, the results are fairly strong and shouldn't be outright dismissed for concerns of variable
measurement.
Now, I report regression analyses focusing on the years in which the CCES survey data
contains variables about the methods by which respondents were contacted by political par-
ties, groups, candidates, or organizations. 2010, 2012, 2014, and 2016 all contain information
about whether or not a respondent was contacted and the form of contact. Specically, the
forms of contact noted are in-person, by phone, via email/text, and via letter/postcard.
Figure 2.7 reports regression output which shows the eects of the four types of contact
on making a political contribution. Results do not support hypothesis 4, that in-person
76
Table 2.7: Logistic Regression: Contact Method
Dependent variable:
donated
2010 2012 2014 2016
(1) (2) (3) (4)
In-Person Contact 0.132
0.208
0.334
0.021
(0.032) (0.034) (0.036) (0.035)
Phone Contact 0.151
0.076
0.316
0.160
(0.034) (0.033) (0.036) (0.031)
Text/Email Contact 1.509
1.913
1.492
1.677
(0.028) (0.030) (0.031) (0.030)
Letter/Postcard Contact 0.130
0.132
0.197
0.125
(0.029) (0.029) (0.032) (0.028)
Constant 1.892
1.798
2.171
1.809
(0.038) (0.038) (0.040) (0.036)
Observations 34,252 31,708 25,952 28,571
Log Likelihood 15,043.810 15,741.250 13,249.620 15,898.180
Akaike Inf. Crit. 30,097.620 31,492.510 26,509.240 31,806.350
Note:
p<0.1;
p<0.05;
p<0.01
77
contact would be the most eective form of contact for producing a contribution. Regarding
in-person contact, there is actually a negative eect in 2012, a statistically signicant but
somewhat modest eect in 2010, a signicantly positive eect in 2014, and a statistically and
substantively insignicant eect for 2016. This is a curious nding considering Gerber and
Green (2000)'s research showing both the eectiveness of in-person canvassing on GOTV
eorts and the lack of eects for direct mail. The lack of eectiveness for in-person contact
may perhaps be explained by the measure used. The forms of contact only ask respondents if
they were generally contacted by a campaign or candidate, not necessarily if the respondents
were being solicited for money. Being contacted by phone on balance has small eects (-0.073
to 0.314) on contributing and being contacted by letter has consistently positive but small
eects on contributing (0.120 to 0.235). By contrast, email/text communication has very
large eect sizes, much greater than any other form of contact (1.532 to 1.844).
I speculate that perhaps because the process of one making a political donation generally
operates via the internet (clicking on a link to a donation page and typing in credit card
information), contact via this same digital method is more eective at producing a donation
than in-person or phone requests. Receiving an email request for money may be more likely
to produce a contribution than other forms of request because an email is extremely likely
to contain a link which one can click and be brought to a donation page that requires credit
card information, which may be stored automatically so that typing out the credit card
information isn't even necessary. In this way, the email/text format lowers the barrier to
making a contribution in a way that in-person, phone call, and letter solicitations do not.
It is also possible that contact via email/text may possibly simply ask for money at higher
rates than other forms of contact, although this supposition is untested.
Respondents had the option to list if they received more than 1 type of campaign contact,
allowing the construction of a variable called \Contact Count" denoting how many forms of
contact a person received. Regression analysis was performed to learn the eects of multiple
forms of contact on likelihood of contributing. The Contact Count feature shows that re-
78
peated solicitations being solicited via multiple forms of communication has a considerably
strong impact on propensity to donate. I include the Contact Count variable analysis as a
factor, showing the eect of each additional form of contact, from 1 to 4, with zero contacts
being the baseline value.
Table 2.8: Logistic Regression: Simple Contact Count Model
Dependent variable:
donated
2010 2012 2014 2016
(1) (2) (3) (4)
Any Form of Contact 1.053 1.618
0.961 2.694
(0.646) (0.411) (0.632) (0.376)
1 Form of Contact 0.212 0.339 0.082 1.434
(0.646) (0.411) (0.632) (0.376)
2 Forms of Contact 0.632 0.063 0.435 0.885
(0.645) (0.411) (0.632) (0.376)
3 Forms of Contact 1.353
0.851
1.266
0.269
(0.645) (0.411) (0.632) (0.376)
4 Forms of Contact 1.819
0.898
1.804
0.032
(0.646) (0.412) (0.633) (0.378)
Constant 2.735
2.531
2.572
2.450
(0.030) (0.029) (0.026) (0.023)
Observations 46,582 44,921 48,753 52,836
Log Likelihood 19,802.200 21,978.740 19,536.940 24,034.550
Akaike Inf. Crit. 39,616.390 43,969.490 39,085.870 48,081.110
Note:
p<0.1;
p<0.05;
p<0.01
As can be seen in the table, 1 form of contact is actually associated with slightly negative
values for 2012 and 2014, and a strongly negative value in 2016. Statistical signicance was
not achieved for 2 forms of contact. For 3 and 4 forms of contact, however, 2012 and 2014
show statistically signicant and substantively strong values. 2012 shows coecients of 1.165
79
for 3 forms of contact and 1.273 for 4 forms of contact. 2014 shows coecients of 1.235 for 3
forms of contact and 1.726 for 4 forms of contact. No statistical signicance was achieved for
2016. The basic conclusion is that email and text is the best way to raise money from people,
but contacting people via multiple methods is more likely to generate a contribution than
only contacting people via one method. In chapter 4, the eectiveness of anxiety-inducing
messages on propensity to donate is explored.
Overall, the CCES and ANES data generally support hypotheses 1, 1a, and 2. However,
hypothesis 1 has only mixed support, as I the predictions made about which variables would
have the highest impact on contribution likelihood were not entirely right. Not only should
strength of partisanship/ideology be considered as having a moderate impact on contribution
likelihood, Republican partisanship and conservative ideology did not have very strong eects
at all, and had much weaker relationships with the propensity to contribute. Over time, the
positive eect sizes of Republican partisanship actually decreased over time, starting at 0.166
in 2006 and ending at 0.073 in 2018. The eect sizes for Democratic partisanship diminished
slightly, beginning at 0.238 in 2006 and shrinking to 0.189. However, this decrease in eect
sizes asymmetrically favors Democratic partisanship as a relevant variable on contribution
decisions, as opposed to Republican partisanship. Education shouldn't be considered a low-
impact factor, but likely a moderate or high-impact factor. Both the ANES and CCES data
show education has large, positive impacts on contribution likelihood. Generally, the eect
of contact on contribution likelihood increased from 0.913
7
to 1.411 in 2016. Hypothesis
4 is rejected, as in-person contact was the method of contact least likely to generate a
contribution. By a large margin, text/email contact is the method of contact that most
increases contribution likelihood, and this eect is robust over the period of 2010 - 2016.
7
again, this is in log odds
80
2.6 Building a Predictive Model
In order to test the eectiveness of these variables at predicting donations, three models were
constructed for 2012, 2014, and 2016 to use variables to predict whether or not somebody
made a donation in the past year. Identical variables were used for 2012 and 2014, and
2016. Logistic regression was performed using donation as the dependent variable for these
purposes. A decision tree model was also tested but it consistently underperformed compared
to logistic regression so it was not used. Independent variables for 2012 and 2014 are:
Contact, Income, Education, Strength of Democratic Partisanship, Strength of Republican
Partisanship, and News Followship Frequency.
Contact is a binary variable indicating 1 if the respondent contributed in the past year or
0 if not.
8
Income refers to a 16 point scale corresponding to an income bracket where higher
numbers mean higher incomes.
9
Education refers to a 6-point scale indicating one's highest
level of education achieved.
10
. Regarding the partisanship variables, Democratic Strength
is a 3-point scale indicating 3 if a respondent is a strong Democrat and 1 for an individual
who leans Democrat; Republican Strength is a 3-point scale indicating 3 if a respondent is a
strong Republican and 1 if the respondent leans Republican. News Followship Frequency is a
4-point scale which indicates the degree to which one follows what's going on in government
and public aairs.
11
The models were trained using a standard 50/50 split between training data and test
data. The model works by using the variables to create a probability that somebody made
a donation, a value from 0-1. Any probability greater than or equal to 0.5 was predicted
8
1 = Strong Democrat, 2 = Not very strong Democrat, 3 = Lean Democrat, 4 = Independent, 5 = Lean
Republican, 6 = Not very strong Republican, 7 = Strong Republican
9
1 = less than $10k, 2 = $10-$20k, 3 = $20-$30k, 4 = $30-$40k, 5 = $40-$50k, 6 = $50-$60k, 7 = $60-
$70k, 8 = $70-$80k, 9 = $80-$100k, 10 = $100-$120k, 11 = $120-$150k, 12 = $150-$200k, 13 = $200-$250k,
14 = $250-$250k, 14 - $250-$350k, 15 = $350-$500k, 16 = $500k+
10
Education is a 6-point scale. 1 = No High School Degree, 2 = High school graduate; 3 = Some college,
4 = 2-year college
11
1 = Hardly at all, 2 = Only now and then, 3 = Some of the time, 4 = Most of the time
81
to be a a donation, and any probability below 0.5 was predicted to be not not a donation.
Accuracy, precision, and recall scores are reported for each model. Accuracy indicates what
percent of the the predictions made were correct, meaning the predicted result was actually
the result in the real test data. Precision refers to the ratio of correctly predicted positive
observations to the total predicted positive observations. In plain terms, that means of all
respondents the model predicts to have donated, how many actually donated? Recall refers
to the ratio of correctly predicted positive observations to all observations in that class. That
is, of all respondents who actually donated, how many did I predict donated?
Figure 2.4: Model Performance, CCES Data Predicting Donations
82
Four models are included for each year that data exists for: 2012, 2014, and 2016.
Each model includes the dependent variable, whether or not a respondent donated, and
a number of independent variables. The rst model for each year includes only income as
an independent variable; the second model includes only the binary contact variable; the
third model includes income, education, strength of Democratic partisanship, strength of
Republican partisanship, and news followship. The fourth model includes all of the variables
in the third model plus contact.
From these results, it is generally conrmed that as a standalone variable, the contact
variable is slightly worse at predicting contribution decisions than income, which is widely
regarded as the most important variable for contributing (Verba et al., 1995). However, con-
tact was found to improve the F1 score of each model it was added to, supporting hypothesis
2.
2.7 Discussion and Conclusion
The hypotheses are reviewed and noted if they are supported, unsupported, or rejected.
• Hypothesis 1: Generally supported. Variables with largest eect sizes on contri-
bution likelihood are contact, income, education, political interest/news consump-
tion/campaign interest, strength of Democratic partisanship, and strength of liberal
ideology.
• Hypothesis 1a: Supported. Regression models show that strength of Democratic and
Republican partisanship are more strongly associated with contributions to parties, and
strength of liberal ideology is more strongly associated with contributions to groups.
However, strength of Republican partisanship generally had lower eects on contribu-
tion likelihood than Democratic partisanship. Conservative ideology had no statisti-
cally signicant eects.
83
• Hypothesis 2: Supported. Contact/solicitation is one of the variables with the largest
eect on contribution odds, although it is shown to not have quite as large of an eect
as income. Despite this, a contacted individual with lower income is as likely to make
a contribution as an uncontacted individual with considerably higher income.
• Hypothesis 3: No hypothesis was made about research question 3, which asked if any
attributes became more or less salient over time. The most notable variables that
changed over time are Republican partisanship, a variable that decreased in relevance
over time. The eects of contact on propensity to contribute increased over time, from
0.913 in 2006 to 1.411 in 2016. The eects of strength of liberal ideology also increased
slightly over time, from 0.221 in 2006 to 0.358 in 2018.
• Hypothesis 4: Rejected. In-person solicitation is the form of contact least likely to
increase odds of contributing. The eect sizes of text/email contact are generally 5 to
15 times larger than other forms of contact.
Overall, the various regression results and descriptive analyses here provide evidence
to support past ndings about why people make political contributions, with two valuable
additions. The most impactful factors that in
uence individuals' contribution decisions
are contact, income, education, political interest/campaign interest/news consumption, and
Democratic partisanship/liberal ideology. The additions made to our understanding of small
donor behavior with this research is the importance of contact/solicitation, Democratic par-
tisanship, and liberal ideology, which have proven to be very important variables in the
pathway towards making a political contribution.
Additionally, being contacted by a campaign is not a simple variable, as there are a
variety of additional sub-variables within the eld of contact. Contact can include form of
contact, whether it be in person, via phone, email, etc. The number of contacts/solicitations
is also important, in addition to the match between solicitee and solicitor. Finally, the actual
message portrayed in that contact and the quality of the message are important.
84
In this empirical chapter, past empirical ndings were re-examined, which revealed several
interesting ndings. Contact was shown to be an incredibly impactful variable on propen-
sity to contribute, the eects of which only increased over time. One interesting nding
was the decline in the importance of Republican partisanship on contribution decisions over
time. The next chapter essentially serves as an extended discussion of this chapter, explor-
ing the dierences in contributing among the parties and ideologies and seeking to explain
them. Chapter 3 also explores dierences in contact among the partisan/ideological groups
to determine if asymmetric partisan contact helps explain lower contribution rates among
Republicans and conservatives.
85
Chapter 3
Beyond the Resource Model
3.1 Abstract
Building upon the ndings of the last chapter, this chapter explores the partisan and ideo-
logical dynamics of small-dollar political contributions. Results show that Democrats began
to contribute at a higher rate than Republicans and conservatives starting after 2010. The
intersection of partisanship, ideology, and campaign contact are explored to provide an ex-
planation for this asymmetric partisan/ideological giving. The simplest conclusion to draw
is that over time, many conservatives and Republicans stopped being asked to give while
liberals and Democrats continued to be contacted by campaigns. Recent data from for Re-
sponsive Politics (2021) supports this, showing that the GOP focused on large individual
donors. Democratic candidates raised on average 33% of their funds from small donors,
while Republicans raised on average 17% of their funds from small donors. Independents
and libertarians raised 23% and 25% from small donors, respectively. Democrats currently
are better able to fundraise from small donors than Republicans are.
86
3.2 The Importance of Democratic Partisanship, Lib-
eral Ideology, and Contact
The regression analyses in the previous chapter showed that contact, Democratic partisan-
ship, and liberal ideology are all impactful variables that increase contribution likelihood.
However, the regression analyses reported in chapter 2 raise a number of additional questions.
Namely, why is Democratic partisanship and liberal ideology more predictive of contributing
than Republican partisanship and conservative ideology? Attempts to answer this question
are provided through a series of descriptive statistics, visualizations, and regression analyses
based on CCES (Cooperative Congressional Election Campaign) data for 2006, 2007, 2008,
2010, 2012, 2014, 2016, 2017, and 2018.
First, descriptive statistics describing the distribution of contributors grouped by parti-
sanship and ideology are reported. Figure 3.1 displays the distribution of contributors by
partisanship over time. Specically, it shows the proportion of donors among each partisan
group. Condence intervals are not included in this gure or others due to the number of
lines making the graph more dicult to interpret, but 95% condence intervals for Figures
3.1 and 3.2 are all within 3% of the value reported.
Clearly visible in Figure 3.1 is a decline in the proportion of donors among Republicans.
While the distribution of contributors among all partisan groups dropped from 2006 to 2018,
the largest drops are among Republicans. For example, In 2006 39% of strong Democrats
were contributors, which dropped to 34.6% in 2018. By contrast, in 2006 37.6% of strong
Republicans were contributors, which dropped to 24.2% in 2018. Notably, the largest drop for
any partisan group was the fall in the proportion of contributors among strong Republicans.
Similar statistics are reported for ideological groups in Figure 3.2.
The same eect is shown in Figure 3.2, where the proportion of donors among conser-
vative individuals saw a large drop over time. While the proportion of donors among very
87
0.1
0.2
0.3
0.4
0.5
2006 2007 2008 2010 2012 2014 2016 2018
Distribution of Donors
colour
Democrats
Independent
Lean Democrat
Lean Republican
Republican
Strong Democrats
Strong Republican
Distribution of Contributors by Party ID, 2006 − 2018
Figure 3.1: Line graph depicting the distribution of contributors by party ID between 2006
- 2018 according to CCES data.
88
0.2
0.3
0.4
0.5
2006 2007 2008 2010 2012 2014 2016 2018
Percent of PID Groups that Contributed
colour Conservative Liberal Moderate Very Conservative Very Liberal
Donation Rates by Ideology, 2006 − 2018
Figure 3.2: Line graph depicting the distribution of contributors by ideology between 2006
- 2018 according to CCES data.
89
liberal individuals actually increased from 39.1% in 2006 to 39.7% in 2018, contributing
among very conservative individuals and conservatives dropped. In 2006, 35.1% of very con-
servative people and 30.5% of conservatives contributed, while in 2018 only 23.7% of very
conservative people and 19.9% of conservatives contributed. These gures clearly show the
drop in contributions among Republicans and conservatives. Additionally, it's apparent that
strong Democrats and very liberal individuals are the top partisan/ideological groups for
contributing by a signicant degree. Seeking an explanation for this asymmetric partisan
giving, dierences in contact by partisanship and ideology are explored in the next section.
3.2.1 Asymmetrical Partisan and Ideological Contact Rates
Theoretically, it is very clear that being contacted by a campaign increases the chances one
will make a contribution. In this section, dierences in contact rates by partisanship and
ideology are explored to determine if dierences in contact among the parties/ideologies
may be driving the dierences in contribution rates. Using similar descriptive statistics and
visualizations, I report gures showing the distribution of contacted individuals by partisan
and ideological groups. Figure 3.3 shows a relatively similar result as the trends in Figures
3.1 and 3.2. The similarities are that the proportion of contacted individuals among strong
Republicans dropped from 2006 to 2016. In 2006, a higher proportion of strong Republicans
(79.7%) were contacted than strong Democrats (77.2%), but by 2016 a lower proportion of
strong Republicans (59.4%) were contacted than strong Democrats (62.0%). Still, results
show that a higher proportion of Republicans were contacted than Democrats, and a higher
proportion of leaning Republicans were contacted than leaning Democrats. This result is
supported by logistic regression analysis in Table 3.1, which shows that being a Republican
is even more strongly associated with the likelihood being contacted than being a Democrat.
Table 3.1 reports the results of a logistic regression analysis where the dependent variable
is being contacted by a campaign, and the independent variables are strength of Democratic
90
0.4
0.5
0.6
0.7
0.8
2006 2007 2010 2012 2014 2016
Distribution of Contacted Indivs
colour
Democrats
Independent
Lean Democrat
Lean Republican
Republican
Strong Democrats
Strong Republican
Distribution of Contacted Individuals by Party ID, 2006 − 2016
Figure 3.3: Distribution of contacted individuals by party ID, 2006 - 2016.
91
partisanship, strength of Republican partisanship, strength of liberal ideology, and strength
of conservative ideology. These regression analyses show that strength of Republican parti-
sanship was the partisanship/ideology feature most strongly associated with increasing the
likelihood of being contacted. This suggests that pure contact dierences between Democrats
and Republicans do not explain dierences in the proportion of contributors among each par-
tisan group.
Table 3.1: Contact DV Results
Dependent variable:
contact
CCES 2006 CCES 2010 CCES 2012 CCES 2014 CCES 2016
(1) (2) (3) (4) (5)
Democratic Strength 0.140
0.131
0.128
0.125
0.128
(0.016) (0.011) (0.011) (0.010) (0.010)
Republican Strength 0.157
0.183
0.188
0.201
0.141
(0.017) (0.013) (0.012) (0.011) (0.011)
Liberal Strength 0.065
0.130
0.164
0.164
0.145
(0.027) (0.023) (0.021) (0.020) (0.019)
Conservative Strength 0.096
0.270
0.089
0.107
0.046
(0.028) (0.022) (0.021) (0.019) (0.018)
Constant 0.667
0.248
0.230
0.200
0.146
(0.030) (0.023) (0.022) (0.021) (0.020)
Observations 26,930 43,747 40,863 43,781 48,730
Log Likelihood 15,442.260 24,842.460 27,700.170 30,174.970 33,361.150
Akaike Inf. Crit. 30,894.520 49,694.920 55,410.330 60,359.930 66,732.290
Note:
p<0.1;
p<0.05;
p<0.01
While contact rate among partisan/ideological groups is not enough to explain the dif-
ferences in contribution rates, the form of contact may have an eect.
92
3.2.2 Contact by Email or Text
Analysis from chapter 2 shows email/text contact increases the odds of contributions more
than any other form of contact, so it is worth exploring as an explanation for partisan
asymmetries in contribution rates. Figure 3.4 shows the dierences in the distribution of
email/text contact among partisan/ideological groups. This does show that the proportion
of Democrats who are contacted via email/text is higher than the proportion of Republi-
cans who are contacted via text. This eect is true for each level of partisanship: strong
Democrats are contacted via email/text at a higher rate than strong Republicans, Democrats
are contacted via email/text at a higher rate than Republicans, and leaning Democrats are
contacted via email/text at a higher rate than leaning Republicans. The same eect is shown
for ideological groups in Figure 3.5. The proportion of very liberal individuals and liberals
who were contacted via email/text were greater than the proportion of very conservative
individuals and conservatives who were contacted via email/text. This suggests that while
base contact rate among partisan/ideological groups is not driving the dierences in contri-
bution rates, it's possible that the partisan/ideological contribution gap is being driven by
dierences in contact rate via email/text. Especially when considering the results of chapter
2, which show how important contact is in contribution decisions, higher rates of contact via
email/text among Democrats/liberals provides a convincing explanation for the dierences
in contribution rates.
93
0.3
0.4
0.5
0.6
0.7
2010 2012 2014 2016
Distribution of Indivs Contacted via Email/Text
colour
Democrats
Independent
Lean Democrat
Lean Republican
Republican
Strong Democrats
Strong Republican
Distribution of Individuals Contacted via Email/Text by Party ID, 2010 − 2016
Figure 3.4: Distribution of Individuals Contacted via Email/Text by Party ID, 2010 - 2016
94
0.4
0.5
0.6
0.7
2010 2012 2014 2016
Distribution of Indivs Contacted via Email/Text
colour Conservative Liberal Moderate Very Conservative Very Liberal
Distribution of Individuals Contacted via Email/Text by Ideology, 2010 − 2016
Figure 3.5: Distribution of Individuals Contacted via Email/Text by Party ID, 2010 - 2016
3.2.3 Eect of party-matched solicitation
Using data from the 2016 ANES, it's possible to distinguish between the eectiveness of
being contacted by a Democrat or a Republican, depending on whether the respondent
95
is a Democrat or Republican. The eect of being contacted by the Democratic party or
Republican party (or both) on the log odds of contributing are reported in Table 3.2.
Table 3.2: Eect of Being Contacted by Democrat vs Republican Party on Donating
Dependent variable:
Contributing
Contacted by Dems 0.513
(0.162)
Contacted by Both 0.684
(0.404)
Contacted by Repubs 0.199
(0.192)
Constant 1.426
(0.125)
Observations 1,167
Log Likelihood 638.385
Akaike Inf. Crit. 1,284.770
Note:
p<0.1;
p<0.05;
p<0.01
Table 3.2 shows that being contacted by the Democratic party has an eect on increasing
the log odds to contribute by 0.513, while being contacted by the Republican party increases
the log odds of 0.199, but is not statistically signicant. The constant in this model is being
contacted by an `Other' party besides Democrats or Republicans. This might suggest that
in addition to Democrats contacting people at higher rates, their messages may also be more
eective at generating contributions. It's possible that because Democrats seem to have a
larger small donor fundraising apparatus and do it more, they've gotten better at eliciting
contributions from people than Republicans have.
