University of Southern California Dissertations and Theses

Habits and Friction: Actual and Perceived Effects on Behavior

(USC Thesis Other)

Habits and Friction: Actual and Perceived Effects on Behavior

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Running head: HABITS AND FRICTION
Habits and Friction: Actual and Perceived Effects on Behavior
by
Asaf Mazar
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
(PSYCHOLOGY)
August 2023
HABITS AND FRICTION
ii
Acknowledgements

First off, I’d like to thank my advisor, Wendy Wood. I can’t begin to describe the roles that
Wendy played—and continues to play—in my life. Her brilliance, kindness, and dedication are
unmatched, and her willingness to go above and beyond to help others is a true inspiration. At
any point during my PhD, she was always there with a smile, happy to advise, chat, and guide.
I’m very fortunate to call her my advisor.

I’d like to thank my committee members: Arthur Stone, for encouraging me to think thoroughly
about methods and study design, and how to map these onto my research questions. I’m also
grateful to Arthur for first guiding me through using intensive longitudinal methods, back when I
started my PhD. I’d like to thank John Monterosso (an honorary committee member), for
invaluable and thought-provoking discussions about the nature of willpower and self-regulation,
as well as insightful comments about designing research to test the hypotheses introduced in this
manuscript. I’m also grateful to Chris Beam and Genevieve Dunton, for providing valuable
feedback on this thesis and the preceding thesis proposal.

I’d like to thank my co-authors and labmates for countless hours of fascinating coffee chats,
dinners, and (yes, even) zoom calls. And finally and most importantly, I’ll be forever grateful to
my parents for supporting and being there for me all along; to Michelle, for being the reason for
it all; and to our darling Ari.
HABITS AND FRICTION
iii
Table of Contents
Acknowledgements……………………………………………………………………………… ii
List of Tables…………………………………………………………………………………….iv
List of Figures……………………………………………………………………………………vi
Abstract…………………………………………………………………………………………vii
Introduction……………………………………………………………………………………… 1
Chapter 1: Illusory Feelings Elusive Habits: People Overlook Habits in Explanations of
Behavior………………………………………………………………………………………….. 5
Abstract……………………………………………………………………………………….. 5
Study 1………………………………………………………………………………………... 7
Study 2………………………………………………………………………………………. 18
General Discussion………………………………………………………………………….. 36
Chapter 2: Americans Discount the Effect of Friction on Voter Turnout…………………. 39
Abstract……………………………………………………………………………………… 39
Significance Statement………………………………………………………………………. 40
Introduction………………………………………………………………………………….. 40
Results……………………………………………………………………………………….. 42
Discussion…………………………………………………………………………………… 47
Materials and Methods……………………………………………………………………… 49
Chapter 3: Actual and Perceived Effects of Friction on App Use Over Time……………... 56
Abstract……………………………………………………………………………………… 56
Introduction………………………………………………………………………………….. 57
Method………………………………………………………………………………………. 79
Results……………………………………………………………………………………….. 90
Discussion…………………………………………………………………………………..105
General Discussion……………………………………………………………………………. 109
References……………………………………………………………………………………... 112
Appendices…………………………………………………………………………………….. 127
Appendix A - Supplementary Materials for Chapter 1…………………………………….. 127
Appendix B - Supplementary Materials for Chapter 2…………………………………….. 136
Appendix C - Supplementary Materials for Chapter 3…………………………………….. 149

HABITS AND FRICTION
v
List of Tables
Table 1.1 Study 1: Means (M), Standard Deviations (SD), and Correlations..……………….....13
Table 1.2 Percent (Count) of Participants Agreeing to Help by Mood and Habit Condition........14
Table 1.3 Study 2: Means (M), Standard Deviations (SD), and Between-Person Correlations....27
Table 1.4 Study 2: Coefficient Estimates for Fixed Effects in the Main Multilevel Model…......29
Table 3.1 Means and Standard Deviations for Key Between-Person Variables...……………….92
Table 3.2 Results for Main Pre-Registered Model Predicting App Use Difference Score............94
Table 3.3 Multiverse Analysis Parameters....................................................................................96
Table 3.4 Results of Multilevel Model Predicting Perceived Efficacy from Condition…............97
Table 3.5 Results of Multilevel Model Predicting Intentions to Limit App Use from
Condition.......................................................................................................................................98
Table 3.6 Results of Multilevel Linear Regression Predicting Intentions to Reduce App Use
from Study Day, Experimental Condition, and Day X Condition Interactions...........................101
Table 3.7 Results of Multilevel Linear Regression Predicting Perceived Ease from
Condition.....................................................................................................................................105
Table S1.1 Study 1: Results of Main Regression Analysis Without Exclusions (N = 1015)......128
Table S1.2 Study 2: Main Analysis Results Using SRHI............................................................129
Table S1.3 Study 2: Main Analysis with Intention and Attitudes as Covariates.........................130
Table S2.1 Perceived Effects of Beliefs and Friction on Turnout only Among Intended In-
person Voters...............................................................………………............................………137
Table S2.2 Actual Effects of Beliefs and Friction on Turnout – Unweighted Sample................139
Table S2.3 Perceived Effects of Beliefs and Friction on Turnout – Unweighted Sample...........140
Table S2.4 Perceived Effects of Beliefs and Friction on Turnout – Post Election Survey......…142
HABITS AND FRICTION
v
Table S2.5 Perceived Effects of Beliefs and Friction on Turnout among those who Reported
Not Voting in Post-Election Survey........................................................................................…143
Table S2.6 Regressions Predicting Support for Policies Affecting Friction from Belief vs.
Friction Importance Ratings....................................................................................................…144
Table S2.7 Variables’ Predicted vs. Actual Explained Variance in Studies S1a and S1b..........148
Table S3.1 Results of Linear Regression Model Predicting Initial (Intake Survey) Intentions
to Reduce Phone Use from Condition.............................................……………………………149
Table S3.2 Results of Multilevel Linear Regression Model Predicting Daily Perceived Ease
from Study day and Experimental Condition......................................................................……150
Table S3.3 Results of Multilevel Linear Regression Model Predicting Daily App Use from
Daily Intentions (person-mean-centered; measured in evening diaries), Study Day, and
Condition.....................................................................................................................................151
Table S3.4 Results of Multilevel Linear Regression Model Predicting Daily Intention – App
Use Difference Score (scaled by person-mean and person-standard-deviation) from Study
Day...............................................................................................................................................152
Table S3.5 Results of Multilevel Linear Regression Model Predicting Daily App Use in the
Friction Conditions from Perceived Efficacy and Experimental Condition................................153

HABITS AND FRICTION
vi
List of Figures
Figure 1.1 Study 2: Likelihood of Drinking Coffee as a Function of Habit and Fatigue..............30
Figure 2.1 Perceived Effects of Beliefs and Friction on Turnout…………………………...…...45
Figure 2.2 Differences Between Vote Planners and Non-Planners……………………………...47
Figure 3.1 Examples for designs of pull (left) and push (right) handles.......................................66
Figure 3.2 Study design.................................................................................................................84
Figure 3.3 App Use Pre- and Post-Intervention in the Control and Friction Conditions...............91
Figure 3.4 Intentions to Reduce Target App Use by Study Day............................................….100
Figure 3.5 Plot of Multilevel Regression Model Predicting Daily Perceived Ease from
Condition and Study Day................................................................…………………………….104
Figure S3.1 Daily Intention - App Use Difference Scores (scaled by person-mean and person-
standard-deviation) by Study Day................................................................................………...154
HABITS AND FRICTION
vii
Abstract
People tend to do (a) what they’ve done in similar situations in the past, and (b) what’s
easiest. In other words, human behavior is strongly shaped by both habit and friction. However,
people’s perceptions may underestimate the pervasive pull of both forces on behavior, with far-
reaching consequences. In Chapter 1, I measured the actual and perceived impact of habits on
behavior. In two studies—a naturalistic study using Ecological Momentary Assessment (EMA)
and an online controlled experiment—I found that participants’ coffee drinking and helping
behavior was substantially driven by habit, but participants underestimated habits’ influence,
implicating inner states (like fatigue and mood) instead. In Chapter 2, I tracked the effects of
friction and beliefs on voter turnout in the 2020 United States election. I found that friction
depressed turnout, but participants underappreciated its effects. Furthermore, those participants
who neglected friction (vs. beliefs) also tended to support policies that make voting more
difficult. In both Chapters 1 and 2, underestimation of habit and friction was remarkably robust,
persisting regardless of financial incentives as well as when making attributions for others (vs.
self). Finally, Chapter 3 compared perceptions and actual effects of a subtle friction intervention
on habitual phone app use. In estimating intervention effects, participants once again overlooked
the potential impact of friction and instead emphasized willpower and motivation. In contrast, I
anticipated that imposition of minor friction may suffice to disrupt habitual phone use. However,
results do not indicate that friction reduced app use more than a control condition. Participants
across conditions exhibited an initial reduction in app use quickly followed by a rapid
rebounding to baseline—after several days or even a single day. Collectively, these accumulated
findings underscore the omnipresent yet overlooked power of habits and friction, while
highlighting the difficulties of creating and maintaining behavior change in field interventions.
HABITS AND FRICTION
1
Introduction
Humans tend to act on behaviors that (a) minimize effort, and (b) they performed in
similar situations in the past. These patterns point to two key forces shaping behavior—friction
and habits. Suggesting that these forces are fundamental to understanding human behavior, they
are not unique to humans, but indeed were at first documented extensively in research on other
mammals such as rats and mice (De Camp, 1920; Hull, 1943).
Friction and Effort in Human Behavior
Friction refers to external features of an environment that make a behavior more difficult
or easy (Mazar et al., 2021). Early on, researchers noted that rodents navigating mazes to reach
food tend to prefer the shortest path (De Camp, 1920). This observation has later been codified
into the law of less work, positing that “organisms will...choose the less laborious behavior”
(Hull, 1943, p. 294). Later research uncovered that humans avoid not only physical demand, but
also cognitive demand (Kool et al., 2010; Shane, 2005). This is not to say that organisms will
always select the path of least resistance—humans and other animals regularly select more
effortful behaviors like climbing mountains and eschwing freely-available food in favor of food
hidden inside a puzzle feeder (the one notable exception being cats, who seem to always be
content with the path of least resistance; Delgado et al., 2022). But rather than undermine the law
of less work, such exceptional instances highlight how effort interacts with other forces to shape
behavior.
Historically, friction has been relatively absent from research on human behavior change.
Theories that arose in the wake of the cognitive revolution of the 1960’s often framed behavior
as “purposive [and] regulated by forethought” (Bandura, 1991, p. 248). Although such framing
does not inherently exclude friction (people could purposefully pursue or avoid friction), in
HABITS AND FRICTION
2
practice theories and empirical evidence alike tended to stress person-level variables—goals,
attitudes, and intentions—at the expense of friction and other situational features.
Situational features received renewed interest with the advent of behavioral economics
and nudges. Encompassing any intervention that does not ban options or change economic
incentives, nudges span a broad spectrum, incorporating informational approaches, social norm
signaling, and making undesirable options non-salient and inaccessible (Thaler & Sunstein,
2009). Perhaps the most striking nudge exemple is the default. In the context of encouraging
registering as an organ donor and putting money away for retirement, making the default choice
the desirable choice dramatically increases participation rates (Johnson & Goldstein, 2003;
Madrian & Shea, 2001).
Default effects highlight how in some cases, even minor increases in friction can exert a
sizable influence on behavior. However, friction interventions other than defaults have received
relatively little attention. One systematic review on the effects of food availability and proximity
on eating identified just 24 studies, most of which were one-time experiments in lab settings
(Hollands et al., 2019). The authors concluded that despite moderate to large effect sizes,
existing knowledge offers only low to very low evidence certainty. In short, although
encouraging evidence suggests that friction can influence short-term behavior in novel lab
settings, it is unclear how friction affects real-world behavior over time.
Habits and the Friction-Habit Interface
When friction is applied to repeated behavior (rather than one-off instances), it would
likely interface with habits. Habits are learned cue-response associations in memory, forming as
a behavior is repeated and rewarded in a given context (Mazar & Wood, 2018; Wood, 2017). As
novel habits become solidified, behavior comes to be directly triggered by stable cues in the
HABITS AND FRICTION
3
response environment, while at the same time becoming less subject to goal-directed control
(Gardner, 2015). For instance, a habit of watching TV after coming home may at first be driven
by a goal of watching a particularly beloved series, but continue even after the original show is
not aired anymore (i.e., the initial goal is removed). Similarly, wearing a seatbelt may be initially
driven by a goal of staying safe or avoiding a ticket, but eventually come to be automated,
without the need for the underlying motivation to be activated every time one enters a vehicle
(Brijs et al., 2011). Participants in another set of studies were randomized to eat either stale or
fresh popcorn while watching movie trailers (Neal et al., 2011). Participants with weak habits
acted as one might expect—they ate more when served fresh rather than stale popcorn.
Participants with strong habits, however, ate similar amounts regardless of whether the popcorn
was fresh or stale, even though the stale popcorn group explicitly rated the popcorn as less tasty.
In other words, even when the goal (eating delicious popcorn) was removed by making popcorn
stale, strong-habit participants continued to eat it out of habit.
Habits can prove notoriously resistant to change (Ersche et al., 2016; Mantzari et al.,
2015). Although pliant in the short-term to vigilant monitoring that inhibits habitual responding
(Quinn et al., 2010), interventions to change habits often follow a triangle-shaped pattern, by
which initial changes are followed by a return to baseline (Wood & Neal, 2016). One reason for
this may be past research’s reliance on cognitive self-regulation strategies. The process model of
self-control differentiates cognitive self-regulation strategies—ones that are implemented inside
one’s mind, such as willpower and reframing—and situational strategies—ones that change the
observable environment (Duckworth, Gendler, et al., 2016). The model additionally
differentiates between strategies that involve selecting environments (such as studying in a
HABITS AND FRICTION
4
library instead of a home) and ones that alter existing environments (such as studying at home
but placing your phone in a different room).
Selecting environments may be an especially effective way of disrupting habits, such as
when sustainability interventions are more effective among people who recently moved to a new
residence (Verplanken & Roy, 2016). In many cases, however, selecting entirely different
environments is often inconvenient or unfeasible, such as in the case of moving to a new address.
In such cases, altering existing environments to disrupt undesirable habits and encourage
desirable ones may present a more effective and practical path to behavior change. In particular,
habit may be particularly disrupted by designing environments that make it impossible to act on
a habit in a purely automatic fashion. As explained in Chapter 3, such changes could induce a
minor change in friction, which in turn may translate into a larger shift in behavior.

HABITS AND FRICTION
5
Chapter 1: Illusory Feelings Elusive Habits: People Overlook Habits in Explanations of
Behavior

Mazar, A., & Wood, W. (2022). Illusory Feelings, Elusive Habits: People Overlook Habits in
Explanations of Behavior. Psychological Science, 33(4), 563–578.
https://doi.org/10.1177/09567976211045345

