University of Southern California Dissertations and Theses

Selling virtue: how human rights NGOs and their donors work together to create a better world ... for themselves

(USC Thesis Other)

Selling virtue: how human rights NGOs and their donors work together to create a better world ... for themselves

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content SELLING VIRTUE:
HOW HUMAN RIGHTS NGOs AND THEIR DONORS WORK TOGETHER TO CREATE A
BETTER WORLD...FOR THEMSELVES
by
Suzie Mulesky
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(POLITICAL SCIENCE AND INTERNATIONAL RELATIONS)
May 2021
Copyright 2021 Suzie Mulesky
Acknowledgements
In 2012, I vacationed in Rome with my then boyfriend (now husband) and his parents.
During that vacation, we hatched a plan to go to graduate school together. With his
advice, I decided to pursue a PhD in political science, and he decided to pursue a PhD
in philosophy of science. Two years later, we graduated from Indiana University together
and moved to Los Angeles where I started my PhD in the Political Science and International
Relations program at USC. He started his PhD in the UCLA philosophy program. Six and
a half years later and with a six month old baby, I defended my dissertation over Zoom
during the middle of a global pandemic. I owe a debt of gratitude to many, many people
for helping me along this journey.
Throughout, I had a secret weapon. Somebody who meticulously edited my work,
battle-tested my arguments, challenged me on every conceivable point, introduced me
to new ideas, disciplines, people, books, and podcasts, encouraged me to find my voice,
pushed me to be more ambitious and courageous, and inspired me to overturn my old
way of thinking about human rights. Imagine having an advisor, coach, line manager,
research assistant, right-hand man, optimizer, advocate, and stay-at-home parent all in
one marvelous, selfless human secret weapon. That was my husband, Spencer. He won’t
get any public credit for the hundreds of hours he poured into helping me develop my
dissertation and generally succeed in the PhD. But he deserves every bit of the credit.
They say behind every great man is a great woman. I understand this phrase on a personal
level because any achievements I earned in the past and will earn in the future owe part
of their existence to my husband’s dedication to my success. Behind me is a great man.
I also thank my dissertation committee members for their support, encouragement, and
mentorship along the way. I thank my advisor, Wayne Sandholtz, for brokering academic
connections, reading all of my drafts, and offering me freedom to develop my arguments.
I thank Pat James for big picture strategizing and detailed comments on my drafts. I thank
Chris Fariss for professional networking, methodological advice, and inspiring my interest
in human rights measurement. I thank Sofia Gruskin for her expertise of nongovernmental
organizations and for connecting me to NGO stakeholders who I interviewed as part of the
field research component of my dissertation.
I thank my mentors, who helped me in many ways beyond the dissertation. Ben Gra-
ham for his collaboration, strategizing, professional development, and general mentorship.
Jonathan Markowitz for his collaboration, general support, and for teaching research de-
sign how it ought to be taught. Both Ben and Jon were some of my biggest cheerleaders in
the program, and if it weren’t for their pep talks and professional encouragement, I would
probably have a lot less self-confidence and sense of direction. I thank Barbara Koremenos
for collaborating with me on several projects and for supporting me through the academic
job market. I also thank Clifford Bob for generously lending his time to read and edit my
work, help me with my job market materials, and professionally support my interest in
pursuing book publication of my dissertation. And Geoffrey Miller for his generosity, for
helping me apply signaling theory to human rights, for encouraging me to write about the
Virtue Economy for a general audience, and for aiding me on the job market.
Many people read parts of my dissertation and offered detailed feedback. I am ex-
tremely grateful for their time and interest. Wayne Sandholtz, Pat James, Chris Fariss,
ii
Sofia Gruskin, Clifford Bob, Michael Spagat, Oliver Scott Curry, Richard Hanania, William
Winecoff, Jason Wu, Eric Posner, Anthony DeMatee, Salih Yasun, and Jonathan Ring.
Lucius Caviola, Stefan Schubert, and Michael Kenwick offered a great deal of advice
and feedback on the survey experimental research design. I am in huge debt to Lucius and
Stefan in particular. They were critical in helping me develop the experiments in Chapter
3. I thank Geoffrey Miller, David Moss, and Stefan Schubert for their conversations about
the theory of donor signaling. I thank Cyrus Samii for his methodological feedback on
multiple regression weights from Chapter 4.
Many people read earlier versions of my dissertation when I was a newer and younger
PhD student. Jennifer Rogla, Stephanie Kang, and I had a dissertation proposal support
group, and they spent many hours with me helping to refine my theory, improve my re-
search design, and aid me with the development of my field research proposal. I also thank
Shiming Yang, Tom Jamieson, Brian Knafou, Victoria Chonn Ching, David Kang, Nicholas
Weller, and Jonathan Markowitz for reading my earlier drafts and providing very useful
feedback on the theory and empirics.
For research support, I thank USC’s School of International Relations for multiple years
of summer research funding, the USC Graduate School for a year-long fellowship, and the
Columbia University Rare Books and Manuscript Library and the International Institute of
Social History in Amsterdam where I conducted archival research for my dissertation.
I am very thankful to have entered my PhD program with a GOAT cohort (greatest of all
time): Stephanie Kang, Shiming Yang, Whitney Hua, Jenn Rogla, Vic Chonn Ching, Brian
Knafou, Yu-Ting Lin, and Ada Li Sarain. They were wickedly fun, and I had a great time
during the four years I lived in LA. Outside of my cohort, I benefited from many conversa-
tions with Joey Huddleston, Taylor Dalton, Therese Anders, Xinru Ma, Mark Paradis, and
Mao Suzuki.
Amy and Ray, my husband’s parents, deserve a great deal of credit as well for their
support. They were there for countless conversations about signaling, altruism, and char-
ity. Not only are they intellectually valuable conversation partners, but they are the kind
of people who drop everything they’re doing to help. They edited my writing, provided a
beautiful home for me to complete the final two and a half years of my PhD, counseled
me throughout the job market process, offered an enormous amount of moral support and
cheerleading, and helped with my newborn baby during maternity leave. I am exception-
ally lucky to have such intelligent and helpful in-laws.
I could not have accomplished all that I did without the help, support, and feedback
from these folks.
iii
Contents
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
1 Introduction 1
Puzzles for Existing Theories of Human Rights NGOs . . . . . . . . . . . . . . . . 1
Introducing a New Theoretical Approach: The Virtue Economy . . . . . . . . . . . 11
Roadmap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2 The Virtue Economy 19
Demand for Social Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Supplying Social Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3 The Lack of Market Demand for a Scientific Approach to Advancing Human
Rights 58
Testing the New Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Explaining the Low Supply of Rigorous Impact Information . . . . . . . . . . . . . 61
Explaining the Dearth of Randomized Trials . . . . . . . . . . . . . . . . . . . . . 75
Explaining the Supply of Overhead Information . . . . . . . . . . . . . . . . . . . 88
Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4 Still Ineffective: Disputing Research on the Effectiveness of Human Rights
NGOs 99
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5 Conclusion 134
Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
Final Word . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
Bibliography 141
Appendix 155
iv
List of Tables
3.1 Welch’s T-test Results from Studies 1-2 . . . . . . . . . . . . . . . . . . . . 72
3.2 Summary of Studies and Hypotheses . . . . . . . . . . . . . . . . . . . . . 78
3.3 Welch’s T-test Results from Studies 3-6 . . . . . . . . . . . . . . . . . . . . 87
4.1 Studies Included in the Replication Analysis . . . . . . . . . . . . . . . . . 104
4.2 Summary of Results from Models Testing the Conditional Effectiveness of
Human Rights NGOs and Purporting a Statistically Significant Positive Effect 116
4.3 Kurtosis of Multiple Regression Weights . . . . . . . . . . . . . . . . . . . . 122
4.4 The Cumulative Impact of Reporting Biases Across 35 Models Estimating the
Effect of Human Rights NGOs . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.1 The Effect of Impact Information on Donations (Studies 1 and 2) . . . . . 169
5.2 The Effect of Randomized Trials on Donations (Studies 3-5) . . . . . . . . 170
5.3 The Effect of Overhead and Impact Information on Donations (Study 7) . 171
v
List of Figures
2.1 Amnesty International Netherlands Catalog . . . . . . . . . . . . . . . . . 52
3.1 Effect of Impact Information on Donations . . . . . . . . . . . . . . . . . . 71
3.2 Mean Donations from Studies 1 and 2 . . . . . . . . . . . . . . . . . . . . 71
3.3 Mean Donations from Study 3 . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.4 Mean Donations from Study 4 . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.5 Mean Donations from Study 5 (A/B Test) . . . . . . . . . . . . . . . . . . . 84
3.6 Mean Donations from Study 5 (RCT Test) . . . . . . . . . . . . . . . . . . 84
3.7 Mean Donations from Study 5 (A/B Test vs. RCT) . . . . . . . . . . . . . . 85
3.8 Mean Donations from Study 6 (Joint Evaluation) . . . . . . . . . . . . . . 86
3.9 Mean Donations from Study 7 (Overhead vs. Impact) . . . . . . . . . . . . 92
3.10 Mean Donations by Income Group (Study 1 Impact Information) . . . . . 94
3.11 Mean Donations by Income Group (Study 7 Impact vs. Overhead) . . . . . 95
3.12 Mean Donations by Income Group (Study 6 A/B and RCT Testing) . . . . . 96
4.1 Marginal Effects: Reproduced from Murdie and Davis (2012) Model 2, Table
1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.2 The Marginal Effect of Human Rights NGO Advocacy on Respect for Physical
Integrity Rights (Murdie and Davis (2012) Model 2, Table 1) . . . . . . . . 108
4.3 Marginal Effects: Reproduced from Murdie and Davis (2012) Model 3, Table
1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
4.4 The Marginal Effect of Human Rights NGO Advocacy on Respect for Physical
Integrity Rights (Murdie and Davis (2012) Model 3, Table 1) . . . . . . . . 111
4.5 Murdie and Davis (2012) Model 1 Table 7 . . . . . . . . . . . . . . . . . . 111
4.6 Murdie and Davis (2012) Model 1 Table 8 . . . . . . . . . . . . . . . . . . 111
4.7 The Marginal Effect of Human Rights NGO Advocacy on Political Repression
from Franklin (2008) Model 2, Table 5 . . . . . . . . . . . . . . . . . . . . 113
4.8 The Marginal Effect of Neighboring Human Rights NGO Membership on
Improvement in Physical Integrity Rights from Bell, Clay, and Murdie (2012)
Model 3, Table 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.9 The Marginal Effect of Neighboring Human Rights NGO Secretariat Offices
on Physical Integrity Rights from Bell, Clay, and Murdie (2012) Model 2
Table A24 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
4.10 Multiple Regression Weights (Represented as a Percentage of Total Weight)
Used to Construct the Effect Estimate from DeMeritt (2012) Model 1, Table 2 119
4.11 DeMeritt (2012) Nominal Sample . . . . . . . . . . . . . . . . . . . . . . . 119
4.12 DeMeritt (2012) Effective Sample . . . . . . . . . . . . . . . . . . . . . . . 119
vi
4.13 Multiple Regression Weights (Represented as a Percentage of Total Weight)
Used to Construct the Effect Estimate from Krain (2012) Model 2, Table 3 121
4.14 Krain (2012) Nominal Sample . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.15 Krain (2012) Effective Sample . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.16 Multiple Regression Weights (Represented as a Percentage of Total Weight)
Used to Construct the Effect Estimate from Bell, Clay, and Murdie (2012)
Model 3, Table 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.17 Bell, Clay, and Murdie (2012) Nominal Sample . . . . . . . . . . . . . . . 123
4.18 Bell, Clay, and Murdie (2012) Effective Sample . . . . . . . . . . . . . . . 123
4.19 The Cumulative Impact of Reporting Biases on the Effect of Human Rights
NGOs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.1 Mean Donations by Income Group (Study 2 Impact Information) . . . . . 172
5.2 Mean Donations by Income Group (Study 3 A/B Test) . . . . . . . . . . . . 173
5.3 Mean Donations by Income Group (Study 4 A/B Test) . . . . . . . . . . . . 174
vii
Abstract
Selling Virtue seeks to explain a variety of puzzling behaviors within human rights philan-
thropy: (1) human rights NGOs routinely claim to have a positive impact despite lacking
rigorous evidence of it, (2) individual donors do not seek and are not responsive to rigor-
ous impact information, (3) accountability organizations do not hold human rights NGOs
accountable to impact despite claiming to, and (4) human rights NGOs continue to garner
money, status, credibility, and authority, despite never providing evidence of their impact
beyond anecdotes that are strongly biased in the positive direction.
I argue that this behavior, while grossly inefficient at achieving the organizations’ stated
missions, is not irrational. Instead, it is created by reputational self-interest and misaligned
incentives. I combine social signaling theory and the reality of market competition to
develop a new framework for analyzing the behavior of human rights NGOs. This new
framework, which I call the Virtue Economy, is a market-oriented model of NGO behavior.
I argue that NGOs act like firms, producing and selling social signals to their donors,
who are best understood as consumers. The primary product on offer is not the effective
production of a public good or public benefit, as is conventionally believed, but rather a
reputational or status benefit in the form of a social signal.
I use this new framework to explain why donors are not interested in or responsive to
information about charity effectiveness and how donor preferences influence the organi-
zational behavior of human rights NGOs. I employ experiments and statistical analysis to
analyze demand-side and supply-side features of the human rights market. The findings
illuminate (1) why human rights NGOs abstain from rigorously evaluating their impact,
(2) why they conceal honest information about their impact from donors and other organi-
zations, (3) why they exaggerate their impact, and (4) why the accountability movement
does not hold human rights NGOs accountable to standards of effectiveness, despite claim-
ing to. Put simply, donors reward this deceptive behavior because it is conducive to social
signaling. I also revisit the statistical literature on the effect of human rights NGOs and
demonstrate that the research suffers from numerous critical problems. A more faithful
reading of the statistical evidence leads to the conclusion that human rights NGOs are
not very effective. This finding should not be surprising, given that human rights NGOs
calibrate their activities to satisfy donor reputational needs, not the needs of the stated
beneficiaries.
viii
Chapter 1
Introduction
We thus find ourselves in a rather peculiar universe where
good intentions are rewarded before they have undergone
the strenuous metamorphosis of being translated into
good deeds, or hard facts.
Christopher Hitchens
Puzzles for Existing Theories of Human Rights NGOs
Academic proponents of human rights nongovernmental organizations (NGOs) have adopted
the view that NGOs, donors, activists, and volunteers are altruistically motivated, collec-
tively working within transnational networks to realize their shared principles by effec-
tively advancing human rights (Keck and Sikkink, 1998; Murdie, 2014; Risse, Ropp and
Sikkink, 1999, 2013; Sikkink, 2017). Critics complain that the human rights movement is
well-intentioned but toothless, with a poor track record of improvement in human rights
to show for it (Hopgood, 2013; Posner, 2014), or that NGOs have organizational self-
interests that deviate from their donors’ altruistic goals (Cooley and Ron, 2002; Prakash
and Gugerty, 2010; Stroup and Wong, 2017). I have some sympathy for both the propo-
nents and the critics, but they collectively suffer from a critical blind spot, which I aim
to move into our field’s center of vision: Human rights NGOs and their donors are funda-
1
mentally self-interested. They give no special priority to maximizing the welfare outcomes of
their recipients or ensuring their programs have a positive impact. To some, this claim may
sound outrageous. But this chapter — and indeed, this dissertation — walks through the
evidence and the theory that arrive at this conclusion.
Amnesty International and Human Rights Watch are the largest and wealthiest human
rights NGOs. Amnesty International’s stated organizational mission is to campaign “for a
world where human rights are enjoyed by all."
1
Similarly, Human Rights Watch “work[s]
to protect the most at risk, from vulnerable minorities and civilians in wartime, to refugees
and children in need."
2
Amnesty enjoys a broad level of support globally, including mil-
lions of supporters and high levels of brand trust (Ron, Ramos and Rodgers, 2005). Both
Amnesty and Human Rights Watch are high-status gatekeepers, garnering more deference
from the media, Western policymakers, peer NGOs, and the public than almost all other
international NGOs (Carpenter, 2014; Stroup and Wong, 2017).
Yet both organizations behave in ways that are inconsistent with their stated goals
and missions. For example, both Amnesty International and Human Rights Watch sys-
tematically ignore scientific evidence about effective solutions at advancing human rights,
causing them to over-invest in ineffective campaigns. Examples include their work to
end child marriage, a problem plaguing hundreds of millions of girls under the age of 18
in South Asia and Sub-Saharan Africa.
3
Many governments and nongovernmental orga-
nizations consider child marriage to be a violation of children’s human rights, and it is
associated with several negative outcomes including social isolation, early high-risk preg-
nancies, domestic violence, and reduced educational attainment.
4
Large-scale randomized
evaluations demonstrate that national child marriage laws and adolescent empowerment
programs (e.g. safe spaces, educational support, and training in life skills, nutrition, re-
1
https://www.amnesty.org/en/who-we-are/
2
https://www.hrw.org/about/about-us
3
The United Nations Population Fund projected that 142 million girls under the age of 18 would marry
between 2011 and 2020. See https://www.unfpa.org/sites/default/files/pub-pdf/MarryingTooYoung.pdf.
4
https://www.povertyactionlab.org/evaluation/empowering-girls-rural-bangladesh
2
productive health, and financial literacy) fail to reduce rates of child marriage. However,
certain interventions are effective at reducing child marriage and teenage childbearing as
well as increasing educational attainment: providing incentives in the form of cooking oil
to families conditional on their delaying their daughters’ marriages. This information is
not obscured and does not sit behind journal paywalls. Rather, this information has been
carefully translated for public consumption in helpful policy briefs.
5
Yet Amnesty and Human Rights Watch have focused their campaigning on the inef-
fective interventions — national legislation and empowerment — not the effective inter-
ventions — conditional subsidies — even in Bangladesh where much of this careful re-
search has taken place. For example, Human Rights Watch’s report on child marriage in
Bangladesh includes recommendations to the Prime Minister and Parliament to reform the
national legislation banning child marriage and to the Ministry of Women and Children Af-
fairs to initiate a nationwide awareness campaign against child marriage and to empower
girls to refuse child marriage. Nowhere is there a recommendation for the government
or international donors to implement an incentive-based program, despite Human Rights
Watch’s acknowledgement that families have an incentive to marry their daughters at ear-
lier ages to pay a lower dowry.
6
Likewise, Amnesty’s campaigning has focused exclusively
on national legislation and empowerment.
7
The story is similar when it comes to efforts to increase school attendance in low-
income countries —- an area well-researched in development economics. Extensive ex-
perimental evidence shows that supplying menstrual supplies, free uniforms, and scholar-
ships to students may sound promising, but has proven ineffective. In contrast, deworming
school-children so they are healthy enough to attend school has proven to be the most ef-
5
For example, see the Abdul Latif Jameel Poverty Action Lab synopsis of field experiments de-
signed to evaluate different types of interventions aimed at reducing rates of child marriage (https://
www.povertyactionlab.org/evaluation/empowering-girls-rural-bangladesh) and the Copenhagen Consensus
Center’s synopsis of this same research (https://www.copenhagenconsensus.com/publication/bangladesh-
priorities-child-marriage-field-et-al).
6
https://www.hrw.org/report/2015/06/09/marry-your-house-swept-away/child-marriage-bangladesh
7
See for example their campaigning in Burkina Faso, another country with high rates of child marriage:
https://www.amnesty.org/en/latest/campaigns/2015/07/my-body-my-rights-burkina-faso/#.
3
fective intervention (MacAskill, 2015). Deworming is four times more effective than the
next best program (iron and vitamin A fortification).
8
The authors of this research won the
2019 Nobel Prize in Economics. Yet of all the rigorously tested programs aimed at increas-
ing school attendance, the only one Amnesty International has advocated for is one of the
few that is shown to have no effect –– providing menstrual supplies. In fact, Amnesty has
never even mentioned the word “deworm" in any of its reports.
9
Despite neither organization utilizing scientific evidence about effective solutions or
conducting rigorous impact evaluations of their own programs, they nonetheless market
themselves as being effective NGOs. For example, in a Spring 2019 letter to its donors,
Amnesty International summarized its 2018 letter-writing campaign, "Letters change lives.
Our global letter writing campaign proves it!" Human Rights Watch says their “support-
ers are making a real positive impact on rights."
10
Similarly, both organizations routinely
appeal to impact when soliciting donations. Amnesty International’s donation page says,
“Your contributions will make a real difference.”
11
Human Rights Watch says, “Donate
now to help defend human rights and save lives.”
12
When justifying its strategic prior-
ities, Amnesty says that “to achieve lasting progress worldwide, we will ensure we al-
ways...identify the most effective ways to create change.”
13
Amnesty also frequently an-
nounces “human rights wins,” for example by noting that “over the last year, almost seven
million Amnesty International supporters have taken action — protesting, writing, peti-
tioning and much more –– to defend and advance human rights everywhere. It’s had a
8
https://www.povertyactionlab.org/sites/default/files/publication/roll-call-getting-children-into-
school.pdf
9
https://www.amnesty.org/en/search/?q=deworm&ref=&year=&lang=en&adv=1&sort=relevance
10
From an email Human Rights Watch sent me in May 2019. The full quote reads, "This year has been
a dark time for human rights around the world. As states across the US pass the most restrictive abortion
laws in decades, women’s basic rights are at stake. But we still have hope. Human Rights Watch supporters
are making a real positive impact on rights. That has never been truer than right now: from today through
June 28 only, we have a powerful matching gift opportunity. Every dollar you donate for human rights will
be worth five."
11
https://www.amnesty.org/en/donate-amnesty/
12
https://donate.hrw.org/page/15328/donate/1?promo_id=1003
13
https://www.amnesty.org/en/latest/campaigns/2016/01/amnesty-goals-2016-2019/
4
huge impact.”
14
However, a feature of human rights philanthropy that is not widely discussed in pub-
lished research is that human rights NGOs rarely, if ever, supply rigorous evidence to justify
these claims or to explain precisely how much impact an individual’s donation is esti-
mated to make. Impact information is almost always conveyed through anecdotal stories
of individuals strongly biased toward positive examples and focused on inputs (e.g. laws
changed, resolutions passed, government promises made) rather than impact.
15
Informa-
tion that is related to impact is often treated as proprietary, rather than itself a public
good. For example, while Amnesty International claims to have a monitoring, evaluation,
and learning system in place, it does not make available to the public or its members the
evaluations of the impact of its priority strategies and programs, as confirmed by Amnesty’s
Global Strategy and Impact Unit.
16
Further, Amnesty has disclosed that its programs do
not have measurable targets for their impact, welfare outcomes, or benefits to recipients,
and it has stated that it has no plans to adopt measurable targets in the future.
17
Its only
measurable target is its movement growth, which is a euphemism for revenue growth.
18
Likewise, Human Rights Watch has “dispens[ed] with the idea that evaluation necessitates
some specific level of rigorous documentation” (Root, 2015). If Human Rights Watch con-
ducts any internal evaluations, they do not make the evaluations available to the public or
to individual donors.
19
As Eric Posner, a pronounced critic of international human rights
14
https://www.amnesty.org/en/latest/campaigns/2018/12/this-years-epic-human-rights-wins/
15
For example, see Amnesty’s impact stories at https://www.amnesty.org/en/latest/impact/ and Human
Rights Watch’s impact stories at https://www.hrw.org/impact.
16
Private email correspondence on April 10, 2019.
17
See Accountable Now’s Independent Review Panel feedback to Amnesty’s 2017 report. Specifically,
note the Panel’s question on page 10, "there are no measurable project related targets built into the global
strategy. Is this something Amnesty is considering introducing in future?" (https://accountablenow.org/
wp-content/uploads/2020/01/Panel-Feedback_Amnesty-International-2017-Report.pdf#page=11). Then,
note Amnesty’s silence on this question in its response: https://accountablenow.org/wp-content/uploads/
2020/01/2019.03.11-Accountable-Now-response-letter-Amnesty-International.pdf
18
https://accountablenow.org/wp-content/uploads/2020/01/2018-AI-Report-to-Accountable-Now-
FINAL.pdf#page=26
19
Researchers have covered similar problems with regard to development aid agencies (Easterly and
Pfutze, 2008), foundations (Reich, 2018), and NGOs operating in the space of international development
assistance (Singer, 2009).
5
practice, has written, “HRW’s budget was $65 million as of 2013. It’s an article of faith
that HRW does good, but there is no evidence whatsoever" (Posner, 2015).
Likewise, accountability organizations claim to hold NGOs accountable to high stan-
dards of program effectiveness but do not actually demand evidence of impact. Amnesty
annually reports against 56 organizational indicators to an accountability organization
called Accountable Now, which was established by Amnesty and several other international
NGOs in response to criticism from governments and corporations. Accountable Now mar-
kets itself as a global platform that supports civil society organizations to be transparent,
responsive to stakeholders, and focused on delivering impact. I analyzed every report
submitted by Amnesty to this accountability group from 2008 to 2017.
20
While Amnesty
consistently describes that it has an evaluation system, it has never provided evidence of
its program effectiveness. In response, Accountable Now has never asked or encouraged
Amnesty to provide data or evidence of its effectiveness, despite giving Amnesty the best
grade in full transparency and accountability.
The story is similar with external charity evaluators, organizations that claim to collect
multiple types of data about nonprofits to improve nonprofit performance. Charity Navi-
gator, the most well-known external charity evaluator, publishes information about charity
overhead, which measures how much the charity spends on administration and fundraising
versus programs and services. However, the overhead ratio is a poor indicator of charity
effectiveness. An ineffective or harmful charity could have a low overhead, and an effec-
tive charity could have a high overhead. Studies have failed to find correlations between
the overhead ratio and a charity’s cost-effectiveness. If anything, pressure to conform to a
low overhead starves nonprofits of necessary administration (Caviola et al., 2014; Gregory
and Howard, 2009; Pallotta, 2008; Sellers, 2018). Knowing that Human Rights Watch
spends 74% of its revenue on program expenses does not provide any information about
20
Amnesty’s accountability reports and responses by Accountable Now’s Independent Review Panel
can be found here: https://accountablenow.org/accountability-in-practice/accountability-reports/amnesty-
international/.
6
how effectively the organization improves human welfare.
21
Finally, most donors do not donate based on charity impact, even though they may
claim to. Consistently, survey and experimental evidence shows that not only do donors
not seek impact information, but they are not responsive to it in the way we would ex-
pect when it is directly provided to them. For example, results from a survey on wealthy
donors reveals that while most donors say they care about nonprofit performance, only 3%
actually compare different charities before deciding where to donate (Camber Collective,
2015).
22
These facts, in combination, are difficult to reconcile with the two predominant the-
oretical approaches to NGO behavior: the ‘principled actor’ approach such as Keck and
Sikkink’s Activists Beyond Borders and the ‘collective action’ approach, such as Prakash and
Gugerty’s Advocacy Organizations and Collective Action (Keck and Sikkink, 1998; Prakash
and Gugerty, 2010). The principled actor approach depicts NGOs and donors as altruisti-
cally motivated, collectively working within transnational networks to realize their shared
principles by effectively advancing human rights (Keck and Sikkink, 1998; Risse, Ropp
and Sikkink, 1999, 2013; Sikkink, 2017). According to this approach, human rights NGOs
should routinely evaluate their impact because the effective advancement of human rights
requires knowledge about which interventions are effective, ineffective, and harmful. A
prerequisite for acquiring this knowledge is to evaluate the impact. Further, information-
sharing is a core theoretical component of transnational advocacy networks. When actors
are embedded in a network of other like-minded, principled actors, they share informa-
tion across the network to support each other’s efforts. Coordinating on effective strategies
and solutions would be instrumental for network effectiveness. Thus, human rights NGOs
21
In October 2020, Charity Navigator — with funding from the Bill and Melinda Gates Foundation —
acquired a small nonprofit founded by economists called ImpactMatters that rates nonprofit organizations on
the basis of their cost-effectiveness (i.e. how much good they accomplish per dollar spent). As of December
2020, only two nonprofits categorized under human and civil rights have impact scores (Paralyzed Veterans
of America and the Pat Tillman Foundation). While numerous nonprofits have impact ratings, most appear
to have a rating of 100 out of 100, reflecting highly inflated ratings.
22
This survey is just one example. Chapter 2 explores the evidence from additional surveys and experi-
ments in more detail.
7
should share information about their positive and negative impact with other organiza-
tions, activists, and volunteers in order to assist the network in converging on effective
solutions and dispensing with ineffective, wasteful solutions. Hiding and distorting this
information makes little sense within the transnational activist framework. Information
sharing, not informational secrecy, should be a prominent feature of the human rights
movement.
There are two potential responses that attempt to reconcile the principled actor ap-
proach with the low supply of rigorous impact evidence. First, some may argue that the
outcomes of interest to nonprofit organizations cannot be studied by means of rigorous
impact evaluations such as randomized controlled trials (Gorvin, 2009; Naughton and
Kelpin, 2015; Root, 2015; Schlangen, 2014). Empirically, this statement is not true as
evidenced by the 2019 Nobel Prize in Economics, awarded to three economists for their
experimental approach to mitigating global poverty as well as the general growth of ran-
domized evaluations (Leigh, 2018; Webber and Prouse, 2018) to study outcomes relevant
to human rights: health and poverty (Banerjee and Duflo, 2012; Vivalt, 2020), education
(Kremer and Miguel, 2004), gender empowerment (Field et al., 2016; Oster and Thorn-
ton, 2011), crime and violence (Blattman, Hartman and Blair, 2014; Blattman, Jamison
and Sheridan, 2017), and even state repression (Slough and Fariss, 2020; Terechshenko
et al., 2019). Conceptually, rigorous impact evaluations include randomized trials, but
they are not limited to RCTs, and they can include other research designs if randomized
trials are inappropriate methodologically. I differentiate rigorous evaluations from non-
rigorous evaluations mainly on the basis of what the evaluations actually measure (inputs
vs. outputs vs. impact), which I discuss in Chapter 3 where I define key terms. For exam-
ple, if an NGO claims to have made a positive impact but only provides anecdotal stories or
only conducts a performance evaluation, I would not consider that to be rigorous impact
information.
In response, some may suppose that rigorous impact evaluations are feasible for service-
8
based interventions but not so (or at least not obviously so) for advocacy-based programs.
While I have sympathy for this distinction, it would have more merit if human rights
organizations such as Amnesty conducted rigorous impact evaluations for their service-
delivery style programs, such as their human rights education programs, issue awareness
campaigns, or letter-writing campaigns. If human rights organizations conducted rigorous
impact evaluations and advertised their rigorous evidence for these types of programs, then
their behavior would be much less puzzling. However, no matter the style of intervention
or complexity of their theory of change, they do not conduct rigorous impact evaluations,
yet they nonetheless do not remain agnostic about their impact.
Second, others may argue that even when impact could be rigorously measured, the
research tools required to conduct the evaluations would be too resource-intensive for
nonprofit organizations (Gorvin, 2009; Naughton and Kelpin, 2015; Root, 2015). This
statement is empirically false for large, wealthy organizations. Based on information pro-
duced by the MEASURE Evaluation Project, an academically managed and USAID funded
organization, the cost can range between $400,000 and $3,500,000 (if the geography
proves logistically difficult and the study period is long) (Evaluation, 2019). Put into per-
spective, Amnesty International’s revenue in 2017 was $354 million, and it spent $173
million on its programs and services (Amnesty International, 2018). Finally, rigorous im-
pact evaluations would help organizations save money and invest in interventions that are
more cost-effective by reducing wasteful spending on interventions that are ineffective or
harmful (Pritchett, 2002).
An explanation based on limited organizational resources is insufficient to explain why
human rights NGOs do not rigorously evaluate their activities or keep such evidence con-
fidential. Every activity costs money, but some costly activities are valuable investments
if organizations enjoy enough benefits that outweigh the costs. The existing explanations
cannot account for why the benefits that accrue from conducting rigorous impact evalua-
tions fail to outweigh the costs of implementing the evaluations.
9
Even if measurement and cost were genuine obstacles, the principled actor approach
cannot explain why (1) human rights NGOs routinely claim to have a positive impact
despite lacking rigorous evidence of it, (2) donors do not seek and are not responsive to
rigorous impact information, (3) accountability organizations do not hold human rights
NGOs accountable to impact despite claiming to, (4) and human rights NGOs continue to
garner money, status, credibility, and authority, despite never providing evidence of their
impact beyond anecdotes.
The collective action approach to NGOs attempts to reconcile some of these puzzles
by representing donors as principals and NGOs as agents in a classic principal-agent rela-
tionship in which altruistic donors delegate the task of producing public benefits to NGOs.
Due to asymmetric information, agency slack, and organizational insecurity, NGOs may
pursue their own organizational interests, such as survival and growth, at the expense of
their donors’ interests (Cooley and Ron, 2002; Prakash and Gugerty, 2010). According
to this approach, we should not necessarily observe human rights NGOs routinely supply-
ing evidence of their impact. Asymmetric information affords human rights NGOs hidden
information about their effectiveness, allowing them to distort their impact to donors by
appearing more effective than they really are. Human rights NGOs would have financial
incentives to claim positive impact without supplying the requisite evidence, if they op-
erate in a competitive environment and face organizational insecurity. Donors, lacking
insight or knowledge into the organization’s effectiveness, would have no recourse other
than to continue supporting the organizations.
While the collective action approach can help to reconcile some of the puzzling features
of human rights philanthropy, it is incompatible with the fact that principals (donors)
exhibit low levels of interest in the outcomes of their agents’ (NGOs’) activities. Surveys
and experiments that examine donor interest in charity effectiveness demonstrate that
donors generally do not search for information about NGO impact: it is not the case that
they are kept in the dark about NGO impact despite searching for the information. When
10
donors are provided direct information about effectiveness in experimental settings, they
are not responsive to it in the way we would expect under the collective action approach.
23
A more promising theoretical approach in accounting for these puzzling features of
human rights philanthropy is to develop a unified theoretical framework that can account
for all the puzzles, rather than developing single ad hoc explanations for each individual
puzzle.
24
The theory that I present in this book, introduced below, can accomplish this task
by accounting for the puzzles introduced thus far as well as additional puzzling donation
patterns that will be discussed in the next chapter.
Introducing a New Theoretical Approach: The Virtue Econ-
omy
Human rights NGOs operate in a fiercely competitive market in which they compete with
thousands of other charitable organizations for funding from individual donors. Market
competition compels human rights NGOs to meet donor demand, otherwise they will lose
funds. Because human rights NGOs must respond to financial incentives in order to sur-
vive, understanding what motivates donors to make contributions to charity is mandatory
for an accurate theory of why human rights NGOs behave in certain ways.
A large body of research spanning psychology, economics, and biology examines why
individuals engage in apparently altruistic behavior and why they donate to charity.
25
This
literature paints a clear picture: altruism is a prestige-seeking activity rather than a truly
“unselfish regard for or devotion to the welfare of others.”
26
Individuals donate to charity
23
For examples, see Karlan and Wood (2017) and Metzger and Günther (2019). For more discussion on
this point, see Chapter 2, which discusses existing research on donor responsiveness to charity effectiveness
and Chapter 3, which presents the data from my survey experiments, demonstrating that donations do not
increase when charities reveal their programs have a positive impact.
24
See Gilady (2018); Simler and Hanson (2018) for a discussion of this approach to theorizing.
25
For examples, see Glazer and Konrad (1996); Griskevicius et al. (2007); Miller (2000); Simler and
Hanson (2018); Zahavi and Zahavi (1997).
26
https://www.merriam-webster.com/dictionary/altruism
11
to send a costly signal —- conscious or subconscious —- that they possess socially desirable
traits such as wealth, a cooperative disposition, generosity, empathy, and group loyalty in
order to attract and maintain allies and mates. Donations are investments in the donor’s
reputation, intended to confer status benefits on the donor rather than to finance the
effective production of public benefits.
In exchange for donor funds, human rights NGOs meet demand by supplying services
that have the spirit of benefiting others (e.g. campaigns, reports, protests, boycotts) but
ultimately serve to benefit the donor’s reputation by helping the donor appear virtuous.
While human rights NGOs routinely claim that they have a large, positive impact, they
rarely, if ever, provide evidence of their effectiveness beyond anecdotal stories of individu-
als that are strongly biased toward positive examples.
My theory of the Virtue Economy focuses on the exchange between nongovernmental
organizations and individual donors, as opposed to institutional donors such as govern-
ments or foundations. When presenting a new theory, it is useful to focus on the most
relevant actors. Individual donors represent the most relevant category of donors for a
theory about signaling because the body of research on the signaling theory of charity
pertains specifically to individual people, not institutions (though there are some notable
exceptions, such as Gilady (2018)). Once the Virtue Economy has been successfully em-
pirically tested against the most plausible or relevant cases, then researchers can evaluate
its ability to generalize to edge cases or more complicated funding models, such as NGOs
funded entirely by foundations or government agencies.
While the Virtue Economy is a general theory that pertains to all manner of nonprofit
social organizations regardless of issue area, I choose to focus the framing, literature,
illustrations, and part of the empirical analysis on human rights organizations for two
reasons. First, human rights organizations represent a most likely case for purely altruistic
donations because donors are geographically distant from any potential beneficiaries of
an NGO campaign. Thus, claiming that even donors to human rights organizations are
12
motivated by self-interest is surprising. In contrast, it would not be surprising to claim
that donors to institutes of higher education are motivated by self-interest, to take one
example. The manifestation of self-interest is more obvious in this case because parents
can help their children attend prestigious universities by donating large sums of money,
to name one possible self-interested motivation. Second, I believe that contributing to the
human rights academic literature is very valuable. It is a corner of social science that could
benefit from theoretical innovation and interdisciplinary connections beyond those already
made with law and sociology. There is also a bit of path dependence on my part. I happen
to have been intellectually raised in the political science human rights literature. It is in this
section of social science that I have spent most of my academic career reading, learning,
and engaging with scholars. This is the literature that I know best, and as such, I personally
can make a bigger impact in this area than I can in other areas I am relatively less familiar
with. Future research will benefit from exploring how well the Virtue Economy generalizes
to other philanthropic sectors. Worth noting, I do not anticipate that the issue area itself
would cause my theory of a virtue economy to generalize or not. In other words, I do not
anticipate that NGOs working on domestic violence or clean water or the environment or
animal welfare would have any more or less reason to sell social signals by virtue of the
issue itself. Instead, if the concept of selling social signals does not generalize well, it is
likely because the funding model is different or the type of self-interested motivation is
different.
The Virtue Economy is a theory about how actors follow their self-interests. Individ-
ual donors follow their self-interested pursuit to build and maintain their reputation, gain
allies, and attract mates. In turn, human rights NGOs follow their organizational self-
interests by supplying what the market demands. Fundamentally, this book presents a
theory about misaligned incentives at two distinct stages, which sustain the inefficiencies
highlighted at the beginning of this chapter. In the first stage, donating to charitable orga-
nizations is primarily a social exercise whereby individuals use conspicuous donations to
13
send costly signals of their socially desirable traits. Donations are primarily driven by sta-
tus, reputational, and social concerns, not a selfless concern for the needs of the beneficia-
ries. Most individual donors lack social incentives to be seen as supporting quantitatively
oriented or data-driven nonprofit social organizations (a point that will be further elab-
orated in the next chapter). In the second stage, the fiercely competitive nonprofit mar-
ketplace compels most human rights NGOs to be responsive to their donors’ preferences,
which creates a negative incentive against adopting a rigorous evidence-based approach to
their programming. Due to the incentive structure, NGO behavior is primarily about these
social and reputational motives, not the achievement of altruistic goals.
My theory should not be interpreted as a criticism of human self-interest or even nec-
essarily a criticism of human rights NGOs. Humans are incurably self-interested, and our
self-interest often serves society well by producing positive externalities. As Adam Smith
famously wrote, "It is not from the benevolence of the butcher, the brewer, or the baker
that we expect our dinner, but from their regard to their own self-interest. We address
ourselves not to their humanity but to their self-love, and never talk to them of our own
necessities, but of their advantages." However, in the context of charity, our self-interest
in developing and maintaining our own reputations for virtue, emotional intelligence, and
group loyalty produces stable waste and gross inefficiencies. This is so because actions
that aid us in appearing virtuous, upstanding, and in possession of good moral character
are not necessarily conducive to effectively, let alone optimally, improving the welfare of
the needy, disadvantaged, or oppressed.
While other challenges to advancing human rights are well-documented (e.g. the self-
enforcement problem with international human rights law and localizing universal norms,
to name only two) (Rodriguez-Garavito and McAdams, 2016), social signaling is arguably
the most serious obstacle, yet it has received no attention. The most serious obstacle to the
NGO advancement of human rights is the misalignment between what is most helpful for
the welfare of those in need and what is most beneficial to the reputation of donors and
14
activists. Social signaling exacerbates the other challenges because it disincentivizes NGOs
to implement, test, and scale cost-effective solutions to these problems. Potentially harmful
programs thrive without serious evaluation, ineffective but showy efforts dominate, and
the rare positive outcome is incidental. The solution to this problem is not to eliminate
social signaling from the equation. Human self-interest will not disappear, but it can be
harnessed. If the principles of highly effective charity practice can be made socially useful
to those in the human rights movement, the competition to signal desirable traits can
take the form of a competition among NGOs to implement cost-effective human rights
interventions.
Roadmap
In the following chapter, I introduce the primary components of the Virtue Economy, my
new theoretical approach for understanding the behavior of human rights NGOs. The
Virtue Economy recognizes the NGO sector as a marketplace with demand-side and supply-
side features. Demand-side refers to the behavior of donors: the motivations individuals
have to donate to philanthropic organizations and the information they seek when making
charitable giving decisions. Supply-side refers to the behavior of NGOs: the services and
programs they implement and the information they provide to the public and their donors.
The first part of Chapter 2 covers the demand-side, exploring a large body of multidis-
ciplinary research about the social and reputational motivations for donating to charity.
This chapter introduces and explains the central motive for donating to charity: to receive
status and reputational benefits. This chapter comprehensively presents the best evidence
from psychology, economics, and evolutionary biology, demonstrating that altruism, par-