Additionally, the model can be updated to include the eect of being contacted by
Democrats or Republicans among respondents who are Democrats or Republicans on the
likelihood of donating. These results show that Democrats contacting a Democrat increases
96
the log odds of contributing by 0.855, while Republicans contacting a Republican increases
the log odds of contributing by .109, a signicantly smaller eect. Additionally, the eect
is statistically signicant for Democrats, but not for Republicans. Adding log odds appro-
priately and converted into probabilities, a Democrat being contacted by Democrats has a
34.0% probability of contributing, while a Republican being contacted by Republicans has
a 23.6% probability of contributing. This is an interesting nding that suggests something
about Democratic solicitation is better at producing contributions from party members than
Republicans. While no speculation is made here, further research on the solicitation strate-
gies of Democrats vs. Republicans would be interesting and helpful.
Table 3.3: Eect of Being Contacted by Democrat vs Republican Party on Donating to
parties
Dependent variable:
Donated
Democrats Republicans
(1) (2)
Contacted by Dems 0.855
0.644
(0.226) (0.383)
Contacted by Both 0.825 0.823
(0.581) (0.669)
Contacted by Repubs 0.278 0.109
(0.394) (0.243)
Constant 1.518
1.228
(0.195) (0.178)
Observations 586 461
Log Likelihood 340.032 241.846
Akaike Inf. Crit. 688.064 491.692
Note:
p<0.1;
p<0.05;
p<0.01
97
3.3 Conclusion
The general conclusions to make from this chapter are that there is some asymmetric par-
tisan and ideological eect occurring that makes it more likely for Democrats and liberals
to contribute than their Republican and conservative counterparts. While I am not able to
oer a full explanation for this eect, the research in this chapter provides enough evidence
to suggest that dierences in partisan and ideological contact/solicitation partially explain
the asymmetry. The core nding is that strong liberals seem to be contacted via email/text
at a higher rate than other ideological subgroups. This may be the explanation behind the
core nding that Democrats and liberals contribute at higher rates than other Republicans
and conservatives. Because contact via email/text is the best form of contact for producing
contributions, it is reasonable that this is the reason Democratic partisanship and liberal
ideology have a stronger relationship with making political contributions. Additionally, re-
gression analysis shows that Democrats being contacted by Democrats has a much stronger
eect on increasing the odds of contributing than Republicans being contacted by Repub-
licans. Taken together, this suggests that either Democrats and liberals possess traits that
make them innately more likely to contribute, or that dierences in email/text contact rates
among the two parties explain the asymmetry in contributing.
Democrats and liberals have been contacted via email/text consistently since 2006, while
conservatives are typically contacted via email/text at lower rates. Additionally, the nal
regression analysis shows that being contacted by a democrat is generally more eective at
producing a contribution than being contacted by a republican. There are many possible
avenues of research to expand upon this work to better understand why Democrats and lib-
eral contribute more than Republicans and conservatives. To begin, research on dierences
in core values between Republicans and Democrats may provide an explanation - for exam-
ple, dierences in values around community or collective action may provide answers. The
data also suggest that online contact may be the starting point of an explanation. Further
98
research could study dierences in fundraising spending and sophistication between the par-
ties, the in
uence of Actblue among Democratic fundraising, and dierences in fundraising
methods/strategies. Chapter four studies the eect of sending anxious emails on fundraising
returns, and Democratic/Republican fundraising is directly compared.
99
Chapter 4
The Anxious Avoidance Model of
Contributor Behavior
4.1 Abstract
As argued in the previous chapters, the most common pathway for an individual to con-
tribute involves somebody who cares about politics being asked to give money by a political
organization. The qualities of these fundraising messages are of great importance, as they
are one of the most common interactions individuals have with campaigns. And perhaps
more than anything else, the main feature of modern political fundraising messages is that
they are designed to produce emotion in people so strong that it compels them to make a po-
litical contribution to a candidate or cause despite the virtual certainty that their individual
contribution will not, by itself, change the outcome of an election. Specically, some of the
most prevalent types of fundraising messages are those crafted to make people so anxious or
worried about the current or potential state of the world that they feel compelled to make
a contribution. Unfortunately, these messages also have the greatest ability to negatively
impact people's psyches and change political behavior and attitudes.
100
4.2 Introduction
No known research examines the eectiveness of using anxiety in campaign messaging to
conduct small dollar fundraising. In the largest recent work on small donor behavior, Magleby
et al. (2018) nd that central to contribution decisions are the candidates themselves and
the messages they send out, with negative and positive messages both being eective. In
2012, around 80% of Obama donors said Obama's personal story was very or somewhat
important in their reasoning for participating in the campaign. Similarly, in 2008, around
75% of McCain donors said one reason they contributed was \the opponent of the candidate I
contributed to would be a bad choice for the country" (Magleby et al., 2018, 345). This work
directly studies the specic messages sent by political campaigns to assess the eectiveness
of anxiety in campaign fundraising messages.
This chapter answers two questions: (1) how many worry-inducing fundraising messages
did political campaigns send in 2020? (2) Does overuse of worry-inducing messages have a
negative impact on fundraising returns?
The rst question is descriptive, so no hypotheses are made about it, but it is expected
that some campaigns will overuse worry-inducing fundraising messages. Regarding the sec-
ond question, I hypothesize campaigns that use too many fear or anxiety-invoking messages
in their fundraising emails will actually have lower total fundraising returns because they
either do not appear credible or generate so much anxiety in their constituents that they
start to avoid campaign communication. The overuse of doomsday-style messaging may cre-
ate a \the sky is falling" eect, where individuals will see overly extreme messages as either
unbelievable or anxiety-inducing, and begin to avoid them as a result. While the intensity
of anxiety is not measured in this work, extreme messages are more likely to be perceived
by recipients as unconvincing when less extreme messages are also received (Fisher, 1988,
101
1985). Dierent methods of email classication could examine the intensity of emotion in
emails.
Fisher (1988) discusses the concept of narrative rationality in storytelling, in which nar-
rative probability and delity are the considerations for judging the merits of stories. If we
consider fundraising emails to be stories told to people to make them have such an emo-
tional reaction that they contribute money, then the fundraising messages are narratives
with probability and delity that impact their acceptance. Narrative probability refers to
how likely the story seems: \whether or not a story coheres or `hangs together,' whether
or not the story is free of contradictions." Narrative delity refers to the truthfulness of
the story, `the degree to which it accords with the logic of good reasons: the soundness of
its reasoning..." Both of these features are impacted by human cognition, and factors such
as partisanship and pre-existing opinions can create rationalizations and reasons to believe
things that others reject. Additionally, research shows that repeated use of negative mes-
saging reduces source credibility (Ernst et al., 2017), although candidates already seen as
credible may use negative messages without losing credibility (O'cass, 2002).
This hypothesis for sending anxiety inducing emails is that at some threshold, recipients
will start avoiding the source of anxiety (the emails) and stop engaging with the campaign.
The expectation on a macro level is that campaigns that overuse worry-inducing messages will
receive lower total fundraising returns, controlling for other variables. The theory of anxious
avoidance suggests that people who make political contributions to avoid or alleviate negative
feelings will shift and begin to avoid/alleviate the negative feelings invoked by fundraising
messages by ignoring the messages themselves.
To test this theory, individual contribution data and committee expenditures data was
collected from the FEC, and fundraising emails for several hundred (N = 382) candidates
running for various federal oces were collected. Then, each candidate's email messages were
classied to determine if they are anxiety-inducing, and linear regression was performed to
102
determine if anxious emails have a positive or negative relationship with fundraising totals.
Control variables were included to account for some of the many other factors that go into
fundraising returns.
4.3 Modern Fundraising Emails: Anxious, Apocalpytic,
and Frantic
Fundraising emails from political candidates and groups frequently sound apocalyptic, fran-
tic, and desperate. An Atlantic article from Wagner (2012) pokes fun of the emails by linking
the words of a commenter: \Those emails...every single one of them made it sound as if me
and my suggested donation of $5 was humanity's last, best hope..."
A Bloomberg Businessweek analysis from Green (2012) describe the Obama fundraising
team's emails as \...strange, incessant, and weirdly overfamiliar." Yet it is exactly these
type of emails that performed well among potential campaign contributors. The Obama
campaign itself reported that some of the subject lines of the emails with the highest returns
are:
1. I will be outspent
2. Some scary numbers
3. Last call: Join Michelle and me
4. Would love to meet you
Of course, in 2020 to describe these emails as \strange" would be a stretch, because
modern fundraising emails from political candidates look very similar to these. Modern
fundraising emails are rife with overly familiar language, use potential contributors' names
and other identifying information such as city/zip code, are eusively personally thankful,
and stress a variety of real-world problems as well as fundraising deadlines. Campaigns also
103
regularly entice people to contribute by oering chances to meet candidates, membership
cards, campaign merchandise/apparel, buttons and memorabilia, etc.
For example, this email from Kamala Harris makes it sound like she and the recipient
have a rich history of working together in California politics, when this is almost certainly
untrue.s
Besides this example, candidates regularly send emails where they \won't sugarcoat
things", have \huge news!", would \love to give you a call, [name]", they \wouldn't be
asking if it weren't important", they are \counting on people like you", ask if you'll be \one
out of 126 needed people from [city/state/zipcode] to contribute", and make many other
requests in such conversational, familiar ways. In an eort to break through to people with
natural language, political fundraising emails from any candidate generally tend to use very
similar language.
Despite some of the Obama team's top fundraising emails being fairly worry-inducing
or relying on giveaways, some of the subject lines are undeniably positive, such as "If you
believe in what we're doing..." and "Thankful every day".
1
In fact, according to the team itself, the campaign writers and analysts were bad at
predicting which emails would have large returns. Amelia Showalter, director of digital
analytics for the Obama 2012 campaign, said in an interview with Bloomberg Businessweek,
\We were so bad at predicting what would win that it only reinforced the need to constantly
keep testing... Every time something really ugly won, it would shock me: giant-size fonts for
links, plain-text links vs. pretty `Donate' buttons. Eventually we got to thinking, 'How could
we make things even less attractive?' That's how we arrived at the ugly yellow highlighting
on the sections we wanted to draw people's eye to" (Green, 2012). The team generally found
that writing a shocking or surprising subject line like `Hell yeah, I like Obamacare' would
1
4 of the top 12 fundraising emails include mention of "Michelle", and another one is simply "The
most popular Obama", which may be referring to Michelle. That's 1/3 of Obama's top fundraising emails
leveraging Michelle's popularity, rather than Barack's.
104
Figure 4.1: This is an image sent from the Kamala Harris campaign for President, sent on
May 4, 2019
105
Figure 4.2: An email notication from Joe Biden, sent by the DNC HQ
get a lot of clicks, but only for a little while. Every novel email style had a shelf-life and the
team would have to come up with something new."
Perhaps one of the most surprising results of the 2012 Obama digital fundraising team
was that there were seemingly no negative consequences to
ooding people's inboxes with
campaign emails. Toby Fallsgra, the campaign's email director, said \At the (end of the
campaign), we had 18 or 20 writers going at this stu for as many hours a day as they could
stay awake... The data didn't show any negative consequences to sending more." Fallsgra
ending by saying \We do know that getting all those e-mails in your inbox is at least mildly
irritating to some people. Even my father would point that out to me."
What becomes apparent after reading through thousands of fundraising emails from
many candidates and fundraising groups is that email fundraisers often do not use a specic
rhetorical strategy that they think will work best. Instead, they use every possible strategy,
basically throwing a variety of emails at the proverbial wall of member inboxes and seeing
what will stick. Even though the prediction made in this chapter is that sending too many
fear or anxiety-inducing emails will actually hurt fundraising goals, it is possible that sending
a great amount of anxiety-inducing emails has no net negative consequences for campaigns.
Because this question has not been answered through scientic inquiry, it is precisely this
nature of digital fundraising that is investigated. The specic research questions investigated
are:
(1) What percentage of fundraising emails sent from political candidates are anxiety-inducing,
or invoke anxiety and fear to generate contributions? Do certain candidates or types of can-
didates try to invoke more worry than others?
106
(2) Does overuse of worry-inducing fundraising emails work to generate larger fundraising
totals, or does it suppress contributions?
It is important to clarify here that all invocation of fear, anxiety, or anger is not necessarily
a manipulative or normatively bad action by a campaign or political group. This dissertation
was written during the global COVID-19 pandemic, a time when many Americans doubt
the severity and even the existence of the pandemic (Allington et al., 2020; Imho and
Lamberty, 2020). As a result, it is clear that there are potential societal benets to be gained
by inducing anxiety in the public to change their behavior. Yet even outside the context
of a pandemic, it is clear that invoking fear, anxiety, or anger is not always bad. People
hold strong political beliefs, and contributing money to defeat a candidate or policy that
would have negative impacts on people's lives is not only reasonable behavior, it is desirable
in a well-functioning democracy. In fact, there is reason to believe that political fears lead
to more knowledgable, trusting citizens who seek out more information (McDermott, 2004;
Marcus et al., 2008). Quoted in Albertson and Gadarian (2015), Garsten (2009, 196) write,
\Deliberation and judgement therefore seem to emerge not in sedate citizens... but instead
in citizens who have been disturbed out of their calm and made attentive by sharp feelings
of anxiety. Partiality and passion together, in the form of anxiety, can prod re
ection."
Indeed, invoking fear or anxiety in fundraising emails or otherwise is not a wholly bad
thing for a well-functioning democratic society. Quite to the contrary, it is desirable in part.
However, large amounts of anxiety and fear in politics can lead to initially invisible negative
consequences for both the people receiving these emails as well as the campaign sta sending
them. Regarding individuals on the receiving end of worry-inducing emails, people will start
to engage in active evasion of the content from these emails. Specically, the very reason that
people pay attention to political fundraising messages and contribute is to alleviate anxiety
or other negative emotions by taking action. However, after contributing or even considering
contributing enough times, people will begin to appreciate that their contribution is simply
not enough to produce the changes that they desire and will instead begin to ignore or avoid
107
the anxiety-inducing messages. As a result, campaigns that overuse worry-inducing emails
will receive lower fundraising returns than campaigns that do not overuse worry-inducing
emails. Simply put, inducing too much anxiety in people will cause them not to alleviate
anxiety by engaging more, but to alleviate anxiety by engaging less.
4.4 Theory: Fundraising Through Avoidance of Nega-
tive Emotions Leads to Avoidance of Contributing
The theory of anxious avoidance explicitly suggests that campaigns that overuse emails which
invoke fear and anxiety to compel action will see diminishing fundraising returns.
2
It is noted here that campaign emails that highlight real-world worry-inducing events may
be engaging in eective and substantively meaningful political communication with potential
voters/contributors. People want to give their money to help elect the right candidates win
and stop bad outcomes from occurring. However, I predict there is some individual threshold
at which overuse of these types of emails, whether or not they are based in reality, will cease to
be eective because people will realize their individual actions are not enough to produce the
changes they seek and the only solution is to escape the sources of anxiety: the fundraising
emails themselves. However, on longer time scales and across many campaigns, the eect
will be visible. This eect may be observed by determining the anxiety or worry-inducing
nature of each email a political campaign sends out and comparing fundraising returns while
controlling for variables like advertising spending, press coverage, and name ID. All other
things being equal, I hypothesize campaigns that send out \too many" anxiety-inducing
emails will see diminishing fundraising returns as their donor population will begin to tune
out or avoid the campaign emails.
2
It is expected that such an eect would be moderated by the level of real worry-inducing events occurring
in the world that the campaign fundraising emails can draw from, but this is beyond the scope of this project
108
Despite accepting that inducing worry is a part of political communication that often
has meaning, it is normatively problematic for campaigns to intentionally generate worry
through hyperbole, exaggeration, discussing events out of context, or by repeatedly call-
ing upon the same events to get more \mileage" out of a worry-inducing real-world event.
Essentially, it is harmful for people's psychological well-being and political attitudes to be
overexposed to campaign emails that overuse worry-inducing rhetoric in their fundraising
messages, whether the worry is appropriate or not. While sending truthful anxiety-inducing
messages is preferable to hyperbolic anxiety-inducing messages, it's possible the same eects
may occur. I theorize that every individual has a threshold of anxiety that, when crossed,
they will change their behavior to avoid the source of anxiety.
At a core level, the theory outlined is well articulated by the following sequence of phe-
nomena.
1. A political organization, group, or campaign observes inciting real-world events.
2. The political organization develops a strategy or plan of action to respond to the
inciting events.
3. The political organization sends out communication to its members (typically the mem-
bers of an email list) to generate certain reactions, most commonly contributing money,
signing a petition, taking a certain position on an issue, or voting a certain way.
4. Individuals respond to the communication (typically emails) in a variety of ways, but
they initially seek to soothe the anxiety by taking action. Anxiety-reducing behaviors
include getting more involved, seeking more information, forming strong opinions, dis-
cussing an issue with others, and making a contribution are all possible ways to soothe
anxiety.
5. Individuals continue to receive and tolerate anxiety-inducing political communica-
tion/information after taking action.
109
6. At some threshold, individuals become unwilling or unable to continue their soothing
behavior of seeking additional information, contributing more, discussing the issue with
others, volunteering, etc.
7. At some threshold, individuals begin to be made overly anxious by receiving more
anxiety-inducing political information. Realizing that they cannot stop the problem
themselves and are unable or unwilling to devote more personal resources to solving
the political problem, they begin to avoid the source of anxiety: political informa-
tion/communication. In practice, this may be the choice to stop reading emails, im-
mediately delete emails, set up a rule to sort the emails into a designated folder/inbox,
or even unsubscribe completely.
8. At some time horizon, people become open to receiving more anxiety-inducing political
communication.
9. A new inciting real-world event begins and the process starts anew.
While it would not be possible for the scope of this project to investigate the veracity
of each of these proposed phenomena, if the theory of anxious avoidance is even partially
correct, these theorized phenomena will have a number of observable consequences. In
this chapter, I will primarily investigate the potential consequence of these phenomena to
campaigns. As a campaign overuses worry-inducing messages, it is expected that they will
receive fewer contributions by causing people to enter the state of anxious avoidance faster
than campaigns that do not incite as much anxiety. Essentially, if you tell people the sky
is falling and they need $10 by midnight from you to prop it up, they might donate. If you
tell people the sky is falling and need $10 by midnight every week for you to prop it up,
they might question if the sky is really falling at all, and if they can really trust you. Or,
if they look up and think the sky really is falling, they may become unconvinced that their
contribution is really going to save the sky and rely on other people with more money to
contribute to the sky protection candidate.
110
The same mechanism by which people avoid anxiety may cause people to stop paying
attention to these campaign emails because they are a source of anxiety, and people know
their contributions will not solve the issue because they do not have enough to give. In prac-
tice, avoidance from campaign communication may be blocking accounts, marking political
emails as spam, or just automatically deleting them upon arrival. In this way, contributing
can be considered an ephemeral anxiety-soothing mechanism that is eventually replaced by
disengaging from the campaign.
On using anxiety-inducing fundraising appeals, Albertson and Gadarian (2015) write,
\...repeated use of threatening appeals may lessen the eectiveness, both because the
public begins to tune out warnings for threats that may not occur (i.e., the Chicken Little
phenomenon) and because the anxious appeal becomes associated with one political party.
When the sky does not fall or when anxiety seems overtly partisan, sections of the public
may refuse to become anxious. During times of crisis, though, anxiety can lead citizens to
set aside partisanship and their own policy predispositions and follow political leaders who
they normally would not, thus making an anxious politics very appealing to political elites."
This passage raises a particularly important point for this work. The primary phe-
nomenon investigated is the \Chicken Little Phenomenon": what are the consequences for
the public and elites from the contemporary system of pervasive anxiety-inducing fundrais-
ing? In this chapter, fundraising emails are classied based on whether or not they are
anxiety-inducing, then those emails are used in analysis to determine the long-term eects
of anxiety on small-dollar fundraising.
4.5 Research Design
This project investigates the prevalence of worry appeals in a sample of fundraising emails,
then through a combination of content analysis and regression analysis, determines the eect
111
of worry appeals on fundraising totals for over 500 campaigns for Congress, Governorships,
and the Presidency.
4.5.1 Research Questions
1. What is the prevalence of worry-inducing emails among 2020 presidential candidates?
2. What is the relationship between use of worry-inducing emails on campaign fundraising
returns?
3. Are campaigns punished for overusing anxiety?
Corresponding to these research questions are hypotheses.
4.5.2 Hypotheses
1. H: A negative curvilinear relationship exists between between worry-inducing messages
on fundraising returns. At some time threshold, worry-inducing messages will begin
to have a negative eect on fundraising.
2. H2: A negative relationship exists between the overuse of anxiety and fundraising
returns.
4.6 Data and Methods
The research proceeded in a four-step process. First, 83,606 campaign emails were collected
politicalemails.org. Second, I thsoe emails were classied for being anxiety-inducing or not
using gradient-boosted decision trees (Chen et al., 2015). Third, contribution and expendi-
ture receipts were collected for all candidates that had email data. Fourth, a variety of linear
regression models were performed to determine the relationship between anxiety-inducing
emails and small-dollar fundraising returns.
112
4.6.1 Collecting the Data: Political Emails
The text content for thousands of emails was collected from https://www.politicalemails.org,
a database of candidate emails created in 2019. The website is reliably scrapable using
standard web-scraping tools, including the `rvest` package for R or the `bs4` library for
Python. It is crucial to note that while Politicalemails.org subscribes to the mailing lists
of many political candidates, it is not an exhaustive database of every email ever sent by a
candidate since 2019. One major reason for this is that political candidates may have several
email lists in addition to their primary email list, and campaigns also run experiments and
A/B tests where they send dierent emails to dierent individuals. While email accounts
used by politicalemails.org are virtually guaranteed to miss some emails sent by candidates,
there is no way to get around this issue, but the number of emails tracked by each candidate
is completely serviceable to the goals of this research.
In total, 83,609 emails from politicalemails.org were classied. A majority of the emails
were classied using a supervised machine learning model that was trained on 6,000 emails
labeled as worry/anxiety-inducing by MTurk workers. To train the supervised classier, the
content of 6,000 emails was uploaded to MTurk workers for labeling. After nobody signed up
to classify emails, the decision was made to greatly truncate the emails to only include the
rst 140 characters. Full reasoning behind that decision can be found in the Appendix for
this chapter, but to summarize, most people only look at emails for a few seconds and don't
even open them, so it was best to study email previews/openers, rather than classifying the
entirety of an email. In addition to being more costly, classifying the entire text of an email
in this case may lead to measurement error, as it is reasonably expected that most people
do not closely read every word of political fundraising emails.
113
4.6.2 Classifying Anxiety
After selecting the exact email text to classify, the next step was to develop and adhere
to a coding scheme to detect anxiety, which diers signicantly from past work. Here the
goal was to detect the intention to induce anxiety, while much research on anxiety or worry
classication is from medical or psychological work that attempts to identify anxiety in
people from what they write. And much of the medical/psychological work is explicit in
asking people to identify their anxiety.
For example, Mera and Ichimura (2008) use a journal-like method, analzying diabetic
patient responses to the question, \Have you developed any symptoms of anxiety? If yes,
tell me the details." They use the freeform responses written by patients and classify anxiety
into 5 types of problems: mental, physical, diet, physical activity, and medicine-related. The
actual classication is done via a machine learning model that uses words collected from a
related web bulletin to nd feature words related to various types of anxiety. Some example
words are: fat, insulin, calorie, surgery, kidney, excessive drinking, and muscle.
Because the purpose of this classier is to detect the invocation of worry or anxiety,
there was less work to go on and the inherently subjective nature of what makes a piece of
text worry-inducing became clear. a variety of options were considered: manual coding of
data, hiring crowdsourced coders to label data for various elements of anxiety according to a
preset coding scheme, and using a pseudo-survey approach with crowdsourced coders. For a
number of practical reasons and the loss of formerly usable crowdsourced coding platforms
for academics,
3
the decision was made to classify emails with a single-question coding scheme
to essentially create a voting-based classier on MTurk, with only one person classifying one
piece of text, as recommended by Barber a et al. (2019). The MTurk workers were given the
following instructions:
3
Notably, the service formerly named Crowd
ower, Figure-Eight, and now Appen removed their academic
pricing services that allowed for aordable crowdsourced labeling of training data for academics. Their costs
are now prohibitively expensive, starting at $10,000.