Abstract
Habits underlie much of human behavior. However, people may prefer agentic
explanations that overlook habits in favor of inner states such as mood. We tested this
misattribution hypothesis in an online experiment of helping behavior as well as an ecological
momentary assessment study of college students’ everyday coffee drinking. Both studies
revealed a substantial gap between attributed and actual influences on behavior: Habit strength
outperformed or matched inner states in predicting behavior, whereas participants’ attributions
for their behavior emphasized inner states. Participants continued to overlook habits even when
incentivized for accuracy, as well as when making attributions for other people’s behavior. We
discuss how this attribution pattern could adversely influence self-regulation.
Habits are key to successful functioning in our day-to-day lives. By automating repeated
behaviors, habits enable people to consistently get enough sleep, stay fit, eat a healthy diet, and
study (Galla & Duckworth, 2015). Furthermore, engaging in routine, habitual behaviors is
associated with a greater sense of security and even meaning in life (Avni‐Babad, 2011;
Heintzelman & King, 2019). Everyday habits also enable multitasking: In an experience
sampling study, 43% of daily behaviors were performed habitually in that they were repeated
frequently in the same location, typically while participants were thinking about something other
than what they were doing (Wood et al., 2002).
HABITS AND FRICTION
6
Habits are mental associations between contexts and responses that develop as people
repeat rewarded responses in a given context (Knowlton & Diedrichsen, 2018). Once habits have
formed, context cues automatically activate the repeated response in mind (Mazar & Wood,
2018). Habit associations are separate from the behavior they produce: For example, when
entering a car, the habitual response of wearing a seatbelt may be activated in mind regardless of
whether one overtly acts on it by buckling up.
Given the many opportunities people have in daily life to observe their own repeated
actions, one might expect lay theories to accurately account for habits. In support, participants in
one study read about an office worker who locked in a colleague by turning the office door knob
counterclockwise (Gershman et al., 2016). Participants demonstrated attribution to habit by
placing less blame on the office worker if their home doors also opened counterclockwise,
implying that the office worker acted out of habit. Also relevant, the actor-observer effect posits
that people readily attribute their own behavior (but not others’ behavior) to environmental
influences (Jones & Nisbett, 1971). Because habits are activated by cues in the environment,
people may consequently ascribe their own behavior to habits.
Nevertheless, there is reason to believe that people overlook habit when accounting for
their own repeated, habitual behaviors. People overvalue introspective thoughts, feelings, and
emotions in self-judgments (Pronin, 2009); and they interpret actions as intentional by default
(Rosset, 2008). Illustrating overattribution to inner states, smokers in one study reported that
their smoking was triggered by negative affect, even though in-the-moment affect assessments
revealed little association between negative affect and subsequent smoking (Shiffman et al.,
1997). In another study, self-described emotional eaters demonstrated little daily association
between negative affect and snacking (Adriaanse, Ridder, et al., 2011). In addition, participants
HABITS AND FRICTION
7
with stronger habits reported greater certainty in their behavioral intentions, even though these
intentions did not predict their future behavior (Ji & Wood, 2007). Taken together, these findings
suggest that people may exaggerate the effect of inner states on behavior while discounting the
role of habit.
In sum, the present article tests a potential bias to overlook the influence of habits on
behavior. Such a bias would be important to document given the many downstream influences of
lay theories about behavior (e.g., McFerran & Mukhopadhyay, 2013). For example, such a bias
may cause people to ineffectively self-regulate by putting too much weight on regulating inner
states such as mood and too little weight on using self-regulation strategies that may better
control habits such as reducing triggering cues (Duckworth, Gendler, et al., 2016).
The Present Research
In two studies, we measure the effects of habits and inner states on a behavior and assess
participants’ attributions for that behavior. Our first study was an experiment that orthogonally
manipulated habit strength and mood to assess their effects on helping behavior. Our second
study used ecological momentary assessment to track coffee drinking over a typical week. In
both studies, miscalibration of the actual and perceived effects of habit would emerge if
participants placed less value on habit than inner states in their attributions than would be
accurate, given the actual effects of habit and inner states on behavior.
Study 1
Participants first recalled a happy, sad, or neutral event and then completed a simple,
supposedly unrelated task that trained either a strong or weak habit to press one of two computer
keys. Immediately after the task, participants pressed one of the keys to indicate if they were
willing to donate a small amount of time to help the researchers. The habit manipulation should
HABITS AND FRICTION
8
lead strong habit participants to select the response that maps onto the extensively-trained key.
Specifically, our hypotheses compared (a) the actual effects of the mood and habit manipulations
on participants’ decision to help or not with (b) participants’ attributions for their helping.
Orthogonally manipulating habits and inner states does not imply that habits and inner
states are always uncorrelated. Indeed, in daily life, habits often align with moods, goals, and
feelings. However, once formed, habitual responses are triggered directly by context cues, with
inner states providing limited input (Wood et al., 2021).
Method
Power Analyses
Power analyses for logistic regression were conducted using the WebPower package in R
(Zhang & Yuan, 2018). Results suggested that 787 participants were required to achieve 80%
power for detecting a difference of 10% in helping behavior between conditions (namely, 45%
help in one condition vs. 55% in another). Because this study used a novel habit-formation task,
this expected difference was chosen because it represents a plausible moderate effect size. Pre-
registration plans including exclusion criteria, materials, data files, and analytic code for both
studies can be accessed at:
https://osf.io/5xfsm/?view_only=ba7cacb9c9e047c1ad60fdd137f6d84a
Participants
We recruited 809 online participants via Prolific (388 male, 392 female, 15 other, 14
declined to answer; Mage= 35.17, SDage= 13.32, range = 18-78). An additional 115 participants
were excluded who did not pass the comprehension check, and an additional 91 participants were
excluded due to extreme PANAS scores (+/-2 SD from their mood condition’s mean; 76
HABITS AND FRICTION
9
participants) or reaction times (median reaction time > 500 ms; 15 participants). Including all
participants in the analyses did not notably alter the results (see Table S2 in Appendix A).
Procedure
After providing informed consent, participants completed an autobiographical emotional
memory task (Mills & D’Mello, 2014) in which they recalled either a happy, sad, or neutral
memory. The description of each emotion condition is given in the bolded text in parentheses
below:
Recall (an event in your life that made you happy/ an event in your life that
made you sad/ the last time that you brushed your teeth). Take some time to really
experience the event and the feelings associated with it. When you are ready, describe the
event below in your own words. You may use between 5-40 words.
Participants had a minimum of 30 seconds to write about the scenario in a text box, after which
they could proceed with the experiment. As a manipulation check, participants then completed
items from the Positive And Negative Affect Schedule (Thompson, 2007).
For the habit-formation task, participants completed 40 trials in which the letter “m” or
“z” appeared on the screen, and participants were required to respond by pressing the
corresponding key. In the strong habit condition, participants practiced either the left (“z”) or the
right (“m”) response on 36 out of 40 trials (90%) and the alternate response on the remaining 4
trials (10%). The specific response (z/m) was counterbalanced across participants. In the weak
habit condition, participants practiced each response equally (20 trials on each side). Participants
were instructed to be as fast and accurate as possible. This task follows habit formation
procedures in prior research (e.g., Hardwick et al., 2019).
HABITS AND FRICTION
10
Immediately following the task, a screen displayed the helping request: “Are you willing
to complete 40 additional trials (~5 minutes) as a favor to us (without additional compensation),”
with “Yes” and “No” responses mapped onto the same keys used in the habit-formation task (“z”
and “m”). The behavioral measure was whether participants agreed or not. For strong habit
participants, the “No” response was always mapped onto the more heavily practiced response.
For weak habits participants, “Yes” and “No” responses were randomly assigned to each key.
The “No” response was over-trained in the strong habit condition because the helping request
(donating five minutes of participants’ time) was taxing and might have spurred participants to
exert control to decline it. Participants then answered a comprehension check to test whether
they understood the help request along with additional measures (see below).
Measures
Positive and Negative Mood. Participants responded to the prompt “indicate the extent
to which you feel the mood below RIGHT NOW.” On scales ranging from 1 (very slightly/not at
all) to 9 (extremely), participants rated positive emotions (”inspired,” ”determined,” ”attentive,”
”proud,” ”alert,” ”active”) and negative ones (“upset,” ”hostile,” “ashamed,” ”nervous,”
”afraid,” ”guilty”) taken from the Positive And Negative Affect Schedule (Thompson, 2007).
Positive and negative item ratings were averaged to create a positive affect score (𝛼 = .82) and a
negative affect score (𝛼 = .80).
Self-Attribution. On scales ranging from 0% (not at all important) to 50% or more
(extremely important), participants rated the extent to which their decision to help or not was due
to habit, “I responded automatically, without thinking,” and mood, “My mood at the time (I felt
good/bad).” The sum of both answers could range from 0-100%. Presentation order of mood and
habit was counterbalanced in both the attribution and incentivized measures.
HABITS AND FRICTION
11
Incentivized Other-Attribution. To minimize judgment biases, participants were
incentivized to provide accurate explanations for others’ behavior. The incentive should
minimize the effects of conversational norms regarding plausible or socially acceptable
explanations for a behavior. Attributions for others’ behavior are additionally informative
because they should be relatively unaffected by self-serving biases that could influence self-
attributions. Thus, on scales ranging from 0% (not at all important) to 50% or more (extremely
important), participants indicated “How important do you think that the following factors are in
determining whether OTHER participants agree or decline to complete additional trials?”
Participants then rated habit and mood, as in the self-attribution measure (above). Accurate
ratings (within 5% of the study results), earned a chance to win a $10 bonus.
Habit Strength: Self-Reported Behavior Automaticity Index (Gardner et al., 2012; a
subset of items taken from the Self-Report Habit Index, Verplanken & Orbell, 2003). On scales
ranging from 1 (strongly disagree) to 7 (strongly agree), participants rated the extent to which
“hitting a key (‘z’ or ‘m’) in the task is something that I…”: (a) “did without thinking,” (b) “did
automatically,” (c) “did without having to consciously remember,” and (d) “started doing before
I realized.” Ratings were averaged to create a perceived automaticity score (𝛼 = .85).
To our knowledge, this is the first use of this measure with a simple finger-movement
task, and experienced automaticity did not differ between the weak (M = 4.09) and strong habit
conditions (M = 4.08), 95% CI [-0.25, 0.23], t(478.9) = -0.07, p = .947, d = -.005, 95% CI [-
0.15, 0.14]. This measure tests a downstream consequence of habit formation—perceived
automaticity. However, perception of automaticity is not an especially sensitive measure in and
of itself and could tap processes other than habit (Hagger et al., 2015; Mazar & Wood, 2018).
Given that even weak habit participants reported relatively high levels of experienced
HABITS AND FRICTION
12
automaticity, our simple key-pressing task produced uniformly high experienced automaticity.
For this reason, we do not discuss this measure further.
Habit Strength: Reaction Time (RT). As in prior research, the strength of habit
associations in key-pressing tasks can be assessed directly through reaction times to respond to
the cue (Hardwick et al., 2019). Due to the skewness common in reaction time distributions,
median rather than mean reaction times were used in all analyses.
Comprehension Check. This measure was included to ensure that participants
understood the help request. Immediately after indicating their response, participants answered:
“What was the request that you just responded to?” by choosing one of the following options: (a)
To continue for an additional 20 minutes, (b) To continue for an additional 5 minutes, (c) To
recommend the study to a friend, or (d) To receive double compensation for my participation.
Answers a, c, and d were coded as incorrect.
Results
Descriptive statistics and correlations among key variables are presented in Table 1. The
proportion of participants in each condition who agreed to help is presented in Table 2.

HABITS AND FRICTION
13
Table 1.1
Study 1: Means (M), Standard Deviations (SD), and Correlations
Variable M (SD) 1 2 3 4 5 6
1. Habit
attribution
11.48
(13.90)

2. Mood
attribution
28.54
(16.11)
.04

3.
Incentivized.
habit other-
attribution
21.35
(13.63)
.49**

.12**

4.
Incentivized
mood other-
attribution
31.99
(12.86)
.04

.56**

.09**

5. Positive
affect
5.26
(1.52)
.05

.11**

.04

.16**

6. Negative
affect
2.13
(1.15)
.13**

.07

.11**

.02

-.20**

7. Reaction
time
298.8
(61.47)
.05

.08*

.01

.06

.15**

-.04

Note. Attributions range from 0 - 50 with higher scores reflecting greater importance, mood
scores range from 1-9 with higher numbers reflecting stronger feelings, and higher reaction times
(in milliseconds) reflects slower responding in the habit training task.
* p < .05
** p < .01

HABITS AND FRICTION
14
Table 1.2
Percent (Count) of Participants Agreeing to Help by Mood and Habit Condition
Habit condition
Weak Strong Total
Mood
condition
Control 60% (50/83) 43% (77/178) 49% (127/261)
Sad 48% (46/96) 47% (87/185) 47% (133/281)
Happy 55% (50/91) 47% (82/176) 49% (132/267)
Total 54% (146/270) 46% (246/539) 48% (392/809)

Note. Higher percentages reflect more participants agreeing to help by working 5 extra minutes.

Regression models were fit to the data using the following predictors: mood condition
(dummy-coded, with the control condition as the reference level), habit condition (effects-coded,
-1 = weak habit, +1 = strong habit), and interactions between mood and habit. This analytic
design was used for all Study 1 analyses.
Manipulation Checks on Mood
The mood induction successfully invoked positive affect (measured via the PANAS): In
the sad memory condition, participants reported less positive affect compared with the control
condition, b = -0.38, 95% CI [-0.65, -0.11], β = -0.25, p = .006. In the happy memory condition,
participants reported more positive affect compared with the control condition, b = 0.42, 95% CI
[0.15, 0.69], β = 0.27, p = .003. Habit condition did not significantly influence positive affect, b
= -0.08, 95% CI [-0.27, 0.12], β = -0.05, p = .424. No interactions emerged between habit and
mood conditions, both ps > .2.
HABITS AND FRICTION
15
Results for negative affect also indicated the success of the mood induction: Participants
in the sad condition reported higher levels of negative affect compared with the control
condition, b = 0.83, 95% CI [0.63, 1.02], β = 0.72, p < .001. Participants in the happy condition
did not differ on negative affect from participants in the control condition, b = 0.03, 95% CI = [-
0.17, 0.22], β = 0.02, p = .786. This is in line with previous work showing the relative
independence of the positive and negative PANAS subscales, reflecting an underlying
independence of positive and negative affect (Thompson, 2007). Habit condition did not
significantly affect negative affect, b < 0.01, 95% CI [-0.14, 0.14], β < 0.01, p = .954. There were
no interactions between habit and mood conditions, both ps > .2.
As an additional mood manipulation check, two coders blind to condition and hypotheses
coded each open-text response to the mood manipulation on (a) whether it displayed negative
affect, and (b) whether it displayed positive affect. Coders’ ratings showed high interrater
reliability: The negative affect ratings showed 97% agreement, Cohen’s k = .95, and positive
affect ratings showed 95% agreement, Cohen’s k = .91. Disagreements between coders were
resolved by discussion. Suggesting that the mood manipulation was successful, most open-text
responses in the negative affect condition showed negative affect (84.34%), compared with few
of the responses in the control (12.26%) and positive (3.75%) conditions. Similarly, most
responses in the positive affect condition showed positive affect (80.15%) compared with fewer
in the control (35.25%) and negative (3.56%) conditions.
Manipulation Check on Habit
The reaction time measure revealed that the habit manipulation was successful: In the
strong habit condition, participants were significantly faster to respond than in the weak habit
condition, b = -20.43, 95% CI [-28.10, -12.77], β = -0.33, p < .001. There was a non-significant
HABITS AND FRICTION
16
trend for slower reaction times for participants in the happy mood condition compared with the
control condition, b = 9.22, 95% CI [-1.47, 19.90], β = 0.15, p = .091. No other effects
approached significance, all ps > .2.
Habit and Mood Effects on Helping - Actual
A logistic regression model tested the actual effects of habit and mood on helping
behavior (Yes / No). Habit significantly influenced this decision, such that participants in the
strong habit condition (who had just extensively practiced the “no” response) were less likely to
agree to help, OR = 0.84, 95% CI [0.73, 0.98], p = .024. Mood condition did not significantly
determine helping, either for the sad condition, OR = 0.94, 95% CI [0.67, 1.32], p = .721, or the
happy condition, OR = 1.02, 95% CI [0.73, 1.44], p = .892. Thus, habit influenced behavior,
whereas mood did not.
1

Habit and Mood Effects on Helping - Attributed
To test the perceived effects of habit and mood, a dependent-samples t-test assessed the
within-person difference in attributions to mood compared with habit. Participants strongly
attributed their behavior to mood over habit, Mdiff = 17.10, 95% CI [15.66, 18.54], t(800) =
23.27, p < .001, d = 1.14, 95% CI [1.01, 1.26]. As anticipated, a strong albeit somewhat smaller
bias in favor of mood remained when participants were incentivized to give accurate attributions
for others’ behavior, Mdiff = 10.61, 95% CI [9.37, 11.85], p < .001, d = 0.80, 95% CI [0.69, 0.91].
Thus, as anticipated, participants’ attributions favored mood over habit more than would be