ticularly toward non-kin, evolved as a prestige-seeking activity rather than a truly unselfish
regard for the welfare of others. The signaling motive for altruism is not unique to par-
ticular types of donors —- such as wealthy donors or donors from Western countries —-
15
and is not even restricted to humans. Evolutionary biologists have documented reputa-
tional motives for altruism in other animals, including most notably social bird species and
primates.
Chapter 2 covers five of the main areas of evidence pointing toward signaling as a key
motivator for altruism and charitable giving.
1. Giving is conspicuous.
2. Giving is motivated by the desire to attract romantic partners.
3. Giving is competitive.
4. Giving is motivated by pain, effort, and sacrifice.
5. Giving is not motivated by the charity’s impact.
The second part of Chapter 2 combines signaling theory with the reality of market
competition to develop a novel theoretical framework for explaining supply-side NGO dy-
namics. In short, I argue that NGOs operate in a fiercely competitive market. NGOs are
‘substitutable,’ and they must respond to financial incentives in order to survive. Thus, un-
derstanding what motivates donors to make contributions to charity is mandatory for an
accurate theory of why NGOs behave in certain ways. This chapter presents the book’s key
thesis: in exchange for donor funds, NGOs supply services that have the spirit of benefiting
others but ultimately serve to benefit the donor’s reputation. NGOs do not need to prove
they are effective organizations in order to attract donors. Instead, they need to appeal to
donor reputation. Because NGOs and their donors are working together to pursue their
self-interest, the welfare outcomes of program recipients take a back seat to their more
central goals: building reputation, earning social credit, winning friends, and attracting
mates. Though this chapter concentrates on the field of human rights, the argument also
applies more generally to all manner of NGOs and nonprofit social organizations.
Chapter 3 applies the signaling framework to explain why donors are not interested in
or responsive to information about charity effectiveness. The demographics of the recipient
16
and the charity’s overhead are more salient because they say something about the laud-
able intentions of the donor in a way that welfare outcomes do not. Indeed, information
about charity impact is counterproductive in many ways to the demonstration of traits like
empathy because an emphasis on quantifiable welfare outcomes, cost-effectiveness, and
scalability makes donors appear cold, calculating, deliberative, and willing to stop helping
as soon as the calculus changes. Further, rigorously generated impact information would
undermine the unrealistic but widely shared belief that almost all charities are highly ef-
fective. Chapter 3 marshals several pre-registered survey experiments to test whether the
low supply of rigorous impact information and the low supply of accountability for NGO
effectiveness can be best explained by the lack of demand among donors for such informa-
tion. My analysis demonstrates that NGOs are rewarded for exaggerating their impact and
concealing their ineffectiveness.
Chapter 4 may read as a detour from the dissertation’s main goal of establishing the
Virtue Economy as a novel theory of NGO behavior, but it is necessary in order to address
a large empirical elephant in the room. Readers aware of statistical studies in the field of
international human rights may ask: if human rights NGOs do not ensure their programs
have a positive impact, then why does academic statistical research demonstrate that hu-
man rights NGOs are generally effective at advancing human rights globally, in particular
physical integrity rights? Many scholars working in this area frame existing statistical stud-
ies on the effectiveness of human rights NGOs as painting a positive, albeit conditional, ef-
fect. Chapter 4 re-examines these studies and finds that they suffer from numerous critical
and devastating problems. I assemble five of the most-cited statistical papers testing the
effectiveness of human rights NGOs on respect for human rights and analyze the relevant
models from each of the studies, totaling 35 models. I find that human rights NGOs are
often more harmful than they are beneficial, but researchers fail to highlight these harm-
ful effects in their abstracts, introductions, and discussions of results. The replication also
reveals that the narrow positive effects are driven by a very small number of countries.
17
Most of the time, however, the impact of human rights NGOs does not achieve statistical
significance. Researchers fail to emphasize the plethora of non-significant results, choosing
instead to direct their attention to the rare statistically significant positive effect or to spin
the non-significant results as consistent with their theory. A more faithful reading of the
statistical evidence leads to the conclusion that human rights NGOs are ineffective, even
conditionally. This finding is not surprising, given that human rights NGOs calibrate their
activities to satisfy donor reputational needs, not the needs of the stated beneficiaries.
Chapter 5 concludes with a discussion about how my argument fits into the broader
debates within the human rights literature. Selling Virtue reshapes the debate about the
efficacy and purpose of NGOs. Believers and critics of human rights NGOs have both
misconstrued the issue because they both assume the sincerity of the overt statements
made by donors and activists about desiring to help victims, rather than probing the latent
but dominant incentive to improve reputation through charity. The social and reputational
needs of donors, volunteers, activists, and NGOs drive the work, rather than the stated
goals of helping the needy or uplifting society.
18
Chapter 2
The Virtue Economy
...you will acquire many friends, and all the best men will
be attracted towards you: for virtue, in however obscure a
position, cannot be hidden, but gives signs of its presence.
Seneca c. 4 BC to 65 AD
Demand for Social Signals
Donations are Social Signals
Donating one’s own resources —- money, time –– to a charitable cause appears to be the
"essence of altruism," an action that benefits others at one’s apparent expense (Griskevicius
et al., 2007, 86). People around the world have a strong drive to donate time and money
to charitable causes. Americans annually spend 2% of GDP donating to philanthropy. In
2019, Americans donated nearly $450 billion to charity, 70% of which was contributed
by individual donors.
1
Amnesty International alone received donations from more than 2
million individuals globally in 2018, totaling nearly 300 million euros with the addition of
legacies, bequests, and grants from foundations (Amnesty International, 2019).
1
https://www.nptrust.org/philanthropic-resources/charitable-giving-statistics/#:~:text=Americans%
20gave%20%24449.64%20billion%20in,a%205.1%25%20increase%20from%202018.&text=Corporate%
20giving%20in%202019%20increased,a%2013.4%25%20increase%20from%202018.&text=Foundation%
20giving%20in%202019%20increased,a%202.5%25%20increase%20from%202018.
19
Philanthropy is a global phenomenon. According to the Charities Aid Foundation 2018
survey, the most “generous" countries, defined as the proportion of the population that
donated, includes Myanmar, Indonesia, Australia, the United Kingdom, New Zealand, The
Netherlands, Norway, Iceland, Ireland, and Malta.
2
In Myanmar, 88% of people donated
money, 78% in Indonesia, and 71% in Australia. Even in poor or developing countries, a
non-trivial percentage of the population donates money to charity. For example, Ethiopia’s
GDP per capita was only $768 in 2017, but 43% of the population donated money.
3
In
Haiti, with a GDP per capita of $765, 54% of the population donated money.
Why do individuals voluntarily give up their resources to strangers? What is the un-
derlying motivation for donating to charitable organizations? According to kin selection
theory — a popular evolutionary theory of altruism — “individuals can promote their own
genetic future by making sacrifices on behalf of others who carry copies of their genes”
(Van Vugt and Van Lange, 2006, 12). The degree of altruistic sacrifice will be strongest for
close relatives, in particular offspring, and will weaken as the genetic relation decreases.
According to direct reciprocity or reciprocal altruism — another popular evolutionary the-
ory of altruism — altruism can be beneficial in the long-term if our altruism is reciprocated
(Trivers, 1971). “Individuals sometimes act benevolently toward others in the expectation
of a reciprocal act of kindness in the future" (Van Vugt and Van Lange, 2006, 14). Kin
selection theory explains why individuals are altruistic toward their own relatives, in par-
ticular their offspring. Direct reciprocity explains why individuals are altruistic toward
other individuals in a dyadic social exchange (i.e., I’ll scratch your back if you scratch
mine). But neither of these theories can explain why individuals are altruistic toward ge-
netic strangers or people they never intend to interact with (Hardy and Van Vugt, 2006;
Roberts, 1998). What benefit do individuals receive by giving away some of their income
2
These results are based on self-reported data, indicating whether the individual donated money to char-
ity within the last month. The CAF World Giving Index 2018 is based on data collected over a five-year pe-
riod (2013-2017) and includes results from 146 countries. https://www.cafonline.org/docs/default-source/
about-us-publications/caf_wgi2018_report_webnopw_2379a_261018.pdf.
3
For comparison, GDP per capita in the U.S. in 2017 was 78 times Ethiopia’s, at $59,928. https://data.
worldbank.org/indicator/ny.gdp.pcap.cd?year_high_desc=false (accessed April 26, 2019).
20
to complete strangers?
A third evolutionary theory of altruism explains the category of altruism observed in
the philanthropic sector. It has several different names, but some of the more commonly
used ones include the signaling theory of altruism, competitive altruism, social signaling,
and the signaling theory of charity.
4
The basic thesis of this theory is that altruism is a
reputation-building exercise, intended to benefit the giver, not the givee.
According to a large body of research that spans multiple fields, primarily psychol-
ogy, economics, and biology, altruism evolved as a costly signal to conspicuously display
an individual’s desirable traits in order to attract and maintain potential allies and mates
(Miller, 2000; Roberts, 1998; Simler and Hanson, 2018; Tessman, 1995; Zahavi and Za-
havi, 1997). Altruism is a costly signal (i.e. an evolved handicap): an individual must
undergo a fitness-reducing behavior with immediate short-term costs in order to credibly
communicate to other individuals the possession of several important traits: wealth, abun-
dant resources, kindness, trustworthiness, self-control, and a cooperative intent, among
others (Hardy and Van Vugt, 2006; Zahavi, 1995). Upon successfully signaling socially
desirable traits, individuals receive social rewards, increasing their status, and in turn,
increasing their survival and reproductive prospects.
Amotz Zahavi, an evolutionary biologist, introduced the handicap principle in 1975
and later extended the handicap principle to altruism in 1995. A generalized definition of
a handicap is a fitness-reducing characteristic, trait, or behavior that helps individuals test
the quality of potential social exchange partners. Evolved handicaps exist across the animal
kingdom to credibly signal all sorts of qualities: genetic fitness, reliable parenting, wealth,
strength, and social bond (Zahavi and Zahavi, 1997). Examples of evolved handicaps
include the following:
Male peacocks grow large, physically expensive tails to attract females.
4
‘Virtue signaling’ is also commonly used especially in public discourse, but I choose not to use this label
because it has been politicized and is often used to denigrate one’s opposing political party. However, I do
believe this term is technically accurate and even more precise than some of the other terms, such as ‘social
signaling.’
21
Male bowerbirds waste their time looking for rare items in the forest to put in like
piles in front of a large structure they have built. This conspicuously wasteful behav-
ior culminates in a mating ritual whereby they showcase their findings and structures
to females.
Humans engage in conspicuous consumption: they spend extra money on luxury
items such as fancy watches to show off their wealth and status.
Similar to other forms of evolved handicaps, humans often give their time and money
to philanthropic organizations such as Amnesty International to credibly show they are
cosmopolitan and progressive, among other traits. As Zahavi and Zahavi (1997) put it,
“to the altruist, the cost —- the waste —- of altruism is no different than the waste of,
say, growing and carrying a large, heavy, decorated tail like the peacock, or building an
elaborate bower like the bowerbird" (150).
A signaling model of altruism implies that the goal of altruism is not necessarily to
maximize the benefit to others or efficiently produce a public good. Instead, the goal
of altruism is communication: to signal to others that an individual possesses traits that
are desirable to allies and mates. The signaler is more interested in being seen providing
assistance and less interested in the effect of that assistance (Gilady, 2018). As Zahavi
and Zahavi (1997) state, “the helper benefits from the act of helping, and the benefits to
others are incidental —- a side effect” (134). Burum, Nowak and Hoffman (2020) argue
that donors evolved to be insensitive to the efficacy of altruistic actions because "social
rewards tend to depend on well-defined and highly observable behaviours" (1245). The
act of giving, they argue, is more readily observable than the efficacy of such giving. While
"everyone knows what it means to give," not everyone knows what it means to give effec-
tively, and even if they did know, they cannot necessarily be sure that everyone else knows
what effective giving looks like. Thus, when norms such as giving are perpetuated through
the practice of social rewards, the behavior must be easily observable and well-defined for
third-parties to know when to confer the social rewards. Moreover, as will be discussed
below, focusing on the efficacy of one’s giving might actually be counter-productive, as
22
it may signal a disposition for deliberation and logic, traits that are considered by many
social communities to be unappealing (Montealegre et al., 2020).
This underlying motive for behaving altruistically —- to attract potential mates and
allies –– is hidden to the signaler. Individuals are not necessarily conscious of the signals
they send and receive. Robert Trivers refers to this as adaptive self-deception: “we deceive
ourselves the better to deceive others" (Trivers, 2011, 3). Under this framework, donations
to charitable organizations are social signals — they are costly investments in an individ-
ual’s reputation intended to confer status benefits on the donor rather than to finance the
effective production of public benefits.
Altruistic signaling (hereafter referred to as simply ‘signaling’) is a reality in any social
system where people make judgments based on our actions, and we want those judgments
to fall in one direction or another. Signaling is an inevitable social dynamic, one that
actually has many benefits. For example, signaling helps build social trust and a sense
of community, and it also provides a self-interested basis for cooperation with non-kin.
The most important evolutionary benefit signaling bestows comes in the form of more
friends and romantic partners for the altruist. Altruism’s reproductive benefits can accrue
through direct mate choice, as argued by Tessman (1995) and Miller (2000), or indirectly
through high status, as argued by Zahavi and Zahavi (1997) and Boone (1998). Tessman
(1995) theorizes that altruism is a courtship display “designed as an honest advertisement
of the capacity and the intention of the altruist to be a reliable mate and parent.” Potential
mates would find altruistic behavior attractive because it signals their “willingness to work
vigorously and selflessly” for the benefit of the sexual partner (157). Miller (2007) argues
that mate choice is a powerful moral filter and that “we fall in love with...mental and moral
traits” (209). Miller’s central contention is that many moral virtues have been sexually
selected for. Traits such as kindness, generosity, magnanimity, heroism, and sexual fidelity
honestly advertise a potential mate’s genetic fitness and parenting abilities. Buss and his
colleagues find that kindness is an internationally top-rated virtue among sexual partners,
23
beating intelligence, health, and physical attractiveness (Buss et al., 1990).
Signaling is also informative for allies and mates when they are trying to figure out who
would make a good friend, parent, or romantic partner. There are two classes of traits that
conspicuous altruistic displays can signal: first, that an individual has an abundance of
resources or has the ability to procure resources, and second, that an individual possesses
socially desirable personality traits or virtues, such as empathy, warmth, and a cooperative
disposition (Griskevicius et al., 2007, 87). This is a constantly evolving research program,
with many academics from evolutionary biology, evolutionary anthropology, evolutionary
psychology, social psychology, and economics contributing different theories to explain
what traits individuals select in their social partners, and therefore which traits individuals
wish to signal. Researchers have proposed various traits or virtues that are theorized to
enhance an individual’s value on the biological market, whether by attracting mates and
allies or deterring rivals. Some researchers, for example Zahavi and Zahavi (1997), argue
that altruistic behavior is a conspicuous display of an individual’s physical fitness – much
like the peacock’s tail. The wastefulness of the altruistic act serves as an honest signal
that the individual is physically fit and can therefore afford to waste resources. Others, in-
cluding Miller and Curry, argue that altruistic handicaps serve as an advertisement of the
individual’s underlying virtues, which can either communicate the individual’s parenting
qualities to potential mates or an individual’s dominance or submission to potential allies
or rivals (Curry, 2008; Miller, 2007). This book will take a very broad approach at deter-
mining which traits or virtues individuals wish to signal. Ultimately, for the purposes of
explaining human rights NGO behavior, it does not matter which specific traits individuals
signal by donating to human rights NGOs. Instead, what matters is that they are signal-
ing, and they donate to human rights NGOs in order to signal. As long as the following
assumptions are true, the signaling theory of altruism can be directly extended to donors
and NGOs, even if the details regarding the specific traits on display are still being worked
out.
24
1. Altruism is a conspicuous display of an individual’s underlying socially desirable
traits.
2. Individuals directly benefit from the act of helping, by reaping a direct gain in repu-
tation, prestige, or social status.
3. Individuals are motivated to engage in altruistic behavior for the direct gains in rep-
utation, prestige, or social status.
When moving into the context of charity, signaling has many downsides. Signaling
leads to billions of dollars of lost opportunity because charities cater to donor preferences
and reputational concerns, not the needs of the beneficiaries. This produces a lot of wasted
time and money on improving donor reputation without improving welfare outcomes for
recipients. This is mainly so because signal reliability depends on conspicuous costs —-
whether we donate or whether we volunteer. It does not depend on whether our donations
actually help people.
Research in social and evolutionary psychology suggests that when individuals select
social exchange partners, they look for information about a potential social partner’s un-
derlying character, including whether they are trustworthy, cooperative, kind, compas-
sionate, warm, empathic, and loyal (Barclay and Willer, 2007; Barclay, 2010; Buss et al.,
1990; Milinski, Semmann and Krambeck, 2002). Actions serve as a rich set of signals
about the person’s underlying character. As a result, individuals evaluate whole persons,
rather than specific acts, when making character evaluations of others (Uhlmann, Pizarro
and Diermeier, 2015). We ask ourselves, ‘is this person good or bad,’ rather than ‘is this
action right or wrong.’ Critically, the judgement of the consequences of an action are often
dissociated from the judgment of an individual’s underlying character. Uhlmann, Pizarro
and Diermeier (2015) find that certain information is more diagnostic of an individual’s
character than others.
The information that is relevant to signaling a positive character is not necessarily
relevant to producing positive consequences for the recipient of one’s assistance. Three
important diagnostic pieces of information for character evaluations include whether the
25
decision to provide assistance was taken quickly and easily (rather than carefully or de-
liberatively), whether the decision was accompanied by a display of genuine emotions
(as opposed to originating from calculation or deliberation), and whether the decision
involved costs for the decision-maker (Uhlmann, Pizarro and Diermeier, 2015).
Consistent with the theoretical work on person-perception, experimental evidence demon-
strates that individuals are more likely to be selected for social interactions when they dis-
play empathy, warmth, and emotions (as opposed to competence and reason), even when
the decisions made by individuals who display logic and deliberation result in better wel-
fare outcomes, such as more lives saved in moral dilemma games (Everett et al., 2018; Lee,
Sul and Kim, 2018; Montealegre et al., 2020; Rom and Conway, 2018; Uhlmann, Pizarro
and Diermeier, 2015). Actions that produce more positive consequences can signal that
one lacks empathy, thus producing an image as a less moral, less empathic, and less trust-
worthy person, despite the action itself producing better outcomes (Rom and Conway,
2018).
Montealegre et al. (2020) find evidence for social disincentives against donating to
charities based on their impact, since "donors who make charitable decisions by deliberat-
ing about the cost-effectiveness of charities are perceived as less moral and less desirable
as social partners than those who decide based on empathizing with the recipients of the
donation" (2).
5
Many readers may naturally surmise that because they do not feel as though their
kindness or altruism is motivated by a selfish concern for reputation, they are therefore
selflessly motivated. This would be a mistake. Our motivation for reputation and status is
hidden to us and very often does not serve as an obvious proximate cause of our altruistic
or charitable behavior. In order to receive social credit for one’s altruism, one needs to
5
If true, this does not necessarily mean that people will never pay careful attention to the consequences
of altruistic actions. There are two other evolutionary paths to altruism: kin-selection and reciprocity. In
both of these systems, the consequences of an altruistic action are important, indicating that people will
be more interested in evaluating whether the altruistic action produces a positive outcome for the intended
beneficiary. Future research would benefit from explicitly testing this implication.
26
behave conspicuously. However, if one behaves conspicuously, society might recognize this
act of altruism as selfishly motivated and punish the individual. For example, consider
Mathew 6:2-4, "So when you give to the needy, do not announce it with trumpets, as the
hypocrites do in the synagogues and on the streets, to be honored by others. Truly I tell
you, they have received their reward in full. But when you give to the needy, do not let your
left hand know what your right hand is doing, so that your giving may be in secret. Then
your Father, who sees what is done in secret, will reward you." This tension is precisely
why researchers in the science of self-deception believe that human selfish motives for
altruism are subconscious. By hiding the true motive, individuals are more successful at
appearing to be genuinely good. Thus, feeling sincere about one’s aid, or feeling genuinely
kind and compassionate, is fully compatible with having a selfish, status-seeking motive
to be kind or helpful. In fact, the more sincere or genuine one feels, the more likely they
will be to pass society’s lurking cheater-detection systems. As a result, we should not be
particularly confident that we are objective observers of our own motivations. And we
should not let our own intuitions dictate the explanations we provide for complex social
behavior. Because our motives to engage in altruistic behavior are so often hidden to us,
introspection alone will never be able to identify the underlying motives of our apparently
altruistic behavior.
Evidence
Conspicuous Giving
If donations are social signals, then they must be conspicuous for the signal to be received
by a third party. Research on donation patterns demonstrates that charitable contribu-
tions are fundamentally conspicuous and social. For example, according to a nonprofit
survey conducted in the Netherlands, door-to-door solicitation was successful in 94% of
the households solicited, compared with direct mail solicitations, at only 28% (Bekkers,
27
2005). According to the 1994 Independent Sector Survey of Giving and Volunteering con-
ducted by the Gallup Organization, 85% of people who were solicited by philanthropic
organizations donated some money or property (80% volunteered) (Bryant et al., 2003).
In experimental settings, individuals are more likely to donate when their former room-
mate solicits a donation on behalf of the university (Meer, 2011). Additionally, individuals
are more likely to give in the presence of two solicitors than in the presence of one. They
are also more likely to give when the solicitors are well-dressed middle-aged females than
casually dressed college-aged females (Jackson and Latané, 1981). Finally, the influence
of social pressure appears to be in place even in the presence of subtle eye cues (as op-
posed to subtle cues of other facial features) (Bull and Gibson-Robinson, 1981; Haley and
Fessler, 2005; Kelsey, Vaish and Grossmann, 2018; Powell, Roberts and Nettle, 2012).
6
This suggests “the existence of automatic cognitive mechanisms for detecting social gaze
and regulating social behavior accordingly” (Nettle et al., 2013, 35).
Most studies indicate that anonymous rates of donation are quite low, though the rate
of anonymous donating varies slightly depending on the study. Glazer and Konrad (1996)
calculate anonymous donations to non-profits on file at the Pittsburgh Business Library.
The authors find that the highest anonymous donation rates were to the Pittsburgh Phil-
harmonic at 1.29%, Carnegie Mellon University at 0.26%, and Yale Law School at 0.21%.
Raihani (2014) examines anonymous donations from an online British fundraising web-
site, finding that approximately 5 percent of donations are anonymous. Rates of anonymity
are higher among large and small donors, suggesting that individuals who give outside the
normative standards (too much and too little) maintain anonymity to avoid social reper-
cussions. Andreoni and Petrie (2004) experimentally manipulate the degree of donation
publicity in a public goods game. When they give participants the option to donate anony-
mously, only 1.1% choose to do so. The Chronicle of Philanthropy publicly registers do-
nations of a million dollars or more. Of the 9,677 donations in its registry, only 10% are
6
This finding has been replicated by several experiments. The likelihood of donating is influenced by
social gaze, although the mean donation size is not (Nettle et al., 2013).
28
anonymous.
7
Finally, I collected data on the International Rescue Committee’s donations
in 2017, and less than 10% were anonymous.
8
Moreover, nonprofits often provide their
donors with ample opportunity to self-advertise in the form of stickers, ribbons, decals,
mugs, tote bags, t-shirts, and hats (Grace and Griffin, 2006).
Hoffman, Hilbe and Nowak (2018) demonstrate that anonymous donating is a high-
risk, high-reward strategy. It successfully “hides” the reputational motive by making it
more difficult for others to infer that the signaler is only donating for reputational bene-
fits. However, it is a risky strategy because it reduces the potential for others to find out
unless the anonymous donor tastefully informs a select few individuals (e.g. spouse, close
friends).
The social incentives for behaving altruistically are similar for wealthy, white donors
and individuals from poor or developing countries. Research across several public health
contexts demonstrates that social image concerns motivate public health outcomes. The
general finding across numerous studies is that when parents are given the ability to con-
spicuously show the rest of their community that they care for their children, compliance
rates with health interventions increase. For example, when adults can signal in the form
of colorful bracelets that they “contributed to protecting their community from worms,"
the take-up of deworming pills increases more than other material incentives (Karing and
Naguib, 2018, 1). The effects of social signaling have been tested and proven successful
for deworming, childhood immunization (Karing, 2018), and now developing aid projects
are underway to test the effect of signaling on prenatal care for pregnant women.
9
7
https://www.philanthropy.com/factfile/gifts/1?DonorDisplayName_cu=&Category=any&
GiftRecipients_RecipOrgDateline_c=&GiftRecipients_RecipStateFull=any&GiftDonors_SourceWealth_
cu=&GiftDonors_aStateFull=any&GiftYear=any (accessed January 13, 2019).
8
https://www.rescue.org/sites/default/files/document/2813/mkt1801annualreportwebfinal.pdf (ac-
cessed January 14, 2019).
9
https://www.poverty-action.org/study/social-incentives-prenatal-care-and-skilled-delivery-sierra-
leone
29
Mating Motive
If altruism evolved by sexual selection, and if donations are intended to attract roman-
tic partners, then several observations should follow. First, people should select for kind,
altruistic partners, meaning that altruistic individuals should be viewed as desirable ro-
mantic partners. Barclay (2010) experimentally tests this hypothesis by comparing how
individuals rate other people based on a series of altruistic versus neutral descriptions and
corresponding photographs. Participants were asked to rate how desirable the hypotheti-
cal person is for various types of relationships and interactions, such as dating and lending
money. Barclay finds that women are more likely to choose prosocial men for everything
from long-term relationships to one-night stands, indicating that women find altruistic
men to be attractive long-term and short-term partners.
Second, mating motives should increase giving. Griskevicius et al. (2007) test this
hypothesis on college students using a survey experiment. They find that when people are
primed to think about impressing a romantic partner, they are more willing to volunteer
their time when it is publicly visible but not when it is private. Van Vugt and Iredale
(2013) also test this hypothesis experimentally using a public goods game. Typically in
iterated public goods games, as participants reach the final round, they donate less money
to the public good. But the authors find that by placing a male observer in the room, men
engage in less free-riding. They also find that by placing an attractive woman in the room,
men actually donate more as they reach the final round, competing to be seen as the most
prosocial, generous person.
Third, altruism should be associated with mating success. Arnocky et al. (2017) eval-
uate this prediction by comparing self-reported and real altruistic behavior with self-
reported sexual success among college students. They find that people who score higher
in dispositional altruism and people who exhibit more altruistic behavior in a dictator
dilemma game have more lifetime sexual partners and more casual sexual partners.
10
10
Arnocky et al. (2017) measure dispositional altruism using a 20-item self-report scale, which includes
30
Fourth, donating to certain charities should signal various beliefs and traits. The dating
site OkCupid allows users to sort potential mates based on which select nonprofit they
support (e.g. the American Civil Liberties Union and Planned Parenthood). OkCupid’s re-
search demonstrates that certain charities are highly predictive of certain traits and beliefs
among their users, and supporting certain charities improves the chances of being mes-
saged and going on dates. For example, somebody who donates to the ACLU is four times
more likely to consider themselves a feminist and twice as likely to “say they could not
date someone who didn’t support the #MeToo movement." Users who show their support
for the ACLU in turn ‘get 15% more likes, and have 10% more conversations, than those
who don’t."
11
Users who say they do not support the ACLU are seven times more likely
‘to think guns should always be allowed in schools’ and five times more likely to respond
‘what wage gap’ to the question, “should we be fighting to close the gender wage gap?"
This case study demonstrates that donating to certain charities can reveal a great deal of
information about somebody. OkCupid in turn provides this service because people want
to choose their mates on this basis. This is proof that altruistic signaling works in the
dating market.
Competitive Philanthropy
Research in economics employs altruistic signaling to reconcile why government spending
does not crowd out private contributions (Andreoni, 1990; Glazer and Konrad, 1996; Nel-
son and Greene, 2003; Vesterlund, 2006). If donating bestows reputational benefits, and
these benefits factor directly into the individual’s utility function, then donors “no longer
view the donations by others as a perfect substitute for their private donation, and hence
they will not generally prefer that donations are made by others” (Vesterlund, 2006, 572).
answers to prompts such as ‘I have given money to a charity,’ and ‘I have donated blood.’
11
https://theblog.okcupid.com/what-it-means-when-your-okcupid-match-has-the-aclu-righttolove-
badge-184a209b6190
31
This also explains the prevalence of competitive giving.
12
Sugin (2018) documents several
areas of competition in philanthropy. One area is the competitive naming rights bidding
system in which donors bid for the rights to place their name in a conspicuous location
on or inside a building, such as universities and libraries. Raihani and Smith (2015) find
that male donors give more money to attractive female fundraisers, and they donate larger
amounts when donating after a large donation made by another male. Barclay and Willer
(2007) demonstrate that people compete to be more generous when they can benefit from
being chosen as cooperation partners. Herrmann, Engelmann and Tomasello (2019) find
that even children engage in competitive altruism. By the age of eight, children display
similar sharing behavior, sharing more balls with other children when they are observed by
a third person and when they know one of them could be picked for a future collaborative
game.
Martyrdom Effect
An important element to signaling is the conspicuous display of sacrifice, such as donating
large sums of money, volunteering ample time, or engaging in physically strenuous activity
such as charity runs. In contrast, openly benefiting from ‘altruistic’ behavior without paying
a cost undermines the credibility of the signal. Even if the interaction is a win-win in which
both the benefactor and beneficiary benefit, most people do not perceive the self-interested
behavior as a signal of prosociality. We naturally view such an individual as only helpful
to others insofar as she can help herself.
Existing research on person-perception demonstrates that the costliness of altruistic
behavior provides diagnostic information about an individual’s underlying character, such
as the donor’s trustworthiness and willingness to help or cooperate, even when the costly
display is not relevant to the improvement in the recipient’s welfare (Uhlmann, Pizarro
12
Competitive giving is a specific application of the more general phenomenon of competitive altruism.
For a discussion of competitive altruism, see Barclay (2004); Barclay and Willer (2007); Hardy and Van Vugt
(2006); Roberts (1998).
32
and Diermeier, 2015). Moral psychologists refer to this phenomenon as the martyrdom ef-
fect, "as it essentially involves people suffering for a cause they believe in and care about”
(Olivola, 2011, Ch. 4). Research by Olivola and Shafir (2013) demonstrates that "will-
ingness to contribute to a charitable or collective cause increases when the contribution
process is expected to be painful and effortful rather than easy and enjoyable” (91). Their
experiments reveal that people are willing to donate more money on average to a 5-mile
charity run than an outdoor charity picnic, even though the act of running five kilometers
does not confer any additional benefits for the charity’s beneficiaries. In a set of observa-
tional studies, Olivola (2011) calculated the donations to endurance fundraisers, finding
that charities can increase the amount of money raised per donor by making the event
more painful, effortful, and challenging.
In contrast, consider the heavily criticized yet highly effective practice known as "earn-
ing to give," in which somebody earns a high-paying salary in order to donate much of it
to a cost-effective charity. While such a strategy involves the donation of large sums of
money for the purpose of saving lives with a high degree of certainty, it simultaneously
involves earning a lot of money. This strategy has earned a lot of criticism, not on the
basis of its consequences for human welfare, but rather for what such a strategy signals
about the underlying character of the participating donors. David Brooks, in his criticism
of earning-to-give, unintentionally makes the case for the signaling theory of altruism:
"When we evaluate our friends, we don’t just measure the consequences of their lives. We
measure who they intrinsically are. We don’t merely want to know if they have done good.
We want to know if they are good” (Brooks, 2013).
Some social psychologists call this "tainted altruism,” because those who engage in
altruistic behavior but personally benefit from doing so appear less altruistic, even if con-
sequentially their behavior is helpful to others. For example, people prefer a doctor that
practices in a poorer country, earns less money, and saves fewer lives to a doctor that
practices in a wealthier country, earns more money, and saves more lives (Olivola, 2011).
33
Personally profiting from even effective charitable behavior backfires socially. "People
have a more general bias against the very notion of seeking personal gains from charity."
The individual who profits is more likely to be viewed as less moral than somebody who
engages in "equivalent self-interested behaviors that produce no charitable benefit" (New-
man and Cain, 2014, 648). For example, when people are given the option to choose a
fundraiser for a charitable cause, they choose the less successful fundraiser so long as the
organizer keeps less of the money. People are less likely to choose Option A in which their
charity earns $1.1 million and the organizer keeps $55,000 than they are to choose Option
B in which their charity earns $1 million and the organizer keeps $10,000 of it. Option A
is clearly the best fundraising route, given that the organization will generate tens of thou-
sands more in donations. The spoiling factor is that the organizer for Option A pockets
more money. Subjects rated the organizer from Option A as less moral than the organizer
from Option B, and they did not rate him as more beneficial to the organization’s cause.
The importance of sacrifice for costly signaling can explain why the successful charity
fundraiser Daniel Pallotta was heavily criticized in the media and was eventually forced
out of business, despite inventing the category of multi-day fundraising events and raising
an unprecedented $305 million in donations for AIDS research and other causes. The
problem, as many in the media vocally expressed, was that he earned $400,000 per year
as the CEO for the for-profit charity fundraiser (Pallotta, 2008; Newman and Cain, 2014).
Disregard for Impact
In an altruistic signaling system, any public benefits generated by donations are side effects
for both the signaler and the signaler’s audience (Zahavi and Zahavi, 1997). If donors give
to charity primarily to signal their socially desirable traits, then what would interest them
is not the production of public benefits but rather the receipt of reputational benefits. For
this reason, philanthropic outcomes should be of little interest to donors and members of
their social community. What is more important to donors is signaling that they care about
34
certain groups or possess certain traits, not whether their donations produce measurably
good outcomes for the recipient.
At minimum, the signaling theory suggests that donors are not interested in rigorous
evidence of charity impact. Ample survey and experimental evidence support the predic-
tion that donors are not motivated to give based on welfare outcomes. Survey research on
donor motivations suggest that most people are quite charitable, but most people do not
make charitable giving decisions with the intent to donate to the most effective charities.
Most donors will say that they donate to make a difference. For example, 94% of high
net worth American donors say they typically give when they believe their gift can make a
difference. But 78.3% do not monitor or evaluate the impact of their donations. Only 37%
of high net worth American donors say one of their top challenges to giving is monitor-
ing their giving to ensure it has its intended impact (Osili, Clark, St. Claire and Bergdoll,
2016).
While most American donors in the top 30% of U.S. households in terms of income say
they care about nonprofit performance (85%), only 30-35% do research on any charitable
gift in the course of a calendar year (Camber Collective, 2015; Hope Consulting, 2010,
2011). Of those that do research, 63% do so to validate the nonprofit they are seeking
to give to (Hope Consulting, 2010), and only 3-10% do comparative research on charities
before donating (Camber Collective, 2015; Hope Consulting, 2010, 2011). Only 6.5%
of donors claim to do comparative research on how much charities are accomplishing
before making a charitable contribution (Camber Collective, 2015). Even this number
is likely inflated. When asked open ended questions about their research purposes and
what information they were looking for, the most common response is that donors want to
make sure their contribution is used to fund program services, not overhead. For example,
one donor responded that they wanted to “make sure that the money was actually being
used for the cause and not to line others pockets,” and another donor reported that they
“researched % going to cause vs. % going to administrative costs.” Of the 96 open-ended
35
responses, none elaborated on how the donor researched the effectiveness of the charity.
13
This finding corresponds with research by Gregory and Howard, who find evidence of
a ‘nonprofit starvation cycle’ in which donors prioritize the overhead ratio and financial
transparency over nonprofit performance, which pressures nonprofits to spend less on
administrative overhead (Gregory and Howard, 2009).
14
In addition, when donors do
research, the evidence suggests their research is not very thorough. 72% of donors that do
some research spend less than one hour researching. Only 2% spend one to two days (less
than 1% of the overall sample) (Hope Consulting, 2011).
Experimental data support the survey evidence: typically, donors are affected by other
factors more than they are by welfare outcomes. For example, Metzger and Günther
(2019) ran an experiment where they vary what kind of information study participants
can purchase before making a donation. This allows them to measure the type of in-
formation people are more likely to seek when making real donations. They find that
participants demand information about the recipient type of the NGO the most, followed
by administrative overhead, and last, the NGO’s aid impact.
Verkaik (2016) conducted two online experiments to see whether UNICEF’s advertise-
ment of impact information versus output information (how the funds are used) increases
donations. Verkaik finds that including output information about UNICEF’s anti-malaria
bed nets increases generosity, but the addition of impact information does not.
15
This is
true even among the highly educated sample of participants. It takes priming the highly
educated to think about deliberation (by having them solve math problems prior to donat-
ing) to increase donations based on impact information.
13
The other responses included that the donor “saw a need," talked with someone from the nonprofit,
is involved with the organization, has family who has been supporting the nonprofit for years, and so on
(Camber Collective, 2015).
14
See also Pallotta (2008) for a more general discussion about the ‘economic apartheid’ created in the
nonprofit sector due to an obsession with charity overhead.
15
Output information reads, “You can give a long-lasting insecticide treated bed net with every 4$ you
donate, which protect the children against malaria during the night, when most infections are incurred."
Impact information reads, “Your bed nets are a highly effective way of combating malaria. Using the nets
has been proven to halve the number of malaria incidences and to drop the child death rate by one fifth"
(Verkaik, 2016, 13)
36
Karlan and Wood (2017) conducted a field experiment, measuring the effect of infor-
mation about an NGO’s effectiveness on donation behavior. They test if individuals who
previously supported the NGO, Freedom from Hunger, were more likely to donate when in-
formed via fundraising letters that the NGO’s projects effectively reduce poverty. The first
group received a letter with a qualitative story about an individual beneficiary. The sec-
ond group received a letter with information about the individual as well as a note that the
NGO’s positive impact had been scientifically proven. They find that, on average, “donation
behavior does not change when previous donors are presented with evidence of the char-
ity’s effectiveness in achieving its goals.” However, this effect is driven by two factors: (1)
“small prior donors donate less in response to being told about the scientifically-measured
effectiveness of the charity,” and (2) large prior donors donate more (Karlan and Wood,
2017, 1).
The low level of interest in impact is not unique to donors in wealthy, developed coun-
tries. Donation behavior to human rights organizations in the Global South also map onto
donation patterns observed in developed democracies. Field experiments conducted by
Absar et al. (2017) in Mexico City confirm that people donate the most to organizations
when their financial transparency is described, not when their impact is elaborated.
16
Caviola et al. (2014) find, experimentally, that some of the preference for the overhead
ratio may be its ease of evaluation. When a charity’s cost-effectiveness is presented in an
easy and accessible manner as that of the overhead ratio, individuals choose the more cost-
effective charities over the lower overhead charities. However, Berman et al. (2018) find
that individuals tend to ignore effectiveness information, even when it is easily compara-
ble, when choosing across different causes (rather than across different charities within
the same cause). Their “results suggest that individuals view charity as a relatively sub-
jective decision...one in which people often feel justified to discount welfare-maximization
concerns in order to choose in accordance with their personal preferences” (Berman et al.,
16
Although impact information does increase donations compared to the control group, it is not clear if
this difference is due to random chance because the authors do not perform an hypothesis test.
37
2018, 835).
17
Finally, surveys and experiments have repeatedly confirmed the finding that human
generosity is insensitive to scope: we are more likely to donate when the charity helps one
identifiable victim than when the charity helps many statistical victims (Kogut and Ritov,
2011).
Alternative Explanations
There are two additional varieties of consumption that likely influence donations to charity
to some degree, and while they also serve to selfishly benefit the donor, they will not be
the focus of this book. First, donors may benefit materially from the donation. Second,
donors may benefit emotionally by feeling good after donating. This is called warm glow
giving. Both varieties will be addressed in turn.
A common argument made about wealthy donors is that they donate to charity in order
to receive tax benefits. For example, when the wealthy families of Europe made competi-
tive and conspicuous bids in 2019 to cover the reconstruction efforts of the Notre Dame af-
ter its spire caught fire and collapsed, many criticized them (McAuley, 2019; Reich, 2019).
Some speculated that the billionaire donors were giving money “to receive a generous tax
break from the state" (McAuley, 2019). This public criticism prompted a response from
Arnault and Pinault, France’s “richest dynasties", who both said they were not looking
for tax benefits. The Pinault family released a public statement saying “The donation for