114
"Worry can be dened as concern about future events in which there is uncertainty about
the outcome and where the individual experiences feelings of anxiety. Read the subject
line and rst 140 characters of an email from a political campaign and make your best
determination if you think the intent of the text is to make its recipient worry.
Remember, these are emails sent from political campaigns, so it is expected you will not
always agree with the things the sender is saying. Try to understand that the text you will be
reading are the subject lines of emails sent from campaigns to their supporters, so your task
is to determine if the political campaign is sending emails to make its supporters anxious or
worried."
Then, for each task MTurk workers were given the subject of an email and the rst 140
characters of the email, and asked if they thought it was intended to induce worry/anxiety
or NOT intended to induce worry/anxiety. Figure 4.3 below shows a preview of the MTurk
job.
Figure 4.3: An example of an actual job MTurk workers were asked to do
The dataset was trained on 6000 email openers and validated the classier on 2600 emails.
To construct the classier, gradient-boosted decision trees with the R package xgboost(Chen
et al., 2015) were used, based on the gradient boosting framework in Friedman et al. (2000);
Friedman (2001). Unfortunately, the max F1 score
4
achieved with these classiers was
0.62, which is adequate but falls short of being optimal. The accuracy = 0.573, the pre-
cision = 0.549, and the recall = 0.711. The xgboost classier creates a prediction score
between 0 and 1 for anxiety, with higher values being theoretically more certain that a text
4
F1 score = weighted average of precision and recall (explained in next footnotes), where the best value
for F1 = 1 and worst value = 0
115
is worry/anxiety-inducing. Picking a prediction `threshold' should be done to maximize the
F1 score, which balances precision and recall.
56
Because the worry/anxiety classication
problem is inherently very subjective, the data aren't very well separated with the labelling
from MTurk coders. As a result, the optimal prediction threshold for maximizing the F1
score was all anxiety prediction scores> 0.1.
In total, 32,943 texts were classied as \not worry/anxiety-inducing" and 53,346 texts
were classied as worry/anxiety-inducing. I report a histogram of the worry/anxiety-prediction
scores below.
Because anxiety may be correlated with negativity, negativity/polarity was added as an
independent variable for each email opener. This was done with an out-of-the-box sentiment
classier, a sentiment score ranging from -1 to 1 was constructed using the polarity function
from the qdap package (Goodrich et al., 2020). The polarity function in qdap works by
tagging each word of a given sentence as either: polarized term, neutral term, negator,
valence shifter (amplier) and valence shifter (deamplier).
7
All the polarity values are summed and modied by their amplier and de-ampliers.
So for the sentence, "this dissertation is quite good", the polarity score would be calculated
by nding the polarity sum (polarity of polarized term `good' = 0.8, polarity of amplier
`quite' = 1, so polarity = 1+ 0.8 = 1.8). Then, the entire sentence is taken into account
to modify the polarity score so that polarity is decreased by the number of words. The
reasoning is that the more dense a sentence is with polarity-charged terms, the stronger the
polarity is. That is, a 10-word sentence with 1 negative word will have a lower polarity score
than a 20-word sentence with 1 negative word because the density of the negative term(s) is
lower.Counting the total terms in my example sentence = 5. The polarity score is calculated
5
Precision = the ratio of correctly predicted positive observations to the total predicted positive obser-
vations.
6
Recall = the ratio of correctly predicted positive observations to all observations in that class
7
Polarized term = words associated with positive/negative valence (ex: sucks). Neutral Term = term
with no emotional context (ex: nd). Negator = words that invert polarized meaning (ex: NOT good). A
valence shifter (amplier) = a word that increase emotional intent (ex: totally sucks). A Valence shifter
(de-amplier) = a word that decreases emotional intent (not bad).
116
0
10000
20000
0.00 0.25 0.50 0.75 1.00
Anxiety Prediction Score
Count
Frequency histogram of anxiety prediction scores in email data (N = 86,289)
Figure 4.4: A histogram displaying the distribution of anxiety prediction scores on a scale
from 0-1. 38% of the emails were classied as having an anxiety likelihood of 0.1, which is
the threshold for classifying an email as anxiety-inducing.
117
using the square root of the total terms/words as the denominator and the polarity score as
the numerator. So the polarity score for this example sentence is 1.8 / sqrt(5) = 0.447.
Figure 4.4 shows a histogram depicting the distribution of anxiety prediction scores on
a scale from 0-1. 38% of the emails were classied as having an anxiety likelihood of 0.1,
which is the threshold for classifying an email as anxiety-inducing. While this may lead
to overclassication of emails as anxiety-inducing, this would only suggest that the eects
of anxious emails on fundraising returns in regression analyses are small than in reality.
Classication of non-anxious emails as anxious would lead to small eect sizes, as the eect
of truly anxious emails is likely somewhat diluted.
After the entire dataset of emails were classied for anxiety and negativity, 61% were
classied as anxiety-inducing and 22.6% were classied as negative. Given the limitations of
the anxiety classier, this result would need to be replicated to be generalized, but the fact
that more than half of all campaign emails were classied as anxiety-inducing is striking.
Even if this result is an overestimate of anxiety-inducing campaign emails, a large percentage
of emails sent by campaigns can safely be considered anxiety-inducing.
4.6.3 Descriptive Statistics
The range of the data for the actual period analyzed is from the 27th week of 2019, or the week
of July 5, 2019 through the 48th week of 2020, or the rst week of December. The number
of campaigns/committees analyzed based on matching contribution and expenditures data
from the FEC as well as emails from politicalemails.org was 346. The number of campaign
contributions used in analysis was 10,490,938. The total number of emails that went into
analysis was 48,134. The unit of analysis is a week in a campaign. For each week, each
campaign has data on their number of emails sent out, the number of emails that week that
were classied as worry/anxiety-inducing, the percentage of emails that were classied as
worry/anxiety-inducing, the number of emails classied as having negative sentiment, the
118
percentage of emails classied as having negative sentiment, the total contributions made to
each campaign under $200, the total expenditures made by the campaign, the expenditures
explicitly labelled as for fundraising/advertising/solicitation purposes, and the expenditures
explicitly labelled for fundraising/advertising/solicitation purposes as well as money spent
on Facebook, Google, and a number of SMS/email marketing rms. A number of time-based
models are included to account for temporal eects, as well as models to account for partisan
dierences. Time is an important factor in the timing of political contributions (Magleby
et al., 2018), so time is centrally considered as an important feature in this chapter.
When considering the substantive eect of these gures on politics, however, it is impor-
tant to know just how often campaigns really do send out anxiety-inducing emails. Figure
4.5 reports the mean number of anxious emails all campaigns send over time. This is done
to determine if campaigns are only sending out anxiety-inducing emails toward the end of a
campaign or if they are sending out anxious emails throughout the entire campaign.
1
2
3
4
5
6
7
8
0 20 40 60
Time
Emails Sent Per Week
colour Avg. Anxious Email Count Avg. Email Count
Mean # of Emails and Anxious Emails sent by Campaigns Over Time
Figure 4.5: Graph displaying the mean number of emails and anxious emails sent by cam-
paigns over the period of study, July 2019 - November 2020. Generally, campaigns send 3-4
emails per week, 2-3 of which were classied as anxiety-inducing.
119
Figure 4.5 shows that during the rst 60 weeks of a campaign, the average campaign
sends between 3 and 4 emails per week, 2-3 of which are classied as anxiety-inducing. This
number increases at the 62nd week of the campaign (September 7, 2020), rapidly increasing
to a peak of nearly 7 anxious emails per week for the average candidate by the last week of
the election (November 3, 2020) before rapidly dropping o once the election occurred.
1
2
3
4
5
6
7
0 20 40 60
Time
Emails Sent Per Week
colour Democrats Republicans
Mean # of Emails and Anxious Emails Sent by Campaigns Over Time
Figure 4.6: Graph displaying the mean number of anxious emails sent by campaigns over
time, grouped by candidates from the Democratic party and Republican party. Democrats
typically send slightly fewer anxious emails per week.
Figure 4.6 shows a similar graph, displaying the average number of anxiety-inducing
emails sent out by candidates of each party. Typically, Republicans send one more anxious
emails per week than Democrats do. And in particular, Donald Trump has the record for
most anxious emails sent per week, with 132 anxious emails sent in the last week of the
election. Figure 4.7 shows the weekly anxious email count for both Donald Trump and Joe
Biden.
Also included here is Figure 4.8, which visualizes the the average weekly contribution
total for candidates from contributions under $200. Generally, average fundraising begins
at time 1 (June 29, 2020) at approximately $5,000 per week and slowly increases to around
120
0
25
50
75
0 20 40 60
Time
Emails Sent Per Week
colour Biden Anxiety Count Trump Anxiety
Number of Emails and Anxious Emails Sent by Biden and Trump Over Time
Figure 4.7: Graph displaying the number of anxious emails sent by Donald Trump and Joe
Biden per week. Joe Biden sent slightly fewer anxiety-inducing emails per week than Donald
Trump did.
$20,000 per week by week 58 of the campaign (August 3, 2020), at which point weekly
contribution totals exponentially increase to a peak of nearly $150,000 in the rst week of
November, 2020.
121
5000
10000
25000
50000
100000
0 20 40 60
Time
Small Contributions Received
colour Small Contributions w/out Presidents Small Contributions w/Presidents
Mean Small Dollar Contributions Over Time
Figure 4.8: Graph displaying the mean weekly contribution total for all small contributions
below $200, July 2019 - November
4.7 Results
Table 4.1 displays four models: three bivariate models showing the eect of anxious email
count, all email count, and negative email count on sub-$200 fundraising returns. The
multivariate model shows the eect of anxious email count, all email count, negative email
count, as well as total spending and spending earmarked as related to fundraising activi-
ties. Specically, the models in Table 4.1 report various log-linear regression models with
campaign funds raised under $200 as the dependent variable, and a number of independent
variables: the number of emails sent by a campaign (email count), the number of negative
122
emails sent by a campaign (negative email count), the number of worry/anxiety-inducing
emails sent that week (anxious email count), total campaign spending, campaign spending
related to fundraising (soliciting, advertising, fundraising, and vendors that engage in those
activities). Total campaign spending is included as a general-purpose control variable. Some
candidates are more popular than others and will raise more because of name recognition and
popularity, the importance of the race, media coverage, etc. Candidates that raise more tend
to spend more, so total spending plays an important role as a control variable in the model,
particularly when considering outlier candidates like Donald Trump and Joe Biden. In later
models, candidates are separated by being in the top 10% of spenders and the bottom 90%
of spenders.
Table 4.1: OLS: Eect of Anxious Emails on Small Donor Fundraising
Dependent variable:
Fundraising Below $200
Anxious Emails Email Count Negative Emails Multivariate
(1) (2) (3) (4)
Anxious Email Count 0.327
0.057
(0.006) (0.014)
Email Count 0.222
0.087
(0.004) (0.009)
Negative Email Count 0.558
0.028
(0.013) (0.016)
Total Spending 0.506
(0.009)
Total Fundraising Spending 0.024
(0.005)
Constant 6.138
6.041
6.469
1.657
(0.025) (0.026) (0.025) (0.082)
Observations 13,493 13,493 13,493 13,493
R
2
0.200 0.218 0.125 0.366
Adjusted R
2
0.200 0.218 0.125 0.366
Note:
p<0.1;
p<0.05;
p<0.01
These basic regression results show relatively small but consistent positive eects of send-
ing anxious emails and sending any emails, but the multivariate model shows that negative
emails have a negative eect on fundraising. Log-linear transformation was performed on
the money variables (the dependent variable, Fundraising Below $200, and the independent
123
variables, Total Spending and Total Fundraising Spending) because these money variables
were largely right-tailed, with most of the campaign contributions being made towards the
end of the campaign to a small number of candidates. Due to the log-linear transformation,
these results can be interpreted as suggesting for each additional anxious email sent out in a
week, campaigns received 5.7% increased funds. Additionally, sending each additional email
is associated with an 8.7% increase in weekly fundraising, and negative emails are associated
with a 2.8% increase in fundraising. The low R
2
values are notable for the three bivariate
models, however, showing that in a vacuum these email statistics are not very explanatory.
Still, the positive relationship between sending all emails and fundraising returns is notable,
although not entirely unexpected given that time is unaccounted for in these simple models.
Additionally, total spending is shown to have a positive eect in the multivariate model,
but curiously campaign spending related to fundraising expenses has a negative eect. No
explanation is oered for this, but it is an interesting nding that future work could explore.
In general, this table shows that at least at some points during a campaign, anxious emails
have a positive relationship with small-dollar fundraising returns.
It is important to consider two key features in campaign contributions: time and high-
prole candidates. Figure 4.9 shows that the top 10% of candidates by total spending raise
far greater sums from small contributors than do the bottom 90% of candidates. While sep-
arating candidates in this way is somewhat imprecise due to a host of confounding variables
and endogeneity concerns, it is clear that there is a large dierence in terms of small-dollar
fundraising returns between the top 10% of candidates and the bottom 90% of candidates.
Additionally, time plays a major feature in campaign contributions. The raw contributions
data show that until approximately the 50th week of a campaign, contributions to candi-
dates are generally steady. It is no coincidence that this period of time coincides with early
June, when Joe Biden gained a majority of delegates to become the de facto Democratic
Presidential candidate.
124
Figure 4.9 shows average weekly fundraising for all candidates, separated by being in
the top 10% and bottom 90% of spending. The gap between candidates in the top 10% of
spending and the bottom 90% of candidates is enormous, and on the magnitude of hundreds
of thousands of dollars every week. Still, when the bottom 90% of candidates are examined
by themselves, temporal eects are observed, albeit on a much smaller scale. So time is
expected to have an eect for all candidates, albeit a considerably smaller one for candidates
in the bottom 90% of spending.
5000
25000
50000
100000
175000
0 20 40 60
Time
small_contributions
colour Small Contributions to Bottom 90% of Candidates Small Contributions to Top 10% of Candidates
Mean Small Dollar Contributions Over Time for the
Top 10% of Spenders and Bottom 90% of Spenders
Figure 4.9: Graph displaying the mean weekly contribution total for all small contributions
below $200, June 2020 - November 2020. Grouped by non-Presidential candidates in the top
10% of spending and bottom 90% of spending.
Because time is expected to aect these models, time-related features and their interac-
tion with the email count features are included in the regression models reported in Table
125
4.2. Specically, the features added are time (the number of weeks it has been into a cam-
paign), time squared, time and time cubed, as well as the interaction of each email count on
time (Anxious Email Count * Time, Anxious Email Count * Time
2
, etc.).
The results in Table 4.2 show that at time 0, the eect of sending each additional anxiety-
inducing email increases a candidate's small-dollar fundraising return by 15.1%, sending each
additional email increases fundraising return by 16.5%, and sending each additional negative
email increases fundraising by 2.2%, although the eect for negative emails does not have
statistical signicance. Additionally, increasing total spending by 1% can be interpreted
to increase fundraising returns by 0.467%. Time cubed is shown to have a negative eect,
as well. These coecients, along with the rest of the results of this model, are dicult to
interpret because of the temporal elements and interactions. To better interpret what these
models suggest about the eects of anxiety-inducing emails on fundraising over time, Figures
4.10 and 4.11 predict the eects of anxious emails over time, holding all other features (except
time) at their expected weekly means. Email count, negative email count, total spending,
and total fundraising spending are held at their timely means while anxiety is varied between
0, 1, 2, 4, 8, 16, and 32 at each point in time. By \held at their timely means", it means the
mean value for each variable was calculated for each time period and used in the model.
126
Table 4.2: Log-Linear Model: Eect of Emails and Time on Fundraising Below $200
Dependent variable:
Fundraising Below $200
Total Model Top 10% Spenders Bottom 90% Spenders
(1) (2) (3)
Anxious Email Count 0.151
0.011 0.276
(0.092) (0.146) (0.106)
Email Count 0.165
0.242
0.066
(0.064) (0.107) (0.074)
Negative Email Count 0.022 0.348
0.168
(0.103) (0.161) (0.120)
Total Spending 0.467
0.173
0.369
(0.009) (0.024) (0.011)
Total fundraising Spending 0.026
0.024
0.028
(0.005) (0.010) (0.006)
Time 0.011 0.026 0.055
(0.015) (0.034) (0.017)
Time Squared 0.001
0.001 0.002
(0.0005) (0.001) (0.001)
Time Cubed 0.00001
0.00001 0.00002
(0.00000) (0.00001) (0.00000)
Anxiety Count*Time 0.008 0.003 0.020
(0.010) (0.015) (0.011)
Anxiety Count*Time
2
0.0002 0.0002 0.001
(0.0003) (0.0004) (0.0003)
Anxiety Count * Time
3
0.00000 0.00000 0.00000
(0.00000) (0.00000) (0.00000)
Email Count*Time 0.0001 0.010 0.026
(0.007) (0.010) (0.008)
Email Count*Time
2
0.00003 0.0002 0.001
(0.0002) (0.0003) (0.0002)
Email Count*Time
3
0.00000 0.00000 0.00000
(0.00000) (0.00000) (0.00000)
Negative Count*Time 0.008 0.032
0.006
(0.011) (0.017) (0.013)
Negative Count*Time
2
0.0003 0.001 0.0001
(0.0003) (0.0005) (0.0004)
Negative Count*Time
3
0.00000 0.00001 0.00000
(0.00000) (0.00000) (0.00000)
Constant 1.268
6.262
2.217
(0.154) (0.400) (0.170)
Observations 13,354 1,875 11,479
R
2
0.371 0.279 0.277
Adjusted R
2
0.370 0.272 0.275
Residual Std. Error 1.988 (df = 13336) 1.661 (df = 1857) 1.935 (df = 11461)
F Statistic 462.622
(df = 17; 13336) 42.182
(df = 17; 1857) 257.734
(df = 17; 11461)
Note:
p<0.1;
p<0.05;
p<0.01
127
7
8
9
10
11
0 20 40 60
time
Predicted Weekly Fundraising
0
10
20
30
Anxious Emails
Predicted Fundraising at Various Anxiety Counts,
Holding All Other Variables Equal.
9
10
11
0 20 40 60
time
Predicted Weekly Fundraising
0
10
20
30
Anxious Emails
Predicted Fundraising at Various Anxiety Counts,
Holding All Other Variables Equal. Top 10% of Spenders
7.5
10.0
12.5
0 20 40 60
time
Predicted Weekly Fundraising
0
10
20
30
Anxious Emails
Predicted Fundraising at Various Anxiety Counts,
Holding All Other Variables Equal. Bottom 90% of Spenders.
Figure 4.10: Fundraising predictions for a candidate at various levels of anxiety over time,
holding all other features at their means. Includes all candidates. Black lines are marked
between Week 63 and 65 of an entire campaign cycle, corresponding to between September
7 - 21.
These prediction graphs show higher anxious email counts in lighter blue colors, from 0
anxious emails per week to 32. The darkest line represents an anxious email count of 0, which
across all models is shown to decrease with eectiveness over time. For candidates in the
128
top 10% of spending, sending as many anxious emails as possible is expected to be the most
eective at fundraising for a large portion of the campaign, except for the nal 8 weeks or so.
Importantly, all three models predict that anxiety has its lowest point of ecacy at the end of
a campaign. The prediction graphs corresponding to the models in Table 4.2 show relatively
strong support for hypothesis 1: there is an overall negative curvilinear relationship between
sending anxious emails and small-dollar fundraising, although the eect is somewhat small.
Still, these results are robust across models and point to a gradual fatigue with anxiety-
inducing emails. For candidates in the top 10% of spending, anxious email count didn't
have a statistically signicant eect, although this could be driven by the smaller sample
size. Hypothesis 2 is somewhat supported by this result, but can be better explored through
isolating models to certain types of candidates. To test hypothesis 2, additional models were
constructed and are reported in Tables 4.3 and 4.4. The models reported are replications
of the regression models in Tables 4.1 and 4.2, keeping the same features but running the
analysis on four dierent data sets. The datsets included are a subset of candidates separated
by party and proclivity for sending anxiety-inducing emails.
The rst set of models reported are in Table 4.3, and replicate the baseline models re-
ported in the rst table of this chapter, Table 4.1. The features accounted for include the
dependent variable, fundraising below $200, and the independent variables: Anxious Email
Count, Email Count, Negative Email Count, Total Spending, and Total Fundraising Spend-
ing. The four specic models are reported in the tables as \Democrat"
8
, \Republican"
9
,
\Low Anxiety"
10
, and \High Anxiety"
11
. However, in these models logarithmic transforma-
tions are not performed on the money variables: Fundraising Below $200, Total Spending,
and Total Fundraising Spending. The variables are kept purely linear because the R
2
in
the models with logarithmic transformations were far outpaced by the R
2
in the pure linear
models for most models. It is likely that the pure linear models outperform logarithmic
8
Representing the model performed on candidates belonging to the Democratic party
9
Representing the model performed on candidates belonging to the Republican party
10
Representing the model performed on candidates in the lower 25% of the Anxious Email Count feature.
11
Representing the model performed on candidates in the upper 75% of the Anxious Email Count feature.
129
models because many candidates have 0s for their money variables, which needed to have
1 added to them to even function. The best-performing models are pure OLS models that
include Joe Biden and Donald Trump, so that is what is reported in Table 4.3, Table 4.4,
and Figure 4.11. In the appendix, a regression table and set of prediction graphs using the
log-linear models can be found in Table 6.3 and Figure 6.9.
Table 4.3: OLS: Eect of Emails and Time on Fundraising Below $200
Dependent variable:
Fundraising Below $200
Republicans Democrats Low Anxiety CandidatesHigh Anxiety Candidates
(1) (2) (3) (4)
Anxious Email Count 4,611.251
6,194.184
2,864.074
1,466.797
(786.367) (2,333.263) (553.769) (745.641)
Email Count 3,821.245
16,253.070
6,241.796
3,490.450
(558.554) (1,534.266) (353.370) (596.896)
Negative Email Count 7,039.481
27,302.140
3,324.064
505.445
(964.594) (2,627.630) (720.625) (699.999)
Total Spending 0.020
0.078
0.070
0.072
(0.001) (0.001) (0.002) (0.002)
Total fundraising Spending 0.282
0.409
0.153
0.013
(0.007) (0.008) (0.006) (0.014)
Constant 1,588.222 58,861.680
12,648.620
1,515.060
(1,331.800) (4,038.893) (852.164) (1,122.527)
Observations 3,186 10,146 3,424 3,405
R
2
0.666 0.506 0.812 0.368
Adjusted R
2
0.665 0.505 0.811 0.367
Residual Std. Error 59,045.790 (df = 3180)278,685.700 (df = 10140) 35,567.850 (df = 3418) 44,043.650 (df = 3399)
Note:
p<0.1;
p<0.05;
p<0.01
The relatively simple multivariate models in Table 4.3 applied to Republicans, Democrats,
low anxiety candidates, and high anxiety candidates shows mixed eects of anxious email
count on fundraising. For Republicans, low anxiety candidates, and high anxiety candidates,
sending an additional email was associated with a fairly large drop in small-dollar fundrais-
ing returns, from approximately -$1500 for high anxiety candidates to -$4600 for Republican
candidates. Democratic candidates saw a positive association between anxious emails and
fundraising returns. All four models showed a large, positive, and statistically signicant re-
lationship between sending any email and fundraising returns. Additionally, all four models
130
showed a negative association between sending negative emails and fundraising returns, al-
though for high anxiety candidates this eect was not statistically signicant. All four models
showed a positive relationship between total spending and sub-$200 fundraising returns, but
once again there were mixed results for fundraising-related spending. The Republican and
low anxiety candidate models show a positive and statistically signicant benet to fundrais-
ing spending, but the Democrat model shows a statistically signicant negative relationship
between fundraising-related spending and sub-$200 fundraising returns. Finally, there was no
statistically signicant relationship between total spending and sub-$200 fundraising returns
for high anxiety candidates.