1
We also computed an exploratory model that included the Habit x Mood interaction,
which essentially replicated the results reported in the text: The hypothesized effect of habit, OR
= 0.71, 95% CI [0.54, 0.92], p = .011, no effect of the sad condition, OR = 0.84, 95% CI [0.59,
1.21], p = .348, or happy condition, OR = 0.96, 95% CI [0.66, 1.39], p = .826, and only an
unexpected nonsignificant interaction between habit and the happy mood condition, OR = 1.38,
95% CI [0.96, 1.99], p = .078.
HABITS AND FRICTION
17
expected given the actual effects of each on behavior. It should be noted that despite the strong
favoring of mood, habits were judged a plausible explanation, especially for others’ behavior,
with habits receiving an importance rating of 21% (maximum 50%), compared with 32% for
mood.
Exploratory Analyses: Intensity of Experience and Attributions
We explored whether participants with stronger moods and stronger habits were more
likely to make attributions to mood and habit, respectively. In general, participants with stronger
internal states were only slightly more likely to attribute their behavior to these states. That is,
attribution to mood was weakly correlated with positive PANAS scores, r(801) = .11, and
marginally with negative scores, r(801) = .07 (see Table 1). In addition, attribution to mood was
weakly correlated with reaction time on the habit formation task, so that participants with slower
RTs gave stronger mood attributions, r(801) = .08. Attributions to habit only showed a slight
positive correlation with negative affect, r(800) = .13.
Discussion
This first experiment provided causal evidence that people’s explanations for their
behavior favor inner states over habits even when that behavior is driven by habit. We
manipulated habit strength via amount of practice at a key-press task and manipulated mood
through a memory recall task. Participants then indicated their willingness to help by pressing a
highly practiced or less practiced key.
In our test of actual influences on helping, habit strength determined helping but current
mood did not. Specifically, participants in the strong habit condition, who had earlier extensively
practiced the “no” response key, were more likely to decline a helping request compared with
participants in the weak habit condition, who practiced the “yes” and “no” response keys equally.
HABITS AND FRICTION
18
Thus, participants’ decisions continued to be influenced by their prior key-pressing habit. In
contrast, participants induced to feel sad or happy helped at comparable rates as participants in a
control condition with no mood manipulation. Note that our hypotheses were not about these
behavioral effects per se, but instead concerned the difference between actual and perceived
effects of habit and mood on behavior.
When explaining their behavior, participants attributed their helping more to current
mood than to habit. Thus, attributions were misaligned with the actual determinants of behavior
by underestimating habit effects and overemphasizing mood. Our design provided a compelling
test of this hypothesis given that participants in the strong habit condition should be aware of
their recent, extensive practice at pressing a particular computer key. However, when
incentivized to make accurate attributions about others’ behavior, participants still revealed a
substantial attribution gap favoring mood over habit. Thus, it does not seem that the attribution
pattern was due to artifacts of social desirability or conversational norms that favor mood
explanations. The incentivized measure also suggested that attributions to habit were meaningful:
Despite the substantial attribution gap favoring mood over habit, participants considered habit to
be a plausible determinant of others’ behavior.
Study 2
Study 1 tested explanations for a single behavior immediately following an emotionally
evocative experience that might make mood especially salient and thus likely to influence
attributions. Study 2 instead investigated the attribution bias with a mundane repeated
behavior—coffee drinking—recorded over the course of a typical week. Specifically, coffee
drinking was assessed in response to an inner state, fatigue, and to habit strength, which are
common reasons for coffee drinking (see pilot below).
HABITS AND FRICTION
19
Given that some of our participants drank coffee very often, we anticipated that habit
would strongly influence actual coffee drinking even more than fatigue. However, as in the first
study, we anticipated that participants’ attributions would emphasize fatigue as much as or more
than habit. Thus, these two hypotheses together concern the correspondence between the actual
and perceived determinants of behavior. In an additional test of our model, we anticipated that
fatigue attributions would be unrelated to within-person associations between fatigue and coffee
drinking. In other words, people’s beliefs about fatigue determining their coffee drinking will be
unrelated to its actual role in driving their individual consumption.
Method
Pilot
To assess lay beliefs about the causes of coffee drinking, 40 college students (22 male, 16
female, 2 genderqueer or other) rated six causes of coffee drinking on scales ranging from 1 (not
at all important) to 5 (extremely important), including: fatigue (“tiredness or low energy”), habit
(“habit or behavior routines”), thirst, taste, social motives (“spending time with friends”) and
coffee after a meal. Fatigue was rated as most important (M = 4.05, SD = 0.96), followed by taste
(M = 3.58, SD = 1.03), habit (M = 3.50, SD = 1.22), social motives (M = 3.17, SD = 1.08),
having coffee after a meal (M = 2.12, SD = 1.22), and thirst (M = 1.70, SD = 0.91). A paired-
samples t-test comparing fatigue and habit attributions (within participants) revealed that
participants attributed coffee drinking to fatigue significantly more than to habit, Mdiff = 0.55,
95% CI [0.14, 0.96], t(39) = 2.72, p = .01, d = 0.50, 95% CI [0.11, 0.88].
Design
To capture experiences and explanations as they naturally unfold in daily life, we used a
combination of surveys, daily morning reports, and ecological momentary assessment (EMA).
HABITS AND FRICTION
20
Participants first completed intake surveys, including measures of habit strength and attributions
for their own coffee drinking. Then, over the course of a week, they reported every two hours on
their fatigue and coffee drinking. They also completed a brief survey every morning immediately
after waking up.
Our analysis predicted coffee drinking at one prompt from fatigue experienced at the
prior prompt. This lagged design minimized any self-report bias that might emerge from
concurrent associations between fatigue and coffee drinking (i.e., “I’m drinking coffee therefore
I must be tired”). After the study week, participants completed a final survey. Finally,
participants completed a follow-up survey once data collection for the study ended.
Power Analyses. Ecological Momentary Assessment designs such as the present one
produce thousands of prompts (level-1 sample size), which tend to produce very high power for
within-person effects. Because most of our research questions could be probed within-person
(e.g. using our novel context-specific habit measure), we aimed for a final sample size of 120,
which is in line with typical sample sizes in ecological momentary assessment studies (compare
with a mean sample size of 99 in a recent systematic review; Wen et al., 2017).
To estimate observed power for our multilevel logistic regression, we simulated a dataset
with log-odds regression coefficients of 0.3 and 0.2 (corresponding to our odds-ratios of 1.35 and
1.22) for a level 2 and level 1 variable, respectively. Simulated sampling from this dataset 1,000
times revealed that 50 participants were sufficient to achieve 90% power for our between-
subjects variable (habit strength) and 99.5% power for our within-person variable (fatigue).
Participants
Participants were a convenience sample of 112 U.S. undergraduate students who received
either course credit or monetary compensation (85 female, 27 male, Mage = 20.85, SDage = 2.85,
HABITS AND FRICTION
21
range = 18-33). The (self-reported) selection criteria were: (a) speaking English fluently, (b)
owning a smartphone, (c) being 18 or older, and (d) drinking coffee once a week or more often.
An additional 35 participants were excluded for drinking coffee once or less often during the
study period, and 4 additional participants were excluded for answering less than 50% of
prompts. Thus, the final sample for analyses was slightly smaller than our preregistered target of
120.
To minimize attrition, compensation was linked to compliance. Paid participants received
$20 for completing 80%-100% of EMA prompts, $15 for completing 50%-80%, and $5 for
completing less than 50%. Participants who received course credit had a similar 3-tier
compensation system.
Procedure
Intake Session. After providing informed consent, participants reported their coffee
drinking habit strength, coffee drinking intentions and attitudes, coffee drinking attributions, and
demographics (see measures below). In addition, to obscure the purpose of the study and limit
reactivity, participants answered an identical set of measures about soft drinks. Participants then
wrote down implementation intentions (Adriaanse, Vinkers, et al., 2011) to overcome potential
obstacles for completing the prompts (e.g., “if my phone beeps when I am with people, then I
will excuse myself and answer the prompt”).
Ecological Momentary Assessment (EMA). For five weekdays (i.e., participants were
not prompted on Saturday and Sunday), participants were prompted to respond 8 times per day at
regular two-hour intervals from 8am to 10pm. Each prompt included items meant to obscure the
purpose of the study, including location (e.g., home, campus) and temperature (hot, cold, or
comfortable). Participants then reported how tired they were, whether they drank coffee in the
HABITS AND FRICTION
22
past two hours, and whether they drank soft drinks in the past two hours. Participants also
completed an exploratory mood item and an open-response item in which they briefly described
their current situation.
In addition, because fatigue on waking up may be particularly important for coffee
drinking, participants completed a prompt every morning when they got out of bed. Morning
prompts included the same items as the regular prompts as well as an item asking whether they
had already drunk coffee (a measure of compliance). Thus, we could measure the prospective
effect of waking fatigue on coffee drinking and avoid the self-report bias that might emerge with
concurrent reports (i.e., “I drank coffee, therefore I must have been tired”).
At the end of the first study day, participants with response rates of 50% or above (4 or
more prompts) were informed of their approximate level of compliance via email (50-75% or 75-
100%). Those with less than 50% compliance were contacted by phone and/or text message to
address potential technical difficulties that might have led to low compliance.
Final Survey. Participants were sent the final survey on the weekend after they
completed the EMA portion of the study. This survey included the context-specific habit
measure, single-event self-attribution measure, and open-text measures asking about self-
regulation and general study feedback.
Follow-Up Survey. Shortly after all data collection was completed, participants were
emailed a survey that included the incentivized self-attribution measure (see below).
Measures: Intake
Additional measures for this study are included in the supplemental materials (Appendix
A).

HABITS AND FRICTION
23
Habit Strength.
Behavior-Frequency-In-Context (BFiC; Galla & Duckworth, 2015; Ji & Wood, 2007).
Participants reported how often they drink coffee on a scale ranging from 1 (Less than once a
week) to 5 (More than 7 times a week; that is, more than once a day). They then rated how often
they drink coffee at the same time of day, and at the same location, on scales ranging from 1
(Never or almost never at the same [time / location]) to 5 (Almost always or always at the same
[time / location]). Each participants’ coffee drinking frequency rating was then multiplied by the
time and location stability ratings separately, and the two Frequency X Context scores were
averaged to create a mean habit strength score.
Self-Report Habit Index (SRHI; Verplanken & Orbell, 2003). This is the complete
version of the brief 4-item SRBAI used in Study 1. Note that we did not include the full SRHI in
Study 1 to minimize participant burden). Participants indicated their agreement with a set of
eleven statements regarding coffee drinking (e.g. “drinking coffee is something that I do without
thinking,” “drinking coffee is something that belongs to my daily routine”), on a scale ranging
from 1 (Strongly disagree) to 7 (Strongly agree).
Self-Attribution. On scales ranging from 0 – 100%, participants rated the extent to
which their coffee drinking was driven by habit (“my past behavior and habits”) and by fatigue
(“my energy levels and tiredness”). Anchors of 0% indicated that coffee drinking was unaffected
by a factor and 100% that coffee drinking was completely determined by a factor. Participants
were instructed to not allow the sum of both ratings to exceed 100%.
Coffee Drinking Intentions and Attitudes. On a scale ranging from 0 (not at all) to 100
(extremely), participants rated their liking of coffee drinking (“how much do you enjoy drinking
coffee?”). On a scale ranging from 1 (Strongly disagree) to 7 (Strongly agree), participants rated
HABITS AND FRICTION
24
their coffee drinking intentions (“I intend to drink coffee _,” with the underscore replaced by the
individual participant’s self-reported frequency of coffee drinking; Ajzen, 2002).
Measures: Ecological Momentary Assessment (EMA)
Fatigue. Participants rated how tired they were on a scale ranging from 1 (not at all) to 6
(extremely).
Coffee Drinking. To ensure that a single coffee consumed over a period of time was
categorized as one episode, participants indicated whether they started drinking coffee in the past
two hours (No / Yes - 1 Drink / Yes - 2 Drinks / Yes - 3 Drinks or more). Answer choices were
categorized into a binary drink/did not drink indicator of coffee drinking.
Mood. In this exploratory measure, participants rated their current mood on a scale
ranging from 1 (unhappy) to 5 (happy).
Situation Description (Open-Text Measure). In a free response, participants briefly
described their current situation (e.g., “going to the gym,” “with friends”). Specifically, for
prompts in which they indicated that they had recently been drinking coffee, they described that
coffee drinking situation. For prompts in which they did not drink coffee, they described the
situation they were in one hour previous. These situation descriptions were then used in the
“Context-Specific Habit Measure” and the “Single-Event Attribution Measure” (see below).
Measures: Final Survey
Context-Specific Habit Measure. As an exploratory measure, we randomly selected for
each participant seven situation descriptions for prompts in which participants reported not
drinking coffee and up to seven situation descriptions in which participants did report drinking
coffee. For each situation description, participants rated (a) how often they drank coffee in that
situation, (b) how automatic they perceived coffee drinking to be in that situation, and (c) the
HABITS AND FRICTION
25
strength of their intentions to drink coffee in that situation (see Supplement in Appendix A for
full description).
Single-Event Self-Attribution. To evaluate attributions for a specific instance of a
behavior, participants were shown their open-text situation description for their own final coffee
drinking event, and rated the extent to which habit and fatigue contributed to drinking coffee at
that time. Item wording and answer choices were the same as the intake attribution measure. To
confirm that participants recalled the specific coffee drinking event, they reported whether they
remembered it, and analyses for this single-event measure only included the 81 participants who
answered affirmatively (72%).
Measures: Follow-Up Survey
Incentivized Self-Attribution. Participants were offered a monetary incentive of $3 if
they accurately estimated the effects of fatigue and habit strength on their own coffee drinking
during the study week. The incentive, along with using their own data as an objective
benchmark, were designed to encourage participants to respond accurately and reduce any
influences from social desirability or conversational norms. A total of 78 (70%) participants
responded to the follow up survey and thus provided this rating.
Results
Means, standard deviations, and between-person correlations for key variables appear in
Table 3. The 112 participants (level 2 sample size) produced 3550 individual observations (level
1 sample size), corresponding to an average response rate of 31.7 out of 40 EMA prompts (79%).
On average, participants drank coffee a little over five times during the five-day period
(M = 5.26, SD = 3.09), or approximately once a day. Scores on both habit strength measures
suggested moderate coffee drinking habits. Furthermore, the two measures were strongly
HABITS AND FRICTION
26
correlated with each other, r = .73, 95% CI [.63, .81]. Choice of habit measure did not have a
noticeable impact on the results, and thus analyses are reported using the behavior-frequency-in-
context scale (see table S2 in Appendix A
for analysis results using the SRHI).
Primary Analyses
Results were analyzed using the following multilevel model:

Level 1:
𝑐𝑜𝑓𝑓𝑒𝑒
!"
= 𝛽
#"
+𝛽
$"
𝑓𝑎𝑡𝑖𝑔𝑢𝑒_𝑐𝑚𝑐
!"
+𝑒
!"

Level 2:

𝛽
#"
= 𝛾
##
+𝛾
#$
ℎ𝑎𝑏𝑖𝑡
"
+𝛾
#%
𝑚𝑒𝑎𝑛_𝑓𝑎𝑡𝑖𝑔𝑢𝑒
"
+𝛾
#&
𝑎𝑡𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛
"
+𝑢
#"

𝛽
$"
= 𝛾
$#
+𝛾
$$
ℎ𝑎𝑏𝑖𝑡
"
+𝛾
$%
𝑎𝑡𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛
"
+𝑢
$"

Where i and j represent observations (i) nested within persons (j); coffee refers to whether the
participant did or did not report drinking coffee in the following prompt (i.e. a lead indicator of
coffee drinking meant to capture the prospective association between fatigue and coffee drinking
in the following two hours); This lagged design controls for response biases associated with
concurrently reporting a predictor and outcome; fatigue_cmc is a person mean-centered fatigue
rating at each EMA prompt, computed by subtracting each participant’s mean level of fatigue
from each fatigue rating, so that positive values reflect higher-than-average fatigue for that
person and negative values reflect lower-than-average fatigue for that person; mean_fatigue is
each person’s mean level of fatigue, habit is each participant’s habit strength score, and
attribution is each person’s attribution of coffee drinking to fatigue.

HABITS AND FRICTION
27
Table 1.3

Study 2: Means (M), Standard Deviations (SD), and Between-Person Correlations

Variable
M
(SD)
1 2 3 4 5 6

1. Habit
strength
(BFiC)
11.79
(5.21)

2. Habit
strength
(SRHI)
4.09
(1.35)
.73**
3. Mean
fatigue
3.44
(0.75)
-.05 .09
4. Habit
attribution
36.99
(27.59)
.41** .46** -.07
5. Fatigue
attribution
69.14
(23.43)
-.34** -.24* -.03 -.37**
6. Coffee
count
a

5.26
(3.09)
.54** .44** -.08 .20* -.14
7. Response
rate
31.70
(4.42)
-.05 -.06 -.01 -.16 -.13 .03

Note. Scores for the BFiC ranged from 1-5 and SRHI ranged from 1-7, with higher numbers
reflecting stronger habits. Mean fatigue ranged from 1-6, with higher numbers reflecting higher
fatigue. Habit attribution and mood attribution ranged from 1-100, with higher scores reflecting
stronger attributions. Response rates reflect the number of prompts answered (out of 40
possible), with higher scores indicating higher response rate. Values in square brackets represent
95% confidence intervals.
a
Total number of coffee drinking events reported during the study period.
* p < .05; ** p < .01
HABITS AND FRICTION
28
Habit and Fatigue Effects on Coffee Drinking - Actual
Model estimates for the primary multilevel model are shown in Table 4. To facilitate
interpretation and reduce multicollinearity, all predictors in all regression analyses below were
standardized to have a mean of 0 and standard deviation of 1. Due to convergence issues with the
original frequentist model, we re-specified the main model as Bayesian. To avoid imposing
restrictive priors on the results, we specified uninformative priors for all model predictors (a
prior slope value of 0 with a standard deviation of 100).
To test whether habit determined coffee drinking as well as or better than fatigue, we
compared the standardized coefficients for habit strength and person-mean centered fatigue
(𝛾
#$
and 𝛾
$#
in the model). As anticipated, participants with stronger habits were more likely to
drink coffee, OR = 1.35, 95% CI [1.16, 1.55]. Yet, participants also drank more coffee when
fatigued (within-person), OR = 1.22, 95% CI [1.08, 1.39]. CIs for all other model effects
spanned 1.00 (see Table 4).

HABITS AND FRICTION
29
Table 1.4

Study 2: Coefficient Estimates for Fixed Effects in the Main Multilevel Model

Variable Odds Ratio 95% CI

1. Habit strength 𝛾01 1.35 1.16-1.55
2. Fatigue (person-mean
centered) 𝛾00
1.22 1.08-1.39
3. Fatigue (average person-
level) 𝛾02
0.96 0.84-1.09
4. Fatigue attribution 𝛾03 1.02 0.89-1.18
5. Habit Strength X
Fatigue (person-mean
centered) interaction 𝛾11
0.95 0.84-1.07
6. Fatigue (person-mean
centered) X Fatigue
Attribution interaction 𝛾12
0.96 0.86-1.09

Note. The 95% CI (Credible Interval) represents the range of values that has a 95% chance of
including the population odds ratio.

HABITS AND FRICTION
30
Figure 1.1
Study 2: Likelihood of Drinking Coffee as a Function of Habit and Fatigue

Note. Percentages reflect the likelihood of coffee drinking by the following EMA prompt as a
function of amount of fatigue in the prior prompt and strength of participants’ habit to drink
coffee. Fatigue scores represent each participant’s mean +/- 1 standard deviation.