Notre-Dame de Paris will not be subject to any tax deduction. The Pinault family considers
that it is out of the question to burden French taxpayers." And “Arnault told shareholders
that his family holding company had already hit its ceiling on tax deductions for charitable
donations" (McAuley, 2019).
There are some common misconceptions about tax benefits and charitable giving, so
17
To address the possibility that human reasoning works similarly in other domains, they find that subjects
do not view investment decisions as a subjective decision.
38
it is worth getting into the details. Even if the Arnault and Pinault families could receive
government subsidies, this does not mean that they are materially benefiting from their
donations. Government subsidies influence the cost of donating; if governments offer a
50% deduction, then a donor can donate $1 while only paying 50 cents. In the end, donors
are giving away some amount of their own money; they are not financially enriching
themselves no matter the tax subsidy rate. Since government subsidies only affect the cost
of donating, they can, at most, affect how much money people donate, but they cannot
explain why people donate in the first place. In order to explain why people donate, one
would need a theory of the benefits people receive, which is the task of this chapter.
Another common argument depicts charitable giving as a consumable good in the same
way entertainment is a consumable good. For example, rich patrons might donate money
to an art museum because they live nearby and frequent the museum. There are indeed
examples of this kind of behavior. Peter Brant, an art collector, created a new nonprofit art
center (although visits are by appointment only), called the Brant Foundation Art Study
Center, located next door to his estate and his private polo club in Connecticut. As the New
York Times reported, “critics wonder whether taxpayers are helping subsidize wealthy col-
lectors’ multi-million-dollar purchases with little public benefit in return" (Cohen, 2015).
This is an interesting form of self-serving, material consumption. And in this case,
the consumption is also conspicuous because the wealthy art benefactor can show off his
wealth and generosity to celebrities and “art-world cognoscenti" who frequent his gallery
and attend his “twice-a-year gala openings" (Cohen, 2015). But it is important to separate
the two forms of selfish consumption at work here. The first is material consumption: the
benefactor creates an art museum next door to his home so that he has a place to store
his art collection and view it whenever he pleases. The second is a reputational story: he
uses the art museum to show off to his peers that he is wealthy, cultured, generous, and
high status. This book focuses on the second version of consumption because it permeates
the charity world more than standard material consumption. I do not doubt that some
39
patrons do materially benefit from charitable donations in the same way illustrated by
this example, but this instance will be limited to specific sectors in philanthropy, such as
art and music, which make up a small share of the charity market, estimated at five per-
cent.
18
The larger, more important consumption story is one of conspicuous consumption
for reputational benefit, which is not restricted to specific philanthropic sectors.
Donors may also emotionally benefit from donating. The theory of warm glow giving
suggests that people donate to charity because it makes them feel good about themselves
(Andreoni, 1990). I do not disagree that donors feel good about themselves when giving.
But feeling good is not a deep or ultimate explanation for why people donate. It begs the
question: why do people feel good when they donate? The signaling theory of altruism
and its application to charity offer a satisfying answer for why we feel good. The warm
glow is probably the good feeling we get believing that other people think well of us.
Readers should interpret warm glow giving and social signaling as complementary.
Further, the warm glow theory cannot explain particular features of charitable giving.
For example, if people donate to charity to consume a private warm glow, then the research
on the conspicuous nature of charitable giving would not make sense. Anonymous dona-
tion rates should be higher if people privately consume a good feeling like they privately
consume good feelings from other products, such as entertainment or food. Likewise, the
research on the mating motive does not make sense from a warm glow perspective. If
people only cared about a warm glow, then why do men donate more money when in the
presence of an attractive female?
Another alternative explanation is that donors do care about public benefits, but (1)
they have cognitive biases that produce irrational behavior, or (2) they have limited time,
so they rely on heuristics when deciding where to donate. When people have limited time,
and they cannot do extensive research on charity effectiveness, they refer to heuristics.
But because these heuristics do not always align with rational expectations, charitable
18
https://givingusa.org/see-the-numbers-giving-usa-2017-infographic/ (accessed May 7, 2019).
40
behavior can appear irrational and guided by cognitive biases.
There are many documented heuristics and biases in charity that might each serve
as an alternative explanation to signaling in specific scenarios. However, they cannot
explain in a comprehensive or holistic way the great many puzzles outlined in this chapter.
In other words, a theory of heuristics or a theory of cognitive biases does not offer a
comprehensive or coherent theory of charity. For example, donors rarely research charities
before donating, and there is scant evidence that they research the cost-effectiveness of the
charity. It might be true that researching the effectiveness of different charities takes time
(more on this in the next chapter), so donors use short-cuts such as the name-brand of the
organization or its overhead ratio. As will be presented in Chapter 3, this explanation is
not supported by the evidence. But even if there were some truth to the explanation, it
does not explain all of the other features of charitable giving, such as the mating motive,
the visibility of giving, and the martyrdom effect. On its own, it acts as a one-shot, post hoc
explanation for a behavioral pattern that can be ex ante predicted by the social signaling
theory.
The signaling theory of altruism has advantages such as parsimony and breadth. Heuris-
tics or cognitive biases serve mainly as post-hoc explanations. The theoretical origin for
most of these biases comes from a public benefits perspective: researchers assume that
donors are primarily motivated by public benefits. When they observe behavior that does
not fit with this theory, they call it a bias. Most of these biases were not developed ex
ante from first principles: they are deviations away from the very theory motivating their
research. If anything, the presence of so many biases and heuristics provides evidence that
donors are not primarily motivated by public benefits. At some point, researchers must
look at the evidence and acknowledge that there are only so many more holes the theory
can endure.
41
Supplying Social Signals
When enough people want to invest in their reputation by sending costly signals to com-
municate their socially desirable traits, a market will emerge that accommodates and ser-
vices this demand. Philanthropy should be viewed as such a market (Glazer and Konrad,
1996; Harbaugh, 1998; Nelson and Greene, 2003; Posner, 2000). Given the importance
of making such a transfer of cash conspicuous in a social signaling system, there will be
demand for institutions that can “certify the transfer of money from donor to donee.” The
modern-day manifestation of these institutions are nonprofits, charities, and NGOs. As
Posner argues, one of their purposes “is to take people’s money in a public way” by either
advertising the donations in programs or plaques or by providing the donor with the means
to self-advertise through decals and clothing. Charities “sell reputation; the price is a cash
gift” (Posner, 2000, 65-66).
This book reconceptualizes human rights NGOs and their donors as operating in a mar-
ketplace where the key product for sale is an altruistic signal. In a virtue economy, donors
pay to signal their virtues, and in exchange, human rights NGOs supply these signals in
the form of apparently well-intentioned advocacy, services, and assistance. Rather than
modeling donors and NGOs as though they are principled actors in an activist network
(Keck and Sikkink, 1998; Risse, Ropp and Sikkink, 1999, 2013) or stuck in a principal-
agent dilemma (Prakash and Gugerty, 2010), NGOs should be modeled as though they are
firms selling products, and donors should be modeled as though they are consumers buy-
ing products. Human rights NGOs execute the sales of their signaling services by creating
lifestyle brands and building brand communities around them. The stated beneficiary of
NGO programs is not a central player in the economic exchange between donor and NGO,
although the identity and other characteristics of the beneficiary can moderate the success
of the social signal. Prakash and Gugerty (2010) view "advocacy NGOs as special types
of firms which function in policy markets" (3). I argue that human rights NGOs are not a
special type of firm, they are quite the ordinary type of firm.
42
By articulating a new theoretical framework for NGOs, this book helps to clarify and
distinguish many concepts and ideas that are conflated in Prakash and Gugerty’s Advocacy
Organizations and Collective Action. While the main thesis of their work is that donors and
NGOs operate in a principal-agent relationship, Prakash and Gugerty nonetheless include
a handful of chapters that introduce concepts that are much more logically consistent with
the Virtue Economy.
For example, Clifford Bob contributes a chapter that discusses the "market for hu-
man rights” (Bob, 2010). He conceives of human rights activism “as a global market-
place in which the supply of [reported] abuses interacts with the demand for rights issues
by donors.” He goes on to write, “human rights advocates provide donors with psychic
and reputational benefits gained from the belief that monetary and material contributions
help reduce distant oppression” (134). His theory mainly concentrates on the interaction
between local rights groups vying for attention from international human rights organi-
zations (what he calls "Market 1"). But his conception of supply and demand within a
human rights marketplace makes his theory compatible and complementary to my con-
ception of the Virtue Economy, which mainly focuses on the donors and the human rights
organizations (what he calls "Market 2"). Thus, if one were to combine my theory with
the theory presented by Clifford Bob, one would have a coherent theory of the entire sup-
ply chain within the human rights marketplace: from donors, to the intermediary human
rights organizations, to the aggrieved groups or the local suppliers of the “raw materials.”
Maryann Barakso also contributes a chapter, which furthers a preliminary conception
of the human rights marketplace by arguing that, like firms, “advocacy organizations de-
velop distinctive brand identities as a means of differentiating themselves to internal and
external audiences in a highly competitive market” (Barakso, 2010, 156). Nongovernmen-
tal organizations engage in branding to attract and retain investors and customers. She
argues that the tactical styles of various advocacy organizations contribute to their brand
identity. For example, Greenpeace operates in a niche within the environmental policy sec-
43
tor partly because of the contentious tactics it uses to raise awareness and demand policy
change.
While Bob’s and Barakso’s theories clearly deviate from the principled actor approach,
they do not belong within the collective action approach, even though they are presented
as such. Their theories of markets, supply, demand, and branding are logically inconsistent
with a principal-agent model in which principals delegate services to agents. Instead,
their theories naturally conceive of human rights NGOs as organizations producing or
manufacturing human rights products (e.g. reports of abuses for Bob (2010) and tactical
repertoires for Barakso (2010)) consumed not by the stated recipients (i.e. the oppressed
or aggrieved populations) but instead by the donors. These arguments correspond to the
Virtue Economy.
The key components of the Virtue Economy are as follows:
1. NGOs compete in a market.
2. The economic exchange between donor and NGO is voluntary, with both sides bene-
fiting from the transaction.
3. Meeting market demand, NGOs supply altruistic signals.
4. NGOs pursue branding and marketing to maximize the signaling value of their ser-
vices.
5. Altruistic signals tend to succeed and propagate throughout the charity market on
the basis of perceived intentions and conspicuous effort, not the objective impact of
the NGO on human well-being.
The following sections take each component in turn.
NGOs compete in a market
Nonprofit organizations operate in a fiercely competitive market in which they compete
with millions of other charitable organizations for funding from individual donors. In the
United States alone, over a million nonprofit organizations are registered with the IRS
44
(McKeever, 2018). Further contributing to market competition is the low barrier to en-
try, with nearly all nonprofit applications for 501(c)(3) status receiving approval, between
97.71% (minimum) and 99.26% (maximum) from 1998 to 2008. Reich, Dorn and Sut-
ton (2009)’s investigation of the IRS 501(c)(3) approval process finds, “oversight of the
creation of nonprofit organizations...is weak, bordering on non-existent" (3-4). The mar-
ket has grown more competitive over time, with the latest decade of data demonstrating
53% growth in the number of public charities from 1995 to 2005 and with the IRS ap-
proving 501(c)(3) status for an additional 50,000 nonprofit organizations annually. The
international nonprofit market is likewise highly competitive. Since 1950, the number
of international nongovernmental organizations has risen from a small handful to nearly
60,000 by 2010 (Bush and Hadden, 2019).
Market competition compels charities to meet donor demand. In a competitive charity
market, there is an abundance of substitutes, which means donors have many options
when deciding where to donate. If they are not satisfied with the product on offer from
one charity, they can easily find substitutes and donate elsewhere. As a result, donors can
incentivize organizational behavior they like and disincentivize behavior they do not like
by simply choosing where to donate their money. In turn, market competition compels
human rights NGOs to follow their financial incentives by supplying what donors want,
otherwise they lose funds. Thus, market competition points to a donor-driven explanation
for human rights organizational behavior.
The market responds swiftly (Risse, 2010). Donors respond quickly by punishing orga-
nizations when they are unhappy. Human rights NGOs, in turn, adjust quickly in response
to donor demand. One example is Amnesty International’s foray into campaigning for the
full decriminalization of sex work. In 2015, Amnesty began advocating that no consenting
adults involved in sex work should be criminalized —- including the buyers. This pol-
icy position angered many public feminists as well as many Amnesty donors, who argued
instead for the decriminalization of the sex worker and the continued criminalization of
45
the buyer (Bazelon, 2016). Many celebrities, writers, feminists, and NGOs publicly de-
nounced Amnesty’s policy by arguing that prostitution is “sexist, racist, and classist,” and
that Amnesty’s policy would “fuel rape culture."
19
As many as 500 members from Amnesty’s
Sweden office left the organization after the policy roll-out (Lofgren, 2015). Soon after
the policy announcement, Amnesty essentially killed the campaign. They stopped issuing
reports on sex work, and aside from a few meager attempts to respond to the backlash,
they did not attempt to change public opinion in favor of their policy.
20
The economic exchange between donor and NGO is voluntary
The collective action approach introduced the firm analogy to the study of NGOs by por-
traying donors as shareholders and NGOs as firms (Prakash and Gugerty, 2010). My frame-
work modifies the firm analogy. Rather than depicting donors as principals who delegate
to NGOs, my framework depicts donors as consumers buying services from human rights
NGOs, analogous to the corporation and consumer rather than the corporation and its
shareholder. Just as companies “produce what people desire, instead of trying to convince
people to buy what the company happens to make” (Miller, 2010, 39) so too do human
rights NGOs. They do not convince potential donors to “buy” whatever they happen to
sell; instead, they convince their donors that they sell what their donors already desire to
buy.
This economic exchange is neither altruistic nor exploitative. Instead, both actors see
gains from trade. By working together, donors and human rights NGOs participate in a
transactional relationship whereby they mutually pursue their self-interests. The princi-
pled actor and collective action approaches both contain some notion of an economic ex-
19
See the Change.org petition https://www.change.org/p/amnesty-international-vote-no-to-
decriminalizing-pimps-brothel-owners-and-buyers-of-sex and https://www.nytimes.com/2016/05/22/
magazine/the-5-816-issue.html and https://www.rollingstone.com/politics/politics-news/5-reasons-to-be-
wary-of-amnestys-prostitution-policy-198762/.
20
https://www.rollingstone.com/politics/politics-news/5-reasons-decriminalization-protects-sex-
workers-rights-91292/
46
change, though the collective action approach conceives of an exchange between principal
and agent in a policy market, with the agent exploiting the principal under informational
asymmetry. According to my framework, the exchange is completely voluntary, and the
human rights NGOs do not engage in corruption or exploitation of their funders. In this
sense, I agree with the principled actor approach’s emphasis on the mutual cooperation
between donors, activists, and organizations within transnational activist networks. My
theory does not imply that donors are always satisfied with their signaling products. But,
if they are unsatisfied, there are thousands of other organizations where they can donate
their money.
Under the principled actor and collective action approaches, donors and human rights
NGOs work together in a private, bilateral exchange. My framework stipulates that the
exchange is bilateral but not private. To the contrary, the exchange is quite conspicuous
because of the social purpose of charitable giving. I add another set of actors, the donor’s
social community. These actors influence the degree to which the donor benefits reputa-
tionally from associating with any given charitable organization. To the extent the donor’s
social community is not impressed with the donation, the donor will find somewhere else
to donate.
NGOs supply social signals
Amnesty International’s December 2018 fundraising letter includes a bumper sticker that
makes the altruistic signal for sale rather obvious: “show the world that you are committed
to protecting human rights. Please display this decal where it will be noticed!" Translated
into explicit signaling terms, Amnesty is telling its donors, “In exchange for your donation,
we are providing you with a social signal. People will now know you embody a package of
desirable traits. Please put this decal where you can receive maximum signaling value!"
I conducted an interview with a high-level researcher from Amnesty International who
suggested that issues can come and go out of fashion in human rights practice. Some-
47
times, a certain way of researching or a certain type of issue is no longer considered “sexy"
anymore. Competition for donations contributes to shifting attention-span. Donors like
to donate to organizations that cover new issues — these organizations are seen as “dy-
namic." For example, “trial observation and fair trials...became less and less of a priority
to Amnesty, to Human Rights Watch, and to others just because partly, not fully, but partly
because it just wasn’t considered that sexy. So if you’re looking at the 2017 annual report
and thinking you’re getting all the up-to-date information on fair trial concerns around the
world, you’re in dreamland."
21
The practice of consumer marketing is helpful for understanding the “sale" of ethical
products in the NGO space. Companies that sell products to individual consumers use
market research to understand what consumers want to buy. There are telling signs that
charities also want to know what donors want to “buy." Consider the fact that most research
in philanthropy is oriented at understanding how to increase fundraising as opposed to
understanding what interventions and programs offer the most cost-effective solutions to
achieving the organization’s stated goals. For example, to support the claim that the study
of philanthropy is thriving, one author notes that in the past 15 years, “there have been
250 dissertations on fundraising in higher education” (Callahan, 2017). It appears that
most of the concern in philanthropy is about how to increase revenue rather than about
how to cost-effectively produce public benefits.
Human rights NGOs are no exception. For example, Amnesty International’s only mea-
surable target is its “movement growth,” a euphemism for revenue growth, and it is its only
goal for which it has conducted randomized trials. Amnesty USA’s Innovation Lab collects
proposals for and funds various fundraising techniques to reach the organization’s goal of
attracting two million supporters by 2020. “The Amnesty Innovation Lab will determine
the most effective strategies for growing our supporter base through testing...Successful
tests will provide the foundation to scale up our growth efforts to meet our goal of two
21
Interview conducted on June 16, 2017.
48
million active supporters by 2020."
22
This language does not appear when Amnesty dis-
cusses its programs or campaigns.
The issue-awareness impulse is strong in human rights work as it is a central method for
the activist’s theory of change. Under the principled actor approach, human rights NGOs
strategically frame issues and persuade actors to adopt their norms, values, and principles.
Under the signaling framework, human rights NGOs implement campaigns for positions
their donors already hold, shining a spotlight on issues where many spotlights are already
shining. They do not necessarily want as many people as possible to be aware of the prob-
lem. Instead, they want as many people as possible to know that their donors care about
the problem. That some in philanthropy have criticized this practice (e.g. “Stop Raising
Awareness Already") provides suggestive evidence that this may be a pervasive practice in
philanthropy, warranting some scholars to argue that “too many organizations concentrate
on raising awareness about an issue" (Christiano and Neimand, 2017). High-visibility, pop-
ular issues will be more likely to attract attention from human rights NGOs than neglected
but worthy causes. An omission would likely be interpreted with suspicion by donors, as it
would threaten to make the organization (and by association, its donors) seem callous and
uncaring. They would be more likely to be depicted as bystanders. The pressure to signal
one’s position and alliance on popular issues may be so powerful that it reaches the level
of moral imperative. It is unsurprising that both Amnesty International and Human Rights
Watch have recently reported on #MeToo, the migrant caravan, Khashoggi, gun violence,
and the Amazon wildfires, among many other well-known issues. Donors want to be able
to signal where they stand on these salient issues. A human rights NGO does not need to
convince donors to change their minds to attract their donation. Instead, human rights
NGOs convince donors that they represent the views donors already hold and will provide
assistance in signaling their commitment to these views.
22
Information about Amnesty USA’s Innovation Lab can be found here https://www.amnestyusa.org/
amnesty-innovation-lab-proposal-form/. Information about the testing and proposal requirements can be
found here https://www.amnestyusa.org/wp-content/uploads/2017/06/Guidelines_Tests.pdf.
49
Organizational growth and status are consistent and complementary to the donor’s
reputation. As Ostrower explains, “while philanthropy needs money to survive, it needs
status to attract money" (Ostrower, 1995, 13). The collective action approach portrays
organizational status as a selfish goal of the NGO, one that subverts the intentions of the
donors. However, donors want to associate with high-status organizations. As illuminated
by Ostrower, donors are prone to compete for the privilege of volunteering or donating to
high-status organizations. An organization’s reputability and prestige will directly reflect
back on the donor. Human rights philanthropy is thriving, and gatekeeping NGOs such as
Amnesty and Human Rights Watch enjoy high levels of status and prestige in the human
rights community, not because they have proven themselves to have the highest levels of
organizational effectiveness in terms of advancing human rights, but rather because they
provide the best social signaling services. In a social signaling market, organizations gain
status, prestige, and donations by working on causes that help their donors signal their
virtue, not by implementing the most cost-effective programs at advancing human rights.
NGOs create lifestyle brands to facilitate the sale of social signals
Human rights NGOs execute the sales of their altruistic signals by creating lifestyle brands
and building brand communities around them. This marketing technique is well-understood
in the context of for-profit businesses (Hassay and Peloza, 2009; Muniz and O’Guinn,
2001). One can search "lifestyle brand" or "brand community" on the internet for millions
of examples of consumption communities, defined as "groups of people with feelings of
shared well-being, shared risks, common interests and common concerns centered on the
consumption of a common object" (Hassay and Peloza, 2009, 26). Examples include Nike,
a company once known for producing running shoes but now is associated with athletic
subculture, or Victoria’s Secret, an American-based lingerie and clothing retailer headquar-
tered in Ohio that brands itself as English upper class and pretends to be headquartered in
London.
50
Firms often create lifestyle associations with their products so that when customers
use their products, other people associate the customer with that particular lifestyle. Con-
sider the beer product Corona. The advertisements do not describe specific features about
the product, such as the price, alcohol content, or whether the beer tastes good. In-
stead, Corona’s advertisements simply associate the product with the beach so that when
a customer brings Corona to a party, everybody understands the customer is bringing a
beach-like party atmosphere without having to say a word about it (Simler and Hanson,
2018).
Nonprofit organizations engage in the same form of marketing and branding (Barakso,
2010). Human rights NGOs create lifestyle brands so when donors make their support for
the organization conspicuous, they advertise a wealth of information about themselves.
Think back to the example of OkCupid’s research about what support (or lack of support)
for the ACLU signals: support goes much further than simply associating with the specific
work of the organization. It can also signal many additional qualities about the donor,
including a variety of beliefs, values, interests, and hobbies. Like Corona, Amnesty Inter-
national has also created a lifestyle brand. Consider a recent Amnesty merchandise catalog
sold in the Netherlands, which features products being used by a young woman riding a
bike in an upper-middle class urban environment (Figure 2.1). By wearing Amnesty’s fair
trade recycled tea bag that says “equality, justice, and freedom," a donor would signal that
she is cosmopolitan, urban youth, environmentally friendly, a local shopper, and upper
middle class. Other products for sale include a fair trade mini herb garden from India and
bracelets made by poor women from Nepal.
In addition to creating lifestyle brands through advertisements, the sale of branded
merchandise, strategic issue-adoption (Bob, 2010), and tactical repertoires (Barakso, 2010),
NGOs also build brand communities around their lifestyle brands. Brand communities are
well-understood in marketing research, including nonprofit marketing (Hassay and Peloza,
2009; Muniz and O’Guinn, 2001). Successful examples in the private sector include Apple,
51
Figure 2.1: Amnesty International Netherlands
Catalog
Tesla, and Harley-Davidson. Brand communities in the consumer space have been proven
to lead to increased market share, increased sales, and lifetime loyalty (Hassay and Peloza,
2009). Amnesty International is probably the best example of a brand community in the
charitable world. As argued by Muniz and O’Guinn (2001), brand communities repre-
sent social relationships among admirers of a brand, they are non-geographically bound,
and successful brand community building leads to near “fanatical loyalty" and lifetime loy-
alty of consumers (Hassay and Peloza, 2009). These features are readily observed with
Amnesty. Social relationships are embedded in the organization’s "tactical repertoire" via
social gatherings, protests, and in-person letter-writing campaigns. Amnesty even mar-
kets itself as a global movement rather than a bureaucratic organization. As for fanatical
loyalty, it is common to run into an Amnesty member who has been a member for life. I
conducted an interview with a high-level human rights lawyer from the Netherlands who
married a fellow Amnesty member and has been a lifetime member.
23
Another component
of brand-building includes shared rituals and traditions, such as Amnesty’s numerous chap-
23
Interview conducted on June 14, 2017.
52
ter meetups and annual in-person letter-writing campaigns that it has been hosting since
its founding in 1961. Brand communities also produce a shared sense of moral respon-
sibility and higher purpose, similar to what religion can provide its adherents. Hopgood
(2006) discusses the similarities between religion and human rights work. He argues that
Amnesty International successfully built a global movement for secular-minded individ-
uals who still craved the community-bonding and higher sense of purpose that religion
can provide. Hopgood even writes, “the social institution [Amnesty] most resembles is a
chapel or a meeting house" (Hopgood, 2006, 3).
When observing specific actions of human rights NGOs, scholars should ask what these
actions are doing to build the organization’s brand community. How does this action
establish the organization as a lifestyle brand? How is this action going to make the donors
look? Will it make them appear cosmopolitan and multicultural? Will it make them appear
to care about the environment? Or will it make them appear to be calculating, deliberative,
and scientific? These are the questions researchers should be asking as social scientists
studying NGOs. And scholars should be explaining human rights NGO behavior in terms
of brand-community building.
NGOs rarely supply evidence-backed programs due to (a lack of) mar-
ket demand
The distinguishing trademark of this new framework for human rights NGOs is the im-
plication that donors do not demand the effective advancement of human rights, and hu-
man rights NGOs therefore do not implement evidence-based solutions to advance human
rights. This claim may seem counter-intuitive, in part because both donors and human
rights NGOs speak the language of “effectiveness", “impact", and “making a difference."
However, this language is performative, surface-level rhetoric more than it is indicative of
a deep commitment to ensuring positive impact.
In the previous chapter, I presented a series of puzzles, most of them pertaining to
53
impact and accountability. To summarize, NGOs do not appear to prioritize effectiveness,
and they are not held accountable to effectiveness-related standards. Why? My market-
oriented approach to NGOs stipulates that if NGOs are not effective or accountable, it must
be because donors do not demand NGO effectiveness or accountability. Specifically, NGOs
do not supply rigorous impact information or conduct rigorous impact evaluations, and
the accountability movement focuses on overhead rather than effectiveness because this
is the behavior from charity that donors demand. Signaling offers the best explanation to
understand why donors do not create demand for effectiveness and accountability.
It should go without saying that to effectively advance human rights, an organization
needs to know which of its programs are effective, which are ineffective, and which are
harmful. In order to know whether a program is effective, an organization must evaluate
the impact of its programs. But this strategy comes with financial risk. An NGO that com-
mits itself to an evidence-based approach turns human faces and stories into a spreadsheet
with numbers. Rather than publishing uplifting stories about individual people, the NGO
would publish tables and graphs depicting abstract, general trends. This method of help-
ing people produces measurable results and objectively improves the welfare of people’s
lives, but it poses financial problems for human rights NGOs that are funded by donors op-
erating in a social signaling system. Rigorous impact evaluations could undermine donor
expectations, make the organization and its donors appear cold and uncompassionate, and
it could take the organization down paths that would alienate its donors and signal traits
that are socially undesirable. So while such a method is vital to doing good, it could have
serious financial downsides for human rights NGOs.
A less risky strategy for the NGO is to make claims about effectiveness without having
done any of the evaluation work to supply rigorous evidence. Donors are not skeptical of
charity effectiveness; rather, they are startlingly optimistic and trusting that most chari-
ties are unrealistically effective. If an organization were sensitive to its financial health,
it would use its evaluations not to learn about its shortcomings and failures, but rather
54
to confirm what its donors already believe. Impact evaluations serve as marketing tools
to help human rights NGOs appear effective rather than be effective. NGOs claim to have
a positive impact because it is literally free to say. As the research on donation behavior
demonstrates, nobody will challenge their claims. By claiming to be making a difference,
maximizing impact, and saving lives, human rights NGOs get to reap the benefits of ap-
pearing to be effective organizations without paying any of the costs. The next chapter
will test this key claim of my theory.
A possible counter-argument might be that human rights NGOs balance competing
incentives or demands as best they can: they cater to their donors but they also try to ef-
fectively accomplish their stated organizational goals. While this sounds like a reasonable
idea, an understanding of market dynamics, in particular market segmentation, leads to a
different prediction. The rare donors who are interested in maximizing the effectiveness
of their charitable contributions would simply not donate to organizations that are only
partially committed to effective work. Instead they would donate to organizations with
proven effectiveness. The market would be segmented, with a variety of NGOs servicing
different consumer needs, rather than each individual NGO servicing the entire distribu-
tion of donor profiles. The handful of NGOs that do scientifically approach their work
exist to meet demand from the small segment of the charity market interested in proven
effectiveness.
Organizations in this segment of the market include a community of donors and non-
profit organizations known as Effective Altruism. Some of these organizations include
GiveWell, The Life You Can Save, Malaria Consortium, Against Malaria Foundation, He-
len Keller International, Evidence Action, SightSavers, SCI Foundation, and GiveDirectly.
While it may seem tempting to view these organizations and their donors as evidence
against the theory presented in this book, they constitute a very small niche in the global
charity market. According to a liberal estimate, Effective Altruist donors account for at
55
most 1-2% of all donors,
24
concentrated in very elite communities such as the San Fran-
cisco Bay Area, London, New York, and Harvard (Dullaghan, 2020). Moreover, Effective
Altruists tend to be young, white, male, nonreligious, liberal, and highly educated (Dul-
laghan, 2019). GiveWell, the nonprofit organization that helps individual donors find the
most cost-effective charities in the world, receives individual contributions one-fifth of
what Amnesty International receives from individual donors.
25
Perhaps Effective Altruists — given their attention to cost-effectiveness and rigorous
evidence of impact — represent a counter-example to the signaling theory of charity. Yet,
intriguingly, I have found that Effective Altruists are the most interested in this theory and
how it may apply to them. For example, at an Effective Altruist conference in 2016, an
evolutionary psychologist from the University of New Mexico who has contributed theoret-
ically and empirically to the signaling theory of altruism gave a talk about what Effective
Altruists might be signaling differently from other types of donors.
26
This chapter only treats NGOs that are funded, at least partially, by individual donors.
It does not treat NGOs that are entirely funded by government donors or grants from
foundations. I will leave it up to future research to determine how well a signaling model
can generalize to these other funding models.
Concluding the Virtue Economy: Shifting Our Priors
This chapter developed a new theoretical framework for explaining the behavior of NGOs,
namely one that analytically privileges market-based economic concepts such as voluntary
24
GiveWell received $61 million from individuals in 2018, which accounts for 0.26% of interna-
tional donations made in the U.S. Assuming this under-estimates giving by Effective Altruists by a fac-
tor of four to eight leads to an estimate of 1-2%. Sources: GiveWell (https://www.givewell.org/
about/impact), NP Source (https://nonprofitssource.com/online-giving-statistics/#:~:text=Giving%20to%
20International%20charities%20decreased,3%25%20of%20all%20donations).)
25
GiveWell received $61 million from individual contributors in 2018. Amnesty International received
$353 million from individuals. Note that I do not count Good Venture’s donation to GiveWell or the incuba-
tion grants to GiveWell because they represent large donations made by foundations and other organizations,
not individuals. I likewise do not count grants to Amnesty. Sources: GiveWell https://www.givewell.org/
about/impact. Amnesty International: https://www.amnesty.org/en/2018-global-financial-report/.
26
https://vimeo.com/236430422
56
exchange, gains from trade, and supply and demand. Research on NGOs would bene-
fit from scholars moving away from a principal-agent or transnational activist model of
the donor-NGO relationship toward a consumer-firm model. In this chapter, I made the
following arguments:
people use their charitable contributions to signal;
the desire to signal moral virtues and other socially attractive qualities pulls donors
away from creating demand for cost-effective charitable behavior;
in response to market demand, NGOs do not develop cost-effective programming
because they are not incentivized to.
Rather than assuming most people and organizations are altruistic with the occasional
self-interested outlier, researchers who study nonprofit organizations should assume most
people and organizations are self-interested with some behaving more or less altruistically
and some behaving corrupt or fraudulent. To be clear, I do not argue that 100% of behavior
is signaling or that 100% of donors are motivated by signaling. Instead, I argue that typical
NGO and donor behavior is motivated by whatever is in their self-interest. In other words,
my theory explains the center of the distribution —- not the outliers on either extreme.
The existing approaches to NGOs assume that 0% of behavior is signaling, and it is this
assumption that my book aims to challenge. It would be more reasonable to assume that
0% of donation behavior in the human rights area is motivated by pure altruism.
57
Chapter 3
The Lack of Market Demand for a
Scientific Approach to Advancing
Human Rights
Populus vult decipi, ergo decipiatur.
People want to be deceived, so let them be deceived.
Testing the New Framework
In this book, I do not empirically test the signaling theory of charity per se because hun-
dreds of papers and books already test the signaling theory. Instead, I test what has not
yet been tested: whether or not donors create financial incentives for NGOs to be effective
and accountable. In other words, I test whether a demand-side explanation can account
for the puzzling supply-side behavior that motivates this book.
My theoretical approach to understanding NGO behavior is analogous to understanding
firm behavior. For example, the answer to the question, ‘why do firework stands close up
shop after July 4?’ has a simple answer: customers are done buying fireworks for the
year. Under my market-oriented framework, NGOs should be theorized in the same way.
If Amnesty International ends its campaign to legalize sex work, the first suspicion should
58
be that its donors did not want what Amnesty was selling. Likewise, if NGOs conceal
information about their impact, they do so likely because their donors do not demand
information about the organization’s impact.
The first chapter presented a series of puzzles. These behaviors can be simplified into
three statements about supply-side NGO behavior:
1. Human rights NGOs rarely supply evidence-backed information about their impact.
2. Human rights NGOs rarely conduct randomized trials to test the impact of their pro-
grams.
3. Accountability organizations supply information about charity overhead rather than
impact.
The first puzzle is the low supply of evidence-backed impact information, including
information about the organization’s cost-effectiveness or information about the organi-
zation’s impact based on rigorous evaluations. According to the theoretical framework
presented in this book, we can explain this behavior by looking at donor preferences:
donor preferences drive the low supply of rigorous impact information. Given that donors
are not particularly interested in realistic impact results, and given that we do not observe
human rights NGOs supplying this kind of information, we should expect that providing
positive but realistic impact results will not increase donations relative to not providing
any impact information. I have other hypotheses that deal with different outcomes that
could come from impact evaluations, and these will be discussed in the next section. While
some outcomes we care about may be difficult to measure, this cannot explain why hu-
man rights NGOs do not supply rigorous information about their impact but nonetheless
claim to have a positive impact. If measurement difficulty were the only reason for the low
supply, then the organizations would remain agnostic about their impact.
The second puzzle is the dearth of rigorous impact evaluations such as randomized
trials. My framework expects that donors do not reward randomized trials in any way. I
expect that individuals will not increase donations to a charity when it conducts a random-
ized trial, compared to a charity that implements an intervention unilaterally, untested.
59
Recall that supply-side explanations cannot explain this puzzle. For example, while ran-
domized trials may be costly to implement, the cost of production cannot explain why an
organization refrains from evaluating itself. If the extra cost of conducting a randomized
trial were compensated by donors, then it would be worthwhile for the organization.
The third puzzle is the accountability movement’s supply of overhead information
rather than impact information. My theory predicts that organizations supply this infor-
mation because overhead is what donors care about, not impact. To see if this is true, I pit
overhead and impact information against each other. I predict that people will donate less
money to a charity that is revealed to have a high overhead (relative to a charity that does
not provide information about its overhead), even when it helps more people.
The following sections, organized by puzzle, test each hypothesis using pre-registered
online survey experiments. In total, I test the hypotheses against the responses from 2,673
unique individuals.
Before proceeding, I define key terms used throughout this chapter: effectiveness, im-
pact, cost-effectiveness, and overhead. The key to evaluating an NGO’s effectiveness is
to collect evidence about the impact of the NGO’s programs. Often, when organizations
evaluate their programs or provide information on their websites regarding their program
effectiveness, they only provide information about the immediate outputs of the program,
not the impact the program had on the lives of the stated beneficiaries. Inputs are the
direct actions the NGO takes (e.g. publishing research, providing educational services,
meeting with government officials, sending letters to prison officials). Outputs refer to
intermediate outcomes caused by inputs that are believed to be related to generating an
improvement in the stated beneficiaries’ lives (e.g. raising public awareness, convincing
the United Nations to pass a resolution, influencing media coverage). Impact refers to the
improvement or decline in the conditions directly influencing the lives of the stated ben-
eficiaries (e.g. reducing rates of political imprisonment, reducing rates of child marriage,
increasing school attendance). If a program produces a positive impact, then it can be said
60
to be effective. However, if a program merely produces a positive output, we must remain
agnostic about the program’s effectiveness.
Cost-effectiveness information will be featured in several of the experimental survey
vignettes, so it is important to differentiate it from other concepts such as impact, effec-
tiveness, and overhead. Cost-effectiveness is different from effectiveness and impact in that
it incorporates information about the impact of a program as well as the program’s cost.
A program could be highly effective at improving people’s lives, but if it is tremendously
expensive, then it may not be cost-effective if other programs can achieve comparable re-
sults at a lower cost. Information about cost-effectiveness would allow donors to infer how
much impact they could make based on their charitable contributions.
Overhead is a measure of how much an organization spends on fundraising and ad-
ministration. The overhead ratio measures how much an organization spends on over-
head compared to programs and services. Many conflate overhead with effectiveness and
impact. Just as a measure of an organization’s input is not equivalent to impact, the over-
head is likewise not equivalent to impact. An organization could have a low overhead,
spending most of its budget on programs and services, but if the programs themselves are
ineffective, then a low overhead will not help the organization achieve positive impact.
Explaining the Low Supply of Evidence-Backed Impact In-
formation: Donors Do Not Reward Realistic Outcomes
As already established in the previous chapter, donors do not seek and are not responsive to
information about charity effectiveness: they do not seek effectiveness information before
giving, and they do not adjust their giving behavior when provided direct information
about effectiveness. For the supply of rigorous evidence of impact to be worthwhile for an
organization, donors would need to reward such information.
In a social signaling system, common knowledge is more relevant than private knowl-
61
edge. Most people are extremely optimistic about charity effectiveness. The social incen-
tives are such that donors do not need to prove to members of their social group that their
money has accomplished what they claim (or what the charity claims) it has accomplished.
This is so for three reasons. (1) Most people believe that most charities produce positive
outcomes. (2) Most people believe that most charities are unrealistically cost-effective.
(3) Most people believe that the difference in cost-effectiveness between charities is small.
This implies that a donor will appear to be doing a lot of good by donating to nearly any
charity that is understood by one’s social community to be engaging in programs that are
in the spirit of benefiting others. Put differently, donors will receive social credit for doing
a lot of good simply by donating. This implies that knowledge about charity effectiveness
is not socially useful for the average donor.
People share a presumption that charities are unrealistically effective. They believe
that few charities or programs are ineffective, and they believe the cost of saving lives
is unrealistically low. In two separate surveys, participants were asked how much they
thought it would cost “a typical charity to prevent one child in a poor country from dying
unnecessarily by improving access to medical care.” In one survey, their median estimate
was $40, and in the other survey, it was $150. The surveys asked the same question about
the most cost-effective charities, for which the median estimate was $25 in one survey
and $100 in the other (Caviola et al., 2020; Greenberg, 2017). That is, according to the