These data softly suggest that anxiety-inducing emails were more strongly associated with
fundraising increases for Democratic candidates, which could make sense given the dynamics
of the 2020 election. Donald Trump and the direction he was taking the Republican party
represented an existential threat to many Democrats, who openly campaigned on stopping
Trump as the biggest threat to the country. As a result, it would make sense for anxious
appeals to work better for Democrats in 2020. To better understand the dynamics of timing
on these four types of candidates, the full models including time, time squared, time cubed,
and the interactions between time features and email count features are reported in Table 4.4.
These models have relatively good ts, with an Adjusted R
2
value of 0.686 for the Republican
model, 0.542 for the Democrat model, 0.819 for the low anxiety candidate model, and 0.374
for the high anxiety candidate model.
Because these models account for a very long period of time and time is an important
factor in campaign contributions, it's best to start by interpreting the interaction terms
between email count features and time features, as the raw coecients represent the eect of
a feature at time 0, which is neither easily interpretable nor meaningful. The most notable
eects reported in these models are (1) the negative eect between anxiety count and time
for low anxiety candidates; (2) the negative relationship between anxiety count and time
131
cubed for Republicans, low anxiety candidates, and high anxiety candidates; and (3) the
positive eect between anxiety count and time cubed for Democratic candidates.
Once again, prediction graphs are replicated for these new models, reported in Figure
4.11. They work in the same way as the graphs in Figure 4.10, holding all features except
anxiety count at their means for each given week of the campaign. The results are very
interesting and illuminating. Vertical black lines are marked to indicate the point in time
at which sending 0 anxiety-inducing emails becomes the optimal fundraising strategy. Be-
ginning with the predictions for Republican candidates: anxious emails are shown to have
a large positive eect until around the 55th week of the campaign, at which point sending
no anxiety-inducing emails predicts the greatest fundraising returns. The Democrat model
shows anxious emails are more eective than non-anxious emails until about the 20th week
of the campaign, at which point sending 0 anxious emails per week predicts the greatest
fundraising returns. This eect lasts until about the 60th week of the campaign, at which
point sending anxious emails predicts greater fundraising returns. It's notable that this
eect is robust and the general shape of the predictions remains even when Joe Biden is
excluded from the data, suggesting Biden alone is not driving this eect. Instead, it's worth
noting that Democrats were the only group in these models where anxiety-inducing emails
were predicted to be most successful at fundraising toward the end of the campaign, in-
creasing in eectiveness until election day. One potential explanation for this eect could
be Democratic fear and anxiety over a Trump re-election, making anxiety-inducing emails
particularly eective at generating campaign contributions. Moving on to the low anxiety
candidates, sending any anxious emails only predicts greater sub-$200 fundraising returns
until approximately week 10 of the campaign, at which point sending 0 anxiety-inducing
emails per week begins to predict the highest fundraising returns, a pattern which remains
until the end of the campaign. This could suggest that candidates who generally don't send
anxiety-inducing emails tend to see the most success just avoiding anxiety altogether. Re-
garding high anxiety candidates, sending anxiety-inducing emails predicts greater or nearly
132
Table 4.4: OLS: Eect of Emails and Time on Fundraising Below $200
Dependent variable:
Fundraising Below $200
Republicans Democrats Low Anxiety CandidatesHigh Anxiety Candidates
(1) (2) (3) (4)
Anxious Email Count 137.278 7,719.591 5,684.701
3,258.098
(6,952.169) (12,659.840) (3,434.718) (4,969.316)
Email Count 4,712.269 5,122.907 8,940.113
1,636.092
(5,095.745) (8,481.873) (2,091.126) (4,087.566)
Negative Email Count 10,691.160 39,384.140
1,094.206 6,107.257
(7,440.158) (14,786.310) (4,339.309) (4,736.319)
Total Spending 0.015
0.073
0.069
0.069
(0.001) (0.001) (0.002) (0.002)
Total fundraising Spending 0.257
0.374
0.157
0.013
(0.007) (0.008) (0.006) (0.014)
Time 351.717 7,475.817
2,683.043
441.095
(823.709) (2,218.119) (477.959) (680.348)
Time Squared 28.778 275.318
77.875
15.416
(23.798) (69.793) (14.962) (20.556)
Time Cubed 0.396
3.164
0.671
0.156
(0.205) (0.633) (0.135) (0.181)
Anxiety Count*Time 317.007 1,783.060 1,162.556
510.306
(652.007) (1,364.680) (355.697) (499.725)
Anxiety Count*Time
2
25.894 90.903
37.781
20.118
(17.766) (41.021) (10.476) (14.350)
Anxiety Count * Time
3
0.373
1.052
0.340
0.216
(0.146) (0.358) (0.091) (0.122)
Email Count*Time 304.138 925.171 1,720.159
266.540
(454.405) (894.442) (216.465) (403.790)
Email Count*Time
2
0.607 24.473 50.632
9.047
(11.971) (26.981) (6.422) (11.495)
Email Count*Time
3
0.102 0.324 0.425
0.105
(0.096) (0.237) (0.056) (0.097)
Negative Count*Time 2,066.477
7,114.453
478.967 941.159
(705.112) (1,572.636) (459.563) (488.513)
Negative Count*Time
2
76.250
296.220
18.475 32.135
(19.464) (46.592) (13.691) (14.078)
Negative Count*Time
3
0.770
3.347
0.173 0.291
(0.161) (0.401) (0.119) (0.119)
Constant 1,041.895 35,557.340
14,273.650
4,219.179
(8,308.804) (19,911.440) (4,381.529) (6,268.952)
Observations 3,186 10,146 3,424 3,405
R
2
0.687 0.543 0.820 0.377
Adjusted R
2
0.686 0.542 0.819 0.374
Residual Std. Error 57,215.190 (df = 3168)268,224.200 (df = 10128) 34,837.200 (df = 3406) 43,815.090 (df = 3387)
Note:
p<0.1;
p<0.05;
p<0.01
133
equivalent fundraising returns than sending no anxious emails until week 55 of the campaign,
at which point sending 0 anxious emails predicts greater fundraising returns. The model for
high anxiety candidates was included to test hypothesis 2, but generally does not support
hypothesis 2 that overuse of anxiety has a negative relationship with sub-$200 fundraising
returns. Hypothesis 1 is rmly supported and is shown to be robust, as for all candidate
groups in this analysis except for Democrats, the eectiveness of anxiety on fundraising
returns declines over time. While the hypotheses are strongly related, they were designed
to show two slightly dierent eects, one of which was not clearly supported. Candidates
who use a lot of anxiety do not appear to be punished for using anxiety more than other
candidates - in fact, they are better able to fundraise using anxiety-inducing emails than low
anxiety candidates.
134
−400000
−200000
0
200000
0 20 40 60
time
Predicted Weekly Fundraising
0
10
20
30
Anxious Emails
Republicans: Predicted Fundraising at Various Anxiety Counts,
Holding All Other Variables Equal.
−500000
0
500000
1000000
0 20 40 60
time
Predicted Weekly Fundraising
0
10
20
30
Anxious Emails
Democrats: Predicted Fundraising at Various Anxiety Counts,
Holding All Other Variables Equal
−100000
0
100000
0 20 40 60
time
Predicted Weekly Fundraising
0
10
20
30
Anxious Emails
Low Anxiety Candidates: Predicted Fundraising at Various Anxiety Counts,
Holding All Other Variables Equal
−100000
0
100000
0 20 40 60
time
Predicted Weekly Fundraising
0
10
20
30
Anxious Emails
High Anxiety Candidates: Predicted Fundraising at Various Anxiety Counts,
Holding All Other Variables Equal
Figure 4.11: Fundraising predictions for a candidate at various levels of anxiety over time,
holding all other features at their means. Includes all candidates.
135
4.8 Conclusion
Taking all of these results into consideration, hypothesis 1 is rmly supported and shown to
be robust, but hypothesis 2 is not clearly supported. However, the theoretical reasoning that
led to the construction of hypothesis two is generally supported. The expectation going into
this research was that candidates that overused anxiety would be punished for sending so
many anxiety-inducing emails. However, the data didn't clearly show that. What the data
did show was that except for Democratic candidates, anxious emails have a robust negative
curvilinear relationship with small-dollar fundraising returns. It's possible this eect for
Democrats was being driven by a unique party-wide anxiety over the re-election of Donald
Trump, but it's also possible this eect was being driven primarily by out-party eects. That
is, it's possible that in Presidential election years, the party out of power is more motivated
by anxious appeals, especially closer to the election. Future research replicating this eect
to test its robustness is warranted.
When comparing the predicted eects of these models to Figure 4.5, which is included
again in Figure 4.12 in the page below for sake of convenience, an interesting result is revealed.
The average campaign sends a generally steady amount of emails and anxiety-inducing emails
per week until approximately the 50th week of a campaign, in June 2020. At this point,
campaigns send increasingly more emails and anxious emails until the end of the campaign.
This is curious because it is also around the same time at which the models reported in
Table 4.4 and Figure 4.11 predict anxiety-inducing emails to raise less money than sending
zero anxiety-inducing emails. If these models can be generalized to the average political
campaign, non-Democrat campaigns are engaging in sub-optimal fundraising behavior by
sending out anxious emails at the end of a campaign.
One notable limitation to this work is the problem of endogeneity: losing campaigns could
send more anxious emails as they drop in the polls because they become more desperate to
136
1
2
3
4
5
6
7
8
0 20 40 60
Time
Emails Sent Per Week
colour Avg. Anxious Email Count Avg. Email Count
Mean # of Emails and Anxious Emails sent by Campaigns Over Time
Figure 4.12: Graph displaying the mean number of emails and anxious emails sent by cam-
paigns over the period of study, July 2019 - November 2020. Generally, campaigns send 3-4
emails per week, 2-3 of which were classied as anxiety-inducing.
raise money to stay in the race. Future work into this topic and any replication of this
work would do well to collect weekly polling numbers for each candidate and indicate if the
candidate is winning or losing at a given point in the race. This would help ensure that any
negative eects of anxiety-inducing emails shown in this work are truly representative of the
eects of anxious emails, and not larger campaign eects.
Further research into this topic could involve closer study of specic campaigns over time
and perhaps even case studies. One interesting line of research would be to replicate and
expand upon this data with a variable that somehow takes into account levels of public
anxiety or number of anxiety-inducing events in the news. Perhaps using newspaper data
to account for the number of anxiety-inducing or negative stories each week could be a
good barometer for how anxious the public might be feeling that week. More ne-grained
classication of the fundraising emails could also serve a similar purpose. For example,
perhaps fundraising emails that cite specic anxiety-inducing news events are more successful
137
at raising money because they tap into already-existing public anxiety. Additionally, emails
that specically mention Trump could be used as an alternative variable.
In conclusion, the data rmly suggest that the eectiveness of anxiety-inducing emails on
fundraising is not a static relationship. Anxiety-inducing emails can not always be considered
the most eective way for campaigns to fundraise. For some types of candidates, such as
Democrats in 2020, anxiety-inducing emails worked best to fundraise towards the end of
the campaign. However, for other types of candidates, anxious emails are only expected to
raise more money than non-anxious emails at the start of the campaign and for some time
afterwards.
The advice for campaigns from this research is that while the average campaign continu-
ously sends the public anxiety-inducing emails, this strategy does not always work, and over
time the eectiveness of anxiety-inducing emails degrades. My suggestion to campaigns is to
pick and choose the moments to utilize anxiety in emails, because while anxiety often does
not work to raise money, it can cause psychological harm in individuals and sub-optimal
fundraising returns for campaigns. Until campaigns choose to send fewer anxious, apocalyp-
tic, and frantic emails, it is likely Americans the country over will still texts and emails like
this one all throughout the year.
138
Figure 4.13: A fundraising text sent at 6:31pm on May 26, 2021
139
Chapter 5
Political Contribution Behavior In
Anxious Individuals
5.1 Abstract
Through an online survey, individual political contribution behavior is studied using re-
spondent anxiety as key explanatory feature. Individuals' level of psychological anxiety is
measured via two survey instruments, the GAD-7 and the IUS-12, and dierences in po-
litical behavior outcomes are measured using anxiety and other political characteristics as
explanatory features. The rst outcome tested is if there exists a positive relationship be-
tween anxiety and contribution likelihood. The second outcome tested is if there exists a
positive relationship between anxiety and likelihood to engage in campaign avoidance be-
havior: ignoring, blocking, unsubscribing from, or unfollowing campaigns.
5.2 Introduction
In this chapter, an online survey is deployed to answer several questions about individual-
level anxiety and its relationship with political fundraising and campaign communication.
140
Two major relationships are explored by asking and answering questions that accompany
them.
1. Relationship 1: The relationship between psychological anxiety/worry and propen-
sity to make political contributions.
Question 1: Are the psychologically anxious/worried more likely to give money
to political causes? In other words, do fundraisers exploit the psychologically anx-
ious/worried to raise money?
Hypothesis 1: Greater levels of psychological anxiety are associated with a greater
likelihood of making political contributions.
2. Relationship 2: The relationship between psychological anxiety/worry and propen-
sity to avoid campaign contact.
Question 2: Are the psychologically anxious/worried more likely to avoid campaign
contact?
Hypothesis 2: Greater levels of psychological anxiety are associated with greater
levels of active avoidance of campaign communication.
The answers to these research questions will allow for a better understanding of the
interaction between campaigns and their constituents. This understanding may help inform
the public about potentially exploitative campaign practices and help protect vulnerable
groups (anxious individuals) from campaign behavior designed to extract their money. If
anxious people do contribute more, informing the anxious public that anxious people are
more susceptible to emotional manipulation may help anxious individuals protect themselves
against distress. For one, informing anxious people that they are more likely to contribute
can help them recognize that they are using campaign contributions as a way to alleviate
anxiety and that they are more susceptible to anxiety-inducing manipulation. If anxious
people are more likely to make political contributions and do not like campaigns using anxiety
to fundraise, larger-scale interventions may be implemented such as the development of email
lters that classify an email as anxiety-inducing or not and automatically
ag or screen out
141
those emails. Regarding the second relationship and question, I predict that anxious people
will avoid campaign communication at higher rates than less-anxious people. It's expected
that anxious people will come to see worry-inducing campaign communication itself as a
form of political anxiety and take measures to avoid campaign contact.
5.3 Methods
5.3.1 The Survey
The aforementioned relationships and research questions are investigated by using data from
a 31-question survey asking about demographics, political behavior and attitudes, and anx-
iety. The survey recruited subjects from MTurk, and 599 usable responses were collected
in total. The entire survey instrument can be found in the appendix of this survey starting
with Figure 6.10. The variables used for regression analyses include standard demographic
and political variables, as well as political contribution features including if a respondent
has ever contributed, as well as instruments that measure a respondent's generalized anxiety
level and tolerance of uncertainty.
The survey was elded between 05/10/2021 - 05/15/2021 by posting an ad on Mturk to
Americans. The rst question participants were asked was their level of interest in politics.
The answer had a 4-point scale, ranging from not interested in politics to extremely interested
(1 = not interested, 4 = very interested). Participants who answered that they had no interest
in politics were screened out of the survey and the rest of their answers were not recorded.
In total, 755 responses were obtained, 599 of which were usable. Because MTurk survey
respondents come from an unrepresentative sample, the results were weighted to be more
representative of the public, as outlined in (Levay et al., 2016). Still, even with weighting it
is best to keep in mind that these results capture aspects of political behavior and attitudes
among an audience of that skews young, Democratic, and male.
142
Two conceptual dependent variables were measured: contributions and campaign avoid-
ance. The primary measure of the rst dependent variable is a question asking if subjects had
ever contributed, pictured in Figure 5.1. Two alternate measures of the dependent variable
for research question 1 are included: contributions in the past 12 months and frequency of
contributions. The primary dependent variable, `ever contributed' is binary
1
, the secondary
variable `contributed in past 12 months' is binary
2
, and the tertiary variable `contribution
frequency' is ordinal. Because the primary and secondary variable are binary and the tertiary
dependent variable is ordinal, logistic regression is needed to assess eects for the primary
and secondary DVs and linear regression is needed to assess eects for the tertiary DV. The
survey instruments for the dependent variables can be found below in Figures 5.1, 5.2, and
5.3.
3
Figure 5.1: Primary DV Instrument: Has the subject ever made a political contribution
Figure 5.2: Secondary DV Instrument: Has the subject made a political contribution in
the past 12 months. Answers `Denitely yes' and `Probably yes' were coded as a 1 for
contributing, and other answers were coded as a 0.
1
Variable measurement: 1 = contributed, 0 = did not contribute
2
Variable measurement: 1 = contributed, 0 = did not contribute
3
Variable measurement for contribution frequency: 1 = never contributed, 2 = I've contributed once or
twice, 3 = I contribute sometimes, 4 = I contribute relatively frequently, 5 = I contribute all the time
143
Figure 5.3: Tertiary DV Instrument: How frequently does the subject makes political con-
tributions
The dependent variable for research question 2 is campaign avoidance, and the instrument
can be found in Figure 5.4. Subjects were given a number of campaign behaviors mainly
related to avoiding campaign contact. Subjects who answered yes to any of the avoidance
behaviors were coded as a yes for a binary variable indicating they had avoided campaign
contact.
4
The resulting binary variable was used for logistic regression on the same variables
as the rst models.
Figure 5.4: Dependent Variable 2 Survey Instrument: A list of campaign avoidance behaviors
subjects have engaged in
Weights
The \anesrake" package was used to perform survey weighting using 2004 ANES data for
representative population weights. The variables used for weighting are age (18-29, 30-
4
This feature was measured via a question that asked respondents to check a box for each behavior they
engaged in. The options were: Ignored emails from a political group, unsubscribed from a political group's
email list, blocked or marked as spam email(s) from a political group, deleted emails, forwarded or shared
emails with others. Campaign avoidance was classied if a user listed that they ignored emails from a
political group, unsubscribed from a political group's email list, or blocked or marked as spam emails from
a political group
144
64, 65+), gender (male), education (have a college degree), liberal ideology, conservative
ideology, Democratic partisanship, and Republican partisanship. Marital status and race
were not included in this survey, so those variables were not used for weighting even though
they are commonly included in weight calculations.
Weighted CCES demographic data (Barney, 2018) were used for the representative uni-
verse by which I weighted my own survey data to make it more representative. These weights
were needed as MTurk audiences tend to dier from the general population on a number of
variables, particularly age, gender, and college education. To evidence the need for these
weights, about 90% of subjects in the survey reported having a college degree and men
outnumbered women 2:1.
Independent Variables
Besides standard demographic and political variables like age and party, the major concept
of interest was subject anxiety. This was measured with two well-validated metrics. The
GAD-7 is a measure of Generalized Anxiety Disorder developed in Spitzer et al. (2006) and
was validated in Swinson (2006); Ruiz et al. (2011) and L owe et al. (2008). The variable
used in regression analysis is a single average of this index. The GAD-7 instrument can be
found in Figure 5.5.
The IUS-12 is a measure of the concept of \intolerance of uncertainty," which was rst
introduced in French in Freeston et al. (1994), translated to English in Buhr and Dugas
(2002), condensed into a 12-item version in Carleton et al. (2007), and validated in Hale
et al. (2016). In Freeston et al. (1994)'s own words, \Intolerance of uncertainty is the ten-
dency of an individual to consider the possibility of a negative event occurring unacceptable,
irrespective of the probability of occurrence." This concept is closely related to anxiety
people may feel about politics, as elections are a prime example of uncertain events with
potentially negative consequences. Individuals with low tolerance of uncertainty may be par-
ticularly driven to both make political contributions and avoid campaign contact to avoid
145
Figure 5.5: The Instrumentation for the GAD-7, a scale to measure general anxiety disorder.
The GAD-7 Avg. variable used throughout this chapter is geneated from the rst of these
two survey questionss. Answering `Not at all' was coded as a 1, `Several days' = 2, `More
than half the days' = 3, and `Nearly every day' = 4.
146
political anxiety. Anxiety may compel people to contribute to soothe political anxiety by
gaining a greater sense of control over the outcome of an election. Additionally, anxiety may
also compel people to avoid campaign contact to escape constant reminders of an election
that may have negative consequences, and which is also not controllable by any one person.
The instrument for the IUS-12 can be found in Figure 5.6. Freeston et al. (1994) note
that the IUS-12 may be used not only as an average score, but rather as a measure of
two distinct types of anxiety. The rst 7 items of the IUS-12 ask about fear and anxiety
regarding future events, and the last 5 items of the IUS-12 ask about uncertainty stopping
or inhibiting actions. The rst 7 items may be folded into a variable called `prohibitory
anxiety' and the nal 5 items may be folded into a variable called `inhibitory anxiety'. I
expect prospective anxiety to be more strongly associated with an increase in likelihood to
contribute and inhibitory anxiety to be more strongly associated with avoiding campaign
contact.
The variables used in the regression models to test hypotheses 1 and 2 are average GAG-
7 score (GAD-7 Avg), prospective anxiety, inhibitory anxiety, political interest, education,
income, liberal ideology, conservative ideology, Democratic partisanship, and Republican
partisanship.
5.4 Results
The average survey respondent was a 37 year-old male college graduate an income between
$50,000 - $60,000. 62% of respondents are Democrats, 24% are Republicans, and 13% are
Independents. 67% of respondents are male and 33% are female. Every usable respondent
reported voting in the last election, with 50 respondents declining to answer. Most respon-
dents (64%) reporting having contributed to a political campaign or group at least once,
with 36% reporting that they have never contributed.
147
Figure 5.6: The instrumentation for the IUS-12, an index to measure intolerance of uncer-
tainty
148
Hypothesis 1 is tested by using the primary and secondary dependent variables for logistic
regression. Results are visualized in Figure 5.7, with the complete regression table reported in
Table 5.1. The results of these regressions show fairly strongly and unequivocally that higher
average scores on the GAD-7 increase the likelihood that individuals make a contribution.
There's also the surprising result that prospective anxiety and inhibitory anxiety, the two
measures of the IUS-12 which were predicted to be strongly related to anxious contribution,
generally aren't shown to have much of an eect. The small eects that these variables do
have are the opposite of my expectations. Prospective anxiety, or fear about what is to come
has a slight negative eect for ever contributing and inhibitory anxiety has a slight positive
eect on contributing, which rejects a portion of my theory. Still, the eects shown in the
IUS-12 measures do not mean that hypothesis 1 is rejected. Given the strength of the GAD-7
average scores on propensity to contribute, these results support hypothesis 1, that anxiety
has a positive relationship with making a campaign contribution. Other variables that have
signicant and positive relationships on contributing are political interest, education, and
Democratic partisanship. While this chapter does not investigate the eect of partisanship
specically on contributions, it is encouraging to see that the ndings from chapter 2 were
replicated here. Surprisingly, liberal ideology has a large negative relationship with making
a contribution in the past 12 months. No explanation is oered for this because it may be
more indicative of the MTurk sample's over-representation of liberals than anything else,
but it is worth noting.
149
male
income
republican
democrat
conservative
liberal
education
political interest
log(age)
inhibitory anxiety
prospective anxiety
GAD−7 Avg.
−1 0 1 2 3
Coefficient Estimate
Model
Ever Contributed
Contributed in past year
Effect of Anxiety on Propensity to Contribute
Figure 5.7: Dot-whisker Plot Visualizing the Eect of Anxiety on Propensity to Contribute.
To look deeper at the data, complete regression tables are reported in Table 5.1. The
constant/intercepts of the model are -7.37 and -10.564 respectively, which represent the log
odds of making a contribution when all of the independent values are at their baseline, or
lowest value. Because log odds are not easily interpretable, these coecients can be converted
into probabilities. In terms of probability, the constants for the model are 0.001 and 0.00005,
150
where the maximum value is 1. The coecients for the GAD-7 Average independent variable
are 0.522 and 0.733 for the `ever contributed' and `contributed in 12mo' models, respectively.