To determine whether these effects held also for the first coffee of the day, we computed
a separate multilevel analysis that predicted coffee drinking on the first scheduled prompt of each
day (i.e., excluding the participant-initiated morning prompts) from waking fatigue as measured
in morning prompts, habit strength, and an interaction between habit and waking fatigue. The
final sample size for this analysis consisted of 100 participants (level-2 sample size) and 307
responses (level-1 sample size, corresponding to participant-days). Out of the original 498
responses, 54 were excluded because participants reported that they already drank coffee by the
HABITS AND FRICTION
31
time that they completed the prompt, and an additional 145 morning reports were excluded
because they were submitted after the first regular EMA prompt of that day, and therefore could
not be used to predict drinking in that prompt. Early morning fatigue was unrelated to coffee
drinking by the following prompt, OR = 0.96, 95% CI [0.65, 1.42], p = .843. Participants with
stronger habits were more likely to drink first thing in the morning, OR = 2.19, 95% CI [1.38,
3.47], p < .001. The interaction between habit strength and fatigue was not significant, p > .2.
Thus, consistent with our expectations, waking fatigue did not influence coffee drinking on the
first prompt of the day, whereas habit strength did.
Given that the present study measured rather than manipulated habit, we examined
whether habit uniquely determined coffee drinking over and above the contribution of attitudes
or intentions concerning coffee drinking. When these two predictors were added to the main
model, the results remained essentially unchanged, with significant main effects of habit strength
and fatigue but no effects for liking or intentions (see Table S3 in Appendix A). Thus, consistent
with our hypotheses, the effect of habit strength on coffee drinking was not due to liking for
coffee or intentions to drink it.
Habit and Fatigue Effects on Coffee Drinking - Attributed
To test our hypothesis that participants would attribute coffee drinking to fatigue more
than habit, we computed a paired-samples t-test comparing the within-person difference between
each participant’s fatigue and habit attributions. As expected, fatigue attributions were
significantly stronger than habit attributions Mdifference = 32.14, 95% CI [24,18, 40.11], t(110) =
8.00, p < .001, d = 1.26, 95% CI [0.84, 1.67].
To ensure that the attribution results were not due to a failure to recall coffee drinking
events or to ambiguity of attributions for multiple instances of a behavior, we evaluated the
HABITS AND FRICTION
32
single-event attribution measure for participants’ last coffee drinking episode. Consistent with
hypotheses, a paired-samples t-test revealed that participants attributed their most recent coffee
drinking event to fatigue more than habit, Mdifference = 25.86, 95% CI [15.65, 36.08], t(80) = 5.04,
p < .001, d = 0.94, 95% CI [0.50, 1.38].
Suggesting that the attribution findings are not due to social desirability or conversational
norms, a paired samples t-test with the incentivized self-attribution measure, designed to
maximize accuracy, revealed significantly stronger fatigue attributions than habit attributions,
Mdifference = 16.01, 95% CI [5.82, 26.20], t(77) = 3.13, p = .002, d = 0.62, 95% CI [0.19, 1.04].
Thus, incentivized participants still overwhelmingly rated fatigue as more important than habit,
even though incentives reduced the size of this effect (a difference of about 16 when incentivized
compared with about 32 in the non-incentivized measure).
Alternative Habit Measure: Effects of Context-Specific Habit -- Actual
The final survey assessed a novel, within-person measure of habit strength to compare
with the within-person measure of fatigue. A multilevel model predicted actual coffee drinking at
each prompt (yes/no) from fatigue (within-person) and context-specific habit (within-person). As
anticipated, context-specific frequency of coffee drinking was an especially strong determinant
of coffee drinking, OR = 1.86, 95% CI [1.52, 2.32]. As in the main analysis, fatigue predicted
coffee drinking as well, OR = 1.24, 95% CI [1.05, 1.49]. The larger effect of habit compared
with fatigue, as well as the non-overlapping credible intervals, reveal that this measure of habit
exerted a stronger effect on coffee drinking than fatigue.
Correspondence Between Perceived and Actual Effects
This study’s repeated longitudinal design allowed us to estimate not only the overall
effect of fatigue, but also whether the actual within-person effect of fatigue on coffee drinking
HABITS AND FRICTION
33
was related to attribution to fatigue. That is, we assessed whether those who actually drank more
in response to fatigue were aware of this relation in that they were more likely to make fatigue
attributions. If participants’ attributions to fatigue are based on shared cultural theories rather
than personal experience, however, we should find that attributions to fatigue are unrelated to the
actual within-person effect of fatigue on coffee drinking. For this analysis, our main multilevel
model tested whether attribution to fatigue moderated the within-person lagged association
between fatigue and coffee drinking (𝛾12 in the model). A positive slope would suggest that
participants with stronger fatigue-coffee drinking associations also gave stronger fatigue
attributions (i.e., more accurate attributions). Supporting our hypothesis, strength of attribution to
fatigue did not moderate the association between within-participant fatigue and coffee drinking
at the next prompt, OR = 0.95, 95% CI [0.82, 1.11]. Thus, participants strongly attributed to
fatigue regardless of its actual effect on their own coffee drinking, consistent with the notion that
attributions draw on shared cultural lay theories (e.g., Wilson et al., 1982).
Exploratory Analyses - Intensity of Experience and Attributions
Correlational analyses provided additional insight into the accuracy of participants’
attributions (see Table 3). First, attributions to fatigue were not correlated with mean fatigue
levels, r(109) = -.03. This lack of effect is consistent with the weak correlations found in Study 1
between mood intensity and attributions, along with prior findings that attributions often reflect
shared cultural lay theories more than individual experience (Wilson et al., 1982).
Second, on a correlational basis, participants with stronger coffee habits made stronger
habit attributions, r(109) = .41 (behavior frequency in context), r(108) = .46 (self-report habit
index). Thus, showing some evidence for accuracy, participants with stronger habits attributed
their coffee drinking more to habit. To probe this effect, we divided our sample into tertiles by
HABITS AND FRICTION
34
habit strength (weak, moderate, and strong). To enable us to compare the attribution measure
with the actual influences on coffee drinking, we computed our main multilevel model predicting
coffee drinking by extracting within-person effects for fatigue and habit for each participant
(using the context-specific habit measure; to match the attribution measure, model slopes in log-
odds units were standardized to have a range of 0-100%). Weak habit participants attributed on
average a 51% difference in favor of fatigue, compared with an actual difference of 18% in the
opposite direction (i.e., in favor of habit). Participants with moderate habits attributed a 39%
difference in favor of fatigue, compared with a 21% difference in favor of habit. Strong habit
participants attributed 4% in favor of fatigue, compared with an actual difference of 33% in favor
of habit. Thus, participants with stronger habits correctly made stronger habit attributions and
weaker fatigue attributions, but they continued to favor fatigue more than was merited by the
actual predictors of their behavior.
Exploratory Analyses - Downstream Effects of Attribution Accuracy on Well-Being
To identify downstream effects of attributions, we assessed whether attribution to habit
over fatigue is associated with a more positive mood in life, as measured using the average of
participants’ mood reports in the ecological momentary assessment prompts. Attribution scores
were calculated as the difference between each participant’s attributions to mood and to habit, so
that positive scores implied attribution to fatigue more than habit, and negative scores implied
attribution to habit more than fatigue. Attribution scores were moderately and negatively
correlated with mood, r(109) = -0.27, 95% CI [-0.44, -0.09], so that attributing to habit over
fatigue was associated with more positive mood.
To explore whether attributions favoring habit exert a unique effect on mood, or whether
this correlation is simply due to people who are less tired and have stronger habits being more
HABITS AND FRICTION
35
happy, we fit a linear regression model predicting mood from participants’ habit strength
(measured using the behavior-frequency-in-context), mean fatigue levels (using the ecological
momentary assessment), and attribution difference scores. Supporting a unique effect of
attribution, stronger attribution to habit over fatigue predicted more positive mood, b = -0.12,
95% CI [-0.23, -0.02], β = -0.23. Lower mean fatigue levels were also associated with more
positive mood, b = -0.22, 95% CI [-0.31, -0.13], β = -0.40. Habit strength did not show a
discernible association with mood, b = 0.04, 95% CI [-0.06, 0.14], β = 0.04. Thus, these
exploratory analyses suggest that greater recognition of habit in one’s own behavior is associated
with higher well-being.
Discussion
In this second study, participants explained the causes of a mundane everyday action—
coffee drinking—and then tracked their momentary fatigue and coffee drinking over the course
of a typical week. Again, our hypotheses compared the actual influences on behavior to
participants’ behavioral attributions. Fatigue and habit strength had comparable effects on actual
behavior. The strong effect of habit maintained across three different measures of habit.
Furthermore, analyses on the first coffee drink of the day and the within-person habit measure
supported our hypothesis that participants would drink in response to habit more than fatigue. If
participants’ attributions were accurate, they should have featured habit as much or more than
fatigue. However, participants miscalibrated these behavioral influences by attributing their
coffee drinking primarily to fatigue rather than habit.
Notably, Study 2 revealed a bias to overlook habit despite design features to reduce
misattribution. The inaccuracy in self-attribution persisted when participants were incentivized
for accuracy or asked at the end of the study about a recent coffee drinking event rather than their
HABITS AND FRICTION
36
coffee drinking in general. Thus, it emerged despite motivation, an objective criterion, and, at
least for our sample of coffee drinkers, adequate opportunity to observe their behavior given the
frequency of consumption (once a day on average). Further attesting to the robustness of this
attribution bias, coffee drinking is a mundane, everyday action that, unlike our first study, is not
commonly preceded by a salient emotion-inducing experience.
General Discussion
In two studies, participants’ attributions overemphasized inner states and
underemphasized habit. Participants’ actual willingness to donate time in a laboratory task as
well as their everyday coffee drinking were determined as much or more by habits than by inner
states (mood and fatigue, respectively). However, participants’ attributions for why they acted
the way they did emphasized inner states more than habit. Thus, participants appear to be both
undervaluing habit compared with its actual influence on behavior and overvaluing inner states
such as mood and fatigue. This pattern is understandable given the disproportionate value
people place on personal introspections (Pronin, 2009) as well as general information- and
motivation-based tendencies to interpret actions as goal-directed (Rosset, 2008). Through these
forces, people may form socially-shared lay theories about behavior that inform their
attributions. This lure of phenomenology not only biases lay theories but also may have oriented
psychological theories to overvalue salient, motivational determinants of behavior (Duckworth,
Gendler, et al., 2016).
The combination of experimental manipulation in Study 1 and naturalistic observation in
Study 2 provides evidence for the causal role of habits as well as the relevance of this attribution
bias in everyday settings. Furthermore, the results replicated across the different measures of
habit strength appropriate in these different tasks: Study 1’s manipulation of practice along with
HABITS AND FRICTION
37
a reaction time measure; Study 2’s self-report measures of behavioral repetition in a given
context (a determinant of habit formation) and experienced automaticity (a consequence of habit
formation); and Study 2’s exploratory within-person, context-specific habit measure tapping
participants’ history of repetition in specific situations.
A number of features of our research would be expected to maximize participants’
accuracy and minimize biases. Each study assessed attributions for specific recent behaviors,
minimizing biased recall. Furthermore, in both studies, attributional biases were evident even
when participants were incentivized for accuracy, as well as when participants were explaining
others’ behavior (Study 1) and regardless of whether participants were explaining a repeated
behavior in general or a specific recent instance (Study 2). These patterns did not stem from
difficulties in interpreting the habit items or plausibly applying them to behavior, as our
participants made habit attributions for their own and other’s behavior even when accuracy was
incentivized. Additionally, in pretesting for Study 2, habit was identified as one of the top
reasons for drinking coffee.
That our participants discounted habit influences on their own behavior may seem at odds
with the actor-observer effect, which proposes that people have a bias to attribute their own
behavior more to environmental factors that they do others’ behavior (Jones & Nisbett, 1971).
Technically, habits are not solely an environmental influence, as they reside both in a person’s
learned associations in memory and in the environment that triggers a habitual response (Wood
& Rünger, 2016). And yet, even if one considers habits an environmental force, a meta-analysis
of the literature showed that the actor-observer effect on attributions emerges largely under
specific conditions, such as for negative events rather than positive ones (Malle, 2006). The
behaviors examined in this paper were presumably either positive (helping) or neutral (drinking
HABITS AND FRICTION
38
coffee). It may be that people would more readily implicate habits in driving undesirable
behaviors—the proverbial “bad habits.”Relatedly, both studies recruited fluent English speakers,
predominantly residing in the U.S. and U.K. If habit underestimation depends on agency beliefs,
it may be smaller in collectivistic cultures that place less emphasis on individuals and more
emphasis on context (e.g., Crandall et al., 2001).
Similar to other lay theories, habit underestimation may have important downstream
effects. For example, it raises questions about the accuracy of people’s reports that lack of
willpower is the primary reason for their failures to lose weight, save money, and exercise
(American Psychological Association, 2012). It also raises questions about the effectiveness of
common self-regulation strategies: If people misattribute the sources of their behavior, then they
may focus on strategies that affect inner states (e.g. reduce coffee drinking by reducing fatigue)
at the expense of situational strategies that may more successfully modify habits. This would
align with the argument that situational self-regulation strategies are relatively non-salient, which
could lead people to overlook these interventions’ potential (Duckworth, Gendler, et al., 2016).
Suggesting that accurate attributions are beneficial, exploratory analyses in Study 2 revealed that
participants who placed more weight on habit in their attributions also reported more positive
mood. It may be that well-being increases not only with habit performance (Heintzelman &
King, 2019) but also with recognizing habits’ elusive yet pervasive role in daily life.

HABITS AND FRICTION
39
Chapter 2: Americans Discount the Effect of Friction on Voter Turnout

Mazar, A., Tomaino, G., Carmon, Z., A., & Wood, W. (in press). Americans Discount the Effect
of Friction on Voter Turnout. Proceedings of the National Academy of Sciences.

Abstract
Whether or not someone turns out to vote depends on their beliefs (such as partisanship or sense
of civic duty) and on friction—external barriers such as long travel distance to the polls. In this
exploratory study, we tested whether people underestimate the effect of friction on turnout and
overestimate the effect of beliefs. We surveyed a representative sample of eligible U.S. voters
before and after the 2020 election (N = 1,280). Participants’ perceptions consistently
underemphasized friction and overemphasized beliefs (mean d = 0.94). A full 91% of
participants’ open-text explanations for Americans’ turnout involved beliefs, compared with just
12% that involved friction. In contrast, the actual effect of beliefs on turnout was comparable to
that of friction, or at most slightly larger (d = 0.24; or about 1/4th of the perceived effect). This
bias emerged across a range of survey measures (open- and close-ended; other- and self-
judgments) and was implicated in downstream consequences such as support for friction-
imposing policies and failing to plan one’s vote.
Significance Statement
Voter turnout depends both on beliefs (such as partisanship) and friction (such as long
lines at a polling place). In a survey of eligible U.S. voters before and after the 2020 election, we
uncovered a mismatch between actual and perceived drivers of turnout: Participants
underestimated the role of friction and overestimated the role of beliefs in shaping turnout.
Furthermore, participants who were prone to this bias tended to support policies that increase
HABITS AND FRICTION
40
friction on voting, such as signature matching requirements. These results inform the ongoing
public debates surrounding voting access.
Significance Statement
Voter turnout depends both on beliefs (such as partisanship) and friction (such as long
lines at a polling place). In a survey of eligible U.S. voters before and after the 2020 election, we
uncovered a mismatch between actual and perceived drivers of turnout: Participants
underestimated the role of friction and overestimated the role of beliefs in shaping turnout.
Furthermore, participants who were prone to this bias tended to support policies that increase
friction on voting, such as signature matching requirements. These results inform the ongoing
public debates surrounding voting access.
Introduction
Casting one’s vote is the defining act of a participatory democracy. And yet, people
sometimes fail to vote when faced with seemingly surmountable barriers, such as longer travel
distances to the polls (Cantoni, 2020), longer wait lines (Stewart & Ansolabehere, 2015),
shortened poll hours (Potrafke & Roesel, 2020), and even rain (Fujiwara et al., 2016). The effects
of such barriers can be particularly decisive for those who have relatively few resources, such as
those with lower income or who do not own a car (Cantoni, 2020; Fujiwara et al., 2016). While
turnout is known to be influenced by such barriers as well as beliefs (such as political ideology;
Jung, 2017; Stevens et al., 2008) little is known about how voters perceive these influences. That
is, what do voters think determines turnout? And do these perceptions influence downstream
outcomes like policy support? To the best of our knowledge, this is the first research to examine
Americans’ such perceptions of what drives turnout.
HABITS AND FRICTION
41
Understanding these perceptions is especially pressing now, as the U.S. witnesses heated
debates surrounding voting rights (Brennan Center for Justice, 2021). Some (mostly
conservatives) support policies that may restrict voting, whereas others (mostly liberals) support
policies that make voting more accessible (Pew Research Center, 2021). If Americans
underappreciate the role of friction in driving turnout, then they may be more likely to support
policies that make voting difficult. Given the tangible impact of friction on turnout, one might
expect Americans to recognize its influence. Yet, people are not always aware of what drives
their behavior (Wilson, 2002) and may overlook friction, especially given Americans’ strong
beliefs in self-control and intentional action (Koehler & Poon, 2006; Pew Research Center, 2014;
Rosset, 2008).
Such a pattern is suggested by the 2020 presidential campaigns’ spending, most of which
went toward media (Open Secrets, 2021b, 2021a) that overwhelmingly targeted beliefs: 98% of
Biden’s Youtube videos and 95% of Trump’s mentioned political beliefs, whereas only 7% and
13%, respectively, mentioned forms of friction (see Section S1 in SI Appendix). More generally,
Americans tend to assume that actions (such as voting) largely reflect intentions (Rosset, 2008)
and interpret their own behavior as driven by beliefs and intentions (Pronin, 2009). Americans,
then, might discount the effect of friction and inflate the effect of beliefs.
The present exploratory research examined eligible U.S. voters’ predictions about the
relative effects of beliefs and friction on turnout. We used a pre-post survey design administered
immediately before and after the 2020 presidential election. Participants were a representative
sample of eligible voters (N = 1,280) from ten election-competitive U.S. states. Participants
estimated the effect on turnout of beliefs (e.g., ideology, voting is a civic duty) and friction (e.g.,
conflicts with work or childcare).
HABITS AND FRICTION
42
To thoroughly explore this question, we used three measures of participants’ perception
of turnout drivers: (a) Open-ended measures. Participants listed “major influence(s)” on turnout,
and then rated the importance of each listed influence, (b) Composite measures. Participants
rated the importance of 12 specific forms of friction (e.g., conflicts with work or childcare) and
beliefs (e.g., ideology, sense of voting as a civic duty) on turnout, and (c) Single-item measures.
Participants rated the overall importance of beliefs and friction. All measures were completed
both with respect to others (Americans’ turnout in general) and self (each participant’s own
turnout).
In the pre-election survey, participants also reported their beliefs (e.g., ideology, voting is
a civic duty) and friction (e.g., whether they own a vehicle). They also indicated support for
friction-related policies, and (as a behavioral measure shown to a randomly-selected half of the
sample) whether they would like to take part in a vote planning exercise.
After the election, participants reported whether they voted and rated any aleatory
election day friction they experienced (i.e., friction that might emerge unexpectedly, such as an
unpredictable work or childcare conflict on election day). They were further informed that one
participant would be randomly selected to receive a $100 bonus if they accurately completed
both parts of the study. To estimate the actual impact of friction and beliefs, we fit models
predicting post-election turnout from the actual friction and belief measures. All study materials,
data, and analytic code are openly available via OSF (see Methods and Materials).
Results
Actual Effects of Beliefs vs. Friction
Table 1 shows the variables used in analyses of the actual effects of beliefs and friction
on turnout. Given that mail-in voting involves different types of friction than in-person voting
HABITS AND FRICTION
43
(e.g., addressing the envelope properly), our analyses focus on in-person voters (987 participants,
77% of the sample). Logistic regression models predicting turnout from beliefs (in one model)
and friction (in a separate model) revealed that each exerted a comparable influence on turnout
(using Tjur’s R
2
; R
2
Belief = .18, R
2
Friction = .19; see Table 2). In a more conservative test, we fit a
narrower friction model excluding variables reflecting knowledge of how/where to vote as well
as prior experience with participants’ intended mode of voting and at their current address.
Although this narrow model explained less variance than the full friction model (as would be
expected when reducing the number of predictors), friction continued to substantially predict
turnout, R
2
= .09.
Note that self-reports inflate the effects of beliefs on turnout, because those with stronger
beliefs are more likely to falsely report having voted (Ansolabehere & Hersh, 2012). Our results,
then, likely overestimate actual belief effects, posing a conservative test of our hypothesis that
people overestimate the effect of beliefs relative to friction.
Perceived Effects of Beliefs vs. Friction
In contrast to the actual effects presented in the previous section, across all perception
measures, participants rated beliefs as markedly more important than friction in driving turnout
(see Table 3 and Figure 1). An especially stark gap emerged in participants’ open-ended
responses listing drivers of Americans’ turnout. A full 91% of participants mentioned at least one
belief, whereas only 12% mentioned friction (Figure 1A). When participants rated on a 10-point
scale the importance of each factor that they listed, a similar pattern emerged, with participants
assigning beliefs far more importance, M = 8.46 (SD = 3.03), than friction on, M = 1.02 (SD =
2.82), d = 1.59 (Figure 1B). When participants were explicitly asked to rate the impact of friction
as well as beliefs, similar (albeit more moderate in magnitude) effects favoring beliefs emerged
HABITS AND FRICTION
44
in: (a) composite ratings of specific types of beliefs and friction sources (Figure 1C), and (b) a
single-item measure of beliefs versus friction (Figure 1D). These analyses included the entire
sample, but qualitatively identical results emerged in analyses of participants who intended to
vote in-person (to facilitate direct comparison with the analyses of actual effects in the previous
section; see Table S1 in SI Appendix). Similarly, attribution to beliefs over friction maintained in
post-election measures, even among participants who did not vote (see Tables S4 and S5 in SI
Appendix).
To directly compare participants’ perceptions to the actual effects of beliefs and friction
on turnout, R
2
scores for actual turnout (computed using linear regression) were converted into
standardized mean difference effect sizes (Borenstein et al., 2021). The differences between
beliefs and friction in predictive turnout ranged from d = 0.04 in favor of friction (when using the
full friction model), to d = 0.17 in favor of beliefs (when using the narrow friction model). In
contrast, measures of perception skewed strongly in favor of beliefs over friction. For example,
when using open-text measures, participants’ estimates revealed a difference score of d = 1.59—
more than 9 times larger than would be justified by even the most conservative model
specification. In sum, the actual effects of friction and beliefs on turnout revealed at most a
small-medium difference in favor of beliefs, whereas participants generally perceived a large
difference in favor of beliefs over friction.