most conservative of estimates, the median respondent believes that a charity of average
cost-effectiveness can save a child’s life in a poor country for only $150. People believe the
most cost-effective charity is roughly 1.5 times more cost-effective than a typical charity,
so they think the difference between the best and the average charity is small.
These beliefs underestimate the true cost of the most cost-effective charities by thou-
sands of dollars. GiveWell conducts detailed cost-effectiveness analyses on global develop-
ment and health charities that can provide rigorous evidence of their impact. According
to GiveWell’s latest estimates, donations to the Malaria Consortium (which implements
62
seasonal malaria chemoprevention programs) saves a life for $2,271.
1
GiveWell identifies
Malaria Consortium as being the most cost-effective charity, which means that it has not
yet identified any other charity in global development and global health that can save a
life for a lower cost. According to GiveWell’s estimates, the most cost-effective programs
in the U.S. cost at least $10,000 per child.
2
The average donor also severely underestimates the variance in cost-effectiveness across
charities. The Disease Control Priorities in Developing Countries (DCP2) offers a compre-
hensive compendium of cost-effectiveness estimates for 108 global health interventions.
According to the DCP2’s estimates, the most cost-effective intervention is 15,000 times
more cost-effective than the least cost-effective intervention and 60 times more so than
the median intervention (Ord, 2013). In addition, Caviola et al. (2020) surveyed global
poverty experts to estimate their beliefs about the difference in cost-effectiveness between
the most cost-effective charity and an average charity. Experts were sampled from fields
including health economics, international development, and charity measurement and
evaluation. The final sample included 45 experts. Half of the experts believe the most
cost-effective charity is at least 100 times more cost-effective than a typical charity.
While most people believe most charities produce positive outcomes, the evidence sug-
gests otherwise. According to the empirical evidence on the impact of social programs,
success of any given program is unlikely. Well-designed, promising ideas fail due to un-
foreseen circumstances or unintended consequences. As explained by the Laura and John
Arnold Foundation, “most [social programs], including those thought promising based on
initial studies, are found to produce few or no effects – underscoring the need to test
many."
3
According to the Foundation’s comprehensive study of randomized controlled tri-
als testing the effectiveness of social programs in the U.S., 75% of the programs were found
to have a weak or null effect. These social programs cover many different issue areas, in-
1
https://www.givewell.org/charities/malaria-consortium
2
https://www.givewell.org/giving101/Your-dollar-goes-further-overseas (accessed March 15, 2019).
3
https://web.archive.org/web/20170428214714/http://www.arnoldfoundation.org/wp-content/
uploads/Request-for-Proposals-Low-Cost-RCT-FINAL.pdf (accessed May 8, 2019).
63
cluding education, employment/training, medicine, and business. In addition, Vivalt’s
analysis of outcomes from international development interventions on AidGrade finds a
similar percentage: “60-70% of intervention-outcomes that were studied have insignifi-
cant meta-analysis results" despite a publication bias in favor of finding positive results
(Todd, 2017).
Given public optimism about charity effectiveness, market competition should compel
most charities to inflate their impact, creating a cycle in which most charities exaggerate
their impact, and most donors believe such claims. If an organization were to deviate from
this pattern, it should under-perform financially compared to its competitors. At best, the
organization’s impact evaluations would confirm what its donors already believe. More
likely though, the organization would undermine its own claims of having a massively
positive impact and lose donors. If this is true, then honesty about charity program results
should not be rewarded by donors.
Hypothesis 1 (H1) Providing positive but realistic impact results will not increase donations
relative to not providing any impact information.
Hypothesis 2 (H2) Providing ineffective program results will not increase donations relative
to not providing any impact information.
I also test an exploratory hypothesis — whether people donate more to charities that
claim to have unrealistically positive impact — to see whether human rights NGOs have
a financial incentive to lie about their impact. Since organizations often exaggerate their
impact (e.g. Amnesty’s letter-writing campaigns say “write a letter, save a life"), I would
expect that such behavior is rewarded by donors, otherwise organizations would stop do-
ing it.
4
The findings here will reveal whether donors are skeptical of incredibly positive
results or whether exaggeration increases donations.
4
https://www.amnestyusa.org/event/write-a-letter-save-a-life-write-for-rights-2017/.
64
Hypothesis 3 (H3) Providing unrealistically positive impact results will increase donations
relative to not providing any impact information.
This section presents results from two preregistered online survey experiments de-
signed to test how individuals respond to different types of impact information. The
studies each test across two separate issue areas, global health and human rights. The
Positly platform is used to recruit participants from Amazon Mechanical Turk (Mturk), a
crowdsourcing website for businesses and researchers to hire remotely located workers to
complete discrete tasks.
5
While many issues have been raised with using Mturk workers
for study recruitment, several social scientists have found that experimental treatment ef-
fects detected with Mturk samples do in fact generalize to other samples and offer more
representative samples than other popular subject recruitment methods such as univer-
sity subject pools (Berinsky, Huber and Lenz, 2012; Goodman, Cryder and Cheema, 2013;
Hauser and Schwarz, 2016; Paolacci, Chandler and Ipeirotis, 2010).
In both studies, subjects are randomly assigned to one of four conditions. The first con-
dition describes a hypothetical charity’s program without providing any information about
the impact of the charity. The second, third, and fourth conditions describe the same
charity’s program and include additional information that the charity rigorously evaluated
itself and found it has a positive impact based on realistic and exceptional figures (Con-
dition 2), the charity found it did not achieve its desired impact (Condition 3), and the
charity found it has a positive impact based on unrealistic and unachievable figures (13x
more effective than any known charity) (Condition 4).
Donors tend to diversify their donations (Baron and Szymanska, 2011; Caviola et al.,
2020). This decision-making context is represented in the experiments by giving partici-
pants the option to allocate their donations between the charity from the treatment group
and another charity option. In both studies, the outcome of interest is measured by the
amount, out of a hypothetical $100, subjects would allocate between two charity options:
5
The end of the chapter discusses potential study limitations with the use of participants from Mturk.
65
the charity from the treatment group and some alternative charity operating in the same
issue area. The section on measurement validation in the appendix includes a discussion
explaining why this choice of wording for the alternative charity is optimal compared to
other options.
At the end of the survey, subjects answer a simple question designed to test their com-
prehension of the survey. Respondents who failed the multiple-choice comprehension
check (5.7% across the two studies) were subsequently removed from the analysis.
Study 1: Medical Intervention and Saving Lives
Study 1 begins by testing how people respond to impact information for a case in which
good data exists on the most cost-effective charities, and good data exists on perceptions
or beliefs of cost-effectiveness among the general public: medical interventions aimed at
saving lives in poor countries. 418 participants, recruited from Amazon’s Mechanical Turk,
were randomly assigned to one of four conditions.
Condition 1 (no impact information): Human Health International operates in de-
veloping countries to provide life-saving medical care to people at risk of preventable
disease.
Condition 2 (realistic positive cost-effectiveness information): Human Health
International operates in developing countries to provide life-saving medical care
to people at risk of preventable disease. Human Health International conducted a
rigorous study which allowed them to measure the effect of their medical care. The
results from the study provided strong evidence that Human Health International
saves the life of one person for every $2,000 spent.
Condition 3 (negative impact information): Human Health International oper-
ates in developing countries to provide life-saving medical care to people at risk of
preventable disease. Human Health International conducted a rigorous study which
allowed them to measure the effect of their medical care. The results from the study
found that Human Health International did not have the desired impact, suggesting
that reform is needed to increase effectiveness.
Condition 4 (unrealistic positive cost-effectiveness information): Human Health
International operates in developing countries to provide life-saving medical care
to people at risk of preventable disease. Human Health International conducted a
rigorous study which allowed them to measure the effect of their medical care. The
66
results from the study provided strong evidence that Human Health International
saves the life of one person for every $150 spent.
Study 2: Issue Awareness and HIV Prevention
Study 2 tests how individuals respond to impact and cost-effectiveness information in
the context of a type of intervention that is common among human rights organizations
(namely, an issue awareness campaign) and a public health issue that has been adopted by
human rights organizations (HIV). To see why this vignette is relevant to the work of hu-
man rights organizations, take Amnesty International’s report for its Stop Violence Against
Women campaign as an example. In this report, Amnesty writes, "The HIV pandemic is
increasingly viewed as a strongly gendered health, development and human rights issue."
6
One of Amnesty’s first recommended remedies in this report is a public awareness cam-
paign. I also chose this vignette because I needed to populate Condition 2 with a realistic
cost-effectiveness figure inspired by existing research (see below for a discussion). For
Study 2, 434 subjects were randomly assigned to one of four conditions.
Condition 1 (no impact information): Human Rights International operates in
developing countries, organizing issue-awareness campaigns to educate teenage girls
about the risk of HIV.
Condition 2 (realistic positive cost-effectiveness information): Human Rights In-
ternational operates in developing countries, organizing issue-awareness campaigns
to educate teenage girls about the risk of HIV. Human Rights International conducted
a rigorous study which allowed them to measure the effect of their issue-awareness
campaign. The results from the study provided strong evidence that Human Rights
International’s issue-awareness campaign prevents one teenage girl from being in-
fected with HIV for every $650 spent.
Condition 3 (negative impact information): Human Rights International operates
in developing countries, organizing issue-awareness campaigns to educate teenage
girls about the risk of HIV. Human Rights International conducted a rigorous study
which allowed them to measure the effect of their issue-awareness campaign. The re-
sults from the study found that Human Rights International did not have the desired impact,
suggesting that reform is needed to increase effectiveness.
6
https://www.amnesty.org/download/Documents/88000/act770842004en.pdf
67
Condition 4 (unrealistic positive cost-effectiveness information): Human Rights
International operates in developing countries, organizing issue-awareness campaigns
to educate teenage girls about the risk of HIV. Human Rights International conducted
a rigorous study which allowed them to measure the effect of their issue-awareness
campaign. The results from the study provided strong evidence that Human Rights
International’s issue-awareness campaign prevents one teenage girl from being in-
fected with HIV for every $50 spent.
The cost-effectiveness information for Condition 2 comes from Dupas (2011). This
study uses a randomized trial to estimate the cost-effectiveness of an issue-awareness edu-
cational program in Kenya intended to prevent HIV infections among teenage girls. Dupas
determines the most realistic estimate is $653 to prevent one case of HIV infection. $50 is
derived by taking 1/13th of the estimate from Condition 2. The 1:13 ratio is the same ratio
from the first study ($150 vs. $2,000). Condition 3 reflects a real example of a similarly
designed program implemented by Evidence Action in Botswana, an international NGO,
that ultimately proved ineffective (Levy, Wang’ombe and Radhakrishnan, 2018).
Results
I analyze the results using Welch’s heteroscedastic method for comparing means, a method
based on Student’s t-test when the variances of the comparison groups are unequal.
7
A po-
tential concern with the study design is that each study makes comparisons across three
groups, even though the statistical test is designed to make comparisons across two groups.
This means there is a greater than 5% probability of finding a difference in means between
groups due to random chance, even though 5% is my chosen significance level for drawing
inferences. Researchers can use several different techniques to correct the actual proba-
bility of making a Type I error. A simple and intuitive technique known as the Bonferroni
method punishes the significance level by the number of comparisons being made by tak-
ing the original significance level (0.05) and dividing by the number of comparisons or
7
Welch’s t-test is a slight modification of Student’s t-test when the variances from both experimental
groups are unequal. Wilcox (2012) offers an insightful discussion of this method.
68
hypotheses (three). The new significance level is 0.0167, which means that the differ-
ence between groups will be inferred as statistically significant if the p-value is lower than
0.0167 rather than the conventional 0.05 level.
Results from Study 1 demonstrate that people donate similar amounts to a charity that
shows it can save a life for $2,000, equivalent to the most effective known charity (M =
$65), as they do to a charity that does not provide any information about its impact (M
= $60.3) (t = 1.15, p = 0.2505, d = 0.16). While people donate 0.16 standard devi-
ations more to the realistically cost-effective condition, this difference is not statistically
significant. People donate less to a charity that provides information that its programs
are ineffective (M = $36.7), compared to a charity that does not provide any information
about its impact (t = 5.42, p < 0.001, d = 0.76). For the revelation of positive impact
information to increase donations compared to not providing any information, a charity
would need to provide evidence of unrealistic, unattainable effectiveness (M = $77.2) (t
= 4.6, p < 0.001, d = 0.63). The results suggest that revealing honest information does
not help, and only stands to harm, the charity’s fundraising.
The behavior of regular, active donors may differ from the general population. In a
smaller study based on the unrealistic positive impact condition, participants were asked,
“In the past 12 months, how much have you donated to charity" (N = 157, excluding sub-
jects who failed the comprehension check). The mean donation (M = $70.9) to Human
Health International among respondents who have donated in the past twelve months
(72%) is not statistically different from the mean donation (M = $69.9) among all respon-
dents including non-donors (t = 0.58, p = 0.561, d = 0.11).
Results from Study 2 again demonstrate that people do not donate more to a human
rights NGO that provides information about its exceptionally positive (realistic) impact (M
= $55.2) compared to an NGO that does not provide any information about its impact
(M = $52.6) (t = 0.63, p = 0.5293, d = 0.09). Also, people donate less to a charity
that provides information showing its programs are ineffective (M = $32.5) compared to
69
a charity that does not provide any information about its impact (t = 4.99, p < 0.001, d =
0.67). For the revelation of cost-effectiveness information to increase donations compared
to not providing any information, a charity would need to provide information claiming
unrealistic and unattainable levels of impact (M = $67.9) (t = 3.97, p < 0.001, d = 0.54).
The evidence again suggests that revealing honest information does not help, and can only
harm, the charity’s fundraising.
Figure 1 plots the estimated effect of each of the three conditions, while Figure 2 plots
the mean donations across the four conditions. Table 1 summarizes the results from the
t-tests for both studies.
One potential concern is that the studies test a null hypothesis that the means across
groups will be equivalent (this is true for H1 and H2 but not H3). This makes the power
of each study (the probability of rejecting the null hypothesis when it should be rejected)
important for interpreting the results, especially the comparison between Conditions 1 and
2. If the studies do not have enough power, then the statistical tests could fail to find a
difference between Conditions 1 and 2, even if there is a difference in means between the
two groups. I conducted a power analysis for each study, which indicates that the studies
are powered with an 80% probability to detect a medium-sized effect of 0.46 standard
deviations.
The appendix includes additional information from robustness tests, demonstrating
the results hold while controlling for other factors, including gender, age, income, and
education. This chapter concludes with considerations about the generalizability of the
findings and the representativeness of the samples.
Discussion
Taken together, the experiments consistently demonstrate that charities face substantial
risk to their fundraising efforts by rigorously evaluating themselves and publicizing the
information. The most likely outcome of a rigorous evaluation —- revealing an ineffec-
70
Figure 3.1: Effect of Impact Information on
Donations
Figure 3.2: Mean Donations from Studies 1 and 2
71
Table 3.1: Welch’s T-test Results from Studies 1-2
Study Comparison N Mean Difference 95% CI Effect Size P t(df)
1
No impact info vs.
Realistic impact
210 $4.7 -$3.4 — $12.8 d = 0.16 P = 0.2505 t(200) = 1.15
No impact info vs.
Null impact
206 -$23.6 -$32.2 — -$15.0 d = 0.76 P< 0.001 t(200) = 5.42
No impact info vs.
Unrealistic impact
212 $16.9 $9.7 — $24.2 d = 0.63 P< 0.001 t(209) = 4.6
2
No impact info vs.
Realistic impact
217 $2.6 -$5.6 — $10.8 d = 0.09 P = 0.5293 t(204) = 0.63
No impact info vs.
Null impact
219 -$20.1 -$28.1 — -$12.1 d = 0.67 P< 0.001 t(210) = 4.99
No impact info vs.
Unrealistic impact
216 $15.3 $7.7 — $22.9 d = 0.54 P< 0.001 t(211) = 3.97
tive program —- would lead to a large reduction in donations. If a program was found
to be among the most cost-effective in the world, the charity running it would see little
change to its donations by advertising this fact. The results from Studies 1 and 2 suggest
they could see a fundraising change of a few percent in either a positive or negative di-
rection. Massively exaggerating impact is the only case in which impact information helps
fundraising.
The shared presumption of unrealistic impact may fuel common deceptive behavior
observed in standard charity marketing practices, including when charities imply that a
certain amount of money donated (typically a small amount, such as $25) will accomplish
a discrete task, such as providing medication. Charities often make claims such as, “donat-
ing $25 is enough to purchase one medication, which can save a life." This is a subtle form
of deception in which the lie is smuggled into the word “can." The information the charity
leaves out is the likelihood of “can." If 1 out of every 100 recipients is saved by taking the
medication, then it would cost $2,500 to save one life, not $25. An honest appeal by the
charity might read, “donating $25 is enough to purchase a medication which has a 1%
chance of saving a life." Sometimes, organizations leave it up to the donor to imagine how
many people they will help by simply providing estimates for how much each unit of an
item costs, such as medications, water filtration systems, and letters to political prison-
ers, without explaining what specific outcomes the organization is trying to improve (e.g.
72
lives-saved, years with clean water, years out of prison). Donors, then, can fill in the blanks
using their own imagination, which is biased in an overly optimistic direction.
8
One possible objection to the current studies is that donors do not reward the most
cost-effective charity because they do not know that the lowest cost-estimate for saving
a life is $2,000. If they did know, they would reward the honest, effective charity. One
may suppose that cost-effective charities would benefit from revealing information about
their impact, but only if they correct donor misconceptions. This is a separate research
question; its answer implicates whether NGO-driven solutions are effective strategies for
changing donor behavior. The purpose of this paper is to test whether donor-driven ex-
planations can account for organizational behavior. Because most donors believe it costs
13 times less than $2,000 to save a life in the best-case scenario, a donor-driven explana-
tion would not expect the most cost-effective charities to be the most popular. For readers
interested in whether donors update their giving behavior after learning about charity
cost-effectiveness, Caviola et al. (2020) find that individuals are prone to update their
giving behavior when told one charity is 100 times more cost-effective than another. The
authors do not probe whether individuals donate to a charity told to be 13 times more
cost-effective. Importantly, the authors set up the comparison as a ratio (e.g. 1.5x and
100x), not in absolute dollar amounts as was done in this chapter.
Caviola, Schubert and Nemirow (2020) find experimental evidence that lay donors
have multiple misconceptions about charity, which reduces the effectiveness of their giving.
The authors demonstrate that correcting these misconceptions improves the effectiveness
of their giving. Assuming this is true, it does not address a key point: the ability of false be-
liefs and misconceptions to be corrected is not evidence that a market exists for false beliefs
to be corrected or that charities even try. Many donors may update their giving when new
information is foisted upon them in an experimental context, but do these donors create
8
For more on this issue, see GiveWell’s discussion of how charities frequently cite misleading cost-
effectiveness figures (https://www.givewell.org/how-we-work/our-criteria/cost-effectiveness#Charities_
frequently_cite_misleading_cost-effectiveness_figures) and (Singer, 2009, 86-87).
73
demand for this information? The surveys and experiments cited in the previous chapter
suggest they do not (Hope Consulting, 2011). It is not clear how a practice of correcting
false beliefs will be sustained beyond the context of an experiment. Which organizations
would have the incentive to supply information to correct false beliefs about charity? Not
the current NGOs that do not evaluate their own impact. This information would have to
come from organizations such as GiveWell that research cost-effectiveness across different
charities or academics who wish to provide a false-belief-correction service. But this brings
us back to the starting point: most people do not search for this information in the first
place. Put another way, this is a demand problem, not a supply problem. A satisfying
explanation for this puzzle cannot rely on false beliefs alone; it must contain some notion
of motivation or incentive, which signaling theory offers.
Another possible concern with the survey vignettes is that they may generalize to pub-
lic health organizations but not to human rights organizations. A common view among
academics who study NGOs is that human rights NGOs engage in policy advocacy by
launching campaigns and conducting legal work, while service delivery organizations try
to improve health and income by directly delivering services. I find that this distinction is
empirically inaccurate. First, ‘human rights’ is a broad concept that covers many outcomes
of interest to service-delivery organizations, such as health, income, education, and even
the environment. This is particularly so when marginalized communities are perceived
to have unequal levels of enjoyment of these outcomes. Thus, cost-effectiveness informa-
tion about programs that are designed to improve health are directly relevant to the work
of human rights organizations. Second, human rights organizations such as Amnesty rou-
tinely implement services that are closer to direct services than they are to policy advocacy,
such as issue awareness campaigns, human rights education programs, and letter-writing
campaigns. There is substantial overlap among NGOs operating in different philanthropic
sectors in terms of the issues that guide their work and the types of programs they imple-
ment. For these reasons, the results from the survey experiments in this section and the
74
following sections are relevant to explaining the behavior of human rights NGOs.
Explaining the Dearth of Randomized Trials: Donors Do
Not Reward Them
Human rights NGOs rarely conduct rigorous impact evaluations such as randomized tri-
als to test the impact of their programs. As the previous section established, conducting
rigorous evaluations and publicizing the information comes with little upside and mostly
downside for a human rights NGO. In addition, donors may view a randomized trial itself
— as opposed to the information it reveals — as unimportant to their donation decisions or
even antithetical to their reputational goals. This section tests whether donation behavior
can explain the dearth of randomized trials. Signaling provides a possible explanation for
why donors do not reward randomized trials.
One possibility is that donors may actually punish organizations for conducting ran-
domized trials. A rigorous, evidence-based approach may make an organization and its
donors appear to lack socially desirable emotions such as empathy and warmth, causing
such an approach to altruism to appear deliberative, cold, calculating, and rational in a
world where most people select social partners for their empathy, warmth, and trustwor-
thiness. Helping people while using a cost-effectiveness rationale could commit a ‘taboo
sacred trade-off,’ placing a price on human life and treating human life as fungible (Tet-
lock, 2003). There is systematic evidence of aversion to experimentation across different
issue areas (e.g. public health, charity), but the evidence is inconsistent depending on
whether one looks at the domain of altruism (Meyer et al., 2019) or corporate practices
(Mislavsky, Dietvorst and Simonsohn, 2019). Research on this topic demonstrates that in-
dividuals disapprove of experimentation when it is used as a method to discover the most
effective means of helping others (Meyer et al., 2019). For example, across several experi-
mental domains, people frequently rate A/B-style randomized trials as inappropriate, even
75
when they rate the implementation of either A or B, untested, as appropriate. Roughly
three times as many people disapprove of the decision to conduct an A/B test, relative to
a unilateral decision to implement either A or B untested. This disapproval is consistent
in multiple domains, including health, genetics, autonomous vehicles, retirement plans,
poverty alleviation, and teacher well-being.
The economic reality for human rights NGOs is one of scarce resources. They necessar-
ily make these trade-offs every time they fund one program and not another or advocate
for one group of people and not another. These decisions are virtually always made implic-
itly, never publicly. If an organization were to routinely implement A/B-style randomized
tests or randomized controlled trials, the trade-off would be explicit and conspicuous,
which could violate this social taboo and risk making the organization (and by reflection,
its donors) appear untrustworthy and socially anomalous. Donors, wanting to signal that
they possess socially desirable traits, would have a social incentive to donate to a different
organization, one that does not use scientific methods to determine whether they should
implement a certain program or help a certain group of people. If human rights NGOs
were to use randomized trials to test the comparative effectiveness of different interven-
tions, the existing research suggests that they might triple disapproval of their organization
and therefore lose a great deal of funding.
Based on this discussion, I expect that individuals will donate less money to a charity
when it conducts a randomized trial, compared to a charity that implements an interven-
tion unilaterally, untested. I test specific versions of this expectation by analyzing how
much individuals donate to organizations that implement A/B trials as well as randomized
trials with control groups. The specific hypothesis for each type of randomized trial is as
follows:
Hypothesis 4 (H4) Individuals will donate less money to a charity when it conducts a ran-
domized A/B trial and implements the intervention that helps more people, compared to a
charity that implements an intervention unilaterally, untested.
76
Hypothesis 5 (H5) Individuals will donate less money to a charity when it conducts a ran-
domized controlled trial and implements the intervention only if it helps more people
than the control group, compared to a charity that implements an intervention unilaterally,
untested.
I expect that randomized trials with a control group will be particularly anathema to
the social purpose of donating, so I expect that
Hypothesis 6 (H6) Individuals will donate less money to a charity when it conducts a ran-
domized controlled trial and implements the intervention only if it helps more people than
the control group, compared to a charity that conducts a randomized A/B trial and
implements the intervention that helps more people.
This section presents results from four online survey experiments to test whether in-
dividuals financially reward charities that (1) conduct A/B tests to determine the most
effective way to help the same group or (2) conduct randomized controlled trials to de-
termine whether the same group should receive the charitable intervention or not. Three
of the studies are pre-registered with AsPredicted (Studies 1, 3, and 4). The second study
was a pilot study and was not pre-registered.
Studies 3-5 use a between-subjects design to test whether individuals financially reward
charities for conducting randomized A/B tests of their programs and implementing the
most effective program. Studies 5 and 6 test whether individuals reward charities for
conducting randomized controlled trials of their programs and implementing the program
only if it outperforms the control group. Study 5 uses a between-subjects design, while
Study 6 uses a within-subjects design. Studies 3-6 test across three separate issue areas,
including medicine, school attendance, and human rights.
In Studies 3 and 4, I randomly assign subjects to one of three conditions. The first
condition describes the unilateral implementation of a program untested (Program A).
The second condition describes the unilateral implementation of an alternative program
77
untested (Program B). The third condition describes the implementation of both programs
(Programs A and B) in which the charity randomly assigns half of the recipients to each
program and chooses whichever program turns out to more effectively improve welfare
outcomes. Studies 5 and 6 add a fourth condition in which a charity conducts an RCT to
test the effectiveness of Program B, choosing only to implement the program if it outper-
forms the control group.
In all four studies, I measure the outcome of interest based on how much money, out
of a hypothetical $100, subjects allocate between two charity options: the charity from the
treatment group and some alternative charity operating in the same issue area.
9
At the
end of the survey, subjects answer a simple question designed to test their comprehension
of the survey. Respondents who failed the multiple-choice comprehension check (4.3%
across the four studies) were subsequently removed from the analysis.
Table 3.2 summarizes the studies and the hypotheses they test.
Table 3.2: Summary of Studies and Hypotheses
Study Description
Study 1
Drug Effectiveness (A/B Test)
Hypothesis 4
Study 2
Educational Interventions in a Low-Income Country (A/B Test)
Hypothesis 4
Study 3
Rape Prevention (A/B and RCT Test)
Hypothesis 4
Hypothesis 5
Hypothesis 6
Study 4
Joint Evaluation (RCT Test)
Hypothesis 5
9
The section on measurement validation in the appendix includes a discussion explaining why this choice
of wording for the alternative charity is optimal compared to other options.
78
Study 3: Drug Effectiveness (A/B Test)
Study 3 begins by applying a nearly identical survey vignette from Meyer et al. (2019)
(modified only slightly to make it relevant to donating) to see if the original finding repli-
cates in the charity context.
10
496 participants were randomly assigned to one of three
conditions.
1. Program A (drug A): Several drugs have been approved by the U.S. Food and Drug
Administration as safe and effective for treating high blood pressure. A director of
a health charity wants to provide good treatment to his patients, so he decides that
his patients who need high blood pressure medication will be prescribed drug A. It is
affordable and patients can tolerate its side effects.
2. Program B (drug B): Several drugs have been approved by the U.S. Food and Drug
Administration as safe and effective for treating high blood pressure. A director of
a health charity wants to provide good treatment to his patients, so he decides that
his patients who need high blood pressure medication will be prescribed drug B. It is
affordable and patients can tolerate its side effects.
3. A/B Test: Several drugs have been approved by the U.S. Food and Drug Adminis-
tration as safe and effective for treating high blood pressure. A director of a health
charity thinks of two different ways to provide good treatment to his patients, so
he decides to run an experiment by randomly assigning his patients who need high
blood pressure medication to one of two test conditions. Half of patients will be pre-
scribed drug A, and the other half will be prescribed drug B. Both drugs are affordable
and patients can tolerate their side effects. After a year, he will only prescribe to new
patients whichever drug has had the best outcomes for his patients.
I analyze the results using Welch’s heteroscedastic method for comparing means. I pool
the data from Program A and Program B and test whether the mean donations from the
pooled group is different from the mean donations from the A/B testing group. I find
that individuals do not reward or punish the A/B testing health charity (t = 0.91, p =
0.3618, d = 0.09); the results fail to reject the null hypothesis that the mean difference
between groups is zero. Donations to the Drug A or Drug B charity (M = $44.6) and
the A/B testing charity (M = $41.7) are not statistically different from one another at the
0.05 significance level. Although donations to the A/B charity are lower than donations to
10
See the supplemental information, page 18, from Meyer et al. (2019) for the wording from the original
study.
79
both the Drug A and Drug B charity, this difference is not statistically significant. A power
analysis indicates that the study is powered with an 80% probability to detect a small-to-
medium effect of 0.31 standard deviations. This evidence is inconsistent with Hypothesis
4 given that average donations do not statistically decrease for the health charity that runs
an A/B test to determine which blood pressure medication is more effective for its patients.
Figure 3.3 plots the mean donations to each health charity.
Figure 3.3: Mean Donations from Study 3
45.3
41.7
43.8
44.6
$0
$10
$20
$30
$40
$50
Drug A Drug B Drug A or B Drug A/B Test
Study 4: Educational Interventions in a Low-Income Country (A/B
Test)
Study 4 tests how individuals respond to an international charity operating in developing
countries to improve school attendance.
11
314 participants were randomly assigned to one
of three conditions.
11
This vignette modifies wording from Meyer et al.’s poverty alleviation vignette. See the supplemental
information, page 17, from Meyer et al. (2019) for the wording from the original study. They find consistent
but slightly weaker evidence for the A/B effect in the context of poverty alleviation.
80
1. Program A (school lunch): Last year, a charity received a large number of dona-
tions. The director of this charity wants to help children in a low-income country
increase their school attendance, so he decides that all children in a low-income
village will receive free school lunches.
2. Program B (transportation): Last year, a charity received a large number of dona-
tions. The director of this charity wants to help children in a low-income country
increase their school attendance, so he decides that all children in a low-income
village will receive free transportation to school.
3. A/B Test: Last year, a charity received a large number of donations. The director
of this charity wants to help children in a low-income country increase their school
attendance, so he decides to run an experiment by randomly assigning children to
one of two test conditions. Half of all children in a low-income village will receive
free school lunches. The other half will receive free transportation to school. After
a year, the director will begin providing everyone in the village whichever resource
(lunches or transportation) turns out to help children attend school more often.
Again, I find that individuals do not reward the international children’s charity that
trials its programs (t = 1.03, p = 0.30, d = 0.13). Donations to the A/B testing charity (M
= $56.7) are not statistically different from donations to either the free school lunch or
free transportation charities (M = $52.8) at the 0.05 significance level, although the mean
donation is greater than either unilateral program in contrast to the results from the first
experiment. Power analysis indicates that the study is powered with an 80% probability to
detect a small-to-medium effect of 0.39 standard deviations. This evidence is inconsistent
with Hypothesis 4 given that average donations do not decrease for the children’s charity
that runs a randomized trial to determine which program helps children attend school
more often. Figure 3.4 plots the mean donations to each charity.
Study 5: Rape Prevention (A/B and RCT Test)
Study 5 tests whether people donate less to a college chapter of a human rights NGO that
conducts a randomized controlled trial and an A/B test to determine how best to reduce
instances of rape on college campus. I based the wording for these vignettes on real
randomized trials of university interventions designed to reduce the prevalence of rape
81
Figure 3.4: Mean Donations from Study 4
56.7
55.2
50.7
52.8
$0
$20
$40
$60
School Lunch Transportation A or B A/B Test
and other forms of sexual assault on campus.
12
626 subjects were randomly assigned to
one of four conditions.
1. Program A (informed consent class): The leader of the college chapter of a hu-
man rights organization wants to reduce rape on her college campus, so she gets
her university’s approval to implement an informed-consent class during first-year
orientation that teaches students how to obtain and give consent.
2. Program B (self-defense class): The leader of the college chapter of a human rights
organization wants to reduce rape on her college campus, so she gets her university’s
approval to implement a self-defense class during first-year orientation that teaches
students how to verbally and physically defend themselves.
3. A/B Test: The leader of the college chapter of a human rights organization wants
to reduce rape on her college campus, so she gets her university’s approval to run
an experiment by randomly assigning students to one of two test conditions during
first-year orientation. Half of all first-year students will take an informed-consent
class that teaches students how to obtain and give consent. The other half will take a
self-defense class that teaches students how to verbally and physically defend them-
selves. After a year, the human rights chapter will implement the program (informed-
consent or self-defense) of the group that experienced the fewest completed rapes.
4. RCT Test (self-defense class vs. control group): The leader of the college chap-
ter of a human rights organization wants to reduce rape on her college campus, so
12
See Santucci (2017) for informed consent and Senn et al. (2015) for self-defense.
82
she gets her university’s approval to run an experiment by randomly assigning stu-
dents to one of two test conditions during first-year orientation. Half of all first-year
students will take a self-defense class that teaches students how to verbally and phys-
ically defend themselves. The other half will not take the self-defense class. After a
year, the human rights chapter will implement the self-defense program if that group
experienced fewer completed rapes compared to the group that did not take the
class.
Similar to the results from the previous two studies, I find that individuals do not
punish or reward the human rights NGO that runs an A/B test on its programs (t = 1.47,
p = 0.1418, d = 0.12). Mean donations to the A/B testing NGO (M = $41.6) are not
different from mean donations to either the NGO that implements an untested informed
consent class or an untested self-defense class (M = $45) at the 0.0167 significance level,
although the mean donation is lower. The chosen significance level is lower than 0.05
because this study makes three comparisons instead of one. Power analysis indicates that
the study is powered with an 80% probability to detect a small-to-medium effect of 0.37
standard deviations. This evidence is inconsistent with Hypothesis 4, given that average
donations do not decrease for the NGO that runs a randomized trial to determine which
university program reduces the rate of sexual violence on campus. Figure 3.5 plots the
mean donations to each NGO.
I find that people donate less on average to the human rights NGO that conducts an
RCT (M = $40.7) compared to the NGO that implements the self-defense class (M =
$49.5) without testing its effectiveness (t = 2.47, p = 0.01424, d = 0.28). This difference
is statistically significant at the 0.0167 level of significance. This evidence is consistent
with Hypothesis 5, given that average donations decrease for the NGO that runs an RCT
to determine if its program should be implemented compared to the NGO that implements
an untested program. However, the results from Study 5 reveal that people do not punish
or reward the human rights NGO that runs an RCT (M = $40.7) compared to an NGO that
runs an A/B test (M = $41.6), in contrast to the expectations from Hypothesis 6 (t = 0.24,
p = 0.8103, d = 0.03). Figure 3.6 and 3.7 plot the mean donations to each NGO.
83
Figure 3.5: Mean Donations from Study 5 (A/B Test)
41.6
40.3
49.5
45
$0
$20
$40
Informed Consent Self Defense A or B A/B Test
Figure 3.6: Mean Donations from Study 5 (RCT Test)
49.5
40.7
$0
$20
$40
Self Defense RCT Test
84
Figure 3.7: Mean Donations from Study 5 (A/B Test
vs. RCT)
41.6
40.7
$0
$10
$20
$30
$40
A/B Test RCT Test
Study 6: Joint Evaluation (RCT Test)
Study 6 tests whether the aversion to control groups in randomized trials found in Study
5 persists when individuals compare the different charities side-by-side. A within-subjects
design is particularly important to test because a joint comparison could generate a new
frame of reference for respondents that the between-subjects studies did not generate,
allowing respondents to compare charities and thus potentially inducing an effectiveness
frame of mind. If this effect were to take place, then the signaling theory would offer little
guidance in explaining donor behavior, and my market-oriented approach to explaining
NGO behavior would be challenged. However, if donations decrease for the RCT-testing
charity, then the results demonstrate a powerful aversion to the use of control groups in
randomized trials in spite of potential countervailing influences coming from an effective-
ness frame of mind.
The wording for the vignettes in Study 6 are identical to the wording from Conditions
2 and 3 from Study 5 (self-defense class versus RCT test). The only difference is that the
study participant chooses how much money out of a hypothetical $100 to allocate between
the human rights chapter that implements an untested self-defense class versus the human
85
rights chapter that tests whether the self-defense class is effective before implementing.
179 subjects were assigned to both conditions.
I analyze the results using a paired t-test, a version of Student’s t-test used when the
same subjects are present in both groups. I find that people donate substantially less to
the human rights chapter that conducts an RCT (M = $23.9) compared to the chapter
that implements the self-defense class (M = $76.1) without testing its effectiveness (t =
10.453, p < 0.001, d = 1.56). In fact, the joint evaluation generates the largest effect
found across all four of the studies with a mean difference of $52.2. This evidence is
consistent with Hypothesis 5, given that average donations decrease for the NGO that
uses an RCT to evaluate the effectiveness of its program, and this difference is statistically
significant. Figure 3.8 plots the mean donations to each NGO.
Figure 3.8: Mean Donations from Study 6 (Joint
Evaluation)
23.9
76.1
$0
$20
$40
$60
$80
Self Defense RCT Test
Table 3.3 summarizes the results from the t-tests conducted for all four studies. The
Appendix includes additional information from robustness tests, demonstrating that the
results hold while controlling for other factors, including gender, age, income, and educa-
tion.
86
Table 3.3: Welch’s T-test Results from Studies 3-6
Study Condition N Mean Effect
3 A 164 $45.3 t(319) = 0.91
B 169 $43.8 p = 0.3618
A/B 163 $41.7 d = 0.09
4 A 100 $55.2 t(196) = 1.03
B 111 $50.7 p = 0.3026
A/B 103 $56.7 d = 0.13
5a A 154 $40.3 t(623) = 1.4708
B 157 $49.5 p = 0.14
A/B 158 $41.6 d = 0.12
5b B 157 $49.5 t(307) = 2.47
RCT 157 $40.7 p = 0.01424
d = 0.28
5c RCT 157 $40.7 t(311) = 0.24
A/B Test 158 $41.6 p = 0.8103
d = 0.03
6 B 179 $76.1 t(178) = 10.453
RCT 179 $23.9 p < 0.001
d = 1.56
Discussion
The evidence across the four studies presents a clear picture. Individuals are not averse to
the process of randomly assigning charitable programs, in contrast to Meyer et al. (2019)’s
argument, so long as the charity’s stated beneficiaries receive a treatment, even if the treat-
ment’s effectiveness is untested. However, individuals are averse to the use of a control
group in randomized trials. This effect is particularly acute when the same individual com-
pares a charity that implements an untested program to a charity that uses a control group
to determine whether the program is effective at all.
The discrepancy between A/B and RCT tests may lie in the fact that the charity will at
least implement some program in the A/B condition, whereas the charity may not imple-
ment a program in the RCT condition. In reality, doing nothing can be better for human
welfare than doing something harmful. However, doing nothing likely sends a bad signal,
suggesting that one does not care about the recipient or is not committed to helping them.
87
The willingness to use a control group, and the willingness to stop providing treatment
may signal that the individual is untrustworthy and lacks compassion and empathy.
Taken together, the four experiments demonstrate that being epistemically rigorous
does not counterbalance the risks of a negative result. Donors do not reward scientific
testing of programs for its own sake. In the case of RCTs, they financially punish the
organization for even running them, compounding the financial punishment that will likely
befall the organization via an ineffective or weak outcome as described in the previous
section. The donor gives no special credit to organizations that test program outcomes, so
the lack of rigorous impact evaluations among human rights NGOs comes as no surprise
when we understand that donors are motivated by reputational goals, and human rights
NGOs cater to their donors due to market competition.
Explaining the Supply of Overhead Information: Donors Do
Not Reward Exceptionally Positive Outcomes in the Face of
a High Overhead
Many organizations that spawned from the accountability movement supply overhead in-
formation rather than impact information. The signaling framework can explain the obses-
sion with overhead. A high overhead signals that the organization and the people working
for it are in it for themselves, not passionately moved to do something for those in need. In
reality, the overhead ratio is not correlated with effectiveness (Caviola et al., 2014). Even
the organizations responsible for publishing information about charity overhead (Charity
Navigator, GuideStar, and the Better Business Bureau’s Wise Giving Alliance) have pub-