Converted into probabilities, each point increase in GAD-7 average equate to a probability
increase of 0.63 and 0.67. Because the model has many other variables, it is impossible for
the model to predict anything having a probability greater than 1, however. For example, the
probabilities are not added up so that a GAD-7 Average score of 2 means individuals have
a probability of 1.26 to contribute. Rather, the log odds must be added and then converted
into probabilities. For example, a GAD-7 Average of 2 would produce a log odds value
of 1.044 and 1.466 for each model, respectively. Converting these values into probabilities
returns 0.74 and 0.81. Higher values of log odds simply create probabilities that converge
on 1, but the probability values still help put things into perspective. In short, the higher
the log odds, the greater the probability. Other impactful variables for ever contributing are
political interest (coef = 0.48, p = 0.62), education (coef = 1.10, p = 0.75), conservative
ideology (coef = 0.71, p = 0.67), and Democratic partisanship (coef = 2.23, p = 0.90).
The fact that this survey shows Democratic partisanship has a positive relationship with
increased contribution likelihood further supports the results of chapter 2 and 3.
151
Table 5.1: Logistic Regression: Eect of anxiety on propensity to contribute
Dependent variable:
Ever Contributed Contributed in past year
(1) (2)
GAD-7 Avg. 0.720
0.790
(0.191) (0.204)
prospective anxiety 0.068
0.049
(0.027) (0.029)
inhibitory anxiety 0.027 0.042
(0.032) (0.032)
log(age) 2.755
2.144
(0.456) (0.498)
political interest 0.512
0.551
(0.146) (0.165)
education 1.028
0.618
(0.143) (0.127)
liberal 0.910
0.436
(0.542) (0.629)
conservative 0.514 0.952
(0.318) (0.334)
democrat 2.271
1.436
(0.601) (0.641)
republican 0.270 0.746
(0.287) (0.313)
income 0.002 0.116
(0.044) (0.049)
male 0.144 0.735
(0.264) (0.304)
Constant 17.122
17.241
(2.062) (2.279)
Observations 532 551
Log Likelihood 260.155 223.331
Akaike Inf. Crit. 546.310 472.663
Note:
p<0.1;
p<0.05;
p<0.01 152
male
income
republican
democrat
conservative
liberal
education
political interest
log(age)
inhibitory anxiety
prospective anxiety
GAD−7 Avg.
−0.25 0.00 0.25 0.50
Coefficient Estimate
Model
Contribution Frequency
Effect of Anxiety on Contribution Frequency
Figure 5.8: Eect of anxiety on contribution frequency
The tertiary variable related to hypothesis 1 is contribution frequency. The results of the
linear regression model are visualized in Figure 5.8, and a full regression table can be found in
Table 5.2. Similar to the results of the `Ever Contributed' and `Contributed in 12mo', GAD-
7 Average scores and education have a positive relationship with contribution regularity.
Oddly enough, Republicans were more likely to report contributing at higher frequencies than
153
Table 5.2: OLS: Eect of anxiety on contribution frequency
Dependent variable:
Contribution Frequency
GAD-7 Avg. 0.361
(0.116)
prospective anxiety 0.026
(0.016)
inhibitory anxiety 0.002
(0.019)
log(age) 0.406
(0.255)
political interest 0.034
(0.093)
education 0.286
(0.062)
liberal 0.542
(0.292)
conservative 0.225
(0.200)
democrat 0.088
(0.287)
republican 0.413
(0.172)
income 0.016
(0.027)
male 0.092
(0.156)
Constant 1.747
(1.070)
Observations 551
R
2
0.202
Adjusted R
2
0.165
Residual Std. Error 1.558 (df = 259)
F Statistic 5.448
(df = 12; 259)
Note:
p<0.1;
p<0.05;
p<0.01
154
Democrats. No explanation is oered for this nding, but it is noteworthy. All other variables
lacked statistical signicance. The results are largely similar, with each additional point on
the GAD-7 being associated with an increase of 0.342 on one's contribution frequency, which
is a scale ranging from 1-4. For example, a GAD-7 Average of 3 is associated with a full
point increase on the contribution frequency scale. So if an individual with an average GAD-
7 score of 0 reported they were a 1 on the contribution frequency scale (meaning they've
contributed once or twice), an individual with all the same characteristics except a GAD-7
Average score 3 points higher (meaning they feel anxious nearly every day) would be expected
to be a 2 on the contribution frequency scale (meaning they contribute sometimes). This
eect of anxiousness on campaign contributions is not very strong, but it is still positive and
statistically signicant.
To test hypothesis 2, that anxiety has a positive relationship with avoiding campaign
communication, similar logistic regressions were run with a change in the dependent variable
to be campaign avoidance. Results are visualized in Figure 5.9, and a complete regression
table is reported in Table 5.3.
155
male
income
republican
democrat
conservative
liberal
education
political interest
log(age)
inhibitory anxiety
prospective anxiety
GAD−7 Avg.
−2 −1 0 1
Coefficient Estimate
Model
Avoid Contact
Effect of Anxiety on Propensity to Avoid Campaign Contact
Figure 5.9: Visualizations of the regression model describing the eects of anxiety on propen-
sity to avoid campaign contact.
These results oer tacit support to hypothesis 2. The only 2 variables with statistical sig-
nicance are GAD-7 Avg. and Democratic partisanship. The constant of this model is -0.40,
which converts into a baseline probability of 0.4, meaning that many people avoid campaign
contact. This is likely the reason that many of these variables do not have statistical signi-
cance: a lot of people avoid campaign contact. Still, the impact of higher generalized anxiety
156
Table 5.3: Logistic Rgression: Eect of Anxiety on Propensity to Avoid Campaign Contact
Dependent variable:
Avoided Campaign Contact
GAD-7 Avg. 0.447
(0.160)
prospective anxiety 0.050
(0.021)
inhibitory anxiety 0.033
(0.026)
log(age) 0.557
(0.341)
political interest 0.053
(0.123)
education 0.169
(0.084)
liberal 0.293
(0.404)
conservative 0.649
(0.268)
democrat 1.357
(0.407)
republican 0.279
(0.231)
income 0.026
(0.036)
male 0.026
(0.209)
Constant 2.289
(1.433)
Observations 551
Log Likelihood 356.200
Akaike Inf. Crit. 738.399
Note:
p<0.1;
p<0.05;
p<0.01
157
scores and Democratic partisanship cannot be ignored. GAD-7 Average has a coecient of
0.39, which translates to an increased probability of 0.6. Democratic partisanship, on the
other hand, has a coecient of -0.98, which converts to a probability of 0.27. To reiterate,
because probabilities are on a scale from 0 to 1, there can be no negative probabilities. The
fact stands, though, that Democrats are actually less likely to avoid campaign contact than
any other groups. Conservative ideology and Republican partisanship have similar direc-
tional eects, although they're not statistically signicant. Overall, these results generally
support both hypothesis 1 and hypothesis 2.
One interesting nding in Table 5.3 is the very strong negative eect shown for Demo-
cratic partisanship on avoiding campaign contact. This suggests that Democrats are consid-
erably less likely than any other partisan or ideological group to unsubscribe from or block
emails from a campaign. The implications of this may help explain the predictions gener-
ated in chapter 4, which found that anxiety actually increases in eectiveness over time for
fundraising from Democrats. If Democrats are less likely to avoid campaigns, that could be
the reason why anxiety is more eective for fundraising from Democrats. While other people
may start avoiding the campaigns, Democrats are more inclined to keep opening campaign
emails.
Something to acknowledge regarding this nding is that among people who avoid cam-
paign contact, not everyone avoids contact because they nd the emails anxiety-inducing.
Figure 5.10 shows the reported reasons for avoiding campaign contact, with the percentage
of respondents agreeing with each reason in the y axis. It shows that the most popular reason
for avoiding campaign emails is nding the campaign emails annoying (41.9%), followed by a
mismatch in beliefs (31.4%), feeling one's contribution wouldn't make a dierence (23.2%),
nding the emails anxiety-inducing (15.9%), and nally nancial limitations from having
already given what one could (0.9%). While nding emails anxiety-inducing was a fairly
common reason for not contributing, nding emails annoying was an even more popular
158
reason. This suggests that many individuals may avoid anxiety-inducing emails not because
they themselves feel anxiety, but because they nd them annoying.
0.0
0.1
0.2
0.3
0.4
Belief Mismatch Emails Annoying Emails Anxiety−Inducing Gave What I Could Wouldn't Make A Difference
reason
percent
Reasons for Actively Avoiding Campaign Emails
Figure 5.10: Bar chart showing the most popular reasons survey respondents listed for
avoiding campaign contact.
While no specic hypotheses are made about reasons why people do not contribute,
the survey provided an opportunity to learn some common reasons why individuals do not
contribute to politics at all. While the descriptive results aren't generalizable to the general
159
population, it is informative to learn why many people say they do not contribute. Figure
5.11 shows common reasons contributors reported for not contributing. The percentage of
respondents that listed a reason for not contributing is depicted in the y axis. The most
common reasons listed for not giving are that people would rather buy other things (22%),
would rather give to charity instead (20.7%), the feeling that their contribution would make
no dierence (10%), a preference to keep one's political views hidden (5.7%) and disliking all
of the candidates (2.7%). While the most common reason for not contributing is a preference
to spend money on other things, the feeling that contributing makes no dierence is also
rather high at 10%. This shows that a sense of ecacy is important in contribution decisions.
160
0.00
0.05
0.10
0.15
0.20
Disliked Candidates Keep Preferences Hidden Makes No Difference Rather Buy Other Things Rather Give to Charity
reason
percent
Most Popular Reasons for Not Contributing
Figure 5.11: Bar chart showing the most popular reasons survey respondents listed for not
contributing.
161
5.5 Discussion
Results showed that higher levels of generalized anxiety are associated with greater likelihood
to contribute, which supports hypothesis 1, as well as greater likelihood to avoid campaign
contact, which supports hypothesis 2. Other variables with signicant positive relationships
to contribution likelihood are political interest, education, and Democratic partisanship,
none of which are surprising and t well with past research and even past ndings within
this dissertation. One surprising nding was the negative relationship between liberal ide-
ology and propensity to have contributed in the past year. The survey took place in May
2021, so the results may have varied had a similar survey been conducted during the 2020
campaign or under dierent contexts. The results didn't support hypothesis 2 as strongly
as they supported hypothesis 1, as the eect sizes for all variables were mainly statistically
insignicant. Still, there is some support to the hypothesis as higher GAD-7 average scores
were associated with increased likelihood to avoid campaign contact. As mentioned earlier,
the baseline probability that a subject had engaged in a campaign avoidance behavior was
0.4 to start, so it is a very common behavior. Increasing the number of survey participants
may help show which variables actually have an impact on campaign avoidance. Overall,
these ndings suggest that the theory of anxious avoidance is generally supported - people
with higher levels of anxiety are more likely to make political contributions than less anxious
individuals even when controlling for major political factors like political interest, education,
partisanship, ideology, and income.
This research shows clearly that individuals with high levels of anxiety are more likely to
contribute to campaigns. These contributions, while small, mean that campaigns can rely
upon anxiety among individuals to prime the fundraising pump. This means that anxious
individuals may bear more of the costs of small dollar fundraising appeals from political
campaigns. Taken together with results from the previous chapter, we can conclude that to
some degree, campaigns rely on anxiety-driven appeals to raise money. Chapter 4 showed
162
that anxious emails are sent out at a constant rate at all times of the year, and increasing
in the months before election day. However, results from this chapter also show that this
campaign strategy may be
eeting. Anxious individuals are much more likely to avoid
campaign contact { measured as ignoring emails, unsubscribing, and similar actions { than
non-anxious individuals. The strategy of inducing anxiety via emails leads to initial higher
donations among anxious people, but also suggests have diminishing returns. Consistent
priming of anxiety is likely to lead to anxious donors turning away from campaigns. The
long-term eectiveness of creating anxiety among potential contributors may not be a viable
long-term strategy, given my results.
The short-term eects of anxiety versus long-term campaign avoidance among anxious
people should be considered. For example, an important question for future research is
understanding if people who are exposed to high degrees of anxiety-inducing communication
turn away from politics in general after some period of time. Political disengagement after
repeated exposure to anxiety-inducing political communication could be a phenomenon akin
to activist burnout, a process in which individuals who are strongly committed to political
goals lose their passion and commitment (Pines, 1994). While political contributors are
dierent from formal political activists, political contributions are a form of political activism
and versions of burnout may apply to political contributors.
Common causes for burnout are exhaustion, cynicism, and inecacy (Maslach and Gomes,
2006). Exhaustion is the individual stress component to political activism. In a political con-
tribution setting, this could come from repeated engagement with anxiety-inducing emails
that are emotionally taxing, as well as exhaustion of nancial resources from contributing.
Cynicism is the detachment or negative feeling towards one's own work. In a political con-
tribution setting, this could come from repeatedly contributing to candidates seen as \the
lesser of two evils", or contributing to stop a bad candidate rather than elect a candidate
who one supports with genuine passion. Inecacy is the sense of a lack of achievement from
one's work. In a political contribution setting, inecacy likely occurs constantly in the form
163
of contributors doubting that their $20 is truly making a dierence, as well as their pre-
ferred candidate losing an election. Inecacy in particular is likely a common feeling among
political contributors, and should be explored more. One idea for a study is examining dif-
ferences in political engagement among individuals who subscribed to campaign email lists
and individuals who did not subscribe to campaign email lists. If anxiety-inducing emails
lead to long-term political disengagement, campaigns should take note and possibly change
their fundraising strategies.
If somebody contributes after having an emotional response to an email, they may expe-
rience cognitive dissonance when dierent versions of essentially the same anxiety-inducing
email arrive in their inbox the next day, week, month, etc. When there is always a fundrais-
ing deadline that contributions need to be \rushed" to, or when there is always a political
crisis that requires donations, it can give contributors the sense that their contribution
doesn't actually matter. This is particularly likely with small campaign contributions, be-
cause contributors already have a good sense that their contribution is not changing an
election outcome. In essence, the risk of sending so many anxiety-inducing campaign emails
is short-term fundraising gains at the cost of long-term, sustainable political contributors.
5.6 Limitations
A major limitation of this study is the lack of representativeness of MTurk samples. Because
MTurk samples skew male, liberal, and young, generalizing results in MTurk studies to
larger populations is not recommended. While weighting the regressions attempted to work
past the core issues of the MTurk sample's representativeness, one should be careful about
making generalizations from these results. Additionally, these survey results bring up an
important limitation to the explanatory power of the theory of anxious avoidance. Many
individuals engaged in campaign avoidance behavior, and avoiding campaigns due to self-
164
reported anxiety was one of the less common reasons given for unsubscribing from or blocking
a campaign. Avoiding campaigns simply because they found their emails annoying.
165
Chapter 6
Conclusion
6.1 Review
The central aim of this research is to answer one fundamental question: why do regular people
contribute to political campaigns, candidates, and causes? This dissertation contains three
major studies, each of which helps answer several facets of this central research question.
6.1.1 Study 1
In chapter two, regression analyses performed on ANES and CCES survey data showed
several demographic, political, and social characteristics that are most strongly associated
with making a political contribution. This study clearly and strongly showed that solicita-
tion/contact, income, education, political interest, frequency of news consumption, strength
of Democratic partisanship, and strength of liberal ideology have large positive eects on
one's propensity to make a political contribution. The contributions made to the politi-
cal behavior literature from this study are the emphasis of Democratic partisanship, liberal
ideology, and contact/solicitation on the propensity to contribute.
166
In chapter three, a mix of descriptive statistics and regression analyses performed on
CCES data show that a greater proportion of Democrats and Liberals are contributors than
are Republicans and conservatives. An interesting part of this nding is that this eect is
not being driven by an increase in contributing among Democrats and liberals, but rather
by a drop in the proportion of Republicans and conservatives that contribute over time.
Exploration into contact rates via email/text oers one potential explanation. Chapter two
shows that contact via email/text is by far the most eective form of contact in terms of
producing small contributions, and chapter three shows that contact via email/text among
Republicans and conservatives is considerably lower than it is among Democrats and liber-
als. Additionally, chapter three found that a stronger association exists between Democratic
contact and contributing than Republican contact and contributing. When Democrats con-
tact individuals (but especially Democrats) they are signicantly more likely to produce a
small contribution than when Republicans contact individuals (even Republicans).
6.1.2 Study 2
Chapter four reported the eects of candidates/campaigns sending out anxiety-inducing
emails on small-dollar (sub-$200) fundraising returns. Results showed that while anxiety-
inducing emails are eective for some candidates during some periods of the campaign,
anxiety-inducing emails have a negative curvilinear relationship with small-dollar fundraising
over time. Specically, in the last several weeks of a campaign, sending any amount of
anxiety-inducing emails is predicted to raise less money than sending 0 anxiety-inducing
emails for all candidates except Democrats. Among Democratic candidates, sending more
anxiety-inducing emails towards the end of the 2020 campaign was modeled to be the best
fundraising strategy. It is unknown if Democrats can always expect anxiety-inducing emails
to be an eective fundraising strategy, or if the eectiveness of anxiety is more linked to
whichever political party is out of power or Trump-specic eects.
167
Donald Trump's election in 2016 caused stress and anxiety among a large number of
Americans (Markowitz et al., 2017), among groups ranging from students (Rogers et al.,
2017a) to Latina/os (Jones et al., 2021) to Democrats and liberals who nd politics important
(Pitcho-Prelorentzos et al., 2018). Additional research could examine the eect of anxiety-
inducing emails among partisans out of power in other years. For example, a replication
or expansion of this research could be conducted in 2024 to better determine if it is only
Democrats who benet from anxiety towards the end of a campaign, if any party out of
power can utilize anxiety for fundraising, or if the eects were specic to Trump and the
larger environment of the 2020 election (including COVID).
6.1.3 Study 3
Results from chapter 5 also support the theory of anxious avoidance as it relates to small-
donor contribution behavior. Findings from a novel survey suggest that higher levels of
general anxiety have a strong positive impact on both an individual's likelihood to make
a political contribution and engage in campaign avoidance. I suggest that the reason why
is because anxious people use contributions as a way to soothe their anxiety and gain a
sense of control over an uncertain political world. However, because there is a nite limit
on how much people are willing to spend for an intangible good, anxious individuals cannot
keep contributing to soothe their anxiety. And still, politically active Americans live in a
country where the average political candidate is sending around 2-3 anxious emails a week,
no matter what the news is. The theory of anxious avoidance suggests the natural response
for an anxious individual under these circumstances is to ignore these campaigns and their
emails that warn of weekly impending doom.
168
6.2 Discussion
This dissertation makes two major contributions: one relates to the demographics of con-
tributors, and the other relates to psychological anxiety. Regarding demographics, the rst
study nds that Democratic partisanship and liberal ideology are better predictors of who
contributes than Republican partisanship and conservative ideology. A possible explanation
of this nding is the observation that Democrats and liberals contact people via email/text
at higher rates than Republicans and conservatives do. This may be the case because the Re-
publican party relies on fundraising from a smaller number of more wealthy donors, which
became a permanently viable strategy after the 2010 Citizens United ruling (Krumholz,
2013).
One possible reason for the dierence in both Democratic contributions and anxiety's
eectiveness in fundraising from Democrats is a result from Table 5.3 in chapter 5. The
table shows the eect of anxiety, education, income, gender, political interest, partisanship,
and ideology on the propensity to avoid campaign contact. Democrats and conservatives both
had statistically signicant negative eects on the likelihood of avoiding campaign contact,
but the eect size for Democrats (-0.989) was more than double that of conservatives (-0.476).
This suggests that Democrats are less likely to block or unsubscribe from campaigns. Greater
Democratic refusal to block/unsubscribe may help explain why anxiety-inducing emails are
more eective for fundraising from Democrats at the end of the campaign. While the theory
of anxious avoidance would suggest that people are likely to avoid things that are anxiety-
inducing, fear of being uninformed or missing something from the news could have driven
Democrats to not avoid anxiety-inducing campaigns.
This result would not be unique: in a study of anxiety-inducing media in Israel, Bodas
et al. (2015) nd that even though many Israelis nd constant news coverage of bombings
to be very anxiety-inducing, they kept watching the news at high rates because they feared
169
missing out on information. In the study, Bodas et al. (2015) surveyed Israelis during Oper-
ation Protective Edge, a ground invasion of Gaza. They found that 87.2% of the population
tuned in to newscasts about the invasion, and 76.7% of viewers increased their news con-
sumption. More than 70% of the Israeli viewers in Bodas et al. (2015) reported nding
the newscasts stressful and also indicated they would not try to avoid watching them. It's
possible that a similar eect occurred during the 2020 election. Coverage of Donald Trump
dominated the media during Trump's entire presidency, so it is feasible that Democrats
were made very anxious by media coverage of Trump and campaign emails, but refused to
tune out because of the fear of missing out on valuable political information. This could
also explain why many Democrats did not unsubscribe or block anxiety-inducing campaign
communication - they didn't want to miss out on anything. This invites additional research
on anxiety and campaign contributions (and other political behavior) in 2024 and beyond.
Comparing the eectiveness of anxiety-inducing emails in campaigns with candidates who are
very anxiety-inducing against candidates who are not very anxiety-inducing would provide
valuable insight.
There are other possible explanations for the asymmetry in partisan/ideological contri-
bution rates, but they are not explored in this dissertation. This research invites further
exploration of why a larger proportion of Democrats and liberals contribute than Repub-
licans and conservatives. Some avenues to explore are dierences in values, dierences in
allocation of expendable money (perhaps Republicans use their money to donate to charities
instead), and dierences in the fundraising strategies and apparati of Republican campaigns.
For example, Democrats could value communal eorts more, Republicans could donate to
their church more, and Republican campaigns could have less sophisticated strategies for
fundraising from small donors. Whatever additional explanations may be, more research on
this partisan asymmetry is warranted.
Regarding anxiety, this dissertation takes a novel approach and seeks to explain small-
dollar political contributions as a function of anxiety. Much of this research is founded in
170
exploring the theory of anxious avoidance in a campaign nance setting, and at the heart of
this theory are contradicting predictions and expectations. The rst prediction of the theory
is that people who are made anxious by politics make political contributions to soothe
their political anxiety. The second prediction of the theory is that people who are made
anxious by politics will eventually stop contributing because over time, repeated exposure
to anxiety-inducing political communication causes individuals to see the communication
itself as a source of anxiety. The evidence presented in chapter 4 shows, to varying degrees,
a robust negative curvilinear relationship between anxiety-inducing emails and fundraising
returns. While dierent models showed varying eects, the models with the best t predicted
that for some parts of the campaign (particularly the beginning), anxiety is more eective
at generating small-dollar political contributions than not using anxiety. However, these
models also showed a drop-o over time, with sending 0 anxiety-inducing emails becoming
the best fundraising strategy for all groups except Democrats by the last month of the
campaign. Some of the models reported in chapter 4 predicted that anxiety would only lose
its eectiveness in the nal several weeks of the campaign, while others predicted a loss in
eectiveness as early as 10 or 20 weeks in. Regardless, the clear result was that sending
anxiety-inducing fundraising emails was only eective at some points of the campaign - and
never at the end of a campaign (except for Democrats).