HABITS AND FRICTION
45
Figure 2.1
Perceived Effects of Beliefs and Friction on Turnout

Note. Difference between the mean perceived effects of beliefs (purple) and friction (yellow) on voter
turnout. The different panels represent: (a) number of open-ended responses participants listed, (b)
importance ratings participants gave for their open-ended responses, (c) composite importance
measures, and (d) single-item importance measures. Effect sizes are in Cohen’s d.

Support for Friction-Related Policies
We next explored the correspondence between underestimating the importance of friction
relative to beliefs (using the composite measure of others’ voting) and support for friction-
HABITS AND FRICTION
46
relevant policies. We averaged the policy support items (e.g., automatic voter registration, exact
signature match) into an index ranging from 1 to 7 (1 = strong opposition to friction-increasing
policies; 7 = strong support for friction increasing policies). This index was supported by an
exploratory factor analysis, which yielded a single latent factor with an eigenvalue greater than 1
(1.79). A regression model predicting support for friction-increasing policies from friction-
beliefs importance difference scores revealed a modest effect, b = 0.15, 95% CI = [0.09, 0.20], .
Thus, participants who underestimated friction (relative to beliefs) were more likely to support
policies that could make voting more difficult. In a more conservative test, this effect held in a
model predicting friction discounting for others controlling for gender, age, income, education,
race, voter fraud concerns, past voting experience, partisanship, and ideology, b = 0.05, 95% CI
= [0.003, 0.10], . In additional conservative tests, these effects largely held when we predicted
each policy support measure individually, rather than as a single factor, with and without
controls (see table S6 in SI appendix). In short, voters who discounted the effect of friction on
others’ turnout tended to support policies that would increase friction on voting.
Planning
Vote planning can increase turnout, (Nickerson & Rogers, 2010) in part by helping
people anticipate friction. Half of our participants were given the option of a vote planning
exercise. Those who opted to plan held stronger political beliefs (indicated by an index averaging
across all belief variables) than those who did not, d = 0.38, suggesting that they might have cast
their ballot regardless. In contrast, planners did not differ from non-planners in aspects of friction
such as perceived voting difficulty (i.e., subjective friction), d = 0.08, or voting knowledge, d =
0.10 (see Figure 2). These results suggest that when deciding whether to plan their vote, our
participants were insensitive to friction: Those who might have benefited the most from vote
HABITS AND FRICTION
47
planning—those who had weak political beliefs or who would have a difficult time voting—were
not more amenable to planning their vote.
Figure 2.2
Differences Between Vote Planners and Non-Planners
Note. Belief strength represents the composite strength of beliefs, calculated as the mean of partisanship
strength, ideology strength, (dis)like for Donald Trump, (dis)like for Joe Biden, support for one’s favorite
candidate, believing that one’s vote matters, and civic duty. Effect sizes are in Cohen’s d.

Discussion
In a large-scale exploratory study of eligible U.S. voters, we compared the actual and
perceived roles of political beliefs (e.g., ideology, voting is a civic duty) and external friction
(e.g., conflicts with work or childcare) on voter turnout in the 2020 election. To the best of our
knowledge, this is the first study examining Americans’ perceptions of what drives voter turnout.
We found that potential voters overwhelmingly attributed turnout to beliefs over friction, with
this tendency emerging both for perceptions of others’ turnout as well as their own.
The friction discounting we observed was particularly striking in participants’ open-
ended responses, with nearly all participants citing beliefs and very few referring to friction. Our
HABITS AND FRICTION
48
participants seemed not to spontaneously think of friction unless explicitly reminded by question
wording. Moreover, even when explicitly prompted to think of friction (in the multiple-choice
items), participants downplayed the effects of friction relative to beliefs. This is also evident in
our analyses of Republican and Democrat presidential campaign videos, which revealed that
both campaigns gave minimal attention to friction. Participants’ interest in vote planning also
proved to be insensitive to friction.
This discounting of friction is notable given the importance and attention that is placed on
elections in the United States. This underestimation may stem from Americans’ strong beliefs
that their actions are primarily driven by autonomy and intentionality (Koehler & Poon, 2006;
Pew Research Center, 2014). It would be interesting to explore differences between these
perceived versus actual effects in less autonomy-focused cultures.
Crucially, the tendency to overlook friction in favor of beliefs in determining voter
turnout was tied to policy support. Discounting friction was linked to increased support for
policies that could dampen turnout, even in conservative analyses that accounted for an array of
demographic and political covariates. This relation between recognizing friction and policy
support informs current debates around voting rights (Brennan Center for Justice, 2021). Liberals
often view conservatives’ support for friction-increasing policies as deliberate attempts at voter
suppression. Yet our research suggests that at least some support for these policies may simply
reflect failure to recognize these policies’ harmful impact on would-be voters.
In conclusion, our work finds that Americans underestimate the impact of friction relative
to beliefs on voter turnout. This robust tendency may inform current debates around voting
rights, requirements, and restrictions.
HABITS AND FRICTION
49
Materials and Methods
All data, code, materials, and auxiliary results are available via the following repository:
https://osf.io/4rk2c/?view_only=4bd874646326482cb50628d8d4b7a797.
The research was approved by INSEAD’s institutional review board, and all participants gave
their informed consent to participate.
Participants.
Sample demographics (weighted and unweighted) can be found in Table 4. Participants
were 1,280 eligible U.S. voters recruited via the Prodege and Prolific Academic online sample
pools. An additional 1,142 participants completed the pre-election survey but not the post-
election survey (53% retention) and were therefore not included in analyses. Participants were
U.S. residents of 10 competitive states: Arizona, Iowa, Maine, Michigan, Nebraska, Nevada,
New Hampshire, North Carolina, Pennsylvania, and Wisconsin. We defined competitive states as
ones in which pre-election polling consistently showed a less than 10% margin between Donald
Trump and Joe Biden (the main presidential nominees). Only competitive states were selected
for analysis because most U.S. states employ a “winner-takes-all” system in the Electoral
College that can dissuade potential voters from turning out for parties or nominees unlikely to
win the majority of their state’s vote.
Iterative Proportional Fitting (also known as raking; Ruschendorf, 1995) was used to
post-stratify the survey sample to match the demographics of the ten included states on gender,
age, race, education, and state, based on data from the 2019 American Community Survey (U.S.
Census Bureau, 2019). Nonetheless, qualitatively identical results were obtained for the
unweighted sample (see Tables S2 and S3 in SI Appendix).

HABITS AND FRICTION
50
Design.
Our study used a pre-post survey design, with participants surveyed before and after the
2020 U.S. presidential election. Data collection for the pre-election survey took place between
October 15-24, 2020, and data collection for the post-election survey took place between
November 3-7, 2020.
Measures.
The primary measures used in the study are listed below. See the OSF repository for
complete materials.
Pre-Election Survey
Beliefs
Partisanship: “Do you think of yourself as a…” (7-point scale; 1 = strong Democrat, 7 =
strong Republican).
Ideology: “In general, how would you describe your own political viewpoint?” (7-point
scale; 1 = very liberal, 7 = very conservative).
Preferred candidate: “If the 2020 presidential election were held today, who, if anyone,
would you vote for if the candidates were:” (Joe Biden, the Democrat; Donald Trump, the
Republican; Another candidate; wouldn’t vote; Don’t know/not sure; Prefer not to say)
Candidate support: If participants reported in the previous item that they intended to vote
for either Biden or Trump, they rated how they would “describe [their] support for [name of
preferred candidate” (3-point scale; 1 = strongly support, 3 = slightly support; Don’t know/not
sure).
Believing that one’s vote matters: “How much do you think that your vote matters?” (7-
point scale; 1 = My vote doesn’t matter at all, 7 = My vote really matters).
HABITS AND FRICTION
51
Civic duty: “To what extent do you think voting is your duty as an American?” (7-point
scale; 1 = not at all my duty, 7 = very much my duty).
Liking for main candidates: “To what extent do you like or dislike Joe Biden” and “to
what extent do you like or dislike Donald Trump” (7-point scale; 1 = strongly dislike, 7 =
strongly like).
Friction
Subjective friction: “How easy or difficult it will be for you to vote in the upcoming
election” (7-point scale; 1 = very easy to 7 = very difficult) and “how much of a hassle it would
be for you to vote in the upcoming election” (1 = not a hassle at all to 7 = a huge hassle).
Time demands: “How many minutes do you expect it will take you to complete your
remaining voting process? This includes every step necessary for you to successfully vote” (0-10
min; 10-20 min; 20-40 min; 40-60 min; 60-80 min; 80-100 min; 100-120 min; 120-150 min; 150-
180 min; 3-6 hours; over 6 hours).
Travel time: “How many minutes do you anticipate it will take to travel to your
designated polling place in the 2020 election?” (under 5 minutes; 5-10 minutes; 10-20 minutes;
20-30 minutes; 30-40 minutes; 40-50 minutes; 50-60 minutes; over an hour).
Intended mode of travel to the polls: “If you were to vote in person in the 2020 election,
how would you get to the polling place?” (by car; walking; by public transit; other).
Vehicle ownership: “Do you own a motorized vehicle (car, truck, van, or motorcycle)?”
(yes; no, but I usually have access to a vehicle (for example via a family member); no).
Work conflict (measured post-election): “Regardless of whether you voted or not, would
you have needed to ask for time off from work to vote?” (yes; no; other).
HABITS AND FRICTION
52
Childcare conflict (measured post-election): “Regardless of whether you voted or not,
would you have needed childcare to vote (that is, to ask someone who doesn't live in your
household to watch over your children)?” (yes; no; other).
Voting knowledge: Participants rated how confident they are that they know how to vote
in-person or by mail, based on their intended mode of voting (7-point scale; 1 = Not at all
confident to 7 = very confident). They also rated whether they know where their polling place is
(yes; probably; no). These measures were not included in the narrow friction model.
Past voting: Participants reported whether they had voted before in a U.S. election,
whether they had voted before in their intended mode of voting, and whether they had voted
before while living at their current address (yes; no; unsure). These measures were not included
in the narrow friction model.
Perceived turnout drivers
Participants reported their perceptions of the importance of beliefs and friction in driving
turnout. Participants completed three types of importance measures: (a) a free response measure
in which participants listed important drivers of turnout and then rated each driver’s importance,
(b) a composite measure in which participants rated the importance of specific types of beliefs
and frictions on a 7-point scale, and (c) a single-item measure in which participants rated the
overall importance of beliefs and friction. Participants completed each importance measure
twice: Once with regard to their own turnout (self-rating) and once with regard to the turnout of
“people in the United States” in general (other-rating).
Free response importance measures. In a series of free response items, participants listed
up to five factors that would impact whether [“you”/”people in the United States”] would vote or
not. We then showed them the factors they had just listed and asked them to rate how important
HABITS AND FRICTION
53
each factor was on a 10-point scale. All responses were coded by two research assistants
(unaware of our hypotheses) as mentioning friction, attitudes, both, or neither.
Composite importance measures. Participants indicated the importance of specific beliefs
(e.g., “How strongly [I / they] identify as Republican or Democrat,” “Believing voting is [my /
one’s] duty as an American”) and specific forms of friction (e.g., “How long it takes to get to the
nearest polling place,” “Weather on election day”) in influencing their own and others’ turnout
(7-point scale; 1 = not at all important, 7 = extremely important). We then averaged these
responses by category (friction or beliefs) to create composite importance indices for beliefs and
for friction.
Single-item importance measures. Participants rated the overall importance of beliefs
(“How strong [your/their] political attitudes are and how important [you/they] think it is to vote”)
and friction (“How easy it is for [you/them] to vote”) in driving turnout, using slider scales
ranging from 0% (Not at all important) to 100% (Extremely important). In the other-rating
single-item measure, participants were incentivized for accuracy. Specifically, they were told
that their estimates will be compared against the actual study’s results, with accurate estimates
(within -+5% of the actual results) awarding participants a chance of winning one of 10 prizes
worth $10 each.
Voting-related items
Intended mode of voting: “If you do vote in the 2020 election, how will you vote?” (In
person; By mail/absentee; Request ballot by mail and drop it off (at drop box or polling place)
on election day; Request ballot by mail and drop it off (at drop box or polling place) before
election day; Don’t know/not sure).
HABITS AND FRICTION
54
Perceived likelihood of voting: Measured using a 7-point scale (1 = definitely not going to
vote to 7 = definitely going to vote), as well percentage point scale items (0-100%, in 10%
intervals) for likelihood of voting as well as likelihood of not voting.
Political variables
Friction-related policy support. Participants were asked “To what extent do you support
or oppose the following policies?,” for six policies: Automatic voter registration, adding polling
places, hiring more poll workers, reducing polling places that serve relatively few people, exact
name match, and exact signature match (last three items reverse-coded; see full text in OSF
repository materials).
Partisanship. “Do you think of yourself as a…” (7-point scale; 1 = Strong Democrat, 7 =
Strong Republican; “Not sure”).
Ideology. “In general, how would you describe your own political viewpoint?” (7-point
scale; 1 = Very liberal, 7 = Very conservative; “Not sure”).
Concerns about voter fraud and voting access. Participants rated their agreement with the
following statements: “Barriers that make it difficult to vote are a serious problem in the United
States right now” and “Voter fraud is a serious problem in the United States right now” (7-point
scale; 1 = strongly disagree, 7 = strongly agree).
Planning
A randomly-selected 50% of participants read the following prompt and responded with
“yes” or “no”: “Would you like to go through a 5-10 minute process to help you plan your vote
and think in advance about overcoming obstacles to voting? Note that participation in this
process is voluntary - it is completely up to you. You will not be paid for time spent on this
process.” Those who agreed were shown the link to a vote-planning website as part of their
HABITS AND FRICTION
55
debrief screen, but were instructed not to click on it, as doing so might prevent their survey
completion from being recorded.
Demographics
Participants reported demographics, including their gender, age, race/ethnicity, highest
level of education attained, and household income.
Post-Election Survey
Turnout
Which of the following statements best describes your vote in the recent (2020) election?
(I did not vote in the election; I tried to vote but did not or was not able to; I voted in the
election)
Aleatory election day friction
Participants reported: whether they worked on election day, whether they would have
needed childcare to vote (regardless of whether they actually voted or not), and whether they
would have needed to ask for time off work to vote (yes; no; other).
Turnout determinant predictions
Participants then completed the same free response importance measure of perceived
drivers of turnout from part 1, but in the past tense (ex. “In your view, what was a major
influence on whether you voted or not?”). As in the pre-election survey, participants completed
this both for themselves and for “people in the United States” in general.

HABITS AND FRICTION
56
Chapter 3: Actual and Perceived Effects of Friction on App Use Over Time
Mazar, A., Babur, B., Duckworth, A., & Wood, W. (in preparation). Actual and Perceived
Effects of Friction on App Use Over Time.

Abstract
People in developed countries spend approximately two months a year on mobile
devices, and many rank excessive smartphone use as a key self-control challenge (Allcott et al.,
2021). However, phone use may be strongly habitual and therefore difficult to regulate
volitionally. The present study tested whether a subtle situational intervention—increasing
friction on app use by moving the app icon to a new location—could decrease app use.
Participants (N = 216 participants, 4,434 participant-days) were randomly assigned to one of four
friction interventions, a motivation intervention, or a control condition, and completed daily
diaries for the next two weeks. Analyses did not detect an effect of either friction or motivation
on app use: app use across conditions tended to reduce immediately post-intervention but quickly
revert to baseline. However, participants estimated that motivation would be far more effective
than friction in reducing app use. In addition, app use reverted to baseline even as intentions to
reduce app use remained strong, suggesting that this rebound effect is not driven by initial
positive intentions faltering over time. The findings highlight the intricacies of generating and
maintaining behavior change over time in field environments.