lished open letters imploring nonprofits and donors to stop using the overhead as a proxy
for charity effectiveness, yet they continue to annually publish data on charity overhead
88
due to demand for the information.
13
The social reality is such that donors have a built-in bias against high overhead, and
they have social incentives to donate to charities with lower overhead, even though over-
head has no bearing on the charity’s effectiveness. Even if the donor possessed private
information that the overhead is not correlated with effectiveness or that high overhead is
necessary to recruit the best talent, what matters for signaling is what other people believe
(or what is public knowledge). Donors would face skepticism in their communities for do-
nating to a charity that has a high overhead. Friends would view them as being scammed
by a greedy huckster pretending to care about a cause. It is easier for the donor to just
donate to a charity with a lower overhead, rather than having to persuade one’s friends
and family that high overhead is not something to be worried about.
Given donor skepticism about high overhead, market competition should compel most
charities to lower their overhead. While this nonprofit starvation cycle is well-documented,
what is not currently understood is how individuals would balance information about
the overhead against direct information about the charity’s impact. This is particularly
important in addressing some arguments that suppose donors may be using the overhead
ratio as a proxy for charity effectiveness when they lack direct information about impact
(Caviola et al., 2014). According to the signaling theory, information about impact is
less relevant, whereas information about overhead is highly salient. Even the most cost-
effective charities should be financially harmed if they have a high overhead. Based on
this discussion, I expect the following to hold:
Hypothesis 7 (H7) People will donate less money to a charity that is revealed to have a high
overhead (relative to a charity that does not provide information about its overhead), even
when it saves more lives with a given amount of money.
13
https://learn.guidestar.org/news/news-releases/2013/2013-06-17-overhead-myth
89
Study 7: Overhead vs. Impact
I implemented one preregistered survey experiment to test whether individuals financially
punish exceptionally effective organizations that have high overhead. I randomly assign
206 subjects to one of two conditions. The first condition is the same as Condition 2
from Study 1 (realistically positive impact). The second condition describes a more cost-
effective version of the same charity, with high overhead information revealed.
1. Condition 1 (positive impact): Human Health International operates in developing
countries to provide life-saving medical care to people at risk of preventable dis-
ease. Human Health International conducted a rigorous study which allowed them
to measure the effect of their medical care. The results from the study provided
strong evidence that Human Health International saves the life of one person for
every $2,000 spent.
2. Condition 2 (exceptional impact + overhead): Human Health International oper-
ates in developing countries to provide life-saving medical care to people at risk of
preventable disease. Human Health International conducted a rigorous study which
allowed them to measure the effect of their medical care. The results from the study
provided strong evidence that Human Health International saves the life of one per-
son for every $900 spent. A large charity evaluator indicated that 50% of donation
revenue goes to overhead expenses such as CEO compensation, fundraising, staff
parties, and retreats.
At the end of the survey, subjects answer a question designed to test their comprehen-
sion of the survey. Respondents who failed the multiple-choice question (11.2%) were
subsequently removed from the analysis.
As with the previous studies, participants are asked how much they would allocate
out of a hypothetical $100 between the charity from the treatment group and some other
option.
Condition 2 clearly describes a more effective organization no matter how the partic-
ipants conceptualize overhead. Even if they think a 50% overhead implies that 50% of
money spent by the charity is purely wasted, thereby doubling the cost of saving a life,
$1,800 is still less than $2,000.
Some may wonder whether information about CEO compensation, staff parties, and
90
retreats would turn away donors rather than overhead itself or a more abstract discussion
of overhead. I chose to include this information because this is how nonprofit organizations
are often treated by the media when their high overhead is exposed. For example, when
the Washington Post investigated the overhead of International Relief and Development
(IRD), previously one of the largest contractors for USAID projects, it chose to expose the
large salaries of the founders as well as the employee retreats and parties. The title of
the article reads, “Nonprofit contractor sent government $1.1 million bill for parties and
retreats."
14
This choice of wording is deliberate. The Post could have chosen “trainings”
and “leadership meetings,” as was the preferred language of IRD.
CEO compensation, staff parties, and retreats are not necessarily at odds with effec-
tively helping the organization’s stated beneficiaries (though they can be, of course). Such
compensation could help attract talented personnel just as is done in the private sector,
which is arguably important for running an effective organization. However, these at-
tributes are at odds in a signaling system where the organization’s actions are used to
make inferences about the underlying character of those who work at the organization
and the donors who make contributions to the organization. These actions signal greed
and selfishness when charity is about signaling virtue, generosity, and sacrifice.
I find that people donate substantially less to an exceptionally effective charity when
it has a high overhead (M = $27) compared to a charity with no overhead information
and less impact (M = $65.8, t = 9.373, p < 0.001 , d = 1.31). The size of the effect is
substantially larger (d = 1.31) than the effect size of null impact information from Study
1 (d = 0.76). Figure 3.9 plots the mean donations to each condition. Section 5 from
the Appendix includes additional information from robustness tests, demonstrating that
the results hold while controlling for other factors, including gender, age, income, and
education.
14
https://www.washingtonpost.com/investigations/nonprofit-contractor-sent-government-11-million-
bill-for-parties-and-retreats/2015/03/12/eeadfdf6-c365-11e4-9271-610273846239_story.html
91
Figure 3.9: Mean Donations from Study 7 (Overhead
vs. Impact)
65.8
27
$0
$20
$40
$60
Impact, $2,000 Impact, $900 w/ 50% overhead
Discussion
This experiment indicates that there is something uniquely punishable about a high over-
head, and it is not because donors use it as a proxy for charity effectiveness. Even when
individuals are provided with information directly about the charity’s impact, donors still
financially punish the most cost-effective charity at saving lives when they are also told
about the charity’s high overhead. By comparing how people respond to impact informa-
tion versus overhead information, the study demonstrates that people are responsive to
information that is more relevant to signaling.
Plausibly, a high overhead is necessary to attract and retain highly talented staff (not
to mention to fund the evaluations themselves). But doing so jeopardizes the financial
health of the organization. Donors would rather fund a completely ineffective charity
than the most effective charity in the world, so long as the most effective one needed a
high overhead to attract top quality employees. And, responding to the incentives, NGOs
race to lower their overhead, even if the cost is the well-being of program recipients.
92
Current theories seeking to explain the behavior of human rights NGOs and their donors
are unequipped to make sense of this result. However, when human rights NGOs and their
donors are viewed as self-interested actors working together to improve their own statuses
and reputations, it is unsurprising that they would rather save face than save lives.
Limitations
Before concluding, it is worth discussing possible limitations with the current studies. One
potential concern is the nature of using convenience samples to make inferences about
a larger population. Every study in this chapter recruited participants from Amazon Me-
chanical Turk (Mturk). Individuals who participate in online tasks on Mturk might be
different from the people who do not participate on Mturk. For example, one such dif-
ference could be income: participants might use Mturk for supplemental income, making
them lower-income and possibly lower-educated on average compared to a representative
sample of Americans. Perhaps lower-income individuals respond differently to scientific
evaluations and cost-effectiveness information than higher-income individuals. Other dif-
ferences could also exist, including age, gender, and other factors not explicitly measured
by the survey questionnaires.
I analyzed whether different income groups respond differently to the treatment con-
ditions. One might wonder whether high net worth individuals respond differently than
the general population. For the impact information studies (studies 1 and 2), only 7 and 6
individuals, respectively, reported earning an income above $200,000, which is the mini-
mum threshold used to select high net worth donors from nationally representative donor
surveys, such as the US Trust’s Study of High Net Worth Philanthropy.
15
Rather than group-
ing the sample based on high net worth status, instead I grouped the sample into three
categories: low, middle, and high income. Low income corresponds to a household income
15
https://scholarworks.iupui.edu/bitstream/handle/1805/11234/high-net-worth_oct_2017-1.pdf?
sequence=4&isAllowed=y
93
Figure 3.10: Mean Donations by Income Group
(Study 1 Impact Information)
$0 $20 $40 $60 $80
No Impact Information
Impact Information: $2,000
Impact Information: Ineffective
Impact Information: $150
Income Group
low
middle
high
of $0 to $39,999; middle income corresponds to $40,000 to $84,999; and high income cor-
responds to $85,000 to $200,000+. For each study, the same pattern holds across income
groups. Figures 3.10, 3.11, and 3.12 plot the mean donations for each income group from
studies 1, 7, and 5. The Appendix includes the graphs from the other studies.
Another potential concern is whether an American sample generalizes to a non-American
sample. Do people outside of the United States respond in different ways to randomized
trials and cost-effectiveness information? If Americans were different from other popu-
lations, then the human rights organizations servicing non-American markets might face
very different financial incentives for adopting rigorous impact evaluations and publiciz-
ing the cost-effectiveness of their programs. The most convincing method for testing the
94
Figure 3.11: Mean Donations by Income Group
(Study 7 Impact vs. Overhead)
$0 $20 $40 $60
Impact, $2,000
Impact, $900 w/ 50% overhead
Income Group
low
middle
high
95
Figure 3.12: Mean Donations by Income Group
(Study 6 A/B and RCT Testing)
$0 $20 $40
Program A
Program B
A/B Test
RCT Test
Income Group
low
middle
high
96
generalizability of the findings from these studies is to recruit participants from outside of
the United States and see whether the empirical patterns are similar or different.
Barring that, I compared whether the findings from these studies are similar to find-
ings from non-American experiments and surveys that analyze similar (but not identical)
treatment conditions. Presently, I know of a couple of such experiments and surveys, and
the evidence looks similar. Samples involve participants from a representative sample of
donors from Puerto Rico (Osili, Ackerman, Bergdoll, Garcia, Kalugyer, Li, Kane and Roll,
2016) and a convenience sample of college students from Switzerland (Metzger and Gün-
ther, 2019). The Puerto Rican survey indicates that 1.2% of Puerto Ricans stopped giving
to a charity because they no longer believed it was effective. The Swiss survey experiment
suggests that 23% of college students pay for information about the charity’s impact before
donating to that organization. These figures compare with the studies presented in this
chapter: 35% and 23% of respondents donated all the money to the cost-effective charity
from studies 1 and 2 respectively, and 8-21% donated all the money to the charities that
implemented a randomized trial (from studies 4-7).
16
The point here is that no survey
or experiment exists in which a majority of the sample sought impact information when
deciding where and whether to donate.
Conclusion
The experimental evidence in this chapter demonstrates that human rights NGOs are fi-
nancially best off by either not revealing their impact or by lying about their impact. This
evidence is consistent with a theory of NGOs that takes into account the self-interested na-
ture of donors and organizations but is unexplained by a theory that assumes that donors
and human rights NGOs are motivated by pure altruism. If the main function of a human
rights NGO is to contribute to a transnational activist network, then sharing negative in-
16
11.7% donated $100 to the A/B test charity in study 4, 21.3% in study 5, 8.9% in study 6, 13% donated
$100 to the RCT charity from study 6, and 8.3% donated $100 to the RCT charity from study 7.
97
formation would be the most important information to share. Publicizing the failure of
a program would prevent other activists and NGOs from wasting their precious resources
on ineffective activities. However, in reality, human rights NGOs guard information about
their impact as though it is a trade secret (Simler and Hanson, 2018, chapter 12), and
other activists do not abandon their pet causes even in the face of strong evidence of a null
or harmful effect. That donors do not reward even charities with the best possible out-
comes more than charities that provide no impact information at all provides an insight
into the relationship between human rights NGOs and their donors. Donors are perfectly
happy to be kept in the dark about impact. While the collective action theories, such as
those described by Prakash and Gugerty (2010), view confidentiality of impact information
as a betrayal of the unsuspecting altruistic donor, this confidentiality appears to be more
readily explained by a donor preference for fantastical impact stories they can uncritically
share with their social circles.
98
Chapter 4
Still Ineffective: Disputing Research on
the Effectiveness of Human Rights NGOs
How easy it is to do harm in wishing to do good! To
foresee with certainty that an establishment will produce
only the effect desired from it, and no effect at variance
with its object...to defend oneself against the prestige of a
seductive project, to take a severe and tranquil view of it
amidst that dazzling atmosphere in which the praises of a
blind public, and our own enthusiasm, show it us
surrounded; this would need the effort of the most
profound genius, and perhaps the political sciences of our
time are not yet sufficiently advanced to enable the best
genius here to succeed.
Anne Robert Jacques Turgot, “Fondation"
Introduction
Readers aware of statistical studies in the field of international human rights may ask: if
human rights NGOs do not ensure their programs have a positive impact, then why does
academic statistical research demonstrate that human rights NGOs are generally effective
at advancing human rights globally, in particular physical integrity rights?
Many scholars working in this area frame existing statistical studies on the effectiveness
of human rights NGOs as painting a positive, albeit conditional, effect (Bell, Clay and Mur-
99
die, 2012; Franklin, 2008; Hendrix and Wong, 2012; Murdie, 2014; Murdie and Davis,
2012; Murdie and Peksen, 2015).
1
In Evidence for Hope, Kathryn Sikkink — one of the
founders of the human rights field — has recently concluded that “Multiple studies, using
a wealth of the best data on the topic, have shown that we can be cautiously optimistic
about the impact of the work of human rights INGOs, but that for the greatest success,
information politics need to be combined with efforts to build strong domestic advocacy
sectors within states, while also bringing pressure to bear from outside" (Sikkink, 2017,
14). Murdie and Peksen echo this conclusion: “a growing body of scholarship has demon-
strated that human rights INGO shaming advocacy has, at least under certain conditions,
had a significant impact on curbing the government’s willingness to commit repression
against it citizens" (Murdie and Peksen, 2015, 3). Murdie, Davis, and Park likewise con-
clude: “Shaming forms the backbone of much of our theoretical understanding of how
human rights improvement occurs...discourse from HROs [human rights organizations] is
critical in getting a regime to start to rethink the use of abusive practices" (Murdie, Davis
and Park, 2020, 84).
Does the statistical evidence support the conclusion that human rights NGO advocacy
has had a significant impact on reducing state repression? Hafner-Burton and Ron cau-
tioned human rights researchers: “students of scientific knowledge production have long
observed a tendency to publish accounts of positive intervention more frequently than to
publish null or negative results. Human rights researchers should take care not to fall into
the same trap" (Hafner-Burton and Ron, 2009, 394).
This chapter re-examines these studies and finds that they suffer from numerous critical
and devastating problems. I assemble the five most cited statistical papers testing the
effectiveness of human rights NGOs on respect for human rights and analyze the relevant
models from each of the studies, totaling 35 models. I examine (1) whether human rights
NGOs are more likely to have a negative or positive impact, (2) whether the estimated
1
In particular, this research tests whether propositions related to human rights naming-and-shaming from
Keck and Sikkink (1998)’s boomerang model are supported by quantitative evidence.
100
effects of human rights NGOs are widespread or narrowly concentrated, and (3) whether
the literature is biased in its reporting of the apparent efficacy of human rights NGOs.
I find that their own analyses reveal human rights NGOs to be more harmful than
beneficial, but researchers fail to highlight these harmful effects in their abstracts, intro-
ductions, discussions of results, and key takeaways. My replication also reveals that the
narrow positive effects are driven by a very small number of countries. Most of the time,
however, the impact of human rights NGOs simply does not achieve statistical significance.
Researchers fail to emphasize the plethora of null results, choosing instead to direct their
attention to the rare statistically significant positive effect or to spin the null results as
consistent with their theory.
The cautious view I present in this paper should not be confused with Sikkink’s descrip-
tion of pessimism: “Pessimism about human rights progress is widespread. Whether on the
news, in the academy, or when one talks to a member of the general public, the standard
view is that all types of human rights abuses in the world are getting worse...Some aca-
demics critique human rights law, institutions, and movements for this perceived lack of
progress" (Sikkink, 2017, 7). This chapter presents a different kind of critique. I do not dis-
miss the effectiveness of NGOs and activists on the basis of stagnating or declining trends
in respect for human rights or human progress more generally. Many researchers have
amassed compelling evidence that demonstrates the great strides in human progress, in-
cluding several cited by Sikkink (2017). This research establishes that the metrics related
to most areas of life that matter to people — health, income, rights, violence, even per-
sonal entertainment — have improved over the last century (Fariss, 2014, 2019; MacAskill,
2015; Pinker, 2012, 2018; Roser, 2019; Rosling, Rönnlund and Rosling, 2018).
Instead, this chapter contributes to this important debate by teasing apart two sepa-
rate empirical arguments that are often conflated. The first empirical question is whether
government compliance with physical integrity rights standards has improved over time.
Most proponents of international human rights practice say yes (e.g. Sikkink (2017)), and
101
some critics say no (e.g. Posner (2014); Hopgood (2013)). A separate and second em-
pirical question is whether human rights NGOs themselves have contributed to this trend.
Proponents say yes. Few critics have weighed in on this point, focusing instead on whether
international human rights law has mattered, rather than NGOs.
2
This chapter demon-
strates that the second empirical claim is not supported by their statistical analysis, even
though the proponents are correct that human welfare has improved.
Moving forward, the evidence presented in this chapter suggests that other political or
economic forces have been more important in contributing to the improvement in global
respect for physical integrity rights. In this sense, the findings from this chapter converge
with Hill and Jones (2014), a study that demonstrates the greater predictive power of do-
mestic legal systems and economic and demographic factors in statistical models of state
repression. A theory of a virtue economy can make sense of these two competing empiri-
cal realities because human rights NGOs calibrate their activities to optimize the signaling
value of donations to their organizations. Thus, it would be surprising and unlikely if hu-
man rights NGOs were effective catalysts of global respect for physical integrity rights. This
conclusion should not be interpreted as bleak; the advancement of the human condition
can be and has been achieved without much help from human rights NGOs.
Analysis
I focus my analysis on quantitative rather than qualitative studies because researchers
feel more secure in their conclusions when backed by quantitative evidence. For example,
Sikkink argues that “employing primarily empirical comparisons with careful use of human
rights data can generate persuasive evidence for the effectiveness of some human rights
2
Since human rights NGOs often advocate for state commitment to and compliance with international
human rights law, it is understandable why critics have chosen to focus on law rather than NGOs per se.
However, for precision’s sake, it is important to separate these two potential causal factors because human
rights NGOs do not always resort to legal remedies. Issue awareness, education, and naming-and-shaming
are common non-legal techniques adopted by human rights NGOs.
102
law and activism" (Sikkink, 2017, 11). Human rights researchers within the quantita-
tive tradition often frame quantitative studies as correctives to biased qualitative research
(Murdie and Davis, 2012; Murdie, Davis and Park, 2020). As recently summarized in Mur-
die, Davis, and Park, “Risse (2002: 274) pointed out some of these [qualitative] studies
‘suffer from methodological problems such as case selection on the dependent variable.’ As
Cingranelli and Richards (2001: 225) lamented, quantitative data is necessary for there
to be convincing ‘scientific evidence of the effectiveness of human rights organizations’"
(Murdie, Davis and Park, 2020, 84).
As a result, this paper focuses on the studies that are considered by human rights
researchers to present the highest standard of evidence in favor of human rights NGO
effectiveness. These studies are listed in Table 4.1. I include only the statistical studies
that estimate the effect of human rights NGOs on state repression. There are many more
studies that estimate the effect of human rights NGOs on other outcomes that are believed
to affect state repression (e.g. public opinion, economic sanctions, protest, foreign direct
investment, humanitarian intervention). I exclude these studies from the analysis because
their dependent variables are not a direct measure of respect for human rights, and the
studies do not make the requisite additional empirical step of demonstrating that NGOs
reduce state repression through protest, sanctions, FDI, etc. I also include only the papers
that find a positive effect, so this replication excludes the research from Hafner-Burton
(2008), Cardenas (2007), and Hill and Jones (2014) that find null or negative uncondi-
tional effects. I also exclude Murdie (2014) — a book-length statistical investigation into
the effects of human rights NGOs — because the work is largely based on the individ-
ual papers already included in the replication analysis. Of the remaining seven studies, I
analyze the five most cited, which totals 1,027 citations.
3
3
This citation count is based on the number of citations reported on Google Scholar as of January 16,
2021. I originally planned to include Hendrix and Wong (2012) as a sixth study, but their replication data
does not exist. Hendrix confirmed this via email. I omit the seventh study on the basis of time constraints.
The book manuscript will include this last study and any additional relevant studies I did not catch or that
may have been published since the dissertation.
103
Table 4.1: Studies Included in the Replication
Analysis
Authors Year Title
Franklin 2008
Shame on You: The Impact of Human Rights Criticism
on Political Repression in Latin America
Murdie and Davis 2012
Shaming and Blaming: Using Events Data to Assess the
Impact of Human Rights INGOs
Krain 2012
J’accuse! Does Naming and Shaming Perpetrators Reduce
the Severity of Genocides or Politicides?
DeMeritt 2012
International Organizations and Government Killing:
Does Naming and Shaming Save Lives?
Bell, Clay, and Murdie 2012
Neighborhood Watch: Spatial Effects of Human Rights
INGOs
I give the studies the full benefit of the doubt by assuming that their models are properly
specified. As such, I do not critique their control variables or estimation strategy. Instead,
I ask, assuming the models are properly specified, does the output of their models support
the conclusion that human rights NGO advocacy, even under certain conditions, has had
a significant impact on curbing state repression? To answer this question, I examine (1)
whether human rights NGOs are more likely to have a positive or negative impact, (2)
whether the positive impact is widespread or narrowly concentrated, and (3) whether the
results suffer from multiple forms of reporting bias. I explain each step in more detail
below.
Are human rights NGOs more likely to have a positive or negative im-
pact?
An important debate in the international human rights literature pertains to what Moyn
defines as a “tiff about whether describing the same thing as mildly uplifting or rather de-
pressing makes more sense” (Moyn, 2018, 127-128). Proponents of the boomerang model
point to the conditional effectiveness of human rights NGOs (Murdie, 2014; Murdie and
Davis, 2012) and the conditional effectiveness of international human rights law (Conrad
104
and Ritter, 2019; Simmons, 2009) as “evidence for hope” (Sikkink, 2017). Yet, skeptics
look at the same evidence unimpressed (Hafner-Burton and Ron, 2009; Moyn, 2018; Pos-
ner, 2014).
For example, in response to the conditional effectiveness of international human rights
law presented by Simmons (2009), both Posner (2014) and Hafner-Burton and Ron (2009)
point out the narrow set of circumstances in which international law could improve human
rights outcomes. “Understood in the best possible light, these studies suggest that a small
number of treaty provisions may have improved a small number of human rights outcomes
in a small number of countries by a small, possibly trivial amount” (Posner, 2014, 78).
Hafner-Burton and Ron run the calculations to estimate how many people could have been
helped by international human rights law assuming that ratification is effective under the
conditions specified by Simmons (2009). They conclude that only a quarter of the world’s
countries could have been helped or 10 percent of the world’s population (Hafner-Burton
and Ron, 2009, 372). Tongue in cheek, Moyn settles that international human rights law
is effective “when Goldilocks consents” (Moyn, 2018, 126).
The statistical studies estimating the effect of human rights NGOs follow a similar pat-
tern in that many of the studies fail to find an unconditional effect of human rights NGOs
but do claim to estimate a positive conditional effect. For example, Murdie and Davis ar-
gue that “mere shaming is not enough" and that the boomerang model explicitly stipulates
conditional propositions: a combination of pressure both internationally and domestically
is necessary to force governments to change their behavior (Murdie and Davis, 2012, 1-2).
They claim to find evidence consistent with this proposition: human rights NGO naming-
and-shaming is effective, conditional on a large presence of domestic rights organizations
and pressure from foreign governments and intergovernmental organizations. Franklin
(2008) also claims that human rights NGOs are effective at curbing repression, but only
when the target state is dependent on foreign capital such as foreign aid and foreign direct
investment.
105
This debate fails to address the possibility that human rights NGOs might also have a
negative impact, in addition to having a narrow positive impact. Several of the papers use
an interaction term to test the conditional effectiveness of human rights NGOs. But effect
estimates for interaction terms can be misleading when not carefully interpreted. Even if
the sign of the effect estimate is positive (suggesting that the effect of human rights NGOs
increases as values along the conditional factor increase), the effect could be negative at
each value of the conditional factor. To address this problem, I replicate every model that
estimates a conditional effect and purports to find a statistically significant positive effect,
and I measure whether human rights NGOs have a positive, null, or negative effect on
respect for human rights at varying levels of the conditional factor. I then calculate what
percentage of the estimation sample is positively or negatively affected by human rights
NGOs in order to more accurately evaluate whether human rights NGOs are more often
beneficial or harmful.
Based on the meta-analysis, I find that when human rights NGOs have a statistically
significant effect on government repression, the effect is more often harmful than benefi-
cial. To illustrate, I looked at the underlying data from Murdie and Davis (2012), the paper
with the most citations in the literature and considered to be the most important empirical
test of human rights NGO effectiveness. Murdie and Davis conclude that advocacy “is an
effective strategy when combined with a strong presence on the ground” of local human
rights organizations.
They base their conclusion on the results from an ordinary least squares regression
model with an interaction term between a measure of international human rights NGO
naming-and-shaming and a measure of the presence of local human rights organizations
within a country. They measure human rights using quantitative data on a country’s respect
for the physical integrity rights of its citizens, i.e. the extent to which the government
engages in torture, disappearances, political imprisonment, and extrajudicial killing. They
control for numerous potential confounding factors, including the size of the country, the
106
country’s wealth, the political regime type, and whether the country is involved in war.
Their results indicate that the relationship is positive and statistically significant. The
sign of the effect estimate on its own does not reveal whether NGOs are beneficial or harm-
ful, so they plot the marginal effect of NGO advocacy to visualize how the effect of NGOs
changes with varying amounts of local rights organizations. This plot is reproduced below
in Figure 4.1.
4
The x-axis is the natural log of the number of domestic rights organizations.
The y-axis is the slope or effect estimate of NGO naming-and-shaming on a country’s phys-
ical integrity rights record. Positive values on the y-axis indicate that NGO naming-and-
shaming is beneficial: NGOs improve a country’s physical integrity rights record. Negative
numbers indicate that NGOs are harmful: they worsen a country’s physical integrity rights
record.
Murdie and Davis predicted that NGOs would only be beneficial with a strong presence
of local rights organizations. However, their plot shows that for some values of domestic
rights organizations, NGOs are actually harmful, yet they do not consider this finding to
contradict their theory. Although they acknowledge that “as the graph highlights, shaming
can actually be a negative strategy for increasing human rights performance if the level of
HRO domestic presence is low," they do not admit that this finding contradicts their theory
or the boomerang model. Instead, they write, "However, as HRO Presence (ln) increases,
HRO shaming can be an effective tool at improving human rights performance” (Murdie
and Davis, 2012, 9-10). Throughout the rest of their paper, they ignore their finding of the
harmful effect of human rights NGOs, failing to wrestle with its falsifying potential and
choosing instead to feature the less frequent positive effect.
To know whether the positive effects are manifested more often than the negative ef-
fects (or vice versa), I calculated the percentage of the sample that is positively or nega-
tively affected by NGO advocacy and plotted this information using a histogram (Figure
4
The original marginal effects plot from Murdie and Davis does not replicate in either R or STATA. The
95% confidence interval should drop below 0 for values that fall below 2.1 ln on the x-axis. Other than this,
the only other difference between the reproduced graph and their original graph is the range on the x-axis.
107
Figure 4.1: Marginal Effects: Reproduced from
Murdie and Davis (2012) Model 2, Table 1
−0.75
−0.50
−0.25
0.00
1 2 3 4 5
Domestic Presence of
Human Rights Organizations (ln)
Marginal Effect of
Human Rights NGO Advocacy
Figure 4.2: The Marginal Effect of Human Rights
NGO Advocacy on Respect for Physical Integrity
Rights (Murdie and Davis (2012) Model 2, Table 1)
108
4.2), which color-codes the observations benefited and harmed by NGOs. The plot displays
that there are very few countries with more than 4.7 ln domestic rights organizations (110
organizations), the threshold needed for NGOs to have a statistically significant positive
impact at the 95% confidence level. Most of the sample is concentrated in the null region
between 3 and 4 ln (between 20 and 90 organizations). Contradicting their prediction,
some of the sample falls below the critical threshold of 2.1 ln (8 organizations) in which
NGOs have a statistically significant negative impact. Their data actually show that only
0.77% of their sample is helped by NGOs (9 country-years). More than four times that —
3.2% — of their sample is harmed by NGOs (38 country-years). The 9 country-years with
a large enough presence of local rights organizations for NGO advocacy to make a posi-
tive impact are the United States during 2000 and France and the United Kingdom during
2000 to 2003. These wealthy, democratic countries are not predicted by the boomerang
model to be the most likely cases of effective NGO advocacy. In fact, Murdie and Davis
erroneously claim that “having a greater-than-average domestic presence of HROs within
a targeted state is crucial for shaming to be an effective strategy for human rights pro-
motion. This level of HRO domestic presence was seen in Brazil, Senegal, and Sri Lanka,
among many others, during this time period” (10). This claim is important to their paper
and the entire transnational activist network model because it suggests that NGO advocacy
is helpful to countries in need, not just rich ones. But this claim is completely false. At no
point do any of these countries see the presence of domestic rights organizations needed
for NGOs to have a statistically significant positive impact.
5
Murdie and Davis also analyze the effectiveness of human rights NGOs conditional on
the advocacy of other governments or intergovernmental organizations. A similar pattern
holds here: their marginal effects plot (reproduced in Figure 4.3) grossly distorts the re-
ality. While the graph appears to indicate that most countries are positively impacted by
5
They claim that the critical threshold is one standard deviation above the mean. But the mean is 3.395,
and the standard deviation is 0.617, so one standard deviation above the mean is 4.01, or 55 domestic rights
organizations — well under the critical threshold of 110 organizations (4.7 ln). None of those countries
listed by Murdie and Davis reach that level at any point.
109
NGOs, in fact the vast majority — 99% of the distribution — sits below 20 on the x-axis,
the threshold needed for NGOs to have a statistically significant positive impact. Only 1%
of the distribution falls above 20 (14 country-years). Once this data is visualized in a less
deceptive manner, readers can see that most of the time, NGOs do not have a positive
impact (Figure 4.4).
Figure 4.3: Marginal Effects: Reproduced from
Murdie and Davis (2012) Model 3, Table 1
−0.1
0.0
0.1
0.2
0.3
0 20 40 60
Indirect Targeting
Marginal Effect of Human Rights NGO Advocacy
To demonstrate the robustness of their findings, Murdie and Davis include in their ap-
pendix a set of models that use an alternative measure for NGO naming-and-shaming,
a variable that measures Amnesty International’s advocacy. While they do not produce
marginal effects plots to unpack their results, they nonetheless claim that “the resulting in-
teraction term is statistically significant and positive...furthering our contention that sham-
ing and blaming has a conditional effect on human rights performance.” As with the previ-
ous two models just discussed, I used their replication data to calculate what percentage of
their sample is positively or negatively benefited from Amnesty International’s advocacy.
I found that 95% of their sample is harmed by Amnesty International, and 0% is helped,
despite the authors’ conclusions (Figures 4.5 and 4.6). An unbiased discussion of these
results would note that if these models are to be taken seriously, then Amnesty should im-
110
Figure 4.4: The Marginal Effect of Human Rights
NGO Advocacy on Respect for Physical Integrity
Rights (Murdie and Davis (2012) Model 3, Table 1)
mediately halt its advocacy in countries with lower levels of domestic rights organizations
or countries with less international pressure until further review of its work to figure out
why its advocacy appears to be backfiring.
Figure 4.5: Murdie and Davis (2012)
Model 1 Table 7
Figure 4.6: Murdie and Davis (2012)
Model 1 Table 8
I used this same method to investigate the results reported in two other papers that test
111
a conditional hypothesis. Franklin (2008) tests a prediction from the boomerang model
that “countries receiving large military and economic aid flows will be more vulnerable
to human rights pressures than those not receiving such flows” (199). The implication is
that human rights criticism will be most effective against the most dependent governments
and either not effective or less effective against the least dependent. What is not implied
here is that human rights criticism will be harmful against the least dependent countries.
Yet that is exactly what Franklin’s model reveals, despite his interpretation that “human
rights criticism significantly reduces repression against subsequent contentious challenges
in countries that have a greater reliance on foreign aid and investment” (203).
Franklin’s model shows there are five times as many observations that are harmed by
NGO advocacy than are helped. NGO advocacy significantly increases repression against
subsequent contentious challenges in countries that have a lower reliance on foreign aid
and investment, which accounts for 74% of the sample. In contrast, NGO advocacy sig-
nificantly reduces repression against subsequent contentious challenges in countries that
have a greater reliance on foreign aid and investment, which accounts for only 14% of the
sample (Figure 4.7).
6
To reiterate, the boomerang model does not anticipate that NGOs
will make repression more likely in countries that have a combined dependence on aid and
FDI of less than 2% of their GNI and GDP.
7
Neither Keck and Sikkink (1998) nor Franklin
(2008) have argued that countries should become more dependent on foreign aid in or-
der to avoid human rights activism making state repression more likely, yet this is what
Franklin’s model implies.
6
In Franklin (2008), foreign capital dependency combines two forms of capital dependence: foreign aid as
a percentage of gross national income (GNI) and foreign direct investment as a percentage of gross domestic
product (GDP). Aid/GNI is a highly skewed variable, with 87% of observations less than 1% aid/GNI. Only
two countries in the sample —- Nicaragua and Guatemala —- have aid accounting for more than 1% of their
GNI. Most of the x-axis is driven by these two countries because values of FDI/GDP sit between 0% and 5%.
In fact, the positive effect of foreign capital dependence is driven entirely by aid dependence, not FDI.
7
Franklin specifies an ordered probit regression model using a dependent variable with four levels. To
calculate what percentage of the estimation sample is negatively or positively impacted by NGO advocacy,
I calculated the marginal effect of NGO advocacy on the probability that a country would not engage in
political repression. If NGO advocacy increases the probability of not committing repression, then they have
a positive impact. In contrast, if NGOs decrease the probability of not committing repression, then they have
a negative impact. This information is used to populate the histogram in Figure 4.7.
112
Figure 4.7: The Marginal Effect of Human Rights
NGO Advocacy on Political Repression from Franklin
(2008) Model 2, Table 5
Bell, Clay and Murdie (2012) analyze whether human rights NGOs positively impact
the human rights conditions in neighboring countries. They argue that “the presence of
HRO [human rights organization] members or volunteers ‘next door’ increases the advo-
cacy mobilization and resources of a domestic population, resulting in improvements in
human rights performance" (355). They predict that the effect of human rights NGOs will
be greater when activists can freely travel between countries. The authors conclude they
“find widespread support for [the] hypotheses: HRO members can form a powerful ‘neigh-
borhood watch’ that can lead to improvements in human rights performance" (355). They
develop three ways to measure their independent variable (neighboring human rights NGO
activity), but only one of them is positive and statistically significant: the average count
(natural log) of the number of human rights NGOs with members or volunteers within the
domestic borders of a country’s neighbors.
8
By using the marginal effects to calculate how
the sample is impacted by neighboring human rights NGO activity, I find that neighboring
8
The two alternative measures that are not statistically significant include neighborhood HRO shaming
— or the average count of human rights NGO shaming events that a country’s neighbors are targeted with in
a given year — and neighborhood human rights NGO secretariat offices — or the average number of human
rights NGOs that have a permanent office within the domestic borders of a country’s neighbors.
113
Figure 4.8: The Marginal Effect of Neighboring
Human Rights NGO Membership on Improvement in
Physical Integrity Rights from Bell, Clay, and Murdie
(2012) Model 3, Table 2
human rights NGO membership has a positive impact on nearly 60% of the sample, and it
has a null effect on the remaining 40% (Figure 4.8).
The authors test the robustness of their findings using an alternative dependent variable
and report again that only one of the measures of human rights NGO activity is positive
and statistically significant. However, this time, it is the measure of neighboring HRO
secretariats that is significant, not neighboring HRO membership. I was unable to replicate
their results. While the authors posted their analysis script online for the tables reported
in the main paper, they did not post their script used to create their appendix tables. I
based my replication on exactly what they reported in their paper and appendix, and my
results show that the interaction term is negative and not statistically significant. This
means that when freedom to move is greater, the effect of human rights NGO secretariat
offices decreases, despite what they reported and predicted. Even though the slope of
the interaction term is negative, I find that neighboring human rights secretariat offices
positively impact human rights in 33% of the estimation sample, but they have a null
impact in 67% of the sample (Figure 4.9).
114
Figure 4.9: The Marginal Effect of Neighboring
Human Rights NGO Secretariat Offices on Physical
Integrity Rights from Bell, Clay, and Murdie (2012)
Model 2 Table A24
Numerous models, using a variety of measures that capture different aspects of human
rights NGO advocacy, demonstrate that human rights NGOs are ineffective most of the
time, but when there is a statistically significant effect, the effect is more often harmful
than beneficial. Table 4.2 summarizes the results discussed in this section. Based on this
analysis, it should not have been surprising to authors of a recent field experiment test-
ing the effect of international shaming by human rights organizations to find a backfire
effect (Terechshenko et al., 2019). Human rights researchers have emphasized the condi-
tional effectiveness of human rights NGOs in order to rescue the NGOs from accusations
of backlash.
9
Yet, these very papers that are said to provide a high standard of evidence
affirming the effectiveness of human rights NGO advocacy actually make a much stronger
case for NGOs as harmful agents operating within the global human rights movement. An
unbiased review of this evidence would have taken these results seriously, probing the im-
plication that these models offer falsifying evidence against Keck and Sikkink’s model of
transnational activist networks.
9
For an example, see Murdie and Davis’ (2012, 4) response to Hafner-Burton (2008).
115
Table 4.2: Summary of Results from Models Testing
the Conditional Effectiveness of Human Rights NGOs
and Purporting a Statistically Significant Positive
Effect
Paper Model
% of estimation sample negatively
impacted by NGOs
% of estimation sample positively
impacted by NGOs
Murdie and Davis (2012) Model 2, Table 1 3.20% (n = 38) 0.77% (n = 9)
Murdie and Davis (2012) Model 1, Table 7 95.30% (n = 1,556) 0.00% (n = 0)
Murdie and Davis (2012) Model 3, Table 1 0.00% (n = 0) 1.02% (n = 14)
Murdie and Davis (2012) Model 1, Table 8 94.70% (n = 871) 0.00% (n = 0)
Franklin (2008) Model 2, Table 5 74.11% (n = 647) 13.86% (n = 121)
Bell, Clay, and Murdie (2012) Model 3, Table 2 0.00% (n = 0) 60.00% (n = 600)
Bell, Clay, and Murdie (2012) Model 2, Table A24 0.00% (n = 0) 33.00% (n = 330)
Is the positive impact of human rights NGOs widespread or narrowly
concentrated?
While many studies that analyze the unconditional effectiveness of human rights NGOs
either fail to find a statistically significant effect or find a statistically significant negative
effect (Franklin, 2008; Hafner-Burton, 2008; Murdie and Davis, 2012), some studies do
in fact find a positive effect that is statistically significant (Bell, Clay and Murdie, 2012;
DeMeritt, 2012; Krain, 2012). To address whether these positive effects are widely dis-
tributed or narrowly concentrated, I analyze which observations are poorly explained by
the control variables and therefore have a bigger impact on the estimation of the effect of
NGOs on human rights, a methodology articulated by Aronow and Samii (2016).
Aronow and Samii demonstrate that multiple regression models differentially weight
observations from the nominal sample. The nominal sample is the original sample of ob-
servations to which researchers wish to fit a regression model. In the nominal sample,
all observations are weighted equally. However, after fitting a multiple regression model,
these observations contribute to the regression effect estimates to different extents based
on how well other variables in the model can account for the values of the observations.
Observations that are well explained by the control variables are given less weight than ob-
servations that are not well explained by the control variables. This weighted sample is the
effective sample that multiple regression uses to estimate the effect. Due to the weighting,
116
“the ‘effective sample’ that regression uses to generate the estimate may bear little resem-
blance to the population of interest, and the results may be nonrepresentative" (Aronow
and Samii, 2016, 251).
Aronow and Samii present a method for characterizing the effective sample by reweight-
ing the nominal sample based on the multiple regression weights, which measure each
observation’s contribution to the effect estimate. The weights are very easy to calculate.
First, the explanatory variable of interest (e.g. human rights NGO advocacy) is regressed
on the control variables. The residuals are then calculated. The value of these residuals
measure the difference between the predicted value and the observed value of the NGO
variable when regressed on the control variables. Multiple regression uses the square of
these residuals to weight the observations. Observations with larger residuals (i.e. the
observations for the NGO variable that are poorly explained by the control variables) are
given more weight when estimating the effect of NGOs on respect for human rights. More