Yet, descriptive statistics report that campaigns send anxiety-inducing emails constantly
throughout a campaign, but particularly towards the end. An important question regarding
campaign behavior is why they constantly send out anxiety-inducing emails, especially to-
wards the end of a campaign, if they don't seem to work? While a denitive answer to this
question is unknowable without proprietary campaign data and decision-making varies by
campaign, it's possible that because anxiety-inducing emails are constantly being sent out
and contributions are constantly received, there appears to be positive feedback for send-
ing out anxiety-inducing emails. That is, if anxiety-inducing emails are being sent out and
contributions are being received, that could give a misleading indication to campaigns that
171
anxiety-inducing emails are driving the contributions. Another possibility is contributions
being driven by new donors: if small donors are essentially tapped out either nancially or
psychologically, a viable campaign strategy could be to `amp up' the anxiety to try to shock
non-donors into giving for the rst time. An exploration of this possible explanation would
require tracking individuals' repeated contributions, which was outside the scope of this
research.
1
It's also possible that campaigns are aware that anxiety-inducing emails cause
people to avoid the campaigns but continue to send anxious emails anyway because the
potential for shocking new donors into contributing is worth the risk of alienating existing
contributors. More research into this topic is warranted, and receiving information directly
from campaigns would be a good source of information. Still, based on some public state-
ments from fundraisers like Whitney (2018, 2017), campaigns could simply be unaware of
or unconvinced of the negative eects of anxiety-inducing emails. Additionally, it's possible
that campaigns are so focused on short-term fundraising gains and making it through an-
other week of the campaign that the alienation of existing contributors just isn't considered.
Yet another possibility is if campaigns are driven by outlier emails. For example, if cam-
paigns nd that their best fundraising hauls come from sending anxiety-inducing emails, that
may encourage them to send many anxiety-inducing emails, even if most anxiety-inducing
emails don't raise more money than non-anxious emails. If a campaign sees that one anxiety-
inducing email produced 10,000 contributions, they may reasonably consider sending many
anxiety-inducing emails in the hopes of getting another high-performing email again.
Another facet to consider in all this is time. Chapter 4 studied contributions and emails
from July 2019 - November 2020, but there are many opportunities to expand upon this re-
search. 2020 was a novel election year with the combination of both Donald Trump's looming
re-election and the COVID-19 pandemic, causing wide-scale anxiety and stress among the
public (American Psychological Association, 2020). It is unclear if the same patterns of
1
Specically, the IRB application for this research stated that all data would be anonymized and iden-
tifying information about individuals would not be collected. The only way to track individuals with FEC
data is by combining contributor name, zip code, state, and employer data into a unique trackable prole.
172
anxiety-inducing emails (and contributor responses to those emails) would appear in 2022,
2024, or beyond, with potentially less anxiety-inducing election conditions. For example,
the eectiveness of anxiety may change in calmer elections without unique candidates like
Donald Trump or unique circumstances such as a global pandemic.
Another aspect of time that should be studied in further research is political disengage-
ment or burnout among political contributors. The discussion section in chapter 5 discussed
burnout and the possibilities for burnout among political contributors as a result of anxiety-
inducing campaign communication. Future research on political contributor engagement
should answer questions such as:
• Does burnout or disengagement occur among political contributors?
• If burnout or disengagement does occur among political contributors, why does it
happen and on what time scales does it happen?
• Do disengaged contributors come back every election cycle, or is there any permanent
or long-lasting attrition among burned-out contributors?
In conclusion, this research has shown that Democrats are more likely to contribute than
Republicans, possibly due to Democrats being contacted more via email/text, the most ef-
fective method of fundraising. Additionally, these ndings show Democrats are less likely
to avoid campaign contact, which may also help explain why Democrats contribute more
than any other group in response to anxious emails at the end of a campaign. However,
this nding should be replicated in 2024 to determine if Democratic contributors respond
more to anxiety-inducing fundraising solicitations, or if any partisans who are out of power
respond more to anxiety appeals in fundraising messages. This research also found that for
all candidates except Democratic candidates, anxiety-inducing fundraising messages have a
negative curvilinear relationship with fundraising returns. This means that over time, the
eectiveness of sending anxiety-inducing emails on small-dollar fundraising returns gener-
ally decreases, particularly in the nal weeks of a campaign. Another important nding is
173
that anxious people are more likely to both contribute and avoid campaign contact, which
helps explain why anxiety-inducing emails have a negative relationship with small-dollar
fundraising returns over time. Finally, this dissertation highlighted the normative harm of
campaigns constantly sending anxiety-inducing emails to a mass audience and invites further
research on any long-term detrimental eects of anxious campaign communication on polit-
ical engagement. To brie
y summarize this dissertation with a broad statement: Democrats
are more likely to make small political contributions than Republicans, anxious people are
more likely to both contribute and avoid campaign contribution, and anxiety-inducing emails
generally lose their eectiveness over time.
6.3 Limitations and Suggestions for Future Research
There are a number of major limitations to this study. First, the data used in this dissertation
are observational, so strong causal claims cannot be made. Conducting an experiment em-
bedded with a campaign would be an ideal way to study the eectiveness of anxiety-inducing
fundraising on various types of individuals. Additionally, conducting free-form qualitative
interviews with contributors and non-contributors to ask them why they contribute or do
not contribute would be enlightening and helpful in generating more research questions that
get at the core of why some people contribute and others do not. This research also touched
on the idea of credibility but didn't approach the subject with depth. Future research into
the fundraising eectiveness of `credible' vs `dubious' candidates would be interesting, and
likely plays an important role in campaign contribution decisions.
There are many questions that are related to this work but were unable to be asked due
to time and budget constraints, or simply being too far outside the scope of this project.
The biggest suggestion made by this research is to simply invite more research on small
contributors and more research on anxiety's role in political behavior. Regarding small
donors, political ecacy is a variable of interest and studies on the role of political ecacy
174
in contribution decisions would be an excellent extension of this research. One important
question is the contribution rates of individuals living in noncompetitive elections. People
who live in districts or states where they feel their vote doesn't mean much may contribute
to races in more competitive locations. For example, it would be interesting to explore
Californians who felt compelled to contribute to the Georgia Senate races in 2020 because
they felt their vote wouldn't mean much. Instead of fullling civic duty through voting,
many individuals feel compelled to fulll their civic duty elsewhere.
This project did not investigate the harm done to individuals as a result of constant
exposure to anxiety-inducing campaign messages. For the work to have a greater social
impact, I thought it better to rst understand if anxiety-inducing emails may actually be
ineective for campaign fundraising, because campaigns have a stronger incentive to raise
money than they do to not cause psycholological harm. Because this work has shown evidence
of anxiety-inducing fundraising decreasing in eectiveness over time, research into the harm
done as a result of exposure to constant anxiety-inducing campaign is warranted. If new
research can show that anxiety-inducing political campaign emails cause harm to individuals
and nd additional evidence of the ineectiveness of anxiety for fundraising, that would
provide a stronger incentive for campaigns to reduce the number of anxiety-inducing emails
they send.
175
Bibliography
Abramowitz, A. I. and Saunders, K. L. (2008). Is polarization a myth? The Journal of
Politics, 70(2):542{555.
Ackerman, B. A. and Fishkin, J. S. (2004). Deliberation day. Yale University Press.
Albertson, B. and Gadarian, S. K. (2015). Anxious politics: Democratic citizenship in a
threatening world. Cambridge University Press.
Aldrich, J. H. (1993). Rational choice and turnout. American journal of political science,
pages 246{278.
Allington, D., Duy, B., Wessely, S., Dhavan, N., and Rubin, J. (2020). Health-protective
behaviour, social media usage and conspiracy belief during the covid-19 public health
emergency. Psychological medicine, pages 1{7.
American Psychological Association (2020). 2020 presidential election a source of signicant
stress for more americans than 2016 presidential race.
Andreoni, J. (1990). Impure altruism and donations to public goods: A theory of warm-glow
giving. The economic journal, 100(401):464{477.
Ansolabehere, S., De Figueiredo, J. M., and Snyder Jr, J. M. (2003). Why is there so little
money in us politics? Journal of Economic perspectives, 17(1):105{130.
176
Apollonio, D. E. and Raja, R. J. L. (2004). Who gave soft money? the eect of interest
group resources on political contributions. The Journal of Politics, 66(4):1134{1154.
Asmundson, G. J. (1999). Anxiety sensitivity and chronic pain: Empirical ndings, clinical
implications, and future directions. Anxiety sensitivity: Theory, research, and treatment
of the fear of anxiety, pages 269{285.
Asmundson, G. J., Vlaeyen, J. W. S., Vlaeyen, J. W., and Crombez, G. (2004). Understand-
ing and treating fear of pain. Oxford University Press, USA.
Badia, P., Mcbane, B., and Suter, S. (1966). Preference behavior in an immediate versus vari-
ably delayed shock situation with and without a warning signal. Journal of Experimental
Psychology, 72(6):847.
Bakir, V. (2020). Psychological operations in digital political campaigns: Assessing cam-
bridge analytica's psychographic proling and targeting. Frontiers in Communication,
5:67.
Barber, M. J., Canes-Wrone, B., and Thrower, S. (2017). Ideologically sophisticated donors:
Which candidates do individual contributors nance? American Journal of Political
Science, 61(2):271{288.
Barber a, P., Boydstun, A. E., Linn, S., McMahon, R., and Nagler, J. (2019). Automated
text classication of news articles: A practical guide. Political Analysis, pages 1{24.
Barlow, D. H. (2004). Anxiety and its disorders: The nature and treatment of anxiety and
panic. Guilford press.
Barney, D. J. (2018). 2016 Subsample Weights Codebook. In CCES 2016 Subsample Weights.
Harvard Dataverse.
Bartlett, J., Smith, J., and Acton, R. (2018). The future of political campaigning. Demos.
177
Baum, M. A. and Groeling, T. J. (2009). War stories-The causes and consequences of public
views of war. Princeton University Press.
Baumeister, R. F., Bratslavsky, E., Finkenauer, C., and Vohs, K. D. (2001). Bad is stronger
than good. Review of general psychology, 5(4):323{370.
Biersack, R., Herrnson, P. S., and Wilcox, C. (1994). Risky Business?: PAC Decisionmaking
in Congressional Elections. ME Sharpe.
Birnbaum, J. H. (2000). The Money Men: The Real Story of Fund-raising's In
uence on
Political Power in America. Crown New York.
Blais, A. (2000). To vote or not to vote?: The merits and limits of rational choice theory.
University of Pittsburgh Pre.
Bodas, M., Siman-Tov, M., Peleg, K., and Solomon, Z. (2015). Anxiety-inducing media:
the eect of constant news broadcasting on the well-being of israeli television viewers.
Psychiatry, 78(3):265{276.
Bonica, A. (2014). Mapping the ideological marketplace. American Journal of Political
Science, 58(2):367{386.
Borkovec, T. D., Alcaine, O. M., and Behar, E. (2004). Avoidance theory of worry and
generalized anxiety disorder.
Bouton, L., Castanheira, M., and Drazen, A. (2018). A theory of small campaign contribu-
tions. Technical report, National Bureau of Economic Research.
Box-Steensmeier, J. M. and Grant, J. T. (1999). All in a day's work: The nancial rewards
of legislative eectiveness. Legislative Studies Quarterly, pages 511{523.
Boyle, M. P., Schmierbach, M., Armstrong, C. L., McLeod, D. M., Shah, D. V., and Pan, Z.
(2004). Information seeking and emotional reactions to the september 11 terrorist attacks.
Journalism & Mass Communication Quarterly, 81(1):155{167.
178
Brader, T. (2006). Campaigning for hearts and minds: How emotional appeals in political
ads work. University of Chicago Press.
Brader, T., Valentino, N. A., and Suhay, E. (2008). What triggers public opposition to
immigration? anxiety, group cues, and immigration threat. American Journal of Political
Science, 52(4):959{978.
Brady, H. E., Schlozman, K. L., and Verba, S. (1999). Prospecting for participants: Rational
expectations and the recruitment of political activists. American Political Science Review,
93(1):153{168.
Brady, H. E., Verba, S., and Schlozman, K. L. (1995). Beyond ses: A resource model of
political participation. American Political Science Review, 89(2):271{294.
Brown, C. W., Powell, L. W., and Wilcox, C. (1995). Serious money: Fundraising and
contributing in presidential nomination campaigns. Cambridge University Press.
Buhr, K. and Dugas, M. J. (2002). The intolerance of uncertainty scale: Psychometric
properties of the english version. Behaviour research and therapy, 40(8):931{945.
Burton, M. J., Miller, W. J., and Shea, D. M. (2015). Campaign craft: the strategies, tactics,
and art of political campaign management. ABC-CLIO.
Cacioppo, J. T. and Berntson, G. G. (1994). Relationship between attitudes and evalua-
tive space: A critical review, with emphasis on the separability of positive and negative
substrates. Psychological bulletin, 115(3):401.
Campbell, A., Converse, P. E., Miller, W. E., and Donald, E. (1960). Stokes. the american
voter. Ann Arbor, MI: University of Michigan Press.
Carleton, R. N., Norton, M. P. J., and Asmundson, G. J. (2007). Fearing the unknown:
A short version of the intolerance of uncertainty scale. Journal of anxiety disorders,
21(1):105{117.
179
Cebula, R. J., Duquette, C. M., and Mixon Jr, F. G. (2013). Battleground states and
voter participation in us presidential elections: an empirical test. Applied Economics,
45(26):3795{3799.
Center for Responsive Politics (2018). Donor demographics.
Chen, T., He, T., Benesty, M., Khotilovich, V., Tang, Y., Cho, H., et al. (2015). Xgboost:
extreme gradient boosting. R package version 0.4-2, 1(4).
Claessens, S., Feijen, E., and Laeven, L. (2008). Political connections and preferential access
to nance: The role of campaign contributions. Journal of nancial economics, 88(3):554{
580.
Clark, P. B. and Wilson, J. Q. (1961). Incentive systems: A theory of organizations. Ad-
ministrative science quarterly, pages 129{166.
Clawson, D., Neustadtl, A., and Scott, D. (1992). Money talks: Corporate PACs and political
in
uence. Basic books.
Confessore, N. and Thee-Brenan, M. (2015). Poll shows americans favor an overhaul of
campaign nancing.
Corrado, A. J., Malbin, M. J., Mann, T. E., and Ornstein, N. J. (2010). Reform in an
age of networked campaigns: How to foster citizen participation through small donors
and volunteers. Washington: Campaign Finance Institute. Available at: http://www.
cnst. org/books reports/Reform-in-an-Age-of-Networked-Campaigns. pdf. Accessed May,
28:2013.
Cox, G. W. and Magar, E. (1999). How much is majority status in the us congress worth?
American Political Science Review, 93(2):299{309.
180
Craske, M. G., Rapee, R. M., Jackel, L., and Barlow, D. H. (1989). Qualitative dimen-
sions of worry in dsm-iii-r generalized anxiety disorder subjects and nonanxious controls.
Behaviour research and therapy, 27(4):397{402.
Crigler, A. N. (2007). The aect eect- Dynamics of emotion in political thinking and
behavior. University of Chicago Press.
Culberson, T., McDonald, M. P., and Robbins, S. M. (2018). Small donors in congressional
elections. American Politics Research, page 1532673X18763918.
Culberson, T., McDonald, M. P., and Robbins, S. M. (2019). Small donors in congressional
elections. American Politics Research, 47(5):970{999.
Cummings, N. and Cummings, G. (2004). Strategy and tactics for campaign fundraising.
Campaigns and elections: American style, pages 67{82.
Damasio, A. R. (1994). Descartes' error: Emotion, rationality and the human brain.
Dillard, J. P. and Shen, L. (2005). On the nature of reactance and its role in persuasive
health communication. Communication Monographs, 72(2):144{168.
Downs, Anthony, e. a. (1957). An economic theory of democracy.
Edelman, M. J. (1985). The symbolic uses of politics. University of Illinois Press.
Endersby, J. W. and Munger, M. C. (1992). The impact of legislator attributes on union
pac campaign contributions. Journal of Labor Research, 13(1):79{97.
Ernst, N., K uhne, R., and Wirth, W. (2017). Eects of message repetition and negativity
on credibility judgments and political attitudes. International Journal of Communication,
11:21.
Evans, D. (1988). Oil pacs and aggressive contribution strategies. The Journal of Politics,
50(4):1047{1056.
181
Evers-Hillstrom, K. (2018). Democrats are rejecting corporate pacs: Does it mean anything?
Eysenck, M. W. (1992). Anxiety: The cognitive perspective. Psychology Press.
Fiorina, M. P. (2016). The political parties have sorted. A Hoover Institution Essay on
Contemporary American Politics. Series, (3).
Fisher, W. R. (1985). The narrative paradigm: An elaboration. Communications Mono-
graphs, 52(4):347{367.
Fisher, W. R. (1988). The narrative paradigm and the interpretation and assessment of
historical texts. The Journal of the American Forensic Association, 25(2):49{53.
Fishkin, J. S. (1997). The voice of the people: Public opinion and democracy. Yale university
press.
for Responsive Politics, C. (2021). Large versus small individual donations. Data re-
trieved from the Center for Responsive Politics at https://www.opensecrets.org/
elections-overview/large-vs-small-donations?cycle=2020&type=A.
Francia, P. L., Goldberg, R. E., Green, J. C., Herrnson, P. S., and Wilcox, C. (1999).
Individual donors in the 1996 federal elections. Financing the 1996 Election, pages 127{
154.
Francia, P. L., Green, J. C., Wilcox, C., Herrnson, P. S., Powell, L. W., et al. (2003).
The nanciers of congressional elections: Investors, ideologues, and intimates. Columbia
University Press.
Freeston, M. H., Rh eaume, J., Letarte, H., Dugas, M. J., and Ladouceur, R. (1994). Why
do people worry? Personality and individual dierences, 17(6):791{802.
Friedman, J., Hastie, T., Tibshirani, R., et al. (2000). Additive logistic regression: a statisti-
cal view of boosting (with discussion and a rejoinder by the authors). Annals of statistics,
28(2):337{407.
182
Friedman, J. H. (2001). Greedy function approximation: a gradient boosting machine.
Annals of statistics, pages 1189{1232.
Frijda, N. H. (1988). The laws of emotion. American psychologist, 43(5):349.
Frost, A. M. (2013). The american donor: an exploration of the modern individual donor.
Garsten, B. (2009). Saving persuasion: A defense of rhetoric and judgment. Harvard Uni-
versity Press.
Gerber, A. S. and Green, D. P. (2000). The eects of canvassing, telephone calls, and direct
mail on voter turnout: A eld experiment. American political science review, 94(3):653{
663.
Gerber, A. S., Huber, G. A., Doherty, D., and Dowling, C. M. (2012). Personality and the
strength and direction of partisan identication. Political Behavior, 34(4):653{688.
Goodrich, B., Kurkiewicz, D., and Rinker, T. (2020). Package `qdap'. Technical report,
Tech. Rep., 2018.[Online].
Gordon, S. C., Hafer, C., and Landa, D. (2007). Consumption or investment? on motivations
for political giving. The Journal of Politics, 69(4):1057{1072.
Graf, J. (2006a). Donors and fundraising in the 2004 presidential campaigns. National Civic
Review, 95(3):35{41.
Graf, J. (2006b). Small donors and online giving: A study of donors to the 2004 presidential
campaigns. Institute for Politics, Democracy & the Internet, Graduate School of . . . .
Grant, J. T. and Rudolph, T. J. (2002). To give or not to give: Modeling individuals'
contribution decisions. Political Behavior, 24(1):31{54.
Grant, T. J. and Rudloph, T. J. (2002). To give or not to give: Modeling individuals'
contribution decisions. Political Behavior, 24(1):31{54.
183
Greco, V. and Roger, D. (2003). Uncertainty, stress, and health. Personality and Individual
dierences, 34(6):1057{1068.
Green, J. (2012). The science behind those obama campaign e-mails.
Gross, J. J. (2013). Handbook of emotion regulation. Guilford publications.
Gross, J. J. and Thompson, R. A. (2007). Emotion regulation: Conceptual foundations.
Grupe, D. W. and Nitschke, J. B. (2013). Uncertainty and anticipation in anxiety: an
integrated neurobiological and psychological perspective. Nature Reviews Neuroscience,
14(7):488{501.
Habermas, J. and Habermas, J. (1991). The structural transformation of the public sphere:
An inquiry into a category of bourgeois society. MIT press.
Hale, W., Richmond, M., Bennett, J., Berzins, T., Fields, A., Weber, D., Beck, M., and
Osman, A. (2016). Resolving uncertainty about the intolerance of uncertainty scale{
12: Application of modern psychometric strategies. Journal of personality assessment,
98(2):200{208.
Hall, R. L. and Wayman, F. W. (1990). Buying time: Moneyed interests and the mobilization
of bias in congressional committees. American political science review, 84(3):797{820.
Hamilton, A., Madison, J., and Jay, J. (2008). The federalist papers. Oxford University
Press.
Hansen, W. L. and Rocca, M. S. (2019). The impact of citizens united on large corporations
and their employees. Political Research Quarterly, 72(2):403{419.
Hasen, R. L. (2016). Plutocrats united: Campaign money, the Supreme court, and the
distortion of American elections. Yale University Press.
184
Herrera, H., Morelli, M., and Nunnari, S. (2016). Turnout across democracies. American
Journal of Political Science, 60(3):607{624.
Herrera, H., Morelli, M., and Palfrey, T. (2014). Turnout and power sharing. The Economic
Journal, 124(574):F131{F162.
Hersh, E. (2015). Hacking the Electorate: How Campaigns Perceive Voters. Cambridge
University Press.
Heyman, D. R. (2016). Nonprot Fundraising 101: A Practical Guide to Easy to Implement
Ideas and Tips from Industry Experts. John Wiley & Sons.
Huey-Burns, C. (2014). The method & madness of those pesky fundraising emails.
Imho, R. and Lamberty, P. (2020). A bioweapon or a hoax? the link between distinct con-
spiracy beliefs about the coronavirus disease (covid-19) outbreak and pandemic behavior.
Social Psychological and Personality Science, 11(8):1110{1118.
Iyengar, S. (1991). Is anyone responsible?: How television frames political issues. University
of Chicago Press.
Jarymowicz, M. and Bar-Tal, D. (2006). The dominance of fear over hope in the life of
individuals and collectives. European Journal of Social Psychology, 36(3):367{392.
Jenkins, N. R. (2021). Minimizing whose in
uence? how rejecting pac contributions aects
contribution patterns.
Johnson, B. N. (2013). Political giving: Making sense of individual campaign contributions.
FirstForumPress, onc.
Jones, B. S., Sherman, J. W., Rojas, N. E., Hosek, A., Vannette, D. L., Rocha, R. R., Garc a-
Ponce, O., Pantoja, M., and Garc a-Amador, J. M. (2021). Trump-induced anxiety among
latina/os. Group Processes & Intergroup Relations, 24(1):68{87.
185
Jones, R. S. (2019). Contributing as participation. In Money, Elections, and Democracy,
pages 27{45. Routledge.
Joulfaian, D. and Marlow, M. L. (1991). Incentives and political contributions. Public
Choice, 69(3):351{355.
Keena, A. and Knight-Finley, M. (2019). Are small donors polarizing? a longitudinal study
of the senate. Election Law Journal: Rules, Politics, and Policy.
Kenski, K., Hardy, B. W., and Jamieson, K. H. (2010). The Obama victory: How media,
money, and message shaped the 2008 election. Oxford University Press.
Kern, M. (1989). 30-second politics: Political advertising in the eighties. Praeger Publishers.
Kessler, R. C., Brandenburg, N., Lane, M., Roy-Byrne, P., Stang, P. D., Stein, D. J., and
Wittchen, H.-U. (2005). Rethinking the duration requirement for generalized anxiety
disorder: evidence from the national comorbidity survey replication.
Kessler, R. C., Keller, M. B., and Wittchen, H.-U. (2001). The epidemiology of generalized
anxiety disorder. Psychiatric Clinics of North America, 24(1):19{39.
Kinder, D. R. (1994). Reason and emotion in american political life. Beliefs, reasoning, and
decision making: Psycho-logic in honor of Bob Abelson, pages 277{314.
Kinder, D. R. (2003). Communication and politics in the age of information.