HABITS AND FRICTION
57
Introduction

“I've uninstalled a ton of apps from my phone that I felt were wasting my time....and I've
turned off notifications on anything that...wasn’t timely and important to me right now.
It’s for the same reason that I don’t keep cookies in my pocket.”

—Justin Rosenstein, ex-Facebook and Google (2020)

Fifteen years ago, something changed in the way humans live their lives. Since 2007,
when the first iPhone was introduced to the world, people across the world have been spending
an ever-larger chunk of their time staring at ever-smaller rectangles. Americans now spend on
average a full quarter of their waking hours on mobile devices (Statista, 2021), with other
countries (both developed and developing) producing similar figures (MMD Labo, 2020; Vivo &
Cybermedia Research, 2019).
And yet, this is not because people wholeheartedly want to use their phones that often.
Americans rate excessive smartphone and social media use as some of their top self-control
challenges, right alongside unhealthy eating and failing to exercise (Allcott et al., 2021). Though
this meteoric rise in smartphone use could be traced to myriad factors, one key factor is likely
smartphones’ remarkable ease of use. Phones and the many rewards that they provide are
accessible at any time and location. Further, digital technology products are crafted and refined
based on an ethos that elevates simplicity above all else (digitaltrends, 2020).
In sum, technology products are designed to minimize friction, or external barriers to
behavior (Mazar et al., 2021). This lack of friction produces phone use patterns that are
HABITS AND FRICTION
58
extremely frequent and fragmented: objective phone use logs find that the median time devoted
to a single digital activity ranges between 10 to 20 seconds (Reeves et al., 2021). When
frictionless technology is coupled with such incessant, repetitive use, it lends itself to habit
formation. The resulting habits can then make once-unfamiliar actions familiar and easy (Avni‐
Babad, 2011). In this account, ease and habit strength reinforce each other: ease facilitates habit
formation and growing habits make phone use ever-easier. This self-reinforcing cycle may
become exceedingly difficult to disrupt, generating the self-control challenge that many
experience as their digital devices beckon.
If undesirable phone use stems from frictionless design, unhindered by situational
barriers, then phone use should decrease when subject to increasing friction. It seems a truism
that when a behavior is made difficult, people are less likely to act on it. And yet, friction has
received relatively little attention in psychology. Although friction does receive more attention in
the field of behavioral economics, such research often emphasizes one form of situational
friction—defaults—and specifically defaults that are administered top-down, such as by a
government or company (Blumenstock et al., 2018; Ebeling & Lotz, 2015). Thus, we still know
little about the effects of other forms of friction, particularly when friction is self-administered.
Perhaps most importantly, we know surprisingly little about how friction shapes repeated
behaviors. For example, one way to impose friction on unhealthy eating is to intervene on food
proximity and availability, making unhealthy foods difficult to access and healthy ones easy to
access. However, the majority of such studies examine how friction influences eating in one-off
lab studies (Hollands et al., 2019). By studying one-off decisions in unfamiliar environments,
such studies fail to capture the dynamics that govern how friction interacts with existing
environments in daily life over time—an important distinction given that many interventions that
HABITS AND FRICTION
59
effectively suppress habits short-term can rapidly lose efficacy (Mantzari et al., 2015; Wood &
Neal, 2016).
The present research bridges these gaps by testing the longitudinal effects of a subtle
friction intervention—changing one’s app icon location on one’s home screen—on habitual app
use. Friction is an especially promising intervention for habitual behaviors since friction is a
situational rather than cognitive intervention, meaning that it involves changing the response
environment in advance of the target behavior (Duckworth, Gendler, et al., 2016). Because habits
are under heightened situational control and diminished intentional control (Gardner, 2015;
Mazar & Wood, 2018), increasing situational friction may be particularly effective at changing
habits, and most importantly, maintaining the change over time.
This research additionally probes the mechanisms that underlie friction by varying the
frequency and dose of friction interventions, examining how these influence intervention uptake
and efficacy over time. Finally, I examine people’s perceptions of each strategy’s effectiveness.
Perceived efficacy is especially important in self-regulation. When interventions are
administered in a top-down fashion by governments and companies, an intervention may be
effective even if people do not perceive it to be effective, or indeed are not even aware of it. On
the contrary, when interventions are self-administered, even effective interventions provide little
benefit if people fail to recognize their efficacy and make use of them.
Situations in Social Psychology
Social psychology often prides itself on its putting situations front and center. Kurt
Lewin’s (1936) field theory, distilled in the famous equation, B = f(P, E), postulated that
behavior (B) is shaped by an interaction between person characteristics (P) and environmental
HABITS AND FRICTION
60
forces (E). Lesser known, Lewin’s applied work went beyond this general formulation, charting
in exquisite detail the gamut of situational forces that shape daily food purchases (Lewin, 1947).
And yet, nearly a century after Lewin, the field still lacks a comprehensive theory of
situations (Kenny et al., 2001; Yang et al., 2009), and particularly a framework for harnessing
situations for behavior change. As a past president of the Society for Personality and Social
Psychology stated, social psychology “has yet to develop a clear, consensual definition or
taxonomy of what situations are, how they might be systematically compared, and which ones
are most influential in what ways” (Reis, 2008, p. 312).
In contrast with the social environment, which social psychology has outlined with
notable success (e.g., credible vs. non-credible sources, Eagly & Chaiken, 1993; descriptive vs.
subjective norms, Schultz et al., 2007), the physical environment is largely uncharted. This is
illustrated in a cursory search through the Behavior Change Technique Taxonomy Database—an
online database of behavior change interventions coded according to 93 established categories
(BCT Taxonomy Interventions, 2016; Michie et al., 2013). At the time of my search (November
2020), the database contained 210 interventions that involved goal setting compared with 21
studies that involved reducing or avoiding cues—a tenfold difference. Similarly, 171 studies
involved providing information about health consequences, compared with 55 studies that
involved restructuring the physical environment.
2

The relative paucity of situational theory may be tied to the English language’s “rich
vocabulary for describing traits but...impoverished vocabulary for describing situations” (Ickes et
al., 1997). For example, a person can be described as “stingy”, but English speakers are at a loss
to concisely describe a situation that makes one less generous. Yang et al. (Yang et al., 2006)

2
With the caveat that this is not a systematic review of all existing behavior change interventions, and the
sample of studies included may not be representative of behavior change interventions as a whole.
HABITS AND FRICTION
61
addressed this challenge by basing their situation taxonomy on Chinese idioms (with Chinese
presumably having a richer vocabulary for situations). Perhaps more concretely, the lack of
theorizing about situations may stem from an incentive structure that rewards researchers who
frame their findings in the broadest terms—which often means the most de-contextualized terms
(Simons et al., 2017; Yarkoni, 2019).
In lieu of situations, psychology historically emphasized change that results from
effortful self-regulation. Many classic self-regulation theories posit that self-control failures
occur when hot, visceral processes overpower cold, cognitive ones (Loewenstein, 1996; Metcalfe
& Mischel, 1999). The roots of such formulations run deep, harking back to ancient Greek and
Christian ideals contrasting carnal and godly elements of the self (Black, 2014; Plato, 347
B.C.E./1967). Such dualistic theories of self-regulation suggest that successful self-regulation is
the result of long-term goals forcefully subduing short-term desires. For instance, the strength
model of self-control assumes that self-control is effortful and costly, making it into a limited
resource (Baumeister et al., 2007).
Mounting evidence, however, suggests that effortful inhibition of impulses plays a rather
minor part in how individuals achieve long-term goals. One meta analysis found only modest
correlations between different measures of self-control (Duckworth & Kern, 2011). Furthermore,
the correlation between self-report and other-report measures of self-control was practically
identical to the correlation of self-report measures with themselves. This seeming lack of
difference between self- and other-reports suggests that self-reports of self-control do not rely on
personal, internal experiences of cognitive inhibition, but rather on outwardly-visible markers of
successful self-regulation (e.g. good grades, frequent exercise)—markers that are available both
for self and others.
HABITS AND FRICTION
62
This result is perhaps not surprising when considering what is by far the most common
self-control measure—the self-control scale (Tangney et al., 2004). High scores on this self-
report measure correlate with nearly every imaginable measure of life success, including better
grades, healthier relationships, and lower rates of psychopathology (Tangney et al., 2004). Even
at face validity, however, the scale’s items seem to tap success at self-regulation—a downstream
consequence of effortful self-control—more than the effortful cognitive self-control process that
is thought to underlie this success. Items that ask participants to report on observable behavior
such as, “I say inappropriate things,” and “I am able to work effectively towards long-term
goals” may thus not reflect effortful inhibition as much as success in working towards long-term
goals.
Further support for this notion of self-control as more than effortful inhibition comes
from Ecological Momentary Assessment (EMA) studies (Shiffman et al., 2008). EMA methods
call for participants to repeatedly report on their immediate or recent experience as they go about
their day. By asking participants to report on recent experience, these methods lessen the impact
of recall biases that may skew conventional surveys (Stone et al., 2007). In contrast with studies
that measure self-control using conventional surveys, experience sampling studies do not find
consistent associations between trait self-control and effortful inhibition in daily life. In one
EMA study, participants reported several times a day on their current desires as well as whether
these desires conflicted with other goals and whether they attempted to inhibit them (Hofmann et
al., 2012). Participants who scored high on the self-control scale reported fewer desires that
conflicted with their goals and consequently less effortful inhibition, meaning that those
participants simply did not experience motivational conflict very often. Another EMA study
corroborated this finding and additionally measured goal attainment, finding that poor goal
HABITS AND FRICTION
63
attainment was unrelated to effortful inhibition and was instead predicted by momentary reports
of temptation (Milyavskaya & Inzlicht, 2017). Simply put, the most successful self-regulators
were the ones who seldom experienced conflicting desires.
Psychology’s emphasis on effortful self-control can account for the dearth of research on
situational self-regulation. And yet developments in other, related fields show welcome progress
that could inspire understanding of situational processes in behavior change. In personality
psychology, for example, the DIAMONDS framework is a situation taxonomy that divides
situations into eight basic factors (e.g., positivity, intellect, and sociality, Rauthmann et al.,
2014). This framework samples a broad array of situations across various populations and has
shown considerable construct and predictive validity (Funder, 2016; Rauthmann et al., 2014).
However, the resulting taxonomy may be most suitable for understanding personality traits and
the situations in which people would or would not act in line with these traits. This is not
surprising given that the Riverside Situational Q-Sort—the measure underlying the taxonomy—
was designed so that its situation categories correspond to personality traits (Funder, 2016).
Thus, it may be of limited use for behavior change. For example, the finding that people are
more likely to “work[] hard” in situations high on duty (Rauthmann et al., 2014, p. 699) does not
provide clear guidance for how to alter situations to, say, help someone study for an upcoming
exam. In sum, behavior change science could usefully extend prior work on situations to
generate systematic interventions for changing behaviors—especially habitual ones.
Situations in Subjective vs. Objective View
Another obstacle to development of situational theory is that even when self-regulation
research does account for situations, it tends to emphasize subjective rather than objective
features of a situation. A situation could be characterized objectively, based on its observable
HABITS AND FRICTION
64
characteristics, or subjectively, via people’s perceptions of the situation. This distinction was
long noted in psychology, with Murray (Murray, 1938) differentiating alpha press, the objective
effect of a situation on behavior, from beta press, a person’s subjective perception of a situation.
Accordingly, the dominant theoretical frameworks in self-regulation research reference
both subjective and objective aspects of a situation. Social cognitive theories, for example, treat
self-regulation as determined by a person, their behavior, and their environment (Bandura, 1991;
Zimmerman, 2000). A similar pattern emerges for the Theory of Planned Behavior (TPB), likely
the most popular model of behavior (one influential TPB review paper has over 100,000 citations
per google scholar, Ajzen, 1991). Although not strictly limited to self-regulation, TPB has often
been used to model self-regulation (Ajzen, 1991). TPB maintains that behavior is caused by
attitudes, norms, and perceived behavioral control (Godin & Kok, 1996). One’s perceived
control over their behavior, in turn, is dictated by both situational constraints (i.e., how much
control they actually have over the behavior) and perception of these constraints.
In practice, however, the situation and its subjective perception are rarely measured
separately. Instead, TPB studies usually measure just perceived behavioral control, with the
assumption that these perceptions are mostly accurate and so “can often be used as a substitute
for a measure of actual control” (Ajzen, 1991, p. 184). Similarly, although social cognitive
theories ostensibly account for environmental effects, one influential overview devotes 9 pages
to a discussion of self-regulation using cognitive variables but a mere 1.5 pages to self-regulation
using situational variables, with the latter section consisting mostly of famous authors’ anecdotes
(Zimmerman, 2000). This may reflect Bandura’s (Bandura, 1991) own assertion that “Most
human behavior, being purposive, is regulated by forethought” (p.248). To sum, in self-
regulation research, situations are as ubiquitous in theories as they are absent in practice.
HABITS AND FRICTION
65
This focus on subjective perceptions makes conceptual sense, at least if one assumes that
we only perceive the objective environment via subjective perceptions. If situational effects are
completely mediated by perception, then we should be able to predict behavior using perceptions
alone. In addition to this rationale, there is also a more practical reason for this pattern—unlike
objective features of a situation, subjective perceptions can be easily measured using the self-
report methods that are endemic in psychology (Sassenberg & Ditrich, 2019).
Despite the conceptual and practical appeal of subjective measures, a systematic
understanding of situational behavior change requires quantifying a situation’s objective features
(Reis, 2008). This is true even if one accepts the assumption that subjective perceptions
completely mediate the effect of situations on behavior. First, studying only perceptions does not
always indicate how to effectively modify situations to change behavior. For example, even if
the effect of situational features on cigarette smoking is completely mediated by the urge to
smoke, knowing a person’s smoking urge does not tell us which elements in the environment
might have caused that urge, or, importantly, how to alter the environment to reduce smoking
urges. Research that does measure situations, on the other hand, can pinpoint antecedents of
smoking, such as periods of waiting and episodes of coffee drinking (Shiffman et al., 2008).
Exploring the situation and its perception as separate constructs—in line with most theoretical
accounts—promises to provide the scaffolding necessary to build a cumulative science of
situations.
Habits and Affordances
How, then, could psychologists begin to trace the pathways by which situations translate
into behavior? The notion of affordances provides a useful tool for understanding person-
situation synthesis. Affordances refer to the notion that objects in the physical environment
HABITS AND FRICTION
66
enable certain behaviors more than others (Gibson, 1966). For example, a door with a single,
wide horizontal bar about arm-height suggests that it opens by pushing; A typical car door
handle, on the other hand, invites pulling (Norman, 2013; see Figure 1). Affordances do not
reside solely within the person or within the environment but instead in an interaction between
the two (Gibson, 1966). Consequently, what the environment affords a person depends on the
person’s abilities. For example, sloped surfaces afford different actions for toddlers (who can
walk) as opposed to infants (who can only crawl; Adolph et al., 1993).
Figure 3.1

Examples for designs of pull (left) and push (right) handles

Credit: User (left) Scott Brody (right), CC BY-SA 4.0, via Wikimedia Commons.