weight goes to observations whose values for NGO advocacy are not well explained by the
control variables.
10
Multiple regression weights provide an accurate measure of the true scope conditions
of the positive effect estimates of human rights NGO advocacy. I will use the multiple
regression weights from each of the relevant studies to calculate the generalizability of the
model’s effect estimates. This method will offer a more precise statement about the extent
to which human rights NGOs have positively impacted the world. The results, in turn,
will contribute to the debate between the critics of human rights practice (e.g. Posner and
Moyn) and the proponents (e.g. Sikkink and Murdie).
Based on the meta-analysis of multiple regression weights, I find that in the few cases
when human rights NGOs are shown to have a statistically significant unconditional pos-
itive impact on human rights, the effect is very narrowly confined to a small subset of
10
The leverage statistic is similar, but leverage is a measure of an observation’s influence on the effect
estimates for all of the independent variables, whereas multiple regression weights measure an observation’s
influence on a specific explanatory variable’s effect estimate.
117
the world. Often, the narrow set of observations that are positively affected by human
rights advocacy are not predicted by the authors themselves to be the most likely cases of
effective NGO advocacy.
For example, DeMeritt (2012) analyzes the effectiveness of international human rights
NGO advocacy at reducing the likelihood and severity of state-sponsored killing. She finds
that NGO advocacy is effective at reducing the civilian death count as well as the proba-
bility of any civilian killings. While DeMeritt tests the hypotheses against a sample of over
550 observations represented by 93 countries spanning 8 years (Figure 4.11), the findings
do not apply to almost any country in the sample.
To see why, Figure 4.10 plots a histogram of the distribution of multiple regression
weights. Each observation’s weight is scaled to aid with the interpretation and compare
across models. A value that is not scaled can be difficult to interpret and compare across
models. For example, a weight of 200 might be low for one model but high for another.
Each observation’s weight is divided by the sum of the total weight, measuring that ob-
servation’s contribution to the effect estimate. Values are therefore bounded between 0
and 1. A scaled value close to zero indicates that the observation contributes very little to
the effect estimate. Higher values indicate the observation contributes more to the con-
struction of the effect estimate. As the histogram illustrates, most observations contribute
nearly 0% of the weight used to construct the estimate of the effect of human rights NGO
advocacy on state-sponsored killing.
Based on the methodology presented by Aronow and Samii, I modified the map of
the nominal sample by shading the countries based on how much weight they contribute
to the effect estimate (Figure 4.12). A darker shade indicates the country contributes
more weight to the effect estimate; a lighter shade indicates the country contributes less
weight. DeMeritt’s findings are highly skewed toward Israel, despite no mention of Israel
as a special or most likely case of effective human rights advocacy. While difficult to
tell based on the map because of the country’s small geographic size, Israel contributes
118
Figure 4.10: Multiple Regression Weights
(Represented as a Percentage of Total Weight) Used
to Construct the Effect Estimate from DeMeritt
(2012) Model 1, Table 2
Figure 4.11: DeMeritt (2012)
Nominal Sample
Figure 4.12: DeMeritt (2012)
Effective Sample
119
the most weight. Israel in 2001 contributes 13.7% of the weight, and Israel in 2002
contributes nearly 9% of the weight. For reference, if all observations contributed equally,
Israel in 2001 would only contribute 0.18%. Even though the country only represents
0.72% of DeMeritt’s nominal sample, Israel represents 25% of the effective sample and is
given nearly six times as much weight as the next most impactful country (Colombia) in
producing the estimate.
To provide another sense of skewness, a mere 1.45% of the sample (8 observations)
contributes over 50% of the weight, and the bottom 242 observations (i.e. the bottom
44% of the sample) combine to contribute just 1% of the weight.
11
The kurtosis of the
distribution (a common measure of a univariate distribution’s skewness) is 114, orders of
magnitude larger than the kurtosis of a normal distribution of 3.
I calculated the multiple regression weights from two other papers that found a sta-
tistically significant positive effect. Krain (2012) analyzes the effectiveness of Amnesty
International’s advocacy at reducing the severity of mass killings in a genocide or politi-
cide. While he finds that Amnesty is not effective at improving general respect for physical
integrity rights, he claims that “naming and shaming has an ameliorative effect on the
physical integrity of people on the ground in cases of genocide or politicide" (Krain, 2012,
586). Krain tests the hypotheses against a sample of over 200 observations represented
by 27 countries spanning 33 years (Figure 4.14), but the findings apply to a much more
restrictive sample (Figure 4.13 and Figure 4.15). The findings are skewed toward Sudan.
The highest-weighted observation — Sudan in 2007 — contributes 9.4% of the weight.
The second highest weighted observation — Guatemala in 1981 — contributes nearly 5%
of the weight. Neither of these countries are discussed by Krain as expected cases of par-
ticularly effective human rights NGO work. Only 18 observations (or 9% of the nominal
sample) contribute over 50% of the weight. The bottom 63 observations (the bottom 31%
11
These numbers are based on the multiple regression weights from Model 1, Table 2. The only difference
between Models 1 and 4 is that Model 4 includes a control variable that measures the natural log of the
population. Model 1 does not include this variable. Otherwise, the multiple regression weights are nearly
identical.
120
of the sample) combine to contribute just 1% of the weight. The kurtosis of the distribution
is 40, 13 times larger than the kurtosis of a normal distribution.
Figure 4.13: Multiple Regression Weights
(Represented as a Percentage of Total Weight) Used
to Construct the Effect Estimate from Krain (2012)
Model 2, Table 3
Figure 4.14: Krain (2012)
Nominal Sample
Figure 4.15: Krain (2012)
Effective Sample
As previously discussed, Bell, Clay and Murdie (2012) analyze whether human rights
NGOs positively impact human rights outcomes in neighboring countries. While they de-
velop three metrics that capture neighboring human rights NGO activity, only one of them
is positive and statistically significant: the number of human rights NGOs with members
121
Table 4.3: Kurtosis of Multiple Regression Weights
Paper Model Kurtosis
DeMeritt (2012) Model 1, Table 2 113.93
Krain (2012) Model 2, Table 3 39.80
Bell, Clay, and Murdie (2012) Model 3, Table 1 469.32
or volunteers within the domestic borders of a country’s neighbors. They conclude that
“HRO members next door...lead to an increased probability of human rights improvements"
(362). While the authors test their hypothesis against a sample of 1,001 observations rep-
resented by 117 countries spanning 9 years (Figure 4.17), the findings do not apply to
most countries in the sample (Figure 4.16 and Figure 4.18). Their results are skewed
toward Portugal and South Korea, which each contribute over 10% of the weight. Even
though both countries only represent 2% of the nominal sample, Portugal and South Ko-
rea represent 20% of the effective sample. The only country that borders South Korea is
North Korea, so while this country contributes the most weight to the effect estimate, it
cannot possibly represent the narrative the authors claim (i.e. their point is not that the
neighborhood effect of human rights NGOs is most impactful when North Korea is the
neighbor). Worth noting, they do not discuss South Korea or Portugal as important cases.
Instead, their introduction includes examples of volunteers from a human rights NGO se-
cretly crossing the Kenyan border into Sudan to aid the Southern Sudanese population as
well as volunteers crossing the Thai border into Cambodia. Yet neither Sudan nor Cam-
bodia are even featured in the top 20 most-weighted countries: Sudan contributes 0.23%
and Cambodia contributes 1.1% of the weight. A little over 8.7% of the observations (87
country-years) contribute over 50% of the weight. One single observation (Portugal in
2003) contributes over 7.5% of the weight alone. It takes 303 of the lowest contributing
observations to accumulate to 1% of the weight. The kurtosis of the distribution is 469,
orders of magnitude larger than the kurtosis of a normal distribution.
The research on the impact of human rights NGOs allegedly demonstrates that human
122
Figure 4.16: Multiple Regression Weights
(Represented as a Percentage of Total Weight) Used
to Construct the Effect Estimate from Bell, Clay, and
Murdie (2012) Model 3, Table 1
Figure 4.17: Bell, Clay, and Murdie (2012)
Nominal Sample
Figure 4.18: Bell, Clay, and Murdie (2012)
Effective Sample
123
rights NGOs can reduce government killing (DeMeritt, 2012), reduce the severity of mass
killings during genocides or politicides (Krain, 2012), and increase the chances of improv-
ing physical integrity rights in neighboring countries (Bell, Clay and Murdie, 2012). While
these optimistic researchers concede that human rights NGOs have a limited impact, a
deeper investigation of their models demonstrates that the positive reach of human rights
NGOs is an idea not supported by their own models. By evaluating the multiple regres-
sion weights from several models that use different measures of human rights NGO activ-
ity, this chapter demonstrates that the positive effect is primarily driven by a minuscule
subset of country-years. The countries that drive the effect are totally different in every
paper and follow no discernible pattern. Importantly, these countries are not concentrated
in Latin America where the transnational activist network canon predicted human rights
NGO advocacy would be particularly transformative (Keck and Sikkink, 1998). Also, none
of these countries were discussed by the authors to be the most likely cases of effective
NGO advocacy. The evidence presented here aligns much more closely with the critics of
international human rights practice (Posner, 2014; Moyn, 2018) who have pointed out
the insignificance of the international human rights regime. They may not have been pes-
simistic enough. Using the most cited papers touted by activists and scholars as the best
evidence of positive human rights impact, the best that can actually be said is that NGOs
generally accomplish nothing, but at least they do not cause that much more harm than
good.
124
Reporting Biases and the Resistance to Falsification
In some of its earlier formulations...their predictions were
testable, and in fact falsified. Yet instead of accepting the
refutations the followers of Marx reinterpreted both the
theory and the evidence in order to make them agree. In
this way they rescued the theory from refutation; but they
did so at the price of adopting a device which made it
irrefutable...and by this stratagem they destroyed its much
advertised claim to scientific status.
Karl Popper, "Science: Conjectures and Refutations"
Thus far, the discussion has concentrated on the few models that have reached sta-
tistical significance. However, this focus masks the larger backdrop of published non-
significant results in which the effect of human rights NGOs has failed to reach statistical
significance, even at the 10% error rate. Often, researchers fail to underscore these mod-
els, choosing instead to highlight the less common statistically significant positive results.
These models, totalling 35, are explored in more detail here (Table 4.4).
Hafner-Burton and Ron (2009) warned of publication bias in the human rights litera-
ture, citing the long-observed tendency for scientific disciplines to publish positive effects
more frequently than non-significant or negative effects. However, publication bias is only
one of several forms of reporting biases that can afflict a scientific literature. Outcome
reporting bias, spin bias, and publication bias can all contribute to the perceived evidence
base, as explained by Vries et al. (2018). In this section, I focus on outcome reporting bias
and spin bias because they can be readily measured and observed based on the existing ar-
ticles in the sample. Publication bias is difficult to observe because it requires one to have
knowledge of the universe of unpublished studies. While some researchers have found cre-
ative ways to overcome this challenge (O’Boyle, Banks and Gonzalez-Mulé, 2017), I focus
on the less well-understood and more subtle forms of reporting bias —- outcome reporting
bias and spin bias.
125
In this chapter, outcome reporting bias refers to the publication of a positive effect within
an article’s tables when the effect is in fact negative or non-significant. I discussed this form
of bias in the first section of this chapter. Outcome reporting bias is of particular concern
when reporting estimated conditional effects because the sign in front of the effect esti-
mate can appear positive even if the marginal effect is more often negative. Spin bias oc-
curs when authors do not faithfully discuss all non-significant or negative results. This can
happen by interpreting non-significant or negative results as consistent with the hypothe-
ses and broader theory or by downplaying the importance of the negative/non-significant
results. “If an article has been spun, treatments are perceived as more beneficial" (Vries
et al., 2018, 2453), even if they are technically reported accurately in tables and figures.
I measure the cumulative effect of outcome reporting bias and spin bias in the following
way. First, I measure the real effects — i.e. the output of the models when I replicate
them. For each model testing the conditional effect of human rights NGOs, I evaluate the
marginal effects and calculate whether a positive or negative effect is more common at the
95% confidence level. I used this method in the first section of this chapter. If the positive
effect is more common, I code the result as positive. Likewise, if the negative result is
more common, I code the result as negative. When the interaction term is not statistically
significant, I code it as a null result. For each of the models testing an unconditional effect,
I code the result as positive (or negative) if the effect estimate is statistically significant and
positive (or negative) at the 95% confidence level. Otherwise, I code it as a null result. I
label these 35 values as the “real effect."
From here, I measure the effect of outcome reporting bias. For each of the 35 models,
I code how the finding is reported in the regression table. If the finding is reported as
positive and statistically significant, I code it as positive. If the finding is reported as
negative and statistically significant, I code it as negative. And if the finding is reported
as failing to reach statistical significance at the 95% confidence level, I code it as a null
126
result.
12
I label these 35 values as the “reported effect."
After this step, I measure the effect of spin bias. For each of the 35 models, I code how
the reported effect is discussed in the paper. If a reported statistically significant negative
effect is discussed as a positive effect, I code it as positive. If a reported non-significant
result is discussed post hoc as consistent with the theory or hypothesis, I code it as spin.
If there is no discrepancy in how the result is discussed and how it is reported, I use the
same code that is given for the reported effect. I label these 35 values as the “discussed
effect."
I created a stacked bar chart to visualize how an evidence-landscape replete with real
null and negative effects transforms into an evidence-landscape entirely populated by posi-
tive effects and effects interpreted as consistent with Keck and Sikkink’s boomerang model
(Figure 4.19). The first bar represents the proportion of real effects that are positive, null,
and negative. The second bar represents the proportion of reported effects. And the third
bar represents the proportion of discussed effects that are positive, null, negative, and spun
in a positive way.
While 66% of the models produce results demonstrating that human rights NGOs are
either harmful or non-significant, only one of these models is honestly discussed as non-
significant (Franklin, 2008). With one exception, all of the real negative or non-significant
effects are discussed as either positive effects, or they are interpreted post hoc as consistent
with the boomerang/spiral model. One of the models that produces a negative effect is
even completely ignored by the author in the discussion of the results, which is indicated
by "NA" in Figure 4.19 (Franklin, 2008).
For an example of spin, Murdie and Davis (2012) test the robustness of their primary
models by including in the same model the interaction terms between human rights NGO
12
Some of the estimates reach statistical significance at the 90% confidence level but not the 95% level.
I code these results as null. I do this for several reasons. First, I want to maintain consistency. Second,
many journals today no longer accept results as significant if they are not significant at the 95% level. Third,
choosing a higher bar does not necessarily work against the authors. Sometimes the estimates would be
coded as negative if I used a lower threshold.
127
Figure 4.19: The Cumulative Impact of Reporting
Biases on the Effect of Human Rights NGOs
0
5
10
15
20
25
30
35
Real Effect Reported Effect Discussed Effect
Count
Positive Spin Null Negative NA
Cumulative Impact of Reporting Biases on the
Effect of Human Rights NGOs
advocacy and (1) the domestic presence of human rights organizations and (2) advocacy
by third-party actors such as governments and intergovernmental organizations. They no
longer find that their key result of interest is statistically significant using two different
measures of the dependent variable, even at the 90% confidence level.
13
While they admit
this, they spin the result as confirmatory of the importance of international human rights
NGOs: “We would argue that this result confirms the importance of continual interna-
tional attention in the later stages of the spiral model (Risse and Sikkink 1999). Without
the attention of these third parties, the ability of HROs to work to impact human rights
performance may be limited" (10). Heads I win, tails you lose. It would be difficult to
conjure a more direct attempt to protect a theory behind the armor of unfalsifiability.
For another example of spin, Bell, Clay, and Murdie test their hypothesis that human
rights NGOs improve human rights outcomes in neighboring countries by measuring hu-
13
As a reminder, their main finding of interest is that international human rights NGO advocacy is an
effective strategy when combined with a strong presence of local rights organizations within a country.
128
man rights NGO activity in three different ways. Only one of these measures is statistically
significant. Yet, they interpret the non-significant results as consistent with their hypoth-
esis. When discussing the discrepancy between the significant effect of domestic human
rights organizations versus the non-significant effect of human rights secretariat offices,
they claim, “these results make sense within our theoretical framework; while HRO sec-
retariats demonstrate that there is some HRO activity in a state, it says little about how
much activity is occurring and whether there are members to aid neighboring populations
as well as repressed populations in the home state. On the other hand, states with large
numbers of HROs with members in their borders can be expected to have enough mem-
bers to help both domestic and neighboring states’ repressed populations" (362). Perhaps
this is so, but one can imagine an alternative interpretation had the results been flipped.
For example, one could paint a plausible story that a permanent secretariat office is pro-
fessionally staffed and better resourced than the mere presence of volunteers or members,
making the presence of secretariats more important in shaping the human rights outcomes
in neighboring countries.
Franklin (2008) likewise engages in spin by interpreting statistically significant harm-
ful effects as positive evidence. Franklin finds that human rights NGO criticism on its own
(i.e. not conditional on another factor) leads to increased repression against protests in the
following months. “Does this mean that criticizing governments’ human rights practices
causes them to respond more harshly to protest? Several political leaders have responded
to human rights criticism verbally with angry defiance, but I doubt that authorities de-
liberately and systematically increase their repressive responses because of criticism. The
more plausible explanation is simply that human rights criticism tends to target the worst
abusers of human rights, which are likely to be repressive in the future" (201-202). While
reverse causation is very plausible, Franklin does not extend this same concern for the
models that produce results consistent with his hypothesis. He does not explain why read-
ers should trust the results from some of the models (the ones that say NGOs do good)
129
but not the others (that say NGOs do harm). This inconsistency is easily missed when one
has a conclusion before analyzing the data. Further, while reverse causation is plausible, it
is also plausible that governments do actually increase repression after being criticized by
NGOs. Basic economics says that as an action becomes costlier, people do less of it.
Every paper engages in either outcome reporting bias or spin bias, except for DeMeritt
(2012). However, this author was only able to avoid this by committing a prior "intel-
lectual crime."
14
While DeMeritt concludes that human rights NGO advocacy saves lives,
the original conference paper used an entirely different variable to measure NGO advo-
cacy, which yielded non-significant results in three of the four models (DeMeritt, 2010).
However, DeMeritt does not report these non-significant results in the published version
of the paper, nor is there any mention of the original variable. Yet the data for the original
variable is in the replication files and able to be run on the published model.
15
The tendency to emphasize rare positive findings and downplay or ignore the more
frequent negative or non-significant findings is common across the studies in the sample.
Table 4.4 summarizes the cumulative effect of outcome reporting bias and spin bias across
the 35 models. While the tables and data visualizations presented in the published papers
clearly show the existence of statistically significant harmful effects and numerous non-
significant results, the abstracts, introductions, and conclusions spin a very different story.
The authors interpret their results to fit their hypotheses, even if the results contradict their
theories, thus rescuing the transnational activist network model from refutation. While
numerous plausible stories can be crafted post hoc to make hypotheses and contradictory
evidence agree, the difficulty of conducting research that warrants scientific status is de-
veloping precise propositions ex ante and then discussing in an honest manner whether
the results are consistent with the propositions. By immunizing itself from refutation, this
body of research has sabotaged its claim to scientific status.
14
Lakatos (1998)
15
Readers can access DeMeritt’s archived paper from the 2010 annual International Studies As-
sociation website: https://convention2.allacademic.com/one/isa/isa10/index.php?&obf_var=4798439&
PHPSESSID=12at9sjmb2sqscj9igaa8b22ig.
130
Table 4.4: The Cumulative Impact of Reporting
Biases Across 35 Models Estimating the Effect of
Human Rights NGOs
Paper Index Independent Variable Dependent Variable Real Effect Reported Effect Discussed Effect
Murdie and Davis 2012 Model 2, Table 1 HRO Shaming x HRO Presence Physical integrity Negative Positive Positive
Murdie and Davis 2012 Model 1, Table 7 (Appendix) AI Shaming x HRO Presence Physical integrity Negative Positive Positive
Murdie and Davis 2012 Model 3, Table 1 HRO Shaming x Indirect Targeting Physical integrity Positive Positive Positive
Murdie and Davis 2012 Model 1, Table 8 (Appendix) AI Shaming x Indirect Targeting Physical integrity Negative Positive Positive
Murdie and Davis 2012 Model 5, Table 1 HRO Shaming x (HRO Presence + Indirect Targeting) Physical integrity Positive Positive Positive
Murdie and Davis 2012 Model 1, Table 10 (Appendix) AI Shaming x (HRO Presence + Indirect Targeting) Physical integrity Negative Positive Positive
Murdie and Davis 2012 Model 2, Table 2 HRO Shaming x HRO Presence Physical integrity improvement Null Positive Positive
Murdie and Davis 2012 Model 3, Table 2 HRO Shaming x Indirect Targeting Physical integrity improvement Positive Positive Positive
Murdie and Davis 2012 Model 5, Table 2 HRO Shaming x (HRO Presence + Indirect Targeting) Physical integrity improvement Positive Positive Positive
Murdie and Davis 2012 Model 4, Table 1 HRO Shaming x HRO Presence Physical integrity Null Null Spin
Murdie and Davis 2012 Model 4, Table 1 HRO Shaming x Indirect Targeting Physical integrity Null Positive Positive
Murdie and Davis 2012 Model 1, Table 9 (Appendix) AI Shaming x HRO Presence Physical integrity Null Positive Positive
Murdie and Davis 2012 Model 1, Table 9 (Appendix) AI Shaming x Indirect Targeting Physical integrity Null Null Positive
Murdie and Davis 2012 Model 4, Table 2 HRO Shaming x HRO Presence Physical integrity improvement Null Null Spin
Murdie and Davis 2012 Model 4, Table 2 HRO Shaming x Indirect Targeting Physical integrity improvement Null Positive Positive
Franklin 2008 Model 2, Table 5 NGO criticism 1 month prior Magnitude of political repression of contentious challenges Negative Negative Spin
Franklin 2008 Model 2, Table 7 NGO criticism 6 months prior Magnitude of political repression of contentious challenges Negative Negative Not discussed
Franklin 2008 Model 2, Table 5 NGO criticism 1 month prior x Foreign capital dependence Magnitude of political repression of contentious challenges Negative Positive Positive
Franklin 2008 Model 2, Table 7 NGO criticism 6 months prior x Foreign capital dependence Magnitude of political repression of contentious challenges Negative Null Null
Krain 2012 Model 2, Table 3 AI background reports Magnitude of genocide severity (0-10) Positive Positive Positive
Krain 2012 Model 3, Table 3 AI press releases Magnitude of genocide severity (0-10) Null Null Spin
DeMeritt 2012 Model 1, Table 2 HRO Shaming Government killing (binary) Positive Positive Positive
DeMeritt 2012 Model 4, Table 2 HRO Shaming Government killing (natural log) Positive Positive Positive
Bell, Clay, and Murdie 2012 Model 1, Table 1 Neighborhood HRO shaming Physical integrity improvement Null Null Spin
Bell, Clay, and Murdie 2012 Model 2, Table 1 Neighborhood HRO Secretariat Physical integrity improvement Null Null Spin
Bell, Clay, and Murdie 2012 Model 3, Table 1 Neighborhood HRO Membership Physical integrity improvement Positive Positive Positive
Bell, Clay, and Murdie 2012 Model 1, Table 2 Neighborhood HRO shaming x Foreign movement Physical integrity improvement Null Null Spin
Bell, Clay, and Murdie 2012 Model 2, Table 2 Neighborhood HRO Secretariat x Foreign movement Physical integrity improvement Null Null Spin
Bell, Clay, and Murdie 2012 Model 3, Table 2 Neighborhood HRO Membership x Foreign movement Physical integrity improvement Null Positive Positive
Bell, Clay, and Murdie 2012 Model 1, Table A23 Neighborhood HRO shaming Physical integrity Positive Positive Positive
Bell, Clay, and Murdie 2012 Model 2, Table A23 Neighborhood HRO Secretariat Physical integrity Positive Positive Positive
Bell, Clay, and Murdie 2012 Model 3, Table A23 Neighborhood HRO Membership Physical integrity Positive Positive Positive
Bell, Clay, and Murdie 2012 Model 1, Table A24 Neighborhood HRO shaming x Foreign movement Physical integrity Null Null Spin
Bell, Clay, and Murdie 2012 Model 2, Table A24 Neighborhood HRO Secretariat x Foreign movement Physical integrity Positive Positive Positive
Bell, Clay, and Murdie 2012 Model 3, Table A24 Neighborhood HRO Membership x Foreign movement Physical integrity Null Null Spin
131
Conclusion
The human rights literature suffers from a “reign of error"
16
in which unjustified optimistic
conclusions about the effectiveness of human rights organizations have become canon-
ized. For example, Murdie and Davis (2012, 13) conclude: “Does shaming by human
rights international non-governmental actors matter? This research provides a resound-
ing, albeit qualified, ‘yes.’" However, based on a full accounting of the statistical models
(including their own) presented in this chapter, a more faithful conclusion is that human
rights NGOs are probably not effective. When they do have a statistically significant effect,
they are harmful more often than they are beneficial. Even among the models outputting
statistically significant positive estimates, the effects are confined to a very narrow set
of countries. And often, these positive effects only appear when the authors have taken
multiple bites of the apple by trying different measures of the independent and dependent
variables. When the statistical results indicate a non-significant effect — or even a negative
effect — the authors consistently spin the results by interpreting them as demonstrating
the power and importance of continuous pressure from human rights NGOs. They are not
willing or able to recognize contradictory evidence when it stares them in the face.
The statistical evidence for the effect of human rights NGOs is consistent with the
Virtue Economy. If donors do not create market demand for a scientific and experimental
approach to advancing human rights, then we should not expect human rights NGOs to
supply programs that are particularly beneficial. Impactful programs are rare even among
the teams of organizations and researchers that do rigorously analyze their effectiveness
(Todd, 2017). Most of the time, human rights NGOs waste their resources on popular but
ineffective efforts. Donors financially reward these gross inefficiencies, sustaining a cycle
of stable waste.
Moving forward, human rights researchers must re-think the assumptions made about
NGOs and their impact. Aggregating across hundreds of organizations is not a valid
16
Jussim (2020)
132
method to detect NGO impact. If researchers are interested in learning which NGOs,
interventions, strategies, and programs are the most effective, they need to turn to the
micro-level by conducting rigorous impact evaluations at the project level to find the rare
but exceptionally effective programs. However, acquiring genuine knowledge of such pro-
grams requires a fidelity to the evidence, even if the evidence disconfirms the researchers’
strongly held notions of the kinds of programs that ought to advance human rights.
133
Chapter 5
Conclusion
Future Research
No book can exhaust every potentially fruitful avenue or cover every argument of relevance
to the book’s main thesis. This book lays the foundation for introducing the signaling
theory of charity to the study of human rights NGOs. I hope that this work will inspire
other researchers to explore implications from the signaling theory presented here. In
particular, I view the psychology of effective altruism as the most important area in need
of future research. Specifically, how can we leverage a better understanding of signaling
to motivate individuals to donate to more effective charitable organizations (and thereby
motivate organizations to become more effective)?
There are many other directions one could take to better understand how donor signal-
ing influences NGOs. Other topics implicated by this framework that I have encountered
throughout my dissertation research include NGO accountability monitors, human rights
research methodology, and cause prioritization.
NGO accountability monitors: who are their customers? What kind of information
do their customers demand? How do accountability monitors aid or stymie social
signals?
Human rights research methodology: is accurate reporting an important concern for
most donors? If activists operate in a virtue economy, what kind of reporting biases
would we expect?
1
1
See Mulesky 2019 for an application of signaling theory to explain the low standard of evidence in
134
Cause prioritization: why do human rights NGOs prioritize certain human rights
issues or victims and neglect others? This topic could be better informed by analyzing
what donating to specific philanthropic sectors, cause areas, recipient types, and
geographic areas signals about the individual donor. Research along this vein would
benefit from the literature in marketing science and brand research.
2
The field could also benefit from comparative research, examining the differences be-
tween NGOs that are funded primarily by private money (individuals, foundations, and
corporations) and NGOs that are funded primarily by public money (governments). While
existing research suggests that even governments may be primarily motivated by repu-
tational concerns when providing foreign aid (Gilady, 2018), governments face a very
different set of incentives than NGOs because they receive funding from supporters and
critics alike, whereas NGOs receive funding only from supporters. The funding structure
for private philanthropy naturally limits the channels of criticism and accountability via
market segmentation, which does not occur for governments. Understanding how NGO
accountability and effectiveness compares among those that rely on public funding from
those that rely on private funding could provide valuable insights into developing a better
understanding of the comparative effectiveness of different funding models.
Implications for the Human Rights Movement
Signaling is an important obstacle to the effective advancement of human rights. The in-
formation that is relevant for doing good — charity effectiveness — is not socially useful
knowledge in a signaling system given the current distribution of donor views. Too much
concern with outcomes can itself be taken as a signal of undesirable traits. The dissociation
between receiving social credit for doing good, on the one hand, and actually doing good,
on the other hand, generates an important suboptimality in human rights philanthropy:
human rights reporting. https://politicalviolenceataglance.org/2019/05/29/fact-checkers-exposed-trump-
but-also-exposed-human-rights-reporting/.
2
See Mulesky 2019 for an application of signaling theory to explain why Amnesty International cam-
paigned on behalf of female Twitter users but not male users. https://quillette.com/2019/01/31/the-virtue-
economy/.
135
human rights NGOs have a financial disincentive to rigorously evaluate their impact. This
means that the very organizations leading the human rights movement are themselves
handicapped by signaling dynamics. It is no exaggeration to say that hundreds of mil-
lions of dollars a year are spent without much attention paid to whether that money was
spent effectively advancing human rights (Posner, 2015). This is a profoundly puzzling
phenomenon when only analyzed from the lens of existing theoretical accounts. How-
ever, when we understand that the stated goals of organizations and their donors are not
the primary goals motivating behavior, then behavior that otherwise appears irrational is
understandable, explainable, and rational. Human rights NGOs are simply responding to
financial incentives, and donors are simply responding to social incentives.
Some complain that donations have been stuck at 2% of GDP for the last four decades,
and most research on philanthropic giving is concerned with discovering how to get more
people to donate more money (Camber Collective, 2015). If donations doubled to 4%
of GDP, we could accomplish twice as much good, they say. But existing research demon-
strates that there is an extreme level of variation in cost-effectiveness across different char-
ities and different types of interventions. This means that most of the value to be gained,
in terms of improving human welfare outcomes, is not by convincing more people to do-
nate more money but rather by convincing existing donors to re-allocate their donations to
more cost-effective charities (MacAskill, 2015; Ord, 2013). If the most cost-effective char-
ities are 100 times more cost-effective than average charities, as development and global
health experts believe, then figuring out how to nudge donations to more cost-effective
charities would generate far more good than convincing people to donate more money to
their preferred charities. There are severe opportunity costs when donors give ineffectively.
So long as donors give as they currently do, human rights NGOs have little incentive to
evaluate the impact of their programs or use impact information to invest in more effective
programs. Highly cost-effective programs exist, and human rights NGOs could invest in
them directly or create new advocacy opportunities to amplify the popularity of the most
136
effective programs. Additionally, human rights NGOs could themselves contribute to this
body of knowledge by running randomized evaluations and sharing the results. The good
news is that there is high upside upon solving this problem: human rights NGOs have
the potential to do orders of magnitude more good, and they do not need any additional
money to do it.
Convincing human rights NGOs to change their current practices requires changing
norms, values, and beliefs among existing donors, or at least making it less socially re-
warding (and more embarrassing) to donate to ineffective causes. If donors demanded
that human rights NGOs be held to high standards of cost-effectiveness, the practice of hu-
man rights would look very different. Human rights NGOs would compete with each other
to prove they were the most cost-effective organization. Such competition would produce
innovation in human rights practice: (1) more knowledge would be gained about how
to effectively advance human rights, and (2) new, innovative solutions would be learned
because human rights NGOs would have the incentive to challenge their pre-existing prac-
tices. A better understanding of social incentives and the social signaling dynamics at play
among members of human rights communities could lead to the development of nudges
that incentivize donors to care about cost-effectiveness and apportion their donations ac-
cordingly, if only to impress their peers. The key puzzle for future research is to figure out
how to make knowledge about program effectiveness socially useful.
Final Word
If there is a mountain of empirically powerful research coming from a myriad of fields that
directly undermines present theories of human rights organizational behavior, how has it
failed to enter into published human rights scholarship, until now? I cannot be absolutely
sure, but I am willing to submit a guess: Human rights researchers fell in love with human
rights organizations before they started studying them. When introduced to theories that
137
described the actors of the human rights movement as purely altruistic, highly effective,
heroic fighters for the universal good of humanity, we were instant converts.
3
To humanize
activists — to truly humanize them — is to recognize that they also operate in accordance
with human nature. In so doing, our picture of human rights, donors, and organizational
behavior will become much more accurate, but the romanticism may fade.
Human rights NGOs do not try to maximize the welfare outcomes of their recipients
or ensure that their programs have a positive impact because they have next to no finan-
cial incentive for rigorously evaluating the effectiveness of their programs. While they
routinely claim that they make a large, positive impact, they rarely, if ever, provide any
rigorous evidence of their effectiveness. Rather than being held accountable for such be-
havior, they are trusted by numerous actors and rewarded with donations, status, prestige,
and authority. As my survey experimental evidence presented in this book demonstrates,
rigorous impact evaluations would serve only to undermine the organization’s financial
health. Claiming to make a positive impact is low-cost for both NGOs and donors because
they receive social credit as if they accomplish an enormous amount of good without hav-
ing to pay any of the costs of producing rigorous evidence of their impact. Why should
organizations sitting at the top of the international human rights hierarchy jeopardize their
status and financial health by holding themselves to a standard that nobody, not even their
donors or even academics studying human rights, demand?
Donors, likewise, do not try to maximize welfare outcomes or ensure their donations
have a positive impact because they have next to no social incentive for scrutinizing char-
ity effectiveness when making charitable giving decisions. Private information about a
charity’s impact is not socially useful knowledge, at least not in most social communi-
ties where people believe that most charities accomplish what they say they do. Further,
adopting certain deliberative, scrupulous approaches to charitable giving risks making the
donor appear cold and calculating, lacking key traits that people covet such as empathy
3
See Caplan (2018) for a similar discussion with regard to signaling, labor economics, and education.
138
and a warm-regard for individuals. By adopting a more scientific approach to their charity,
donors may appear less trustworthy, even if their approach produces better consequences
in terms of welfare outcomes for recipients.
Signaling is the blind spot that both proponents and critics of human rights NGOs have
missed. They have missed the social subtext of donating, and in so doing, have taken
the stated goals of human rights NGOs and their donors too literally. The proponents
are correct insofar as they portray NGOs and donors working collectively to pursue their
shared goals. However, in a virtue economy, they work together to pursue their self-
interest. Donors invest in their reputation by donating to human rights NGOs, paying for
the ability to show their social communities that they are prosocial, open, multi-cultural,
compassionate, empathic, and politically liberal. The human rights NGOs, for doing their
part to respond to market demand, receive money, organizational status, and authority.
The critics are correct insofar as they bring attention to the human rights movement’s
inefficacy (Hopgood, 2013; Posner, 2014, 2015) or NGO self-interest (Cooley and Ron,
2002; Prakash and Gugerty, 2010; Stroup and Wong, 2017). However, the critics that focus
on NGO self-interest miss that donors are also fundamentally motivated by reputational
concerns. The critics that paint the status of human progress as worsening over time
are empirically incorrect in their pessimistic outlook on general trends (Cingranelli and
Filippov, 2018). Respect for human rights has improved over the last several decades
(Fariss, 2014, 2019) for nearly every metric that is relevant to human welfare (Pinker,
2018; Rosling, Rönnlund and Rosling, 2018; Sikkink, 2017).
The debate on human rights efficacy should not be focused on general trends in hu-
man progress, which could be influenced by a myriad of factors unrelated to human rights
activism, such as economic growth, democratization, globalization, technological advance-
ment, and improvement in medicine and public health. Instead, the focus should be on
how relevant efficacy is to NGO decision-making. Human rights NGOs make decisions
daily about which programs they will fund, where they will operate, and what they will
139
do. If efficacy is not a core component of their work, then the debate about the challenges
to the effective advancement of human rights will be misguided. A key challenge to ad-
vancing human rights is that the very organizations at the vanguard of the human rights
movement are financially incentivized to cater to their donors’ reputational needs, not to
ensure their programs have a positive impact. In the naïve view of the principled actor
approach, human rights NGOs are effective because they do what donors want. In the
view of the collective action critics, NGOs are not effective because they do not do what
donors want. In my view, NGOs are not effective because they do what donors want.
140
Bibliography
Absar, Kassira, David Crow, James Ron, Gerardo Maldonado, Juan Pablo Bolaños, José
Kaire and Andrea Martinez. 2017. “Will Publics Pay to Protect Rights? An Experimental
Study of Mexico City Inhabitants’ Willingness to Donate to Local Human Rights Orga-
nizations and of These Groups’ Ability to Use This Data.” A Report of the Human Rights
Organizations Project at the University of Minnesota’s Humphrey School of Public Affairs .
URL: https://www.openglobalrights.org/userfiles/file/MexicoCity2016_Report_final3.pdf
Amnesty International. 2018. “2017 Global Financial Report.”.
URL: https://www.amnesty.org/en/2017-global-financial-report/
Amnesty International. 2019. “2018 Global Financial Report.”.
URL: https://www.amnesty.org/en/2018-global-financial-report/
Andreoni, James. 1990. “Impure Altruism and Donations to Public Goods: A Theory of
Warm-Glow Giving.” The Economic Journal 100(401):464–477.
Andreoni, James and Ragan Petrie. 2004. “Public Goods Experiments without Confiden-
tiality: A Glimpse Into Fund-Raising.” Journal of Public Economics 88(7):1605–1623.
Arnocky, Steven, Tina Piché, Graham Albert, Danielle Ouellette and Pat Barclay. 2017.
“Altruism Predicts Mating Success in Humans.” British Journal of Psychology 108(2):416–
435.
Aronow, Peter M. and Cyrus Samii. 2016. “Does Regression Produce Representative Esti-
mates of Causal Effects?” American Journal of Political Science 60(1):250–267.
Banerjee, Abhijit and Esther Duflo. 2012. Poor Economics: A Radical Rethinking of the Way
to Fight Global Poverty. New York: PublicAffairs.
Barakso, Maryann. 2010. Brand Identity and the Tactical Repertoires of Advocacy Or-
ganizations. In Advocacy Organizations and Collective Action, ed. Aseem Prakash and
Mary Kay Gugerty. Cambridge University Press pp. 155–176.
Barclay, Pat. 2004. “Trustworthiness and Competitive Altruism Can Also Solve the ’Tragedy
of the Commons’.” Evolution and Human Behavior 25:209–220.
Barclay, Pat. 2010. “Altruism as a Courtship Display: Some Effects of Third-party Generos-
ity on Audience Perceptions.” British Journal of Psychology 101:123–135.
141
Barclay, Pat and Robb Willer. 2007. “Partner Choice Creates Competitive Altruism in Hu-
mans.” Proceedings of the Royal Society of London, Series B 274:749–753.
Baron, Jonathan and Ewa Szymanska. 2011. Heuristics and Biases in Charity. In The Science
of Giving: Experimental Approaches to the Study of Charity, ed. Daniel M. Oppenheimer
and Christopher Y. Olivola. New York: Psychology Press pp. 215–235.
Bazelon, Emily. 2016. “Should Prostitution Be a Crime?” The New York Times .
URL: https://www.nytimes.com/2016/05/08/magazine/should-prostitution-be-a-crime.
html
Bekkers, Rene. 2005. “It’s Not All in the Ask. Effects and Effectiveness of Recruitment
Strategies Used by Nonprofits in the Netherlands.” Nonprofit and Voluntary Sector Quar-
terly 40(5):924–973.
Bell, Sam R., K. Chad Clay and Amanda Murdie. 2012. “Neighborhood Watch: Spatial
Effects of Human Rights INGOs.” Journal of Politics 74(2):354–368.
Berinsky, Adam J., Gregory A. Huber and Gabriel S. Lenz. 2012. “Evaluating Online Labor
Markets for Experimental Research: Amazon.com’s Mechanical Turk.” Political Analysis
20(3):351–368.
Berman, Jonathan Z., Alixandra Barasch, Emma E. Levine and Deborah A. Small. 2018.
“Impediments to Effective Altruism: The Role of Subjective Preferences in Charitable
Giving.” Psychological Science 29(5):834–844.
Blattman, Christopher, Alexandra C. Hartman and Robert A. Blair. 2014. “How to Promote
Order and Property Rights under Weak Rule of Law? An Experiment in Changing Dispute
Resolution Behavior through Community Education.” American Political Science Review
108(1):100–120.
Blattman, Christopher, Julian C. Jamison and Margaret Sheridan. 2017. “Reducing Crime
and Violence: Experimental Evidence from Cognitive Behavioral Therapy in Liberia.”
American Economic Review 107(4):1165–1206.
Bob, Clifford. 2010. The Market for Human Rights. In Advocacy Organizations and Collec-
tive Action, ed. Aseem Prakash and Mary Kay Gugerty. Cambridge University Press.
Boone, J. L. 1998. “The Evolution of Magnanimity: When Is It Better to Give Than to
Receive?” Human Nature 9(1):1–21.
Brooks, David. 2013. “The Way to Produce a Person.” The New York Times .
URL: https://www.nytimes.com/2013/06/04/opinion/brooks-the-way-to-produce-a-
person.html
Bryant, W. Keith, Haekyung Jeon-Slaughter, Hyojin Kang and Aaron Tax. 2003. “Participa-
tion in Philanthropic Activities: Donating Money and Time.” Journal of Consumer Policy
26(1):43–73.
142
Bull, Ray and Elizabeth Gibson-Robinson. 1981. “The Influences of Eye-Gaze, Style of
Dress, and Locality on the Amounts of Money Donated to Charity.” Human Relations
34(10):895–905.
Burum, Bethany, Martin A. Nowak and Moshe Hoffman. 2020. “An Evolutionary Explana-
tion for Ineffective Altruism.” Nature Human Behaviour 4:1245–1257.
Bush, Sarah Sunn and Jennifer Hadden. 2019. “Density and Decline in the Founding of
International NGOs in the United States.” International Studies Quarterly 63:1133–1146.
Buss, David M., Max Abbott, Alois Angleitner, Armen Asherian, Angela Biaggio, Angel
Blanco-Villasenor, M. Bruchon-Schweitzer, Hai-Yuan Ch’U, Janusz Czapinski, Boele De-
raad, Bo Ekehammar, Noha El Lohamy, Mario Fioravanti, James Georgas, Per Gjerde,
Ruth Guttman, Fatima Hazan, Saburo Iwawaki, N. Janakiramaiah, Fatemeh Khos-
roshani, Shulamith Kreitler, Lance Lachenicht, Margaret Lee, Kadi Liik, Brian Little,