Kinder, D. R., Abelson, R. P., and Fiske, S. T. (1979). Developmental Research on Candidate
Instrumentation: Results and Recomendations. Center for Political Studies, Institute for
Social Research, University of . . . .
Klandermans, B. (1984). Mobilization and participation: Social-psychological expansisons
of resource mobilization theory. American sociological review, pages 583{600.
186
Kosinski, M., Stillwell, D., and Graepel, T. (2013). Private traits and attributes are pre-
dictable from digital records of human behavior. Proceedings of the national academy of
sciences, 110(15):5802{5805.
Krumholz, S. (2013). Campaign cash and corruption: Money in politics, post-citizens united.
social research, 80(4):1119{1134.
La Raja, R. J. (2014). Political participation and civic courage: the negative eect of
transparency on making small campaign contributions. Political Behavior, 36(4):753{776.
La Raja, R. J. and Wiltse, D. L. (2012). Don't blame donors for ideological polarization of
political parties: Ideological change and stability among political contributors, 1972-2008.
American Politics Research, 40(3):501{530.
Lanzetta, J. T. and Driscoll, J. M. (1966). Preference for information about an uncertain
but unavoidable outcome. Journal of Personality and Social Psychology, 3(1):96.
Lasswell, H. D. (1927). Propaganda technique in the world war. Ravenio Books.
Lau, R. R. (1982). Negativity in political perception. Political behavior, 4(4):353{377.
Laugesen, N., Dugas, M. J., and Bukowski, W. M. (2003). Understanding adolescent worry:
The application of a cognitive model. Journal of abnormal child psychology, 31(1):55{64.
Lawton, T., Mcguire, S., and Rajwani, T. (2013). Corporate political activity: A literature
review and research agenda. International Journal of Management Reviews, 15.
Lerner, J. S. and Keltner, D. (2000). Beyond valence: Toward a model of emotion-specic
in
uences on judgement and choice. Cognition & emotion, 14(4):473{493.
Lerner, J. S. and Keltner, D. (2001). Fear, anger, and risk. Journal of personality and social
psychology, 81(1):146.
187
Levay, K. E., Freese, J., and Druckman, J. N. (2016). The demographic and political com-
position of mechanical turk samples. Sage Open, 6(1):2158244016636433.
Lewis, L. M. and Arkedis, J. (2014). Political Mercenaries: The Inside Story of How
Fundraisers Allowed Billionaires to Take Over Politics. Macmillan.
Lippmann, W. et al. (1943). Us foreign policy: Shield of the republic.
Lipsitz, K. (2009). The consequences of battleground and \spectator" state residency for
political participation. Political Behavior, 31(2):187.
Litmus Software (2017). The ultimate guide to preview text support.
Litmus Software (2019). The state of email engagement.
Lodge, M. and Taber, C. S. (2005). The automaticity of aect for political leaders, groups,
and issues: An experimental test of the hot cognition hypothesis. Political Psychology,
26(3):455{482.
L owe, B., Decker, O., M uller, S., Br ahler, E., Schellberg, D., Herzog, W., and Herzberg,
P. Y. (2008). Validation and standardization of the generalized anxiety disorder screener
(gad-7) in the general population. Medical care, pages 266{274.
Luo, M. and Zeleny, J. (2008). Obama, in shift, says he'll reject public nancing.
Lupia, A. and Menning, J. O. (2009). When can politicians scare citizens into supporting
bad policies? American Journal of Political Science, 53(1):90{106.
MacKuen, M., Wolak, J., Keele, L., and Marcus, G. E. (2010). Civic engagements: Resolute
partisanship or re
ective deliberation. American Journal of Political Science, 54(2):440{
458.
Magleby, D. B., Goodlie, J., and Olsen, J. A. (2018). Who donates in campaigns?: The
importance of message, messenger, medium, and structure. Cambridge University Press.
188
mailchimp (2017). Email marketing becnhmarks and statistics by industry.
Malbin, M. J. (2013). Small donors: Incentives, economies of scale, and eects. In The
Forum, volume 11, pages 385{411. De Gruyter.
Malbin, M. J., Brusoe, P. W., and Glavin, B. (2012). Small donors, big democracy: New
york city's matching funds as a model for the nation and states. Election Law Journal,
11(1):3{20.
Marcus, G. E. and MacKuen, M. B. (1993). Anxiety, enthusiasm, and the vote- the emo-
tional underpinnings of learning and involvement during presidential campaigns. American
Political Science Review, pages 672{685.
Marcus, G. E., Neuman, W. R., and MacKuen, M. (2000). Aective intelligence and political
judgment. University of Chicago Press.
Marcus, G. E., Neuman, W. R., MacKuen, M., and Crigler, A. N. (2008). The aect eect:
Dynamics of emotion in political thinking and behavior. University of Chicago Press.
Markowitz, J., Zhang, J., Liu, Y., Sun, L., Hao, Q., Wang, D., Guo, L., Zhang, B., Shimura,
A., Takaesu, Y., et al. (2017). Anxiety in the age of trump. Comprehensive Psychiatry,
74(1).
Maslach, C. and Gomes, M. E. (2006). Overcoming burnout.
Mason, J. W., Hartley, L. H., Kotchen, T. A., Mougey, E. H., Ricketts, P. T., and Jones,
L. G. (1973). Plasma cortisol and norepinephrine responses in anticipation of muscular
exercise. Psychosomatic Medicine.
Mason, L. (2015). \i disrespectfully agree": The dierential eects of partisan sorting on
social and issue polarization. American Journal of Political Science, 59(1):128{145.
McAulie, T. and Kettmann, S. (2007). What A Party!: My Life Among Democrats: Pres-
idents, Candidates, Donors, Activists, Alligators and Other Wild Animals. Macmillan.
189
McDermott, R. (2004). The feeling of rationality: The meaning of neuroscientic advances
for political science. Perspectives on politics, 2(4):691{706.
McDonald, M. P. (2004). Up, up and away! voter participation in the 2004 presidential
election. In The Forum, volume 2. De Gruyter.
Mera, K. and Ichimura, T. (2008). User's comment classifying method using self organiz-
ing feature map on healthcare system for diabetic. In Proceedings: Fourth International
Workshop on Computational Intelligence & Applications, volume 2008, pages 31{36. IEEE
SMC Hiroshima Chapter.
Merolla, J. L., Montalvo, J. D., Zechmeister, E. J., and Zizumbo-Colunga, D. (2012). Terror-
ism and democracy in latin america and the caribbean. Claremont Graduate University,
unpublished manuscript.
Milbrath, L. W. and Goel, M. L. (1977). Political participation: How and why do people get
involved in politics? Rand McNally College Pub. Co.
Milkis, S. M. and Rhodes, J. H. (2009). Barack obama, the democratic party, and
the future of the" new american party system". In The Forum, volume 7, page
0000102202154088841297. De Gruyter.
Milyo, J., Primo, D., and Groseclose, T. (2000). Corporate pac campaign contributions in
perspective. Business and Politics, 2(1):75{88.
Monat, A., Averill, J. R., and Lazarus, R. S. (1972). Anticipatory stress and coping reactions
under various conditions of uncertainty. Journal of personality and social psychology,
24(2):237.
Mondak, J. J. and D Halperin, K. (2008). A framework for the study of personality and
political behaviour. British Journal of Political Science, 38(2):335{362.
190
Mutch, R. E. (2016). Campaign nance: What everyone needs to know. Oxford University
Press.
Nabi, R. L. (1999). A cognitive-functional model for the eects of discrete negative emotions
on information processing, attitude change, and recall. Communication theory, 9(3):292{
320.
Nacos, B. L., Bloch-Elkon, Y., and Shapiro, R. Y. (2011). Selling fear: Counterterrorism,
the media, and public opinion. University of Chicago Press.
Nickerson, D. W. and Rogers, T. (2014). Political campaigns and big data. Journal of
Economic Perspectives, 28(2):51{74.
O'cass, A. (2002). Political advertising believability and information source value during
elections. Journal of Advertising, 31(1):63{74.
O'Keefe, E. (2014). From shaming to semi-stalking, democrats
ood inboxes for last-minute
campaign cash.
O'Rourke, B. (2019). This is a campaign of people. all people, no pacs. all people, no special
interests. all people, no corporations. very grateful to be running this race with you.
Osgood, C. E., Suci, G. J., and Tannenbaum, P. H. (1957). The measurement of meaning.
Number 47. University of Illinois press.
Overacker, L. (1974). Money in elections. Ayer Publishing.
Overton, S. (2004). The donor class: campaign nance, democracy, and participation. U.
Pa. L. Rev., 153:73.
Ovtchinnikov, A. V. and Pantaleoni, E. (2012). Individual political contributions and rm
performance. Journal of Financial Economics, 105(2):367{392.
191
Pervin, L. A. (1963). The need to predict and control under conditions of threat. Journal
of Personality.
Petrova, M., Sen, A., and Yildirim, P. (2017). Social media and political donations: New
technology and incumbency advantage in the united states.
Pew Research Center (2019). Internet/broadband fact sheet.
Pines, A. M. (1994). Burnout in political activism: An existential perspective. Journal of
Health and Human Resources Administration, pages 381{394.
Pitcho-Prelorentzos, S., Kaniasty, K., Hamama-Raz, Y., Goodwin, R., Ring, L., Ben-Ezra,
M., and Mahat-Shamir, M. (2018). Factors associated with post-election psychological
distress: The case of the 2016 us presidential election. Psychiatry research, 266:1{4.
Primo, D. M. (2002). Public opinion and campaign nance: Reformers versus reality. The
independent review, 7(2):207{219.
Ramsden, G. P. and Donnay, P. D. (2001). The impact of minnesota's political contribution
refund program on small-donor behavior in state house races. State and Local Government
Review, 33(1):32{41.
Redlawsk, D. P., Civettini, A. J., and Lau, R. R. (2007). Aective intelligence and voting:
Information processing and learning in a campaign. The aect eect: Dynamics of emotion
in political thinking and behavior, pages 152{179.
Rhodes, J. H., Schaner, B. F., and Raja, R. J. L. (2018). Detecting and understanding
donor strategies in midterm elections. Political Research Quarterly, 71(3):503{516.
Ridout, T. N. and Searles, K. (2011). It's my campaign i'll cry if i want to: How and when
campaigns use emotional appeals. Political Psychology, 32(3):439{458.
Riker, W. H. and Ordeshook, P. C. (1968). A theory of the calculus of voting. American
political science review, 62(1):25{42.
192
Rogers, J., Franke, M., Yun, J.-E. E., Ishimoto, M., Diera, C., Geller, R. C., Berryman, A.,
and Brenes, T. (2017a). Teaching and learning in the age of trump: Increasing stress and
hostility in america's high schools. UCLA IDEA.
Rogers, T. and Moore, D. A. (2014). The motivating power of under-condence:'the race is
close but we're losing'.
Rogers, T., Moore, D. A., and Norton, M. I. (2017b). The belief in a favorable future.
Psychological science, 28(9):1290{1301.
Roller, E. and National Journal (2014). A eulogy for the dccc's ridiculous fundraising emails:
It's time to move on.
Roseman, I. and Evdokas, A. (2004). Appraisals cause experienced emotions: Experimental
evidence. Cognition and emotion, 18(1):1{28.
Rosenstone, S. J., Hansen, J. M., and Reeves, K. (1993). Mobilization, participation, and
democracy in america.
Rozin, P. and Royzman, E. B. (2001). Negativity bias, negativity dominance, and contagion.
Personality and social psychology review, 5(4):296{320.
Ruiz, M. A., Zamorano, E., Garc a-Campayo, J., Pardo, A., Freire, O., and Rejas, J. (2011).
Validity of the gad-7 scale as an outcome measure of disability in patients with generalized
anxiety disorders in primary care. Journal of aective disorders, 128(3):277{286.
Sarbanes, John P. (2019). For the People Act of 2019.
Saxton, G. D. and Wang, L. (2014). The social network eect: The determinants of giving
through social media. Nonprot and voluntary sector quarterly, 43(5):850{868.
Schlozman, K. L., Verba, S., and Brady, H. E. (2012). The unheavenly chorus: Unequal
political voice and the broken promise of American democracy. Princeton University Press.
193
Schuman, H., Presser, S., and Ludwig, J. (1981). Context eects on survey responses to
questions about abortion. Public Opinion Quarterly, 45(2):216{223.
Sinclair, B. (2012). The social citizen: Peer networks and political behavior. University of
Chicago Press.
Smith, C. A. and Ellsworth, P. C. (1985). Patterns of cognitive appraisal in emotion. Journal
of personality and social psychology, 48(4):813.
Sorauf, F. J. (1988). Money in American elections. Glenview, Ill.: Scott, Foresman/Little,
Brown College Division.
Sorauf, F. J. (1992). Politics and money. American Behavioral Scientist, 35(6):725{734.
Spielberger, C. D. (1972). Anxiety as an emotional state. Anxiety-Current trends and theory,
pages 3{20.
Spitzer, R. L., Kroenke, K., Williams, J. B., and L owe, B. (2006). A brief measure for assess-
ing generalized anxiety disorder: the gad-7. Archives of internal medicine, 166(10):1092{
1097.
Stapinski, L. A., Abbott, M. J., and Rapee, R. M. (2010). Evaluating the cognitive avoidance
model of generalised anxiety disorder: Impact of worry on threat appraisal, perceived
control and anxious arousal. Behaviour Research and Therapy, 48(10):1032{1040.
Steenbergen, M. R. and Ellis, C. (2006). Fear and loathing in american elections: Context,
traits, and negative candidate aect. In Feeling politics, pages 109{133. Springer.
Swinson, R. (2006). The gad-7 scale was accurate for diagnosing generalised anxiety disorder.
Evidence-based medicine, 11(6):184{184.
Tallis, F. and Eysenck, M. W. (1994). Worry: Mechanisms and modulating in
uences.
Behavioural and Cognitive Psychotherapy, 22(1):37{56.
194
Thompson, S. C. (1981). Will it hurt less if i can control it? a complex answer to a simple
question. Psychological bulletin, 90(1):89.
Tversky, A. and Kahneman, D. (1991). Loss aversion in riskless choice: A reference-
dependent model. The quarterly journal of economics, 106(4):1039{1061.
Valentino, N. A., Hutchings, V. L., Banks, A. J., and Davis, A. K. (2008). Is a worried citizen
a good citizen? emotions, political information seeking, and learning via the internet.
Political Psychology, 29(2):247{273.
Verba, S., Schlozman, K. L., and Brady, H. E. (1995). Voice and equality: Civic voluntarism
in American politics. Harvard University Press.
Voigt, K., Ziegler, M., Gr unert-Fuchs, M., Bickel, U., and Fehm-Wolfsdorf, G. (1990). Hor-
monal responses to exhausting physical exercise: the role of predictability and controlla-
bility of the situation. Psychoneuroendocrinology, 15(3):173{184.
Wagner, D. (2012). The obama campaign's desperate fundraising emails worked.
Webster, B. A., Wilcox, C., Herrnson, P. S., Francia, P. L., Green, J. C., and Powell, L.
(2001). Competing for cash: The individual nanciers of congressional elections. In Playing
Hardball: Campaigning for the US Congress. Prentice Hall Upper Saddle River, NJ.
Westen, D. (2008). The political brain: The role of emotion in deciding the fate of the nation.
PublicAairs.
Whitney, M. (2017). The demcoratic party's looming fundraising crisis.
Whitney, M. (2018). How beto o'rourke raised a stunning $38 million in just three months.
Wilcox, C. (2008). Internet fundraising in 2008: A new model? In The Forum, volume 6.
De Gruyter.
195
Wittchen, H.-U., Zhao, S., Kessler, R. C., and Eaton, W. W. (1994). Dsm-iii-r general-
ized anxiety disorder in the national comorbidity survey. Archives of general psychiatry,
51(5):355{364.
Witte, K. (1992). Putting the fear back into fear appeals: The extended parallel process
model. Communications Monographs, 59(4).
Witte, K. and Allen, M. (2000). A meta-analysis of fear appeals: Implications for eective
public health campaigns. Health education & behavior, 27(5):591{615.
Zajonc, R. B. (1980). Feeling and thinking: Preferences need no inferences. American
psychologist, 35(2):151.
Zakowski, S. G. (1995). The eects of stressor predictability on lymphocyte proliferation in
humans. Psychology and Health, 10(5):409{425.
Zychal, K. (2019). Who will be on stage in the democratic debates? it depends on how you
measure popularity.
196
Appendix
Chapter 1 Appendix
Figure 6.1: Flowchart Describing the Relationship Between Variables that Lead to a Political
Contribution
197
Chapter 2 Appendix
Measurement of CCES variables
I report here the survey measures from the CCES for each of the variables used in regression
analysis.
income Thinking back over the last year, what was your family's annual income? [categor-
ical variable with 21 options, ranging from \Less than $10,000" to $250,000 or more",
generally proceeding in increments of $10,000]
contact Did a candidate or political campaign organization contact you during the 2016
election? [categorical variable with options \Yes" and \No"]
education What is the highest level of education you have completed? [categorical variable
with options: \No HS", \High school graduate", \Some college", \2-year", \4-year",
\Post-grad"]
gender Are you male or female? [categorical variable with options: \Male" and \Female"]
political interest / news consumption Some people seem to follow what's going on in
government and public aairs most of the time, whether there's an election going on
or not. Other's aren't that interested. Would you say you follow what's going on
in government and public aairs ... [categorical variable with options: \Most of the
time", \Some of the time", \only now and then", \Hardly at all", \Don't know"]
party id [Identication of party id on a 7-point scale: Strong Democrat, Not Very Strong
Democrat, Lean Democrat, Independent, Lean Republican, Not Very Strong Republi-
can, Very Strong Republican]
ideology In general, how would you describe your own political viewpoint? [categorical
variable with 5 options: Very liberal, Liberal, Moderate, Conservative, Very Conser-
vative]
198
Regression Models Delineated by Income of Respondent
Table 6.1: Low Income Results
Dependent variable:
donated
2006 2008 2016 2017 2018
(1) (2) (3) (4) (5)
Income 0.098
0.113
0.090
0.229
0.170
(0.017) (0.012) (0.012) (0.034) (0.013)
Contact 0.940
1.390
(0.080) (0.042)
Education 0.080
0.208
0.176
0.202
0.206
(0.024) (0.017) (0.013) (0.029) (0.013)
Gender 0.441
0.130
0.206
0.034 0.213
(0.064) (0.049) (0.037) (0.083) (0.039)
Political Interest 1.525
1.079
(0.080) (0.064)
News Consumption 0.601
0.938
0.994
0.584
(0.049) (0.036) (0.073) (0.022)
Democratic Strength 0.171
0.197
0.049
0.060 0.143
(0.035) (0.027) (0.020) (0.043) (0.022)
Republican Strength 0.144
0.117
0.071
0.022 0.105
(0.041) (0.032) (0.023) (0.052) (0.025)
Liberal Strength 0.197
0.283
0.555
0.501
0.414
(0.054) (0.041) (0.033) (0.072) (0.031)
Conservative Strength 0.040 0.014 0.025 0.308
0.118
(0.064) (0.046) (0.038) (0.082) (0.037)
Constant 7.423
7.696
6.889
7.409
5.230
(0.267) (0.203) (0.149) (0.307) (0.126)
Observations 8,815 11,891 22,416 6,839 22,541
Log Likelihood 3,406.641 6,746.058 9,400.293 2,025.887 9,227.837
Akaike Inf. Crit. 6,833.282 13,512.120 18,820.590 4,069.773 18,473.670
Note:
p<0.1;
p<0.05;
p<0.01
199
Chapter 4 Appendix
Reasoning Behind the Truncation Decision
Making a truncation decision to precisely choose how many characters/words to limit the
text of the emails to aects both the classication of the email text as well as the core
theory of anxious avoidance. After studying the problem I found two reasonable truncation
strategies based on two dierent assumptions about precisely how the theory of anxious
avoidance works in practice.
Assumption 1: No Open / No Click: assume the core eect of my theory results
from individuals ignoring an anxiety-inducing email opener (the email opener is the sender,
subject line, and rst words of an email).
Assumption 2: Click / Open: assume the core eect of my theory results from individ-
uals clicking into and reading a small portion of an email and ignoring/avoiding the email
after seeing that it is worry-inducing.
The core dierence is either assuming that individuals are clicking / opening these emails
or glancing at the opener and deciding not to open them. I am making the assumption that
the in-practice logistical way by which the theory of anxious avoidance works in a campaign
fundraising settings is through individual recipients being turned o by worry-inducing text
in the opener. If people see that an email is designed to be worry/anxiety-inducing right at
the start, they are less likely to click it because they do not want to be made anxious about
it.
To work around this, I truncated the text of the emails to the rst 100 words or so
(truncating the text at the rst period after the 100th word in an eort to keep complete
sentences) and included the subject line. In the end, workers were asked to classify the subject
line of an email and the rst 100 words of an email for being worry/anxiety-inducing.
200
Table 6.2: High Income Results
Dependent variable:
donated
2006 2008 2016 2017 2018
(1) (2) (3) (4) (5)
Income 0.167
0.163
0.086
0.106
0.111
(0.017) (0.014) (0.009) (0.014) (0.009)
Contact 0.912
1.447
(0.071) (0.043)
Education 0.103
0.146
0.100
0.196
0.120
(0.018) (0.015) (0.012) (0.021) (0.012)
Gender 0.350
0.026 0.173
0.057 0.208
(0.056) (0.048) (0.036) (0.063) (0.035)
Political Interest 1.266
0.972
(0.071) (0.064)
News Consumption 0.690
1.028
1.158
0.875
(0.052) (0.039) (0.070) (0.030)
Democratic Strength 0.265
0.232
0.151
0.245
0.195
(0.032) (0.027) (0.020) (0.034) (0.021)
Republican Strength 0.192
0.055
0.029 0.084
0.073
(0.034) (0.028) (0.022) (0.040) (0.022)
Liberal Strength 0.227
0.317
0.468
0.400
0.336
(0.053) (0.046) (0.035) (0.056) (0.031)
Conservative Strength 0.051 0.061 0.143
0.071 0.099
(0.054) (0.043) (0.038) (0.066) (0.034)
Constant 7.614
8.192
7.236
7.804
5.798
(0.284) (0.255) (0.179) (0.302) (0.157)
Observations 7,780 11,458 21,005 7,191 20,957
Log Likelihood 4,151.101 7,281.252 9,776.560 3,175.456 10,544.220
Akaike Inf. Crit. 8,322.202 14,582.500 19,573.120 6,368.913 21,106.440
Note:
p<0.1;
p<0.05;
p<0.01
201
I have several justications for this truncation:
1. If people are not willing to be paid to read the entirety of these emails, it seems likely
that people are generally unwilling to read them for free on their personal time as the
intended recipients.
2. Research shows that people do not spend much time looking at marketing emails
(Litmus Software, 2019). While marketing emails are not precisely the same as political
campaign emails, they are similar enough to be a good indicator of how much time
people typically spend looking at political campaign emails.
This combination of feasibility problems and acknowledgment of most-likely real world
behavior leads me to truncate the emails for classication. This does change things: for
example, I may no longer claim to be classifying campaign emails, but rather classifying
campaign email openers. I believe this will yield a classier more suited to my theory of
anxious avoidance. Because I hypothesize that campaign returns have a negative relationship
with the dissemination of worry-inducing emails, I believe that classifying only the opener
of an email will be better at labeling emails that actually were ignored. I assume that the
more time a user spends with a campaign email, the more likely they are to contribute. The
less time a user spends with a campaign email, the less likely they are to contribute.
So instead of measuring the worry-inducing status of an entire email, which can be very
long (average character count for the emails in my sample = 1900), I measure the worry-
inducing status of an email opener, which I directly limit to be short. While I fully predict
there to be emails that start o benign and begin to induce anxiety towards the middle or
end of the email, I expect most of the observable eects of my hypothesis will be sourced from
the subject line and rst words of an email., that leaves a decision between two assumptions:
(1) assuming people are not clicking/opening these emails, or (2) assuming people are only
reading the email for a short while and get the information they need in a matter of seconds.