Ability alone, however, cannot adequately explain how elements in the physical
environment give rise to specific behaviors. After all, a dizzying array of behaviors could be
performed with any object: For instance, a paper clip could be used to hang Christmas ornaments
or clear a clogged spray bottle. A person in 18th century Europe or Japan might readily use
breadcrumbs as an eraser, whereas someone in 21st century California would never entertain that
possibility (similarly, rubber erasers were discovered by mistake when a British engineer absent-
mindedly grabbed a piece of rubber instead of breadcrumbs; Garber, 2013).
As these examples suggest, what people do in a situation strongly depends on what they
did in that situation in the past (as well as what others have done in the past in that same
HABITS AND FRICTION
67
situation). To fully understand situation effects on behavior, then, researchers must understand
people’s past responses and habits.
Habits link situation and person by explaining how recurring context cues can activate
well-practiced behaviors. In this way, habits can constrain a plethora of possible choices into one
or a few well-rehearsed responses. Supporting this role of habits is a set of studies in which
participants had to act flexibly by deciding which foods they should eat after dental surgery, or
cook for a friend with particular preferences (Morris et al., 2021). Thought records suggested
that participants generated a small set of their favorite foods, from which they then selected a
qualifying food based on the given constraints (e.g., foods that are easy to chew after the dental
surgery).
Friction
Early research on habit learning by Hull (1943) suggested that behavior in both human
and non-human animals tends to follow the law of less work, so that all things being equal,
“organisms will...choose the less laborious behavior” (p.294). This is a basic principle in
psychology, at least since De Camp (1920) noted that rats tend to choose the shortest of two
paths when navigating mazes in search of food. Whereas behaviorists’ definition of effort
focused on physical effort, likely because they usually studied rats and pigeons, later research
noted that humans similarly eschew cognitive effort (Inzlicht et al., 2018; Kool et al., 2010).
The effort needed to perform a behavior depends not only on person-level factors such as
skill and ability but also on situational friction. In physics, friction is a force that resists the
motion of objects in contact with each other. Physical friction wields awesome power—in one
illustration, train wheels’ structure minimizes friction—a design that allowed early locomotives
HABITS AND FRICTION
68
to transport heavy loads despite having power equivalent to that of a modern lawn-mower
(Brooklyn Historic Railway Association, n.d.).
Similarly, situations can make some behaviors easier and others more difficult (see
affordances above). However, friction is under-utilized in behavior change science. As described
above (see Situations in Social Psychology), interventions that harness information and goal-
setting far outnumber those that harness situational methods.
This is not to say that friction is absent from behavior change research. Some examples
involve delay interventions, such as slowing down elevator doors’ closing to increase stair use
(Houten et al., 1981) and increasing time delay on obtaining unhealthy snacks from a vending
machine (Appelhans et al., 2018). Further, a storied literature in choice architecture examines the
effect of defaults—one form of friction—on behavior. This literature finds that people often stick
with the default option, even when they can select a different option with nearly no additional
effort (Blumenstock et al., 2018; Ebeling & Lotz, 2015). However, such studies usually test the
effect of friction imposed top-down, such as when a cafeteria makes certain foods less easy to
reach (e.g., Rozin, 2011). In contrast, less is known about how people self-administer nudges. In
particular, we know little about how this self-regulation dynamic plays out in repeated actions
over time, in contrast with one-time decisions such as deciding to set up automatic retirement
savings deposits. Understanding this self-regulation dynamic is important because in many cases,
it is not practical or ethical to administer a one-time top-down nudge.
In contrast with friction’s relative absence from the behavior change literature, it features
heavily in applied disciplines. Clinical approaches for treating substance abuse frequently
involve increasing friction on obtaining and using drugs. For example, one potent antecedent of
drug use is simply “having money in your pocket,” which therapists may mitigate by advising
HABITS AND FRICTION
69
patients to carry only minimal amounts of cash or safeguard money with loved ones (Carroll,
1998, p. 49). Friction is widely recognized in industry settings as well. Hooked, an influential
book on developing technology products, exhorts companies to focus first and foremost on
reducing friction over increasing motivation, claiming that “reducing the effort required to
perform an action is more effective than increasing someone’s desire to do it” (Eyal, 2014, p.
80).
Friction and Habits
The interaction between habits and friction presents a unique example of person-situation
synthesis. Well-practiced and routine behaviors are accompanied by a sense of ease (Hackbarth
et al., 2003), safety, and well-being (Avni‐Babad, 2011). Well-practiced behaviors even feel
shorter, with experts under-estimating task completion more than their less experienced
counterparts, despite experts’ having ample opportunity to experience the task and calibrate time
judgments (Avni-Babad & Ritov, 2003; Boltz et al., 1998).
This feature of habit performance can lead to a dissociation between objective and
subjective friction: A change in the environment might increase perceived friction even when the
actual, objectively required effort stays the same or even decreases. For instance, initial attempts
to encourage residents of a Canadian city to Grasscycle—to allow grass clippings to naturally
decompose on the lawn after mowing instead of collecting and disposing of them—fell flat even
though this new behavior required objectively less effort (White et al., 2019). This phenomenon
also emerges when established websites redesign their appearance. When Facebook streamlined
its design in 2008, the change increased engagement among non-frequent users as expected, but
decreased it for frequent users, for whom habitual sequences of behavior were apparently
disrupted (Anderson & Wood, 2021). Indeed, a tenet of web design states that “users spend most
HABITS AND FRICTION
70
of their time on other sites” (Yablonski, 2021): Because users have set habits and expectations
when they visit a website (e.g., a clickable logo at the upper left corner; Orbit Media Studios,
2015), and because unfamiliar designs require deliberate thought to navigate, bad-but-familiar
designs may outperform better-but-novel ones.
Even minor amounts of friction can cast a sizable pull on behavior: Users often abandon
web pages that take more than a few seconds to load (Google, 2018). In another demonstration
that involved increasing tax collection rates in the UK, directing individuals to a tax form
increased response rates by 4% compared with directing them to a webpage that contained that
form—an effort reduction of only a single mouse click (Service et al., 2015).
Minor friction may prove particularly effective for disrupting habitual behaviors. As
discussed above, when able to act habitually, people tend to default to their habits. But when
minor friction is present, it could prevent the habit sequence from proceeding as usual, forcing a
person to make an active decision. If the behavior is driven by a salient goal, one might decide to
achieve that goal flexibly using a different action. But when the behavior is mindless and not
supported by a goal, friction could steer people towards performing any one of alternative
behaviors, or simply fall back on doing nothing. This is the same logic used to prevent operators
of nuclear reactors to unintentionally operate controls (Preischl & Hellmich, 2013) or, to use a
more mundane example, when your computer asks you to confirm that you really want to delete
a file. These minor obstacles can mitigate unintentional responding without significantly
hindering intentional responding. All these suggest that introducing even miniscule amounts of
friction to a situation could sizably reduce performance of undesirable habits.
The effect of such minor friction is expected to increase the more the behavior is habitual
rather than goal-driven. For example, if a person has a strong goal of opening a phone app to
HABITS AND FRICTION
71
navigate or send a message to a friend, small obstacles like the app icon being in a slightly
different location should not matter, and the person should be able to flexibly respond and use
the app. On the other hand, if someone initiates the sequence of using a phone app habitually,
without being driven by a strong goal, minor impediments may be enough to derail the behavior.
Habits in Behavior Change
Habits form as people repeat a response in a given context (Mazar & Wood, 2018; Wood
& Rünger, 2016). The repeated behavior gradually becomes tied to cues in the performance
context, establishing a cue-response association in memory. Once these associations are
solidified, cues in the environment trigger the associated behavior automatically, with little need
for conscious involvement (Orbell & Verplanken, 2010; Verplanken & Orbell, 2022). Habits
permeate daily life: One study found that participants spent 43% of their waking hours repeating
familiar actions in similar situations, usually with minimal thought (Wood et al., 2002).
Undesirable habits pose a unique problem for behavior change: Such habits can persist
despite contrary intentions (Ji & Wood, 2007; Ouellette & Wood, 1998), attitudes (Itzchakov et
al., 2018), planning interventions (Webb et al., 2009), monetary incentives (Mantzari et al.,
2015), and sometimes, even electric shocks (Ersche et al., 2016). A number of psychological
processes underlie the robustness of habits. First, all things being equal, changing repeated
behaviors is necessarily more difficult than changing one-off behaviors, simply because it
requires influencing multiple instances of a behavior.
But when repeated behavior is habitual—meaning that it is driven automatically by
learned cue-response associations—additional challenges arise. As habitual behavior comes
under contextual control, it simultaneously decreases in intentional control (Danner et al., 2008;
Wood et al., 2005). More often than not, this is a welcome benefit: Habit formation automates
HABITS AND FRICTION
72
repeated behavior, freeing up mental resources that could be used elsewhere. When habits are
undesirable, however, they may consequently prove more intractable than non-habitual
behaviors. This is not to mean that people are forced to uncontrollably act on their habits: Like
other automatic processes (e.g., color naming in the Stroop task, MacLeod, 1992), a habit in
mind can be overridden with conscious control given adequate time, attention, and motivation
(Luque et al., 2020; Wood et al., 2021). However, at some inevitable points in daily life when
time pressure is high, attention is distracted, or motivation wanes, people tend to revert to their
default, habitual responses (Hardwick et al., 2019; Neal et al., 2013; Rebar et al., 2014; Schwabe
& Wolf, 2009).
Because habit memory traces are slow to fade even when not acted on (Bouton et al.,
2011), inhibiting habitual responses requires vigilant monitoring (Quinn et al., 2010) until the
habit association weakens. Yet such monitoring can prove unsustainable long-term.
Consequently, effective habit change requires supplementing interventions that alter persons
(e.g. increase their motivation, encourage them to plan) with interventions that alter situations
(Duckworth, Gendler, et al., 2016). Situational strategies may be particularly effective because
they involve changing the response environment in advance, long before goal-conflicting urges
arise; by the time urges do arise, the environment is set to impede goal-incongruent behavior,
consequently steering a person towards goal-congruent behavior (Duckworth et al., 2018).
Situational interventions such as adding friction may perform especially well at habit reduction,
in comparison with person-level interventions, given that person-level interventions alone may
change attitudes toward a habitual behavior without influencing the behavior itself (Itzchakov et
al., 2018).

HABITS AND FRICTION
73
Lay Perceptions of Situational Influences
In some cases, environments can be altered top-down in a way that requires little if any
input from the population being intervened on. This is exemplified in policies in the United
States that increased turnout by automating voter registration (Brennan Center for Justice, 2019)
3

or that increased the prevalence of Type 2 diabetes by adding refined sugars to food products
throughout the 20th century, mostly unbeknownst to consumers (Gross et al., 2004). In such
situations, an intervention may exert its effects regardless of what people think about it.
Consequently, how those interventions are perceived is of lesser importance. In many cases,
however, practical and ethical concerns dictate that people knowingly and sometimes repeatedly
initiate or at least accept an intervention. In such cases, if people underestimate an intervention’s
efficacy, they may fail to make use of it despite its potential benefits.
This marks a challenge for situational interventions, given a storied literature
demonstrating a tendency in western cultures to overlook situational influences (Choi et al.,
1999). If people underestimate situational influences on themselves, it follows that they may
similarly underestimate the efficacy of situational interventions. Ironically, situational self-
regulation may fail exactly because they are so effective. Duckworth (2016) argued that because
successful situational self-regulation nips undesirable urges at the bud, it may produce little
conscious conflict, which could then drive people to overlook its efficacy. If the urge to check
your phone is effectively subdued when it is in a different room, it is difficult to notice the
counterfactual—that this would not be the case if instead your phone was right in front of you.
Thus, these interventions may fall victim to their own efficacy. Inasmuch as decision
making is guided by perceived utility more than experienced utility (Mitchell et al., 1997), the

3
Citizens in many U.S. states are required to actively register to vote, sometimes with additional requirements such
as registering a signature that exactly matches the one that they provide with their ballot.
HABITS AND FRICTION
74
efficacy of situational interventions may paradoxically cause people to under-value and
consequently under-utilize them. In support, one study on smartphone use reduction found that a
situational intervention—setting one’s phone to display in grayscale—worked better than a self-
monitoring condition (Zimmermann & Sobolev, 2020). An opposite pattern, however, emerged
in participants’ estimates made before the study, so that participants predicted that self-
monitoring would be more effective.
The Present Study
In this study, I tested the effects of adding situational friction to smartphone app use.
Smartphone use is perhaps one of the most strongly habitual behaviors in current daily life. In
early 2020, U.S. adults spent on average nearly 4 hours a day on their phones, with similar
estimates in countries such as India and Japan (BARC & Nielsen, 2020; MMD Labo, 2020;
Nielsen, 2020). In contrast with other behaviors, smartphone use is immediately accessible at
nearly any time and place. This feature gives rise to fragmented use patterns characterized by
frequent brief episodes in which a habit of constantly checking one’s phone may instigate further
use (Oulasvirta et al., 2012).
This project compared the efficacy of friction to a person-centered motivational
intervention—self-affirmation followed by a persuasive message (McQueen & Klein, 2006).
Prior research demonstrated strong, long-lasting effects of self-affirmation on behavior. For
example, completing only three to five brief self-affirmation exercises increased students’ GPA
at a 2-year follow-up (Cohen et al., 2006). Thus, self-affirmation provides a high bar against
which to compare our situational interventions. Because habitual behavior is under increased
situational control and reduced intentional control (Gardner, 2015), situational interventions
should be more effective than motivational interventions at reducing app use. Additionally,
HABITS AND FRICTION
75
situational interventions should be particularly effective because they defuse goal-conflicting
urges before they arise (Duckworth, Gendler, et al., 2016). However, because situational
interventions do not target intentions or attitudes, we expect that they would curb app use
without increasing intentions or attitudes towards reducing app use.
The present study consists of a two-week randomized controlled trial with 6 arms: 4
friction conditions (split themselves into a 2 X 2; frequency: daily vs. weekly; app icon in new
location: same page vs. folder), as well as a motivation condition and a control condition. The
sample consisted of smartphone owners who wanted to reduce the use of a specific phone app.
Participants completed a series of survey measures and were then randomly assigned to a
condition. Participants in the friction condition changed the location of the app icon on their
phone screen, with two intervention features randomly assigned: frequency (every day vs. once a
week), and app icon location (moving the icon to a new location on the same page, or moving the
icon into a separate folder). Participants in the motivation condition completed a self-affirmation
writing exercise and then watched a persuasive video appeal. Participants in the control condition
did not undergo an intervention. All participants then completed daily diaries measuring their
intentions and attitudes toward reducing app use. Finally, at a two-week follow-up, participants’
app activity was recorded using the smartphone’s objective phone logs.
Research Questions and Predictions
Below, I present my main research questions and predictions. First, I test whether
imposing minimal friction on locating and entering an app could reduce app use.

Question 1: What are the effects of a situational intervention (increasing friction)
on reducing phone app use?
HABITS AND FRICTION
76
Prediction 1: The situational friction intervention will reduce phone use,
compared with a control condition.

In addition, because habits are triggered automatically by context cues and are
minimally sensitive to goals, situational interventions should be more effective than
motivational interventions at reducing phone use.

Question 2: What are the effects of a situational intervention (increasing friction)
on reducing phone app use, compared with a motivational intervention?
Prediction 2: The situational intervention will outperform the motivational
intervention in reducing phone app use.

The next research question addresses the mechanism of situational interventions.
If situational interventions work by preventing automatic responding (friction), their
effect on phone use should function independently from intentions and attitudes, so that
participants in these conditions should not have stronger intentions and attitudes towards
reducing phone use (compared with control). Moreover, the motivational condition may
strengthen intentions and attitudes to reduce phone use even if these do not translate into
changes in actual app use (per Prediction 1).

Question 3: What are the effects of a situational friction intervention (vs.
motivational intervention) on intentions and attitudes to reduce phone app use?
HABITS AND FRICTION
77
Prediction 3: The situational intervention will not outperform the motivational
intervention in increasing intentions and attitudes to reduce phone app use.

Finally, because people in western cultures tend to underestimate situational
influences and perhaps over-estimate motivational ones, I expect efficacy of situational
interventions to be underestimated compared with motivational interventions.

Question 4: How does perceived efficacy of situational and motivational
interventions compare with actual efficacy.
Prediction 4: Participants will underestimate the efficacy of friction and
overestimate the efficacy of motivation.

Secondary research questions
Apart from these primary research questions, the data can address several secondary
questions. First, whereas the effect of situational interventions on app use should not depend on
intentions and attitudes (see Prediction 2 above), the effect should depend on the extent to which
app use is mindless and habitual. Therefore, situational interventions should reduce habitual app
use but not necessarily more intentional app use. Given that most of our participants will likely
have strong app use habits, however, it remains to be seen whether any one of our habit measures
(see Measures, below) will be able to capture sufficient variance in habitual vs. intentional app
use.
Another question of interest regards time effects. It is a truism in habit change that
initiating change is easier than maintaining it, with a common pattern of initial promising
HABITS AND FRICTION
78
changes followed by a return to baseline levels of behavior within weeks or even days (Allcott &
Rogers, 2014; Thøgersen & Møller, 2008). Thus, it may be that all interventions are equally
effective on Day 1, with app use in the motivation condition rebounding to baseline by the end of
the study, with lesser rebounding in the situational conditions. On the other hand, a competing
hypothesis is that the friction intervention will prove effective initially, but participants quickly
become habituated to the new home screen layout, forming a novel habit of accessing the app via
its new location. Such a possibility seems especially likely given that most of our participants use
their target app very frequently.
Finally, two other exploratory questions concern the effects of different types of
friction. The research design varies orthogonally the frequency and dose of friction
administered. Participants in the friction conditions move their app icon location on either
a daily or weekly frequency. On the one hand, changing the app icon location daily
should make it very difficult for app use to form into a habit again, given that by the time
a new habit can take hold, the app location is changed again. On the other hand, such
frequent intervention could spur participants to form a new habit of accessing the app in a
different way, or simply become adept at locating the app at various locations.
Furthermore, even if the intervention does not lose its efficacy over time as noted
above, the burden of having to re-apply the intervention every day could prove onerous.
This burden could lead participants to reduce their compliance even as the intervention
remains effective. Such sub-optimal compliance is especially likely if participants under-
estimate how effective the friction manipulation was (see Research Question 4).
Another question concerns the amount of friction to be imposed. The law of least
effort states that people seek the path of least effort. Subjective effort, however, can
HABITS AND FRICTION
79
depend on prior experience; anecdotally, all else being equal, habitual, well-practiced
actions should feel easier than less familiar ones. If so, it follows that one could increase
friction on a behavior by requiring it to be performed in a novel manner—even if the
objective effort required to enact the behavior remains the same. In the present study, this
is achieved by moving an app icon from one location on a home screen to another
location on the same screen. If this intervention does reduce phone use, this would
suggest that merely disrupting a habit (without increasing objective effort) can effectively
curb habitual app use.
In contrast, it may be that curbing phone use requires increasing the objective
amount of effort required to access an app. For this reason, another condition in the
present study involves moving the app icon into a folder. Doing so increases the
difficulty of using an app, even though the difference is only a single tap. If only this in-
folder condition reduces app use, it would suggest that merely disrupting habits (without
increasing objective effort) does not suffice for reducing habitual app use. Instead, habit
disruption needs to be accompanied by increasing the effort required to perform the
habitual behavior. Note, however, that this operationalization is not perfect, in that
moving the app icon into a folder both increases friction on app use and makes the app
icon less salient. Specifically, in the current iOS version as of this writing, placing an app
icon in a folder decreases the size of an app icon.
Method
Study pre-registration can be found in the following link: https://aspredicted.org/K9Y_7V8