Stanislaw Mika, Mariam Moadel-Shahid, Geraldine Moane, Maritza Montero, A. C.
Mundy-Castle, Toomas Niit, Evaristo Nsenduluka, Ryszard Pienkowski, Anna-Maija
Pirttilä-Backman, Julio Ponce De Leon, Jacques Rousseau, Mark A. Runco, Marilyn P.
Safir, Curtis Samuels, Rasyid Sanitioso, Robert Serpell, Nico Smid, Christopher Spencer,
Meri Tadinac, Elka N. Todorova, Kari Troland, L. Van Den Brande, Guus Van Heck, L.
Van Langenhove and Kuo-Shu Yang. 1990. “International Preferences in Selecting Mates:
A Study of 37 Cultures.” Journal of Cross-Cultural Psychology 21(1):5–47.
Callahan, David. 2017. “In a New Journal, Another Sign That Philanthropy Scholarship Is
Gaining Steam.” Inside Philanthropy .
URL: https://www.insidephilanthropy.com/home/2017/11/8/in-a-new-journal-more-
evidence-that-philanthropy-scholarship-isnt-a-lost-cause-after-all
Camber Collective. 2015. Money for Good 2015: Revealing the Voice of the Donor in
Philanthropic Giving. Technical report.
URL: http://static1.squarespace.com/static/55723b6be4b05ed81f077108/t/
56957ee6df40f330ae018b81/1452637938035/\protect\T1\textdollarFG+2015_Final+
Report_01122016.pdf
Caplan, Bryan. 2018. The Case Against Education: Why the Education System Is a Waste of
Time and Money. Princeton University Press.
Cardenas, Sonia. 2007. Conflict and Compliance: Responses to International Human Rights
Pressure. University of Pennsylvania Press.
Carpenter, Charli. 2014. ‘Lost’ Causes: Agenda Vetting in Global Issue Networks and the
Shaping of Human Security. Ithaca, NY: Cornell University Press.
Caviola, Lucius, Nadira Faulmüller, Jim. A. C. Everett, Julian Savulescu and Guy Kahane.
2014. “The Evaluability Bias in Charitable Giving: Saving Administration Costs or Saving
Lives?” Judgment and Decision Making 9(4):303–316.
143
Caviola, Lucius, Stefan Schubert, Elliot Teperman, David Moss and Nadira S. Faber. 2020.
“Donors Vastly Underestimate Differences in Charities’ Effectiveness.” Judgment and De-
cision Making 15(4).
Caviola, Lucius, Stefan Schubert and Jason Nemirow. 2020. “The Many Obstacles to Effec-
tive Giving.” Judgment and Decision Making 15(2).
Christiano, Ann and Annie Neimand. 2017. “Stop Raising Awareness Already.” Stanford
Social Innovation Review .
URL: https://ssir.org/articles/entry/stop_raising_awareness_already
Cingranelli, David and Mikhail Filippov. 2018. “Are Human Rights Practices Improving?”
American Political Science Review 112(4):1083–1089.
Cohen, Patricia. 2015. “Art Collectors Gain Tax Benefits From Private Museums.” The New
York Times .
URL: https://www.nytimes.com/2015/01/11/business/art-collectors-gain-tax-benefits-
from-private-museums.html
Conrad, Courtenay R. and Emily Hencken Ritter. 2019. Contentious Compliance: Dissent
and Repression under International Human Rights Law. Oxford University Press.
Cooley, Alexander and James Ron. 2002. “The NGO Scramble: Organizational Insecurity
and the Political Economy of Transnational Action.” International Security 27(1):3–39.
Curry, Oliver Scott. 2008. The Conflict-Resolution Theory of Virtue. In Moral Pyschology:
The Evolution of Morality: Adaptations and Innateness, ed. Walter Sinnott-Armstrong.
Cambridge, MA: MIT Press pp. 251–261.
DeMeritt, Jacqueline H. R. 2012. “International Organizations and Government Killing:
Does Naming and Shaming Save Lives?” International Interactions 38(5):597–621.
DeMeritt, Jacqueline H.R. 2010. “International Organizations and Government Killing:
Does Naming and Shaming Save Lives?” Working Paper Presented at the 2010 Interna-
tional Studies Association Annual Convention .
Dullaghan, Neil. 2019. “EA Survey 2019 Series: Community Demographics & Character-
istics.”.
URL: https://forum.effectivealtruism.org/posts/wtQ3XCL35uxjXpwjE/ea-survey-2019-
series-community-demographics-and
Dullaghan, Neil. 2020. “EA Survey 2019 Series: Geographic Distribution of EAs.”.
URL: https://forum.effectivealtruism.org/posts/cvkqyxepf4W2whYSK/ea-survey-2019-
series-geographic-distribution-of-eas
Dupas, Rascaline. 2011. “Do Teenagers Respond to HIV Risk Information? Evidence from
a Field Experiment in Kenya.” American Economic Journal: Applied Economics 3:1–34.
144
Easterly, William and Tobias Pfutze. 2008. “Where Does the Money Go? Best and Worst
Practices in Foreign Aid.” Journal of Economic Perspectives 22(2):29–52.
Evaluation, MEASURE. 2019. “FAQ: How Much Will an Evaluation Cost?”.
URL: https://www.measureevaluation.org/resources/publications/fs-15-156
Everett, Jim A.C., Nadira S. Faber, Julian Savulescu and Molly J. Crockett. 2018. “The
Costs of Being Consequentialist: Social Inference from Instrumental Harm and Impartial
Beneficence.” Journal of Experimental Social Psychology 79:200–216.
Fariss, Christopher J. 2014. “Respect for Human Rights has Improved Over Time: Modeling
the Changing Standard of Accountability.” American Political Science Review 108(2):297–
318.
Fariss, Christopher J. 2019. “Yes, Human Rights Practices Are Improving Over Time.”
American Political Science Review 113(3):868–881.
Field, Erica, Rachel Glennerster, Nina Buchmann and Kyle Murphy. 2016. Cost-Benefit
Analysis of Strategies to Reduce Child Marriage in Bangladesh. Technical report Copen-
hagen Consensus Center.
URL: https://www.copenhagenconsensus.com/sites/default/files/field_child_marriage.pdf
Franklin, James C. 2008. “Shame on You: The Impact of Human Rights Criticism on
Political Repression in Latin America.” International Studies Quarterly 52(1):187–211.
Gilady, Lilach. 2018. The Price of Prestige: Conspicuous Consumption in International Rela-
tions. University of Chicago Press.
Glazer, Amihai and Kai A. Konrad. 1996. “A Signaling Explanation for Charity.” The Ameri-
can Economic Review 86(4):1019–1028.
Goodman, Joseph K., Cynthia E. Cryder and Amar Cheema. 2013. “Data Collection in a
Flat World: The Strengths and Weaknesses of Mechanical Turk Samples: Data Collection
in a Flat World.” Journal of Behavioral Decision Making 26(3):213–224.
Gorvin, Ian. 2009. “Producing the Evidence that Human Rights Advocacy Works: First
Steps towards Systematized Evaluation at Human Rights Watch.” Journal of Human
Rights Practice 1(3):477–487.
Grace, Debra and Deborah Griffin. 2006. “Exploring Conspicuousness in the Context of
Donation Behaviour.” International Journal of Nonprofit and Voluntary Sector Marketing
11(2):147–154.
Greenberg, Spencer. 2017. Social Science as Lens on Effective Charity: Results From Four
New Studies. San Francisco, CA: .
URL: https://www.eaglobal.org/talks/social-science-as-lens-on-effective-charity-results-
from-four-new-studies/
145
Gregory, Ann Goggins and Don Howard. 2009. “The Nonprofit Starvation Cycle.” Stanford
Social Innovation Review .
URL: https://ssir.org/articles/entry/the_nonprofit_starvation_cycle
Griskevicius, Vladas, Joshua M. Tybur, Jill M. Sundie, Robert B. Cialdini, Geoffrey F. Miller
and Douglas T. Kenrick. 2007. “Blatant Benevolence and Conspicuous Consumption:
When Romantic Motives Elicit Strategic Costly Signals.” Journal of Personality and Social
Psychology 93(1):85–102.
Hafner-Burton, Emilie M. 2008. “Sticks and Stones: Naming and Shaming the Human
Rights Enforcement Problem.” International Organization 62(4):689–716.
Hafner-Burton, Emilie M. and James Ron. 2009. “Seeing Double: Human Rights Impact
through Qualtiative and Quantitative Eyes.” World Politics 61(2):360–401.
Haley, Kevin J. and Daniel M. Fessler. 2005. “Nobody’s Watching?: Subtle Cues Affect Gen-
erosity in an Anonymous Economic Game.” Evolution and Human Behavior 26(3):245–
256.
Harbaugh, William T. 1998. “What Do Donations Buy?: A Model of Philanthropy Based
on Prestige and Warm Glow.” Journal of Public Economics 67(2):269–284.
Hardy, C. L. and M. Van Vugt. 2006. “Nice Guys Finish First: The Competitive Altruism
Hypothesis.” Personality and Social Psychology Bulletin 32:1402–1413.
Hassay, Derek and John Peloza. 2009. “Building the Charity Brand Community.” Journal
of Nonprofit & Public Sector Marketing 21(1):24–55.
Hauser, David J. and Norbert Schwarz. 2016. “Attentive Turkers: MTurk Participants Per-
form Better on Online Attention Checks than Do Subject Pool Participants.” Behavior
Research Methods 48(1):400–407.
Hendrix, Cullen S. and Wendy H. Wong. 2012. “When Is the Pen Truly Mighty? Regime
Type and the Efficacy of Naming and Shaming in Curbing Human Rights Abuses.” British
Journal of Political Science 43:651–672.
Herrmann, Esther, Jan M. Engelmann and Michael Tomasello. 2019. “Children Engage in
Competitive Altruism.” Journal of Experimental Child Psychology 179:176–189.
Hill, Daniel W. and Zachary M. Jones. 2014. “An Empirical Evaluation of Explanations for
State Repression.” American Political Science Review 108(3):661–687.
Hoffman, Moshe, Christian Hilbe and Martin A. Nowak. 2018. “The Signal-Burying Game
Can Explain Why We Obscure Positive Traits and Good Deeds.” Nature Human Behaviour
2(6):397.
Hope Consulting. 2010. Money for Good: The US Market for Impact Investments and
Charitable Gifts from Individual Donors and Investors. Technical report.
URL: https://assets.aspeninstitute.org/content/uploads/2012/04/ANDE_
MFGSummaryNote_15AUG10.pdf
146
Hope Consulting. 2011. Money for Good II: Driving Dollars to the Highest-Performing
Nonprofits: Fact Base and Research Report Full Version. Technical report.
URL: https://www.guidestar.org/ViewCmsFile.aspx?ContentID=4038
Hopgood, Stephen. 2006. Keepers of the Flame: Understanding Amnesty International.
Ithaca, NY: Cornell University Press.
Hopgood, Stephen. 2013. The Endtimes of Human Rights. Ithaca, NY: Cornell University
Press.
Jackson, Jeffrey M. and Bibb Latané. 1981. “Strength and Number of Solicitors and the
Urge toward Altruism.” Personality and Social Psychology Bulletin 7(3):415–422.
Jussim, Lee. 2020. “How to Create Scientific Myths Without Really Trying.”.
URL: https://www.psychologytoday.com/blog/rabble-rouser/202001/how-create-
scientific-myths-without-really-trying
Karing, Anne. 2018. “Social Signaling and Childhood Immunization: A Field Experiment
in Sierra Leone.” Working Paper .
URL: https://economics.yale.edu/sites/default/files/jmp_socialsignaling.pdf
Karing, Anne and Karim Naguib. 2018. “Social Signaling and Prosocial Behavior: Experi-
mental Evidence in Community Deworming in Kenya.” Working Paper .
URL: https://drive.google.com/file/d/164wP0_dqOFfKvUenHVixlY3Js7oFPFhg/view
Karlan, Dean and Daniel H. Wood. 2017. “The Effect of Effectiveness: Donor Response
to Aid Effectiveness in a Direct Mail Fundraising Experiment.” Journal of Behavioral and
Experimental Economics 66:1–8.
Keck, Margaret E. and Kathryn Sikkink. 1998. Activists Beyond Borders: Advocacy Networks
in International Politics. Cornell University Press.
Kelsey, Caroline, Amrisha Vaish and Tobias Grossmann. 2018. “Eyes, More Than Other
Facial Features, Enhance Real-World Donation Behavior.” Human Nature 29(4):390–
401.
Kogut, Tehila and Ilana Ritov. 2011. The Identifiable Victim Effect: Causes and Boundary
Conditions. In The Science of Giving: Experimental Approaches to the Study of Charity, ed.
Daniel M. Oppenheimer and Christopher Y. Olivola. Taylor & Francis.
Krain, Matthew. 2012. “ J’accuse! Does Naming and Shaming Perpetrators Reduce the
Severity of Genocides or Politicides?” International Studies Quarterly 56(3):574–589.
Kremer, Michael and Edward Miguel. 2004. “Worms: Identifying Impacts on Education
and Health in the Presence of Treatment Externalities.” Econometrica 72(1):159–217.
Lakatos, Imre. 1998. Science and Pseudoscience. In Philosophy of Science: The Central
Issues, ed. Martin Curd and J. A. Cover. W. W. Norton & Company.
147
Lee, Minwoo, Sunhae Sul and Hackjin Kim. 2018. “Social Observation Increases Deonto-
logical Judgments in Moral Dilemmas.” Evolution and Human Behavior 39(6):611–621.
Leigh, Andrew. 2018. Randomistas: How Radical Researchers Are Changing Our World. New
Haven, CT: Yale University Press.
Levy, Karen, Julie Wang’ombe and Diksha Radhakrishnan. 2018. “Ambiguous Results and
Clear Decision-making: A Sugar-Daddy Awareness Program Evaluated in Botswana Will
Not Be Scaled Up.”.
URL: https://www.evidenceaction.org/no-sugar-not-scaling/
Lofgren, Emma. 2015. “Swedes Reject Amnesty in Protest of Sex Vote.” The Local .
URL: https://www.thelocal.se/20150813/swedes-reject-amnesty-in-protest-of-sex-vote
MacAskill, William. 2015. Doing Good Better: How Effective Altruism Can Help You Help
Others, Do Work that Matters, and Make Smarter Choices About Giving Back. New York:
Penguin Random House LLC.
McAuley, James. 2019. “Billionaires raced to pledge money to rebuild Notre Dame. Then
came the backlash.” Washington Post .
URL: https://www.washingtonpost.com/world/europe/billionaires-raced-to-pledge-
money-to-rebuild-notre-dame-then-came-the-backlash/2019/04/18/7133f9a2-617c-
11e9-bf24-db4b9fb62aa2_story.html
McKeever, Brice. 2018. The Nonprofit Sector in Brief 2018: Public Charities, Giving, and
Volunteering. Technical report Urban Institute.
URL: https://nccs.urban.org/publication/nonprofit-sector-brief-2018#the-nonprofit-
sector-in-brief-2018-public-charites-giving-and-volunteering
Meer, Jonathan. 2011. “Brother, Can You Spare a Dime? Peer Pressure in Charitable
Solicitation.” Journal of Public Economics 95(7):926–941.
Metzger, Laura and Isabel Günther. 2019. “Making an impact? The relevance of infor-
mation on aid effectiveness for charitable giving. A laboratory experiment.” Journal of
Development Economics 136:18–33.
Meyer, Michelle N., Patrick R. Heck, Geoffrey S. Holtzman, Stephen M. Anderson, William
Cai, Duncan J. Watts and Christopher F. Chabris. 2019. “Objecting to experiments
that compare two unobjectionable policies or treatments.” Proceedings of the National
Academy of Sciences 116(22):10723–10728.
Milinski, M., D. Semmann and H. Krambeck. 2002. “Donors to charity gain in both indirect
reciprocity and political reputation.” Proceedings of the Royal Society of London, Series B
269:881–883.
Miller, Geoffrey. 2000. The Mating Mind: How Sexual Choice Shaped the Evolution of Human
Nature. Norwell, MA: Anchor Books.
148
Miller, Geoffrey. 2007. “Sexual Selection for Moral Virtues.” Quarterly Review of Biology
82:97–125.
Miller, Geoffrey. 2010. Spent: Sex, Evolution, and Consumer Behavior. New York: Penguin
Books.
Mislavsky, Robert, Berkeley J. Dietvorst and Uri Simonsohn. 2019. “The Minimum Mean
Paradox: A Mechanical Explanation for Apparent Experiment Aversion.” Proceedings of
the National Academy of Sciences 116(48):23883–23884.
Montealegre, Andres, Lance Bush, David Moss, David Pizarro and William Jimenez-Leal.
2020. “Does Maximizing Good Make People Look Bad?” Working Paper .
Moyn, Samuel. 2018. “Beyond the Human Rights Measurement Controversy.” Law and
Contemporary Problems 81(4):121–137.
Muniz, Albert M., Jr. and Thomas C. O’Guinn. 2001. “Brand Community.” Journal of
Consumer Research 27(4):412–432.
Murdie, Amanda. 2014. Help or Harm: The Human Security Effects of International NGOs.
Stanford University Press.
Murdie, Amanda and David R. Davis. 2012. “Shaming and Blaming: Using Events Data to
Assess the Impact of Human Rights INGOs.” International Studies Quarterly 56:1–16.
Murdie, Amanda, David R. Davis and Baekkwan Park. 2020. “Advocacy output: Auto-
mated coding documents from human rights organizations.” Journal of Human Rights
19(1):83–98.
Murdie, Amanda and Dursun Peksen. 2015. “Women’s Rights INGO Shaming and the
Government Respect for Women’s Rights.” Review of International Organizations 10:1–
22.
Naughton, Emma and Kevin Kelpin. 2015. “When evaluating human rights progress, focus
also on the journey.”.
URL: https://www.opendemocracy.net/openglobalrights/emma-naughton-kevin-kelpin/
when-evaluating-human-rights-progress-focus-also-on-jour
Nelson, Phillip J. and Kenneth V. Greene. 2003. Signaling Goodness: Social Rules and Public
Choice. Ann Arbor, MI: University of Michigan Press.
Nettle, Daniel, Zoe Harper, Adam Kidson and Melissa Bateson. 2013. “The Watching Eyes
Effect in the Dictator Game: It’s Not How Much You Give, It’s Being Seen to Give Some-
thing.” Evolution and Human Behavior 34(1):35–40.
Newman, George E. and Daylian M. Cain. 2014. “Tainted Altruism: When Doing
Some Good Is Evaluated as Worse Than Doing No Good at All.” Psychological Science
25(3):648–655.
149
O’Boyle, Ernest Hugh, George Christopher Banks and Erik Gonzalez-Mulé. 2017. “The
Chrysalis Effect: How Ugly Initial Results Metamorphosize Into Beautiful Articles.” Jour-
nal of Management 43(2):376–399.
Olivola, Christopher Y. 2011. When Noble Means Hinder Noble Ends: The Benefits and
Costs of a Preference for Martyrdom in Altruism. In The Science of Giving: Experimen-
tal Approaches to the Study of Charity, ed. Daniel M. Oppenheimer and Christopher Y.
Olivola. New York: Psychology Press.
Olivola, Christopher Y. and Eldar Shafir. 2013. “The Martyrdom Effect: When Pain and Ef-
fort Increase Prosocial Contributions.” Journal of Behavioral Decision Making 26(1):91–
105.
Ord, Toby. 2013. The Moral Imperative toward Cost-Effectiveness in Global Health.
Technical report Center for Global Development.
URL: http://amirrorclear.net/files/the-moral-imperative-towards-cost-effectiveness-in-
global-health.pdf
Osili, Una, Chelsea Clark, Mallory St. Claire and Jonathan Bergdoll. 2016. The 2016 U.S.
Trust Study of High Net Worth Philanthropy: Charitable Practices and Preferences of
Wealthy Households. Technical report.
URL: https://scholarworks.iupui.edu/handle/1805/11234
Osili, Una, Jacqueline Ackerman, Jonathan Bergdoll, Silvia Garcia, Adriene Davis Kalu-
gyer, Yannan Li, Addison Kane and Abe Roll. 2016. Giving in Puerto Rico. Technical
report.
URL: https://scholarworks.iupui.edu/bitstream/handle/1805/10915/Giving%20in%
20Puerto%20Rico%20-%20full%20report%208_31_2016%20final%20to%20printer.
pdf?sequence=1&isAllowed=y
Oster, Emily and Rebecca Thornton. 2011. “Menstruation, Sanitary Products, and School
Attendance: Evidence from a Randomized Evaluation.” American Economic Journal: Ap-
plied Economics 3(1):91–100.
Ostrower, Francie. 1995. Why the Wealthy Give: The Culture of Elite Philanthropy. Princeton
University Press.
Pallotta, Dan. 2008. Uncharitable: How Restraints on Nonprofits Undermine Their Potential.
Tufts University Press.
Paolacci, Gabriele, Jesse Chandler and Panagiotis Ipeirotis. 2010. “Running experiments
on Amazon Mechanical Turk.” Judgment and Decision Making 5(5):411.
Pinker, Steven. 2012. The Better Angels of Our Nature: Why Violence Has Declined. Penguin
Books.
Pinker, Steven. 2018. Enlightenment Now: The Case for Reason, Science, Humanism, and
Progress. Penguin Random House LLC.
150
Posner, Eric. 2015. “What’s the best use for Human Rights Watch’s budget?”.
URL: http://ericposner.com/whats-the-best-use-for-human-rights-watchs-budget/
Posner, Eric A. 2000. Law and Social Norms. Cambridge, Mass.: Harvard University Press.
Posner, Eric A. 2014. The Twilight of Human Rights Law. Oxford University Press.
Powell, Kate L., Gilbert Roberts and Daniel Nettle. 2012. “Eye Images Increase Charita-
ble Donations: Evidence From an Opportunistic Field Experiment in a Supermarket.”
Ethology 118:1–6.
Prakash, Aseem and Mary Kay Gugerty. 2010. Advocacy Organizations and Collective Action.
Cambridge University Press.
Pritchett, Lant. 2002. “It Pays to be Ignorant: A Simple Political Economy of Rigorous
Program Evaluation.” Policy Reform 5(4):251–269.
Raihani, Nichola J. 2014. “Hidden altruism in a real-world setting.” Biology Letters 10(1).
Raihani, Nichola J. and Sarah Smith. 2015. “Competitive Helping in Online Giving.” Cur-
rent Biology 25(9):1183–1186.
Reich, Rob. 2018. Just Giving: Why Philanthropy is Failing Democracy and How It Can Do
Better. Princeton University Press.
Reich, Rob. 2019. “Donors have pledged nearly a billion euros to restore Notre Dame. You
may not want to thank them.” Washington Post .
URL: https://www.washingtonpost.com/politics/2019/04/19/donors-have-pledged-
nearly-billion-euros-restore-notre-dame-you-may-not-want-thank-them/?utm_term=
.ef2699d8f884
Reich, Rob, Lacey Dorn and Stefanie Sutton. 2009. Anything Goes: Approval of Nonprofit
Status by the IRS. Technical report Center on Philanthropy and Civil Society Stanford
University: .
URL: https://pacscenter.stanford.edu/publication/anything-goes-approval-of-nonprofit-
status-by-the-irs/
Risse, Thomas. 2010. Rethinking advocacy organizations? A critical comment. In Ad-
vocacy Organizations and Collective Action, ed. Aseem Prakash and Mary Kay Gugerty.
Cambridge University Press.
Risse, Thomas, Stephen C. Ropp and Kathryn Sikkink. 1999. The Power of Human Rights:
International Norms and Domestic Change. Cambridge: Cambridge University Press.
Risse, Thomas, Stephen C. Ropp and Kathryn Sikkink, eds. 2013. The Persistent Power
of Human Rights: From Commitment to Compliance. Cambridge: Cambridge University
Press.
151
Roberts, G. 1998. Competitive altruism: From reciprocity to the handicap principle. In
Proceedings of the Royal Society of London, Series B: Biological Sciences. Vol. 265 pp. 427–
431.
Rodriguez-Garavito, Cesar and Sean McAdams. 2016. “A Human Rights Crisis? Unpacking
the Debate of the Future of the Human Rights Field.”.
URL: https://papers.ssrn.com/abstract=2919703
Rom, Sarah C. and Paul Conway. 2018. “The strategic moral self: Self-presentation shapes
moral dilemma judgments.” Journal of Experimental Social Psychology 74:24–37.
Ron, James, Howard Ramos and Kathleen Rodgers. 2005. “Transnational Information
Politics: NGO Human Rights Reporting, 1986-2000.” International Studies Quarterly
49(3):557–587.
Root, Brian. 2015. “Can rights organizations use low-burden self-reflection for evalua-
tion?”.
URL: https://www.openglobalrights.org/can-rights-organizations-use-lowburden-
selfreflection-for-evaluation/
Roser, Max. 2019. “The short history of global living conditions and why it matters that
we know it.”.
URL: https://ourworldindata.org/a-history-of-global-living-conditions-in-5-charts?
fbclid=IwAR0G8DRKqBC2DbP1ekM8i7syS0Lgwu1yutTz0qutTjvGcaMAYQc_x55u3yE
Rosling, Hans, Anna Rosling Rönnlund and Ola Rosling. 2018. Factfulness: Ten Reasons
We’re Wrong About the World - and Why Things Are Better Than You Think. New York:
Flatiron Books.
Santucci, Jeanine. 2017. “A new tactic to teach freshman about sexual assault: theater.”
USA Today .
URL: https://www.usatoday.com/story/college/2017/09/14/a-new-tactic-to-teach-
freshmen-about-sexual-assault-theater/37435345/
Schlangen, Rhonda. 2014. Monitoring and Evaluation for Human Rights Organizations:
Three Case Studies. Technical report Center for Evaluation Innovation.
URL: https://www.evaluationinnovation.org/wp-content/uploads/2014/01/CEI-HR-
Case-Studies.pdf
Sellers, Ron. 2018. “The Overhead Ratio Mindset.” Philanthropy Journal News .
URL: https://pj.news.chass.ncsu.edu/2018/05/21/the-overhead-ratio-mindset/
Senn, Charlene Y., Misha Eliasziw, Paula C. Barata, Wilfreda E. Thurston, Ian R. Newby-
Clark, H. Lorraine Radtke and Karen L. Hobden. 2015. “Efficacy of a Sexual As-
sault Resistance Program for University Women.” New England Journal of Medicine
372(24):2326–2335.
Sikkink, Kathryn. 2017. Evidence for Hope: Making Human Rights Work in the 21st Century.
Princeton University Press.
152
Simler, Kevin and Robin Hanson. 2018. Elephant in the Brain: Hidden Motives in Everyday
Life. Oxford University Press.
Simmons, Beth A. 2009. Mobilizing for Human Rights: International Law in Domestic Poli-
tics. Cambridge University Press.
Singer, Peter. 2009. The Life You Can Save: How To Do Your Part To End World Poverty. New
York: Random House.
Slough, Tara and Christopher J. Fariss. 2020. “Misgovernance and Human Rights: The
Case of Illegal Detention without Intent.” American Journal of Political Science .
Stroup, Sarah and Wendy Wong. 2017. The Authority Trap: Strategic Choices of Interna-
tional NGOs. Cornell University Press.
Sugin, Linda. 2018. “Competitive Philanthropy: Charitable Naming Rights, Inequality, and
Social Norms.” Ohio State Law Journal 79.
Terechshenko, Zhanna, Charles Crabtree, Kristine Eck and Christopher J. Fariss. 2019.
“Evaluating the influence of international norms and shaming on state respect for rights:
an audit experiment with foreign embassies.” International Interactions 45(4):720–735.
Tessman, Irwin. 1995. “Human Altruism as a Courtship Display.” Oikos 74(1):157–158.
Tetlock, Philip E. 2003. “Thinking the unthinkable: sacred values and taboo cognitions.”
Trends in Cognitive Sciences 7(7):320–324.
Todd, Benjamin. 2017. “Is it fair to say that most social programmes don’t work?”.
URL: https://80000hours.org/articles/effective-social-program/
Trivers, R. L. 1971. “The Evolution of Reciprocal Altruism.” Quarterly Review of Biology
46:35–57.
Trivers, Robert. 2011. The Folly of Fools: The Logic of Deceit and Self-Deception in Human
Life. New York: Basic Books.
Uhlmann, Eric Luis, David A. Pizarro and Daniel Diermeier. 2015. “A Person-Centered
Approach to Moral Judgment.” Perspectives on Psychological Science 10(1):72–81.
Van Vugt, M. and P. A. M. Van Lange. 2006. Psychological adaptations for prosocial behav-
ior: The altruism puzzle. In Evolution and Social Psychology, ed. M. Schaller, J. Simpson
and D. Kenrick. New York: Psychology Press.
Van Vugt, Mark and Wendy Iredale. 2013. “Men behaving nicely: Public goods as peacock
tails.” British Journal of Psychology 104(1):3–13.
Verkaik, Dave. 2016. “Do Donors Really Care About Impact Information? A Dual Process
Account.” Working Paper .
URL: https://www.researchgate.net/publication/318722670_Do_Donors_Really_Care_
About_Impact_Information_A_Dual_Process_Account
153
Vesterlund, Lise. 2006. Why Do People Give? In The Nonprofit Sector: A Research Hand-
book. Yale University Press pp. 568–587.
Vivalt, Eva. 2020. “How Much Can We Generalize from Impact Evaluations?” Journal of
the European Economic Association Forthcoming.
URL: http://evavivalt.com/wp-content/uploads/How-Much-Can-We-Generalize.pdf
Vries, Y. A. de, A. M. Roest, P. de Jonge, P. Cuijpers, M. R. Munafò and J. A. Bastiaansen.
2018. “The cumulative effect of reporting and citation biases on the apparent efficacy of
treatments: the case of depression.” Psychological Medicine 48(15).
Webber, Sophie and Carolyn Prouse. 2018. “The New Gold Standard: The Rise of Random-
ized Control Trials and Experimental Development.” Economic Geography 94(2):166–
187.
Wilcox, Rand. 2012. Modern Statistics for the Social and Behavioral Sciences: A Practical
Introduction. Taylor & Francis Group.
Zahavi, A. 1995. “Altruism as a handicap - the limitations of kin selection and reciprocity.”
Journal of Avian Biology 26:1–3.
Zahavi, A. and A. Zahavi. 1997. The Handicap Principle: A Missing Piece of Darwin’s Puzzle.
Oxford University Press.
154
Appendix
Chapter 3: Experiments
Methods
All studies were conducted through Positly, an online study recruiter that solicits high qual-
ity participants from Amazon Mechanical Turk. All participants had American IP addresses
and were over 18 years of age. Participants were compensated for their time (Study 1-2:
$0.36, Study 3-6: $0.25, Study 7: $0.36). Participants who had taken one of the prior
experiments or pilots were excluded from participating in future experiments. Participants
from MTurk were recruited by Positly and redirected to the survey questionnaire hosted
on Qualtrics. Upon agreeing to continue with the study after reading the welcome and
informed consent page, participants were randomly assigned to the experimental condi-
tion. Afterwards, they answered a simple and easy question designed to test their com-
prehension. Those who failed the comprehension check were omitted from the analysis.
Participants from Studies 1 and 2 were debriefed at the end of the survey if they were
assigned to the fourth condition. They were told “the principal investigator does not know
of any charity that can save a life [or prevent a case of HIV] for as little as $150 [$50].
The best estimates suggest that the lowest cost is $2,000 to save a life from preventable
disease in low-income countries [$650 to prevent a case of HIV infection within one year
in low-income countries]." All studies reported in the main text were preregistered with
AsPredicted prior to the collection of data. All studies received exempt status by the Uni-
155
versity of Southern California’s Internal Review Board.
Measurement Validation
All seven studies measure how much money out of $100 individuals hypothetically allocate
to the charity from the treatment group and another charity option. Before implementing
the final survey instruments, I pilot-tested various phrases to use for the alternative charity
option. In the end, I chose the language of “some other charity" as the optimal alternative
option. This wording is optimal because the median individual in the baseline condition
evenly splits their $100 donation between the two charities (M = $47.7), which allows
me to detect any movement in either direction. As pilot-tested, if I were to choose a well-
known charity as the alternative (e.g. Doctors Without Borders), the donations in the
baseline condition would have started out below $50 (M = $35), making it more difficult
to detect a negative effect.
I chose this approach rather than asking individuals to allocate between the hypotheti-
cal charity and pocketing the money for themselves. Charities compete with other charities
for funding, and I want to reflect this reality in the experiments. Probing how much more
(or less) money individuals want to give to the charity from the treatment group compared
to a competitor directly speaks to the realities of market competition. Relatedly, donors
tend to diversify their donations, so the decision of how much money to donate to any
particular charity is often made against the backdrop of allocating donations across many
different charities. As pilot-tested, if I were to have participants allocate $100 between
the charity and themselves, the donations in the baseline condition would have started out
below $50 (M = $21.9).
Survey Instruments
Each study began with a welcome and informed consent page:
156
Welcome to our study on charity.
Thank you for your interest in helping us with this study. Your participation will consist
of a simple online questionnaire and should take approximately 3 minutes. Do not hesitate
to ask any questions or mention concerns about the study either before, during, or after
your participation by emailing sjcaldwe@usc.edu.
Before you begin, please read the following information to confirm you are happy to
take part. This is a requirement by the university.
1. I have read and understood the above participant information.
2. I have had the opportunity to ask questions about the study, and if I have, I have
received satisfactory answers to these questions.
3. I understand how to raise a concern or make a complaint.
4. I understand that my participation is voluntary and that I am free to leave the study
at any time, without giving any reason, without penalty.
5. I understand that data collected during the study may be looked at by authorized
individuals from the University of Southern California where it is relevant to my taking
part in this research.
6. I permit these individuals access to my research records.
7. I agree to results of this research study being reported in student dissertations,
peer-reviewed journals, or at scientific meetings, but I know that I will not be named or
identified in these publications.
8. I understand that this project has been reviewed by, and received ethics clearance
through, the University of Southern California Office for the Protection of Research Sub-
jects.
9. I confirm I am over 18 years of age.
10. I agree to take part in this study.
If you agree with all above points, please continue. Clicking below indicates that you
have read the description of the study, and you agree to participate in the study.
157
Please read the text and the questions very carefully. There will be comprehension
questions designed to check how carefully you read. If you give incorrect answers to these
questions, your responses will be excluded from the study’s analysis.
Study 1
Condition 1: Human Health International operates in developing countries to provide
life-saving medical care to people at risk of preventable disease.
You have the opportunity to donate $100 to the following charities. How much of the
$100 would you allocate to each one? (Please only enter numbers. The numbers must
sum to 100).
Condition 2: Human Health International operates in developing countries to provide
life-saving medical care to people at risk of preventable disease. Human Health Inter-
national conducted a rigorous study which allowed them to measure the effect of their
medical care. The results from the study provided strong evidence that Human Health
International saves the life of one person for every $2,000 spent.
You have the opportunity to donate $100 to the following charities. How much of the
$100 would you allocate to each one? (Please only enter numbers. The numbers must
sum to 100).
Condition 3: Human Health International operates in developing countries to provide
life-saving medical care to people at risk of preventable disease. Human Health Inter-
national conducted a rigorous study which allowed them to measure the effect of their
medical care. The results from the study found that Human Health International did not
have the desired impact, suggesting that reform is needed to increase effectiveness.
158
You have the opportunity to donate $100 to the following charities. How much of the
$100 would you allocate to each one? (Please only enter numbers. The numbers must
sum to 100).
Condition 4: Human Health International operates in developing countries to provide
life-saving medical care to people at risk of preventable disease. Human Health Inter-
national conducted a rigorous study which allowed them to measure the effect of their
medical care. The results from the study provided strong evidence that Human Health
International saves the life of one person for every $150 spent.
You have the opportunity to donate $100 to the following charities. How much of the
$100 would you allocate to each one? (Please only enter numbers. The numbers must
sum to 100).
Comprehension check: According to the information you have been provided, how
does Human Health International help people in developing countries?
by providing medical care
by organizing information-awareness campaigns
by transferring cash
Debrief for all participants: Debrief: Human Health International is a hypothetical
international health charity. You have not been provided $100, and no amount of money
you indicated will be given to either “Human Health International" or “Some other inter-
national health charity."
Debrief for participants assigned to Condition 4: Debrief: While you have been told
that Human Health International, a hypothetical charity, can save a life for every $150
spent, this number is fabricated. The principal investigator does not know of any health
charity that can save a life for as little as $150. The best estimates suggest that the lowest
cost is $2,000 to save a life from preventable disease in low-income countries.
159
Study 2
Condition 1: Human Rights International operates in developing countries, organizing
issue-awareness campaigns to educate teenage girls about the risk of HIV.
You have the opportunity to donate $100 to the following charities. How much of the
$100 would you allocate to each one? (Please only enter numbers. The numbers must
sum to 100).
Condition 2: Human Rights International operates in developing countries, organizing
issue-awareness campaigns to educate teenage girls about the risk of HIV. Human Rights
International conducted a rigorous study which allowed them to measure the effect of
their issue-awareness campaign. The results from the study provided strong evidence that
Human Rights International’s issue-awareness campaign prevents one teenage girl from
being infected with HIV for every $650 spent.
You have the opportunity to donate $100 to the following charities. How much of the
$100 would you allocate to each one? (Please only enter numbers. The numbers must
sum to 100).
Condition 3: Human Rights International operates in developing countries, organizing
issue-awareness campaigns to educate teenage girls about the risk of HIV. Human Rights
International conducted a rigorous study which allowed them to measure the effect of
their issue-awareness campaign. The results from the study found that Human Rights
International did not have the desired impact, suggesting that reform is needed to increase
effectiveness.
You have the opportunity to donate $100 to the following charities. How much of the
$100 would you allocate to each one? (Please only enter numbers. The numbers must
160
sum to 100).
Condition 4: Human Rights International operates in developing countries, organizing
issue-awareness campaigns to educate teenage girls about the risk of HIV. Human Rights
International conducted a rigorous study which allowed them to measure the effect of
their issue-awareness campaign. The results from the study provided strong evidence that
Human Rights International’s issue-awareness campaign prevents one teenage girl from
being infected with HIV for every $50 spent.
You have the opportunity to donate $100 to the following charities. How much of the
$100 would you allocate to each one? (Please only enter numbers. The numbers must
sum to 100).
Comprehension check: According to the information you have been provided, how
does Human Rights International help people in developing countries?
by providing blood pressure medication
by organizing information-awareness campaigns
by transferring cash
Debrief for all participants: Debrief: Human Rights International is a hypothetical
international health charity. You have not been provided $100, and no amount of money
you indicated will be given to either “Human Rights International” or “Some other inter-
national human rights charity.”
Debrief for participants assigned to Condition 4: Debrief: While you have been
told that Human Rights International, a hypothetical charity, can prevent one case of HIV
infection for every $50 spent, this number is fabricated. The principal investigator does
not know of any charity that can prevent a case of HIV infection for as little as $50. The
best estimates suggest that the lowest cost is $650 to prevent a case of HIV infection within
one year in low-income countries.
161
Study 3
Condition 1: Several drugs have been approved by the U.S. Food and Drug Administration
as safe and effective for treating high blood pressure. A director of a health charity wants
to provide good treatment to his patients, so he decides that his patients who need high
blood pressure medication will be prescribed drug A. It is affordable and patients can
tolerate its side effects.
If you had the opportunity to donate $100, how much of the $100 would you allocate
to the following? (Please only enter numbers. The numbers must sum to 100).
Condition 2: Several drugs have been approved by the U.S. Food and Drug Adminis-
tration as safe and effective for treating high blood pressure. A director of a health charity
wants to provide good treatment to his patients, so he decides that his patients who need
high blood pressure medication will be prescribed drug B. It is affordable and patients can
tolerate its side effects.
If you had the opportunity to donate $100, how much of the $100 would you allocate
to the following? (Please only enter numbers. The numbers must sum to 100).
Condition 3: Several drugs have been approved by the U.S. Food and Drug Adminis-
tration as safe and effective for treating high blood pressure. A director of a health charity
thinks of two different ways to provide good treatment to his patients, so he decides to
run an experiment by randomly assigning his patients who need high blood pressure med-
ication to one of two test conditions. Half of patients will be prescribed drug A, and the
other half will be prescribed drug B. Both drugs are affordable and patients can tolerate
their side effects. After a year, he will only prescribe to new patients whichever drug has
had the best outcomes for his patients.
162
If you had the opportunity to donate $100, how much of the $100 would you allocate
to the following? (Please only enter numbers. The numbers must sum to 100).
Comprehension check: According to the information you have been provided, what
outcome does the director of the charity want to improve?
school attendance
high blood pressure
violent crime
Study 4
Condition 1: Last year, a charity received a large number of donations. The director of this
charity wants to help children in a low-income country increase their school attendance,
so he decides that all children in a low-income village will receive free school lunches.
If you had the opportunity to donate $100, how much of the $100 would to allocate to
the following? (Please only enter numbers. The numbers must sum to 100).
Condition 2: Last year, a charity received a large number of donations. The direc-
tor of this charity wants to help children in a low-income country increase their school
attendance, so he decides that all children in a low-income village will receive free trans-
portation to school.
If you had the opportunity to donate $100, how much of the $100 would to allocate to
the following? (Please only enter numbers. The numbers must sum to 100).
Condition 3: Last year, a charity received a large number of donations. The director
of this charity wants to help children in a low-income country increase their school atten-
dance, so he decides to run an experiment by randomly assigning children to one of two
163
test conditions. Half of all children in a low-income village will receive free school lunches.
The other half will receive free transportation to school. After a year, the director will be-
gin providing everyone in the village whichever resource (lunches or transportation) turns
out to help children attend school more often.
If you had the opportunity to donate $100, how much of the $100 would you allocate
to the following? (Please only enter numbers. The numbers must sum to 100).
Comprehension check: According to the information you have been provided, what
outcome does the director of the charity want to improve?
school attendance
high blood pressure
violent crime
Study 5
Condition 1: The leader of the college chapter of a human rights organization wants to
reduce rape on her college campus, so she gets her university’s approval to implement an
informed-consent class during first-year orientation that teaches students how to obtain
and give consent.
If you had the opportunity to donate $100, how much of the $100 would you allocate
to the following? (Please only enter numbers. The numbers must sum to 100).
Condition 2: The leader of the college chapter of a human rights organization wants
to reduce rape on her college campus, so she gets her university’s approval to implement
a self-defense class during first-year orientation that teaches students how to verbally and
physically defend themselves.
164
If you had the opportunity to donate $100, how much of the $100 would you allocate
to the following? (Please only enter numbers. The numbers must sum to 100).
Condition 3: The leader of the college chapter of a human rights organization wants
to reduce rape on her college campus, so she gets her university’s approval to run an
experiment by randomly assigning students to one of two test conditions during first-year
orientation. Half of all first-year students will take an informed-consent class that teaches
students how to obtain and give consent. The other half will take a self-defense class
that teaches students how to verbally and physically defend themselves. After a year, the
human rights chapter will implement the program (informed-consent or self-defense) of
the group that experienced the fewest completed rapes.
If you had the opportunity to donate $100, how much of the $100 would you allocate
to the following? (Please only enter numbers. The numbers must sum to 100).
Condition 4: The leader of the college chapter of a human rights organization wants
to reduce rape on her college campus, so she gets her university’s approval to run an
experiment by randomly assigning students to one of two test conditions during first-
year orientation. Half of all first-year students will take a self-defense class that teaches
students how to verbally and physically defend themselves. The other half will not take the
self-defense class. After a year, the human rights chapter will implement the self-defense
program if that group experienced fewer completed rapes compared to the group that did
not take the class.
If you had the opportunity to donate $100, how much of the $100 would you allocate
to the following? (Please only enter numbers. The numbers must sum to 100).
Comprehension check: According to the information you have been provided, what
outcome does the leader of the human rights chapter want to improve?
suicide
high blood pressure
165
rape
Study 7
Condition 1: Human Health International operates in developing countries to provide life-
saving medical care to people at risk of preventable disease. Human Health International
conducted a rigorous study which allowed them to measure the effect of their medical
care. The results from the study provided strong evidence that Human Health International
saves the life of one person for every $2,000 spent.
You have the opportunity to donate $100 to the following charities. How much of the
$100 would you allocate to each one? (Please only enter numbers. The numbers must
sum to 100).
Condition 2: Human Health International operates in developing countries to provide
life-saving medical care to people at risk of preventable disease. Human Health Inter-
national conducted a rigorous study which allowed them to measure the effect of their
medical care. The results from the study provided strong evidence that Human Health
International saves the life of one person for every $900 spent. A large charity evaluator
indicated that 50% of donation revenue goes to overhead expenses such as CEO compen-
sation, fundraising, staff parties, and retreats.
You have the opportunity to donate $100 to the following charities. How much of the
$100 would you allocate to each one? (Please only enter numbers. The numbers must
sum to 100).
Debrief to all participants: Debrief: Human Health International is a hypothetical
international health charity. You have not been provided $100, and no amount of money
166
you indicated will be given to either “Human Health International” or “Some other inter-
national health charity.”
Comprehension check: According to the information you have been provided, how
does Human Health International help people in developing countries?
by providing medical care
by organizing information-awareness campaigns
by transferring cash
Pilots
Study 1: Study 1 is an exact replication of a pilot experiment I ran in which 474 Mechan-
ical Turk workers were randomly assigned to one of four conditions, each describing a hy-
pothetical international health charity but varying the type of impact information revealed
about the charity. The experiment demonstrates that participants only reward the reve-
lation of impact information when the information suggests the charity is unrealistically
effective. Participants did not reward the charity for revealing honest impact information
(realistically positive results or ineffective, null results).
Study 3: Study 3 is an exact replication of a pilot experiment I ran in which 329 Me-
chanical Turk workers were randomly assigned to one of three conditions, each describing
a hypothetical charity but varying whether the director of the charity implements a pro-
gram unilaterally, untested, or whether the director randomly assigns two programs to the
recipients and chooses whichever program has the best outcome. The participants were
then asked how much money they would donate to the charity. The experiment demon-
strates that individuals donate less to a charity that conducts A/B testing compared with a
charity that implements one program unilaterally without testing.
Study 4: Study 4 is an exact replication of a pilot experiment I ran in which 314 Me-
chanical Turk workers were randomly assigned to one of three conditions, each describing
167
a hypothetical charity but varying whether the director of the charity implements a pro-
gram unilaterally, untested, or whether the director randomly assigns two programs to the
recipients and chooses whichever program has the best outcome. The participants were
then asked how much money they would donate to the charity. The experiment demon-
strates that individuals do not donate less to a charity that conducts A/B testing compared
with a charity that implements one program unilaterally without testing.
Study 7: Study 7 is an exact replication of a pilot experiment I ran in which 427 Me-
chanical Turk workers were randomly assigned to one of four conditions, each describing
a hypothetical international health charity but varying the type of impact and overhead
information revealed about the charity. The experiment demonstrates that when a charity
has high overhead, participants will donate less to it even when it saves more than twice
as many lives per dollar as a charity with no overhead information revealed about it.
Test results
This section displays the linear regression results with demographic control variables,
demonstrating that the main effects presented in Chapter 3 hold while controlling for
factors such as age, gender, income, and education.
Income
Chapter 3 includes a discussion of possible limitations to the studies. One possible limita-
tion is that the kinds of people who select into Mturk tasks are different from the kinds of
people who do not complete Mturk tasks. One such difference could be income. Chapter
3 includes plots demonstrating that mean donations across the treatment conditions are
similar for low, middle, and high income groups. This section of the Appendix includes the
plots from the other studies not included in the main body of Chapter 3.
168
Table 5.1: The Effect of Impact Information on
Donations (Studies 1 and 2)
Donations
Study 1 (Saving Lives) Study 2 (HIV Prevention)
(1) (2) (3) (4)
Impact information: $2,000 4.714 5.172
(4.198) (4.110)
Impact information: $650 2.626 2.417
(4.189) (4.194)
Impact information: ineffective 23.583