I will brie
y outline what entails these two assumptions.
202
Because more attention is generally paid to the opener of an email than its middle or
end, analyzing the entirety of an email for its anxiety-inducing content and using that as
a variable that impacts contributions could lead to measurement error. This is because in
many cases, classifying the entire text of an email for its worry/anxiety-inducing status would
be analyzing text that was not actually seen/read. The no-click assumption is a strong one,
as Mailchimp
2
reports the opening rate
3
for political emails is 23%, with an average open
rate across all industries at 21.33% (mailchimp, 2017).
In an analysis of over 10 billion emails opened between between April 2018 - April 2019,
Litmus Software (2019) reports that the average time spent reading an email was 13.4 sec-
onds. They further report that 61% of opened emails are read for 8 seconds or more, 23.5%
of emails are skimmed for 2-8 seconds, and 15% of emails get less than 2 seconds of attention.
Keep in mind that these statistics were reported for emails that were opened. This means
that the large majority of emails are never opened, and nearly 40% of opened emails are read
for only a few seconds. In an ideal world I would classify both the rst few sentences of an
email in addition to its opener, but for now, empirical evidence from marketing rms that
deal with email behavior every day is that the majority of people are spending very little
time engaging with the political emails in their inbox. In addition to my theory that people
are avoiding emails because they are anxiety-inducing, truncating emails to only classify the
email opener as worry/anxiety-inducing is justied.
However, an email opener is not standardized in any way. People use a variety of email
clients to view their inboxes, and depending on the software and their personal settings will
see a variable number of characters of an email. Litmus Software (2019) reports the following
usage statistics for various mail clients: Gmail - 28%, Apple iPhone, 28%, Outlook - 9%,
Apple iPad - 9%, Apple Mail - 8%, Yahoo! Mail - 6%, Google Android - 3%, Outlook.com
- 2%, Samsung Mail - 2%, Thunderbird - 1%. Further breaking that down, for 2018 - 2019
2
a large email marketing rm
3
Email Open rate: a measure of how many people on an email list open/view a particular email
203
centering
Figure 6.2: Several gmail email openers on an iPhone XS Max
they report 42% of emails were opened on mobile devices, 40% opened via webmail, and
18% opened on desktops.
Included below are several screenshots to give a visual sense of the varying nature of
character counts of email openers or email previews. The character count varies across
software client (Apple Mail, Gmail, Outlook, etc.) as well as email client (mobile, web, and
desktop, and device screen size). Additionally, users can resize their windows on desktops
to dynamically change the amount of characters shown in an email opener/preview. And
even more of a complication, for many mail clients the users can choose how many lines of
an email opener to view. For example, Apple Mail defaults to preview 2 lines of text from
the email, but can be changed to preview anywhere between 0 and 5 lines of text.
204
Figure 6.3: Several Apple mail email openers on an iPhone XS Max
Figure 6.4: Gmail client email opener for a shorter window: note fewer words appear as
preview
Figure 6.5: Gmail client email opener for a full-size window
205
Figure 6.6: Gmail client email opener for a very narrow window: note that the text is
changed to resemble the mobile Gmail app user interface
Figure 6.7: Apple desktop mail opener for a large window
206
Figure 6.8: Apple desktop mail opener for a large window with the settings customization
to view 5 lines of preview text
These screenshots illustrate the wide variety and dynamism of just how many characters
are visible in email text. Adding to this complication is that many email clients have a
two-column user interface, where the left column contains basic information about an email
and the right column contains the entirety of a single selected email. Apple's mail client for
macOS and iPadOS shows this clearly. On email clients with larger screen real estate, the
email client software defaults to showing the entirety of an email, so users are immediately
exposed to both the email opener and the entirety of the email itself at the same time. While
not all email clients behave this way, many do, so it complicates the truncation decision
further.
No matter what, there will be a tradeo in choosing a truncation point. A number
of email marketing companies suggest between 40 - 130 characters, 30-80 characters, 100
characters, etc. Litmus Software (2017) compiled the average number of characters for each
email client.
So to calculate the preferred email preheader length:
For a fullscreen window of gmail on a browser running on a 12-inch laptop, Gmail pre-
headers contained approximately 100 characters. Litmus reports the average mumber of
characters displayed for Apple Mail on desktop is 140, for iOS native mail app about 90
characters, while other mobile preheaders were about 50 characters. AOL webmail av-
207
eraged 75 characters. While there is no simple decision, I opted to truncate the email
opener/preview/preheader at 140 characters. I chose the strictest possible standard for
email headers without trying to include complete sentences. I considered making the cuto
mark the rst complete sentence longer than 80 (and 50) characters, but functionally this
made the average character count of the email openers 225 (and 205) characters, which I
believe is too long.
A 140 character-limit for preview text (plus the subject line) is a reasonable limit to
capture what typical email users are actually seeing to determine whether or not they are
going to open and engage with an email. Plus, the character count should give MTurkers
enough context to determine if the email preview is worry/anxiety-inducing. While there is
no perfect solution, I believe this ts the connes of both my theory of anxious avoidance as
well as the feasibility limits of MTurk.
4
Regression Analysis
Table 6.3 reports results of the same regression results as Table 4.4 in chapter 4, except
using logarithmic transformations on the money variables. The Adjusted R
2
for the linear
and logarithmic models are compared below:
• Log-Linear Republican Model: R
2
= 0.3734
• Linear Republican Model: R
2
= 0.686
• Log-Linear Democrat Model: R
2
= 0.4248
• Linear Democrat Model : R
2
= 0.542
• Log-Linear Low Anxiety Candidate Model: R
2
= 0.4054
• Linear Low Anxiety Candidate Model : R
2
= 0.819
• Log-Linear High Anxiety Candidate Model: R
2
= 0.3847
• Linear High Anxiety Candidate Model : R
2
= 0.374
4
In the book \Nonprot fundraising 101," Heyman (2016) suggest to keep emails very short to keep users'
attention.
208
Notably, for only the high anxiety candidate model do the log-linear models have a greater
model t than the pure linear models. To compare the predictions made by each model, the
prediction graphs generated from these models like in Figure 4.11 in chapter 4 are replicated
here in Figure 6.9 for comparison.
8
10
12
0 20 40 60
time
Predicted Weekly Fundraising
0
10
20
30
anxietyCount
Republicans: Predicted Fundraising at
Various Anxiety Counts, Holding All Other Variables Equal.
7
8
9
10
11
0 20 40 60
time
Predicted Weekly Fundraising
0
10
20
30
anxietyCount
Democrats: Predicted Fundraising at
Various Anxiety Counts, Holding All Other Variables Equal. Top 10% of Spenders
6
8
10
12
0 20 40 60
time
Predicted Weekly Fundraising
0
10
20
30
anxietyCount
Low Anxiety Candidates: Predicted Fundraising at
Various Anxiety Counts, Holding All Other Variables Equal. Bottom 90% of Spenders.
6
9
12
15
18
0 20 40 60
time
Predicted Weekly Fundraising
0
10
20
30
anxietyCount
High Anxiety Candidates: Predicted Fundraising at
Various Anxiety Counts, Holding All Other Variables Equal. Bottom 90% of Spenders.
Figure 6.9: Fundraising predictions for a candidate at various levels of anxiety over time,
holding all other features at their means. Includes all candidates. Black lines are marked
between Week 63 and 65 of an entire campaign cycle, corresponding to between September
7 - 21.
@
209
Table 6.3: Log-Linear Model: Eect of Emails and Time on Fundraising Below $200
Dependent variable:
Fundraising Below $200
Republicans Democrats Low Anxiety Candidates High Anxiety Candidates
(1) (2) (3) (4)
Anxious Email Count 0.068 0.146
0.015 0.370
(0.308) (0.083) (0.224) (0.237)
Email Count 0.295 0.175
0.672
0.197
(0.225) (0.056) (0.137) (0.195)
Negative Email Count 0.326 0.032 0.355 0.325
(0.330) (0.097) (0.284) (0.226)
Total Spending 0.688
0.401
0.537
0.551
(0.024) (0.010) (0.021) (0.019)
Total fundraising Spending 0.0004 0.035
0.004 0.017
(0.011) (0.006) (0.012) (0.010)
Time 0.023 0.006 0.098
0.069
(0.036) (0.015) (0.031) (0.033)
Time Squared 0.001 0.001 0.002
0.003
(0.001) (0.0005) (0.001) (0.001)
Time Cubed 0.00002
0.00001 0.00001 0.00002
(0.00001) (0.00000) (0.00001) (0.00001)
Anxiety Count*Time 0.002 0.010 0.009 0.044
(0.029) (0.009) (0.023) (0.024)
Anxiety Count*Time
2
0.00001 0.0003 0.001 0.001
(0.001) (0.0003) (0.001) (0.001)
Anxiety Count * Time
3
0.00000 0.00000 0.00001 0.00001
(0.00001) (0.00000) (0.00001) (0.00001)
Email Count*Time 0.008 0.001 0.038
0.051
(0.020) (0.006) (0.014) (0.019)
Email Count*Time
2
0.00003 0.0001 0.001
0.001
(0.001) (0.0002) (0.0004) (0.001)
Email Count*Time
3
0.00000 0.00000 0.00001 0.00001
(0.00000) (0.00000) (0.00000) (0.00000)
Negative Count*Time 0.018 0.009 0.042 0.021
(0.031) (0.010) (0.030) (0.023)
Negative Count*Time
2
0.0003 0.0002 0.001 0.0004
(0.001) (0.0003) (0.001) (0.001)
Negative Count*Time
3
0.00000 0.00000 0.00001 0.00000
(0.00001) (0.00000) (0.00001) (0.00001)
Constant 2.023
2.075
1.117
0.706
(0.416) (0.150) (0.325) (0.334)
Observations 3,186 10,146 3,424 3,405
R
2
0.377 0.426 0.408 0.388
Adjusted R
2
0.373 0.425 0.405 0.385
Residual Std. Error 2.536 (df = 3168) 1.759 (df = 10128) 2.283 (df = 3406) 2.093 (df = 3387)
Note:
p<0.1;
p<0.05;
p<0.01
210
Chapter 5 Appendix
5/5/21, 3:32 PM Qualtrics Survey Software
Page 1 of 7 https://usc.ca1.qualtrics.com/Q/EditSection/Blocks/Ajax/GetSurvey…xtSurveyID=SV_9ymqvC9q2DnFIuV&ContextLibraryID=UR_1FdwUGVybvPMYkJ
Less than high school degree
High school graduate (high school diploma or equivalent including GED)
Some college but no degree
Associate degree in college (2-year)
Bachelor's degree in college (4-year)
Master's degree
Doctoral degree
Professional degree (for example: JD, MD)
Prefer not to say
Male
Female
Nonbinary
Other (please write)
Prefer not to say
Less than $10,000
$10,000 to $19,999
$20,000 to $29,999
$30,000 to $39,999
$40,000 to $49,999
$50,000 to $59,999
Demographics
1. What is your year of birth?
2. What is the highest level of school you have completed or the highest degree you have received?
3. What is your gender?
4.
What was your total combined household income in 2019 before taxes? This question is completely confidential and
just used to help classify the responses, but it is very important to the survey.
Figure 6.10: The begining of the survey deployed in chapter 5. 211
5/5/21, 3:32 PM Qualtrics Survey Software
Page 2 of 7 https://usc.ca1.qualtrics.com/Q/EditSection/Blocks/Ajax/GetSurvey…xtSurveyID=SV_9ymqvC9q2DnFIuV&ContextLibraryID=UR_1FdwUGVybvPMYkJ
$60,000 to $69,999
$70,000 to $79,999
$80,000 to $89,999
$90,000 to $99,999
$100,000 to $149,999
$150,000 or more
Prefer not to say
Gig worker / Freelance
Not working (temporary layoff from a job)
Not working (disabled)
Other, please write
Not working (retired)
Not working (looking for work)
Prefer not to answer
Working (self-employed)
Working (paid employee)
Yes
No
Republican
Democrat
Independent
Other (please write)
Click to write Choice 5
Strong
5. Which statement best describes your current employment status?
6. Did you vote in the last election?
7.
Generally speaking, do you usually think of yourself as a Republican, a Democrat, an Independent, or something
else?
8. Would you call yourself a strong Republican or a not very strong Republican?
212
5/5/21, 3:32 PM Qualtrics Survey Software
Page 3 of 7 https://usc.ca1.qualtrics.com/Q/EditSection/Blocks/Ajax/GetSurvey…xtSurveyID=SV_9ymqvC9q2DnFIuV&ContextLibraryID=UR_1FdwUGVybvPMYkJ
Not very strong
Strong
Not very strong
Republican
Democratic
Prefer not to say
No
Yes
I don't remember / Prefer not to say
I would rather spend my money on other things
I would rather give money to charity than politics
I don't think it makes a difference
I prefer for people not to know my political preferences
Don't like most candidates
Other, please write
Prefer not to say
9. Would you call yourself a strong Democrat or a not very strong Democrat?
10.
Do you think of yourself as closer to the Republican or Democratic party?
11. Here is a 7-point scale on which the political views that people might hold are arranged from extremely liberal
(left) to extremely conservative (right). Where would you place yourself on this scale?
Political Ideology
12. Have you ever contributed to a political campaign, organization, or group?
13. Below are some common reasons for not contributing. Check any that apply to you:
1 2 3 4 5 6 7
213
5/5/21, 3:32 PM Qualtrics Survey Software
Page 4 of 7 https://usc.ca1.qualtrics.com/Q/EditSection/Blocks/Ajax/GetSurvey…xtSurveyID=SV_9ymqvC9q2DnFIuV&ContextLibraryID=UR_1FdwUGVybvPMYkJ
Extremely likely
Moderately likely
Slightly likely
Neither likely nor unlikely
Slightly unlikely
Moderately unlikely
Extremely unlikely
Prefer not to say
Definitely yes
Probably yes
Might or might not
Probably not
Definitely not
Prefer not to say
I have never made a political contribution
I've only contributed once or twice
I contribute sometimes
I contribute relatively frequently
I contribute all the time
Prefer not to say
Yes
No
I don't know
Prefer not to say
14. How likely are you to contribute to a political campaign, organization, or group in the future?
15. Have you made a political contribution in the last 12 months?
16. How frequently do you contribute to political candidates, campaigns, causes, or groups?
17. Do you consider yourself a regular contributor to political causes/campaigns?
18. How interested in politics do you consider yourself?
214
5/5/21, 3:32 PM Qualtrics Survey Software
Page 5 of 7 https://usc.ca1.qualtrics.com/Q/EditSection/Blocks/Ajax/GetSurvey…xtSurveyID=SV_9ymqvC9q2DnFIuV&ContextLibraryID=UR_1FdwUGVybvPMYkJ
Definitely yes
Probably yes
Might or might not
Probably not
Definitely not
Prefer not to say
Never or almost never
Occasionally
Frequently
Every day
Prefer not to say
Yes
Maybe
No
Prefer not to say
0 is not interested at all,
7 is very interested
19. How frequently do you follow current events and news?
0 is not following at all,
7 is closely following
20. Do you receive emails from political groups that ask you for money?
22. How often do you receive emails from political groups asking you for money?
23. Have you ever made a political contribution by following a link from an email sent by a political campaign?
0 1 2 3 4 5 6 7
0 1 2 3 4 5 6 7
215
5/5/21, 3:32 PM Qualtrics Survey Software
Page 6 of 7 https://usc.ca1.qualtrics.com/Q/EditSection/Blocks/Ajax/GetSurvey…xtSurveyID=SV_9ymqvC9q2DnFIuV&ContextLibraryID=UR_1FdwUGVybvPMYkJ
Ignored emails from a political group
Unsubscribed from a political group's email list
Blocked or marked as spam email(s) from a political group
Deleted them
Forwarded or shared emails with others
None of the above
Prefer not to say
The emails didn't match my political beliefs
I didn't feel like my contributions would make a difference
The number of emails was annoying
The content of the emails was causing me to worry, stress out, feel anxious, etc.
I had already given what I was willing to contribute
Other, please write
Prefer not to say
Almost Never
Sometimes
Often
All the time
Prefer not to say
24. Below are some behaviors related to political email lists. Check any of the following behaviors you've done:
25. Why did you ignore, unsubscribe, delete, or block communication?
26. How frequently do you see ads for political campaigns, candidates, or groups online or on social media?
28. Over the last 2 weeks, how often have you been bothered by the following problems?
Not at all Several days More than half the days Nearly every day
Feeling nervous, anxious or on
edge
Not being able to stop or control
worrying
Worrying too much about
different things
Trouble relaxing
Being so restless that it is hard
to sit still
Becoming easily annoyed or
216
5/5/21, 3:32 PM Qualtrics Survey Software
Page 7 of 7 https://usc.ca1.qualtrics.com/Q/EditSection/Blocks/Ajax/GetSurvey…xtSurveyID=SV_9ymqvC9q2DnFIuV&ContextLibraryID=UR_1FdwUGVybvPMYkJ
Not difficult at all
Somewhat difficult
Very difficult
Extremely difficult
irritable
Feeling afraid as if something
awful might happen
30. If you checked off any problems, how difficult have these problems made it for you to do your work, take care of
things at home, or get along with other People?
31. You will find below a series of statements which describe how people may react to the
uncertainties of life. Please use the scale below to describe to what extent each item is
characteristic of you. Please circle a number (1 to 5) that describes you best.
1 2 3 4 5
Unforeseen events upset me greatly.
It frustrates me not having all the information I need.
One should always look ahead so as to avoid surprises.
A small, unforeseen event can spoil everything, even with
the best of planning.
I always want to know what the future has in store for me.
I can’t stand being taken by surprise.
I should be able to organize everything in advance.
Uncertainty keeps me from living a full life.
When it’s time to act, uncertainty paralyses me.
When I am uncertain I can’t function very well.
The smallest doubt can stop me from acting.
I must get away from all uncertain situations.
32. Do you have any other comments on the survey? We are particularly interested in any suggestions for
improvements if we conduct the survey again. if so, please write your comments below. if no, please leave blank.
The next and final screen will include your survey completion to enter on Amazon Mechanical Turk.
Not at all
characteristic
of me
Somewhat
characteristic
of me
Entirely
Characteristic
of me
217
Below, I report the logistic regression results for every single GAD-7 and IUS-12 question to
determine if there are any types of anxiety or intolerance of uncertainty that are particularly
related to increased propensity to make political contributions. It is interesting to note
that when the only variables in the logistic regression model are the individual component
of the IUS-12 and the GAD-7, very few variables are shown to have individual statistical
signicance.
Table 6.4: Logistic Regression: The eect of GAD-7 and IUS-12 items on contribution
likelihood
Dependent variable:
ever.contributed
ius.12.1 0.019 (0.122)
ius.12.2 0.039 (0.125)
ius.12.3 0.386
(0.119)
ius.12.4 0.002 (0.115)
ius.12.5 0.011 (0.112)
ius.12.6 0.097 (0.121)
ius.12.7 0.013 (0.114)
ius.12.8 0.119 (0.128)
ius.12.9 0.206 (0.126)
ius.12.10 0.151 (0.131)
ius.12.11 0.069 (0.122)
ius.12.12 0.227
(0.120)
gad7.index.1 0.251 (0.167)
gad7.index.2 0.003 (0.151)
gad7.index.3 0.068 (0.157)
gad7.index.4 0.330
(0.150)
gad7.index.5 0.071 (0.142)
gad7.index.6 0.169 (0.149)
gad7.index.7 0.314
(0.161)
Constant 0.515 (0.422)
Observations 558
Log Likelihood 319.665
Akaike Inf. Crit. 679.331
Note:
p<0.1;
p<0.05;
p<0.01
218
When the only variables in the logistic regression model are the individual component
of the IUS-12 and the GAD-7, very few variables are shown to have individual statistical
signicance. The questions that do have an impact are question three of the IUS-12, question
12 of the IUS-12, question 4 of the GAD-7, and question 7 of the GAD-7. However, question
3 on the IUS-12 actually has a negative relationship with contribution likelihood. Question
12 on the IUS-12 and questions 4 and 7 on the gad-7 both have fairly small but positive
relationships with the likelihood to contribute.
219
Abstract (if available)
Abstract
Small donor fundraising is an increasingly important part of campaign finance, yet it is understudied. Specifically, small donor political fundraising relies heavily on emails, which are often frantic and anxiety-inducing by design. I contribute to the body of research on campaign fundraising in three studies, placing focus on the impact of two understudied variables: emotional campaign email solicitation and psychological anxiety. In the first study, I conduct regression analysis on survey data from the ANES and CCES from 2006–2018 to determine the factors most associated with making political contributions. Findings show that campaign contact/solicitation, Democratic partisanship and liberal ideology are among the most predictive factors in determining who contributes. In the second study, I classify over 86,000 campaign emails for anxiety and assess their relationship with campaign fundraising returns over time using logistic and polynomial regression. The third and final study uses survey data from a participant pool recruited through MTurk to measure the relationship between psychological anxiety and campaign contribution/avoidance. Survey results suggest that higher levels of anxiety are associated with a greater likelihood to both contribute to campaigns as well as avoid campaigns. Observational analysis suggests that while anxiety-inducing emails do have a positive relationship with fundraising returns at some points during a campaign, the relationship is subject to diminishing returns among all candidates except Democrats.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Challenges from below: the origins of status competitions in world politics
PDF
Three text-based approaches to the evolution of political values and attitudes
PDF
La eleccion de la pandemia: analyzing Latino political behavior during the 2020 election
PDF
Voices unheard, stories untold: Black women, police violence and political participation
PDF
Bureaucratic politics and power building in the administrative state
PDF
The strategy of firms towards political elites and consumers in polarized western democracies
PDF
Essays on sovereign debt
PDF
Analysis and prediction of malicious users on online social networks: applications of machine learning and network analysis in social science
PDF
Can election reforms improve democracy?
PDF
A theory of status-quo terrorism: democracies in conflict and their proclivity to outsource repression
PDF
Techniques of strategic political communication: the persuasive devices of the two most followed political YouTubers in the Chinese community
PDF
Local votes and outside money: campaign contribution geographic origins and their impact on Los Angeles City Council election outcomes
PDF
The erosion of democracy in the post–Cold War world: understanding the phenomenon and its causes
PDF
Economic policy-making in a changing environment: the politics of success in Peru vis-à-vis China
PDF
Politics is something we do together: identity and institutions in U.S. elections
PDF
Running for office? Three different models in U.S. political campaigns & elections
PDF
Signaling identity: how race and gender shape what representatives say online
PDF
The allocation and accessibility of aid: the role of recipient government agency in dictating donor decisions
PDF
Identity salience and political violence: centering social identity in the study of intrastate conflict
PDF
The representation of LGBT interests in the United States: a multimethod approach
Asset Metadata
Creator
Timm, Joshua
(author)
Core Title
The small dollar political donor: why regular folks give money to politics
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Political Science and International Relations
Degree Conferral Date
2021-12
Publication Date
11/24/2021
Defense Date
11/24/2021
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
Anxiety,campaign fundraising,OAI-PMH Harvest,political anxiety,political contributions,political donors,political fundraising,small donors
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Grose, Christian (
committee chair
), Barberá, Pablo (
committee member
), Hollihan, Tom (
committee member
), Lo, James (
committee member
)
Creator Email
joshuartimm@gmail.com,jtimm@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-oUC17239462
Unique identifier
UC17239462
Legacy Identifier
etd-TimmJoshua-10261
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Timm, Joshua
Type
texts
Source
20211124-wayne-usctheses-batch-900-nissen
(batch),
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Repository Email
cisadmin@lib.usc.edu
Tags
campaign fundraising
political anxiety
political contributions
political donors
political fundraising
small donors