23.975

20.078

20.373

(4.239) (4.150) (4.170) (4.198)
Impact information: $150 16.957

15.828

(4.178) (4.089)
Age 0.472

0.054
(0.120) (0.126)
Education 0.423 0.217
(1.029) (1.047)
Female 10.514

5.622

(2.953) (3.011)
Income 0.702 0.536
(0.867) (0.930)
Impact information: $50 15.338

15.048

(4.199) (4.208)
Constant 60.286

77.861

52.560

48.510

(2.968) (6.778) (2.955) (7.071)
Observations 418 418 434 434
R
2
0.189 0.236 0.146 0.155
Adjusted R
2
0.183 0.223 0.140 0.141
Residual Std. Error 30.415 (df = 414) 29.658 (df = 410) 30.855 (df = 430) 30.846 (df = 426)
F Statistic 32.164

(df = 3; 414) 18.126

(df = 7; 410) 24.534

(df = 3; 430) 11.128

(df = 7; 426)
Significance levels

p<0.1;

p<0.05;

p<0.01
169
Table 5.2: The Effect of Randomized Trials on
Donations (Studies 3-5)
Donations
Study 3 (Drug Effectiveness) Study 4 (School Attendance) Study 5 (Sexual Assault Prevention)
(1) (2) (3) (4) (5) (6)
Drug A 3.630 2.914
(3.613) (3.573)
Drug B 2.105 1.063
(3.586) (3.553)
Age 0.361

0.126 0.316

(0.125) (0.145) (0.110)
Education 0.824 0.290 2.706

(1.034) (1.235) (0.991)
Female 4.000 9.631

10.859

(2.923) (3.510) (2.900)
Income 1.957

0.115 0.096
(0.866) (1.036) (0.851)
School Lunch 1.490 1.782
(4.323) (4.310)
Transportation 6.012 6.153
(4.213) (4.232)
Informed consent 1.251 1.026
(3.631) (3.537)
Self-defense 7.934

8.520

(3.613) (3.544)
Constant 41.706

68.214

56.670

58.546

41.589

60.026

(2.559) (7.165) (3.034) (7.973) (2.551) (6.324)
Observations 496 496 314 314 469 469
R
2
0.002 0.038 0.007 0.032 0.016 0.081
Adjusted R
2
0.002 0.027 0.001 0.014 0.012 0.069
Residual Std. Error 32.666 (df = 493) 32.198 (df = 489) 30.793 (df = 311) 30.595 (df = 307) 32.063 (df = 466) 31.124 (df = 462)
F Statistic 0.509 (df = 2; 493) 3.247

(df = 6; 489) 1.114 (df = 2; 311) 1.714 (df = 6; 307) 3.774

(df = 2; 466) 6.761

(df = 6; 462)
Significance levels

p<0.1;

p<0.05;

p<0.01
170
Table 5.3: The Effect of Overhead and Impact
Information on Donations (Study 7)
Donations
(1) (2)
$900 w/ 50% overhead 38.742

36.947

(4.133) (4.123)
Age 0.404

(0.172)
Education 1.949
(1.475)
Female 6.339
(4.086)
Income 2.428

(1.189)
Constant 65.771

83.518

(2.894) (8.550)
Observations 206 206
R
2
0.301 0.350
Adjusted R
2
0.298 0.333
Residual Std. Error 29.657 (df = 204) 28.890 (df = 200)
F Statistic 87.849

(df = 1; 204) 21.512

(df = 5; 200)
Significance levels

p<0.1;

p<0.05;

p<0.01
171
Figure 5.1: Mean Donations by Income Group
(Study 2 Impact Information)
$0 $20 $40 $60
No Impact Information
Impact Information: $650
Impact Information: Ineffective
Impact Information: $50
Income Group
low
middle
high
172
Figure 5.2: Mean Donations by Income Group
(Study 3 A/B Test)
$0 $10 $20 $30 $40 $50
Drug A
Drug B
Drug A/B Test
Income Group
low
middle
high
173
Figure 5.3: Mean Donations by Income Group
(Study 4 A/B Test)
$0 $20 $40 $60
School Lunch
Transportation
A/B Test
Income Group
low
middle
high
174

Abstract (if available)

Abstract Selling Virtue seeks to explain a variety of puzzling behaviors within human rights philanthropy: (1) human rights NGOs routinely claim to have a positive impact despite lacking rigorous evidence of it, (2) individual donors do not seek and are not responsive to rigorous impact information, (3) accountability organizations do not hold human rights NGOs accountable to impact despite claiming to, and (4) human rights NGOs continue to garner money, status, credibility, and authority, despite never providing evidence of their impact beyond anecdotes that are strongly biased in the positive direction. ❧ I argue that this behavior, while grossly inefficient at achieving the organizations’ stated missions, is not irrational. Instead, it is created by reputational self-interest and misaligned incentives. I combine social signaling theory and the reality of market competition to develop a new framework for analyzing the behavior of human rights NGOs. This new framework, which I call the Virtue Economy, is a market-oriented model of NGO behavior. I argue that NGOs act like firms, producing and selling social signals to their donors, who are best understood as consumers. The primary product on offer is not the effective production of a public good or public benefit, as is conventionally believed, but rather a reputational or status benefit in the form of a social signal. ❧ I use this new framework to explain why donors are not interested in or responsive to information about charity effectiveness and how donor preferences influence the organizational behavior of human rights NGOs. I employ experiments and statistical analysis to analyze demand-side and supply-side features of the human rights market. The findings illuminate (1) why human rights NGOs abstain from rigorously evaluating their impact, (2) why they conceal honest information about their impact from donors and other organizations, (3) why they exaggerate their impact, and (4) why the accountability movement does not hold human rights NGOs accountable to standards of effectiveness, despite claiming to. Put simply, donors reward this deceptive behavior because it is conducive to social signaling. I also revisit the statistical literature on the effect of human rights NGOs and demonstrate that the research suffers from numerous critical problems. A more faithful reading of the statistical evidence leads to the conclusion that human rights NGOs are not very effective. This finding should not be surprising, given that human rights NGOs calibrate their activities to satisfy donor reputational needs, not the needs of the stated beneficiaries.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

By word, not deed: advancing a theory of rhetorical compliance in international law

PDF

The global market for wombs: a study of the transnational surrogacy industry in Mexico

PDF

The IO contestation project: a multi-method approach towards understanding the ways states contest international organizations.

PDF

Perspectives on state capacity and the political geography of conflict

PDF

Refugee protection and empowerment joint venture: understanding how and why public and private actors cooperate to tackle the global refugee crisis

PDF

Deciding modalities of global health governance: What facilitates or hinders public-private partnerships?

PDF

Pathways to nuclear weapon reversal: exploring mechanisms and understanding non-proliferation policy

PDF

Forced march to modernity: State-imposed cultural change and regime stability in 20th-century east Asia

PDF

Politics is something we do together: identity and institutions in U.S. elections

PDF

The fabled fourth estate: media freedom, democracy and human rights

PDF

When passions run high: emotions and the communication of intentions in face-to-face diplomacy

PDF

Rights projects in a globalized world

PDF

A framework for research in human-agent negotiation

PDF

California's adaptation to sea level rise: incorporating environmental justice communities along the California coastline

PDF

Essays in health economics and provider behavior

PDF

A multitheoretical multilevel explication of crowd-enabled organizations: exploration/exploitation, social capital, signaling, and homophily as determinants of associative mechanisms in donation-...

PDF

Essays on health insurance programs and policies

PDF

Organized intraethnic conflict and cooperation in divided stateless nations: the competitive dynamics of rival ethnonationalist factions

PDF

Acculturation team-based clinical program: pilot program to address acculturative stress and mental health in the Latino community

PDF

A roadmap for changing student roadmaps: designing interventions that use future “me” to change academic outcomes

Asset Metadata

Creator Mulesky, Suzie (author)

Core Title Selling virtue: how human rights NGOs and their donors work together to create a better world ... for themselves

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Political Science and International Relations

Publication Date 03/29/2021

Defense Date 01/11/2021

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Charity,effective altruism,human rights,nongovernmental organizations,OAI-PMH Harvest,philanthropy,research methodology,social signaling

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Sandholtz, Wayne (committee chair), Fariss, Christopher (committee member), Gruskin, Sofia (committee member), James, Patrick (committee member)

Creator Email sjcaldwe@usc.edu,susiejcaldwell@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-433240

Unique identifier UC11667204

Identifier etd-MuleskySuz-9368.pdf (filename),usctheses-c89-433240 (legacy record id)

Legacy Identifier etd-MuleskySuz-9368.pdf

Dmrecord 433240

Document Type Dissertation

Rights Mulesky, Suzie

Type texts

Source 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 a...

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