HABITS AND FRICTION
80
Participants
The sample (N = 216 participants, 4,434 participant days) consisted of n = 136 from a
Prolific online research panel and n = 80 from undergraduate subject pools; 129 female, 82 male,
5 other gender; mean (SD) age = 28.13 (11.67) years. An additional 23 participants did not
complete the study and therefore were not included in analyses. An additional 7 participants
were not included as well because they were missing baseline data. Finally, another 29
participants selected a target app that they rarely used to begin with (<= 10 minutes a day during
the baseline measurement week), and were therefore excluded from analyses (this exclusion
criteria was not pre-registered).
Participants qualified for the study if they (a) reported (in a separate pre-screen survey)
that they had at least some desire to spend less time on their phone, namely by scoring a 3 or
higher on the item, “Would you be interested in reducing the time you spend on your phone?” (1
= Definitely not, 7 = Definitely yes); (b) owned an iPhone (because participants’ phone use was
recorded using iOS’s built-in Screen Time app), (c) reported using a smartphone app multiple
times a day, and (d) did not report mainly accessing apps by searching for the app name (because
the study intervention relied on breaking a habit of clicking on an app icon rather than searching
for the app name). iPhone users do not substantially differ from Android users (the main
competing phone operating system) on a range of demographic and personality variables, with
the exception that iOS users may be slightly more affluent on average (Götz et al., 2017; but see
also Ubhi et al., 2017).
Participants were compensated based on compliance: $5 for completing only the intake
session or completing less than 50% of the daily diaries; $15 for completing 50%–80% of daily
HABITS AND FRICTION
81
diaries; and $20 for completing over 80% of daily diaries. Those who participated for course
credit received a comparable gradual compensation scheme.
Power analyses using the pwr package in R revealed that 45 participants per cell would
be required to obtain 80% power to detect a medium effect (f = 0.25) in a one-way ANOVA.
Thus, we aimed to recruit 50 participants per condition or 300 in total, yielding an approximate
final sample size of 270 participants (given 10% attrition). Eventually, due to budget and
practical concerns, data collection was halted somewhat earlier, at around 260 participants. Note,
however, that because the pre-registered analyses involved a multilevel linear model rather than
a one-way between-subjects ANOVA (as in the power analyses), these power analyses should
provide a conservative power estimate, given that the study design employs a mixed design that
measures within-person change before and after intervention, and therefore removes noise
generated by individual differences. In addition, to ensure that any null results found are not the
result of insufficient power, I ran analyses collapsing across certain conditions, so that cell sizes
exceeded 50 per cell (see Results section).
Design
The study design is outlined in Figure 2. The study used a mixed design, with time (pre-
vs. post- intervention) as a within-person factor and condition (friction vs. motivation vs. control)
as a between-person factor. Further, four friction cells, represented a 2 X 2 design orthogonally
manipulating friction intervention frequency (daily vs. weekly) and icon location (same page vs.
folder).
Qualifying participants met with a research assistant for an intake session, followed by a
final session approximately 14 days later. In the 13 intervening days, participants completed
daily diaries every evening and responded to an email every morning (see intervention below).
HABITS AND FRICTION
82
Intake and final sessions were held in-person for undergraduate participants and on Zoom for
Prolific participants.
Procedure
Intake Session
After providing their informed consent, participants selected their target app—the app
that they focused on during the study period. Participants selected from a multiple-choice list of
popular smartphone apps or named another app if it was not listed. Because harmful effects of
app use may stem more from passive use (e.g., scrolling as opposed to actively using social
media to connect with others; Verduyn et al., 2015), the research assistant instructed participants
to select an app that they did not mainly use to directly message others. Participants then
answered a series of questionnaires (see Measures) and received the relevant intervention for
their condition.
Daily Diaries
For the duration of the study (the 13 days between the intake and final sessions),
participants completed diary surveys every evening. Participants accessed the evening survey
link via a shortcut on their home screen. Participants were instructed to complete the evening
survey at 7pm each day or as close to 7pm as possible, up until midnight.
Morning Emails
All participants received emails every morning with content varying based on
experimental condition. Participants in the daily condition (as well as participants in the weekly
condition on day 7 of the study) received emails that instructed them to change the location of
their target app icon and upload a screenshot of the new location. To ensure that all conditions
received an equal frequency of contact from the research team, participants in the control and
HABITS AND FRICTION
83
motivation conditions (as well as participants in the weekly condition on all days but day 7 of the
study) received emails asking them to confirm that they are still committed to the study (“I
confirm that I am taking part in this study and am committed to reducing my [app] use.”).
Final Session
Two weeks after their intake session, participants completed a final survey that included
all measures from the intake survey, as well as measures tapping how effective they thought the
study was at reducing their app use, as well as how easy or difficult it is for them to reduce their
app use. Afterwards, a research assistant guided them through their Screen Time data, with
participants reporting to the research assistant their Screen Time data between 7 days pre-intake
(to provide a baseline measure of phone use) through the day prior to the final session.
Follow-Up Survey
One month after intake, participants received a survey assessing their perceived app use
since the study ended and whether they had implemented any strategies for reducing app use
since the study ended. In addition, participants provided Screen Time records for the previous
week.
Intervention
Friction. After completing the intake survey, all friction participants were guided by a
research assistant in moving their target app icon to a different location. All friction participants
additionally turned off Siri Suggestions for their target apps (Siri Suggestions are prompts to
open an app or webpage based on time of day, recent activity, location, etc.). The participants
also disabled their apps from appearing in Spotlight Search, meaning that participants were not
able to access the app by searching for the app name. This step was included to block this
alternative way of accessing the app, given that this alternative would not be impacted by the
HABITS AND FRICTION
84
friction intervention. In addition, friction condition participants were randomly assigned to a
level of intervention frequency (daily vs. weekly) and icon location (same page vs. folder).
Frequency (daily vs. weekly). Participants were instructed to either move their app icons
every day or once a week (i.e., twice during the study period: once during the intake session, and
once exactly one week after intake).
Icon Location (same page vs. folder). Participants were instructed to either move their
app icon to a different location on the same home screen page (same page conditions), or place
the icons in a new folder to add an additional (although still minimal) layer of friction (folder
conditions).
Figure 3.2
Study Design

Motivation. Participants in the motivation condition completed a self-affirmation
exercise by selecting a value from a list and writing one to two paragraphs explaining why it is
HABITS AND FRICTION
85
important to them (McQueen & Klein, 2006). They then watched a ~4-minute video appeal
concerning the negative psychological effects of smartphone use.
Control. Similar to participants in the other conditions, participants in the control group
attended the intake and final sessions, and completed the evening surveys as well as morning
emails.
Key Measures
The section presents the key measures in the study. Additional measures can be found in
the following section. In the quoted descriptions below, “[app]” refers to each individual
participant’s target app.
Intake Session Measures
Perceived Efficacy of Self-Regulation Strategies. Participants rated the efficacy of
several self-regulation strategies on a 0-100% slider scale (0 = would not influence my app use to
100 = would get me to completely stop using the app). An answer of 50%, for example, indicated
that a method could effectively reduce one’s app use by half. The default position of each slider
was 0%. The strategies were “Using Willpower,” “Getting motivated to reduce [app] use,”
“Moving the [app] icon to a different location on the same page,” and “Reducing [app]
notifications.” These options were designed to gauge the perceived efficacy of motivational and
situational strategies.
Self-Regulatory Success. Participants completed the brief Self-Control Scale (Tangney
et al., 2004). While this scale has not shown to be correlated with effortful self-regulation (see
Introduction), it may yet tap into success at self-regulation given its correlations with desirable
life outcomes (Tangney et al., 2004). The scale is a 13-item list including statements like “I am
HABITS AND FRICTION
86
good at resisting temptation” rated on a 5-point scale (1 = Not at all like me to 5 = Very much
like me).
Daily Measures
Perceived Self-Regulatory Success. As a measure of how well participants felt they
controlled their behavior each day, they rated the statement “Today, I used [app]...” on a 5-point
scale (1 = Less than I’d like, 5 = Much more than I’d like).
Final Survey
In the final survey, participants again responded to all measures included in the intake
survey, except for the Brief Self-Control Scale and the measure tapping default phone
notifications. In addition, the final survey included the measures below.
Phone Use. Daily phone use data were taken from participants’ iPhones using the built-
in Screen Time app, starting 7 days before the intake session (to provide a weeklong baseline
measurement of phone use) and ending a day prior to the final session.
The study’s main outcome was the time (in minutes) that participants spent on the target
app. A secondary outcome was the time that participants spent on their phone as a whole. Three
additional Screen Time metrics were recorded only for the post-intervention period: number of
overall phone pickups (i.e., unlocks); number of times that the target app was the first app used
after pickup; and number of target app notifications.
Additional Measures
In addition to the key measures described above, participants also completed the
Following measures.

HABITS AND FRICTION
87
Intake Session
Habit Strength. There is no gold standard for measuring app use habit strength. The
most common measures of habit rely either on self-reports of frequency and use context or
perceived automaticity. Self-reports of frequency may be skewed in that participants strongly
underestimate their app use. Self-reports of use context pose additional challenges in that phones
can be used across a wide range of times and places, in contrast with habits with more narrowly-
defined contexts (e.g. morning coffee). Finally self-reports of automaticity are problematic since
perceived automaticity can diverge from actual automaticity (Labrecque & Wood, 2015); in
addition, perceived automaticity captures forms of automaticity other than habits. Finally, given
the current ubiquity of phone use, all phone use habit measures may suffer from a ceiling effect
by which habits are uniformly strong across the majority of the sampled population.
These caveats aside, participants completed on an exploratory basis the two most
common measures of habit, the Behavior Frequency in Context scale (BFiC) and the Self-Report
Habit Index (SRHI; Verplanken & Orbell, 2003)).
Behavioral Frequency in Context (BFIC). Participants responded to the item “How
many minutes do you spend on [app] on your phone every day?” on a scale ranging between 0-6
hours (in 10 minute intervals). In addition, participants responded to the following context
stability items on a 7-point scale (1 = Never or almost never, 7 = Always or almost always):
“how often is [app use] at the same time of day,” “how often is [app use] at the same location,”
and “how often is [app use] in the same mood?”.
Self-Report Habit Index. Participants responded to a nine-item scale containing
questions regarding the perceived automaticity of smartphone use (e.g., “using [app] is
HABITS AND FRICTION
88
something that would require effort not to do”) on a 7-point scale (1 = Strongly disagree, 7 =
Strongly agree).
Intentional vs. Habitual Use. Participants rated what proportion of their phone use is
intentional (“you have a goal in mind before you start reaching for your phone”) or unintentional
(“you start reaching for your phone without a goal in mind”) on a 0-100% slider scale (0% =
completely unintentional, 100% = completely intentional). The default slider position was 50%.
Norms. On a 7-point scale (1 = Strongly disagree, 7 = Strongly agree), participants rated
the social norms surrounding their target app use, including descriptive norms (“Most people
who are like me use [app] regularly”) and injunctive norms (“my friends expect me to use [app]
regularly”).
Default Phone Notifications. Participants indicated whether the majority of their phone
apps are programmed to send them a sound alert, vibration, both, or neither.
Enjoyment. On a 7-point scale (1 = Not at all, 7 = Extremely), participants responded to
the prompt “How much do you enjoy using [app]?”
Intervention Motivation. In a multiple-choice item, participants indicated their most
important reason for wanting to reduce their screen time (e.g., “to be more mindful and in the
moment,” “to avoid comparing myself to others”). Participants then indicated which outcome
was more important to them: reducing minutes spent on the target app or reducing the number of
times they opened it, or both. Finally, participants indicated whether or not they tried to reduce
their smartphone use before (yes, no)—and if so, they briefly described their strategy in an open-
text item.
Well-being. Participants completed a shortened version of the Positive and Negative
Affect Schedule (Thompson, 2007), concerning their emotional state for the past 2 weeks by
HABITS AND FRICTION
89
rating 9 different emotions along a 5-point scale (1 = Not at all, 5 = A lot). They then rated their
sleep quality over the past week on an 11-point sliding scale (0 = worst, 10 = best).
Daily Surveys
Intentions. Participants indicated, “I want to limit my [app] use tomorrow,” on a 7-point
scale (1 = Strongly disagree, 7 = Strongly agree).
Expectations. Participants reported, “I expect to limit my use of [app] tomorrow,” on a 7-
point scale (1 = Strongly disagree, 7 = Strongly agree).
Attitudes. Participants recorded their attitudes toward reducing their app use in the
moment (“Right now, reducing my [app] use feels...”) on 7-point semantic differential scales
(difficult/easy; bad/good, Osgood, 1962).
Intervention Efficacy. Participants in the friction conditions indicated, “how effective
was moving the app icon at reducing your app use today?” (1 = not at all effective, 7 = extremely
effective).
Affect. Participants completed a shortened (9-item) version of the Positive and Negative
Affect Schedule, regarding their experience that day (e.g., “Today, I felt Proud,”) on a 5-point
scale (1 = not at all, 5 = a lot).
Procrastination. On 5-point scales (1 = not at all, 5 = definitely), participants completed
3 items relating to their procrastination that day: “today, I [put off tasks that I intended to do; had
an important task to get done, but instead did things that were more fun; delayed starting to work
on a task or finishing one].”

HABITS AND FRICTION
90
Final Session
Perceived Behavior Change. Participants estimated on a 0–100% scale how much they
reduced their app use as a result of the intervention. An answer of 50%, for instance, meant that
they reduced their app time by one half.
Follow-up Survey
Perceived Efficacy. Participants rated the effect of the intervention since the study ended
on a 0–100% scale (0 = did not reduce my [app] use to 100 = got me to completely stop using
[app]). Participants rated their phone use relative to their use before the study started.
Intervention Follow-Ups. Participants in the motivation and friction groups indicated
whether they continued any features of the interventions (e.g., watching more videos or moving
their apps).
Weekly Average Phone Use. Participants reported their average phone and target app
use for the first full week after the intervention, based on their objective ScreenTime app logs.
Results
Means and standard deviations for key variables are in Table 1. During the baseline week
(prior to the study), participants used their target app about 1 hour per day, and their phones for 6
hours per day. Participants reported very strong habits for phone use, with an average SRHI
score of 5.55 on a 7-point scale. App use for participants in both the control condition and the
friction conditions can be found in Figure 3. Descriptively, participants in the friction condition
display a marked decrease in app use one day post-intervention. However, a similar pattern
appears for participants in the control condition.
HABITS AND FRICTION
91
Figure 3.3
App Use Pre- and Post-Intervention in the Control and Friction Conditions

Note. The vertical red line represents the beginning of the study period. Black horizontal bars
represent mean intentions for that day. Light boxes represent 95% confidence intervals for the
mean.
HABITS AND FRICTION
92
Table 3.1
Means and Standard Deviations for Key Between-Person Variables
Variables Mean Standard Deviation
Baseline app use (minutes) 64.99 52.61
Baseline phone use (minutes) 381.66 192.18
Perceived efficacy of willpower 42.36 22.21
Perceived efficacy of motivation 52.51 21.97
Perceived efficacy of changing app icon location
(same page)
17.86 18.41
Perceived efficacy of changing app icon location
(different page)
30.37 25.86
Self Report Habit Index 5.55 0.76
Self-regulation success (Brief Self-Control Scale) 2.96 0.42

Note. Perceived efficacy ranged from 0-100. Self Report Habit Index scores ranged from 1-7.
Self-regulation success scores ranged from 1-5. Measures of perceived efficacy, self-report habit
index, and self-regulation success were measured during the intake survey. Baseline measures
reflect pre-intervention phone use.
Results - App Use
Results for the main pre-registered model can be found in Table 2. This linear multilevel
model (days nested within participants) predicted daily minutes-on-app difference score (daily
minutes-on-app minus average minutes-on-app during baseline week) from condition, dummy-
HABITS AND FRICTION
93
coded with the control condition as the reference group, study week (0 = week 1, 1 = week 2),
and the interactions between study week and all dummy-coded condition variables (see model
equations below).

Level 1:
𝒂𝒑𝒑_𝒖𝒔𝒆
!"
= 𝛽
#"
+𝛽
$"
𝑤𝑒𝑒𝑘
!"
+𝑒
!"

Level 2:

𝛽
#"
= 𝛾
##
+𝛾
#$
𝑚𝑜𝑡𝑖𝑣𝑎𝑡𝑖𝑜𝑛
"
+𝛾
#%
𝑓𝑟𝑖𝑐𝑡𝑖𝑜𝑛_𝑑𝑎𝑦_𝑠𝑎𝑚𝑒
"
+𝛾
#&
𝑓𝑟𝑖𝑐𝑡𝑖𝑜𝑛_𝑑𝑎𝑦_𝑑𝑖𝑓𝑓
"
+
𝛾
#'
𝑓𝑟𝑖𝑐𝑡𝑖𝑜𝑛_𝑤𝑒𝑒𝑘_𝑠𝑎𝑚𝑒
"
+𝛾
#(
𝑓𝑟𝑖𝑐𝑡𝑖𝑜𝑛_𝑤𝑒𝑒𝑘_𝑑𝑖𝑓𝑓
"
+𝑢
#"

𝛽
$"
= 𝛾
$#
+𝛾
$$
𝑚𝑜𝑡𝑖𝑣𝑎𝑡𝑖𝑜𝑛
"
+𝛾
$%
𝑓𝑟𝑖𝑐𝑡𝑖𝑜𝑛_𝑑𝑎𝑦_𝑠𝑎𝑚𝑒
"
+𝛾
$&
𝑓𝑟𝑖𝑐𝑡𝑖𝑜𝑛_𝑑𝑎𝑦_𝑑𝑖𝑓𝑓
"
+
𝛾
$'
𝑓𝑟𝑖𝑐𝑡𝑖𝑜𝑛_𝑤𝑒𝑒𝑘_𝑠𝑎𝑚𝑒
"
+𝛾
$(
𝑓𝑟𝑖𝑐𝑡𝑖𝑜𝑛_𝑤𝑒𝑒𝑘_𝑑𝑖𝑓𝑓
"
+𝑢
$"

Results did not uncover a significant main effect of any condition on app use. That is, no
condition reduced app use more than the no-treatment control condition. Interactions were
similarly nonsignificant, with the exception of a marginally significant interaction between week
and the motivation condition, suggesting that the increase in app use between the first and second
weeks of the study tended to be stronger for participants in the motivation condition (compared
with the control condition).

HABITS AND FRICTION
94
Table 3.2
Results for Main Pre-Registered Model Predicting App Use Difference Score
Predictors b ꞵ 95% CI p
Intercept -11.72 -0.01 -22.70 – -0.73
Study week (1 or 2) -3.53 -0.07 -12.42 – 5.37 0.437
Motivation -0.33 -0.01 -16.74 – 16.08 0.969
Friction (Weekly + Different Page) -12.46 -0.26 -28.39 – 3.48 0.125
Friction (Daily + Same Page) -4.93 -0.10 -20.48 – 10.62 0.534
Friction (Weekly + Same Page) 6.18 0.13 -9.37 – 21.73 0.436
Friction (Daily + Different Page) 4.38 0.09 -10.73 – 19.48 0.570
Week X Motivation 12.52 0.26 -0.74 – 25.77 0.064
Week X Friction (Weekly + Different
Page)
3.88 0.08 -9.02 – 16.78 0.555
Week X (Daily + Same Page) 3.88 0.08 -8.73 – 16.49 0.547
Week X (Weekly + Same Page) 9.91 0.21 -2.68 – 22.51 0.123
Week X (Daily + Different Page) 4.44 0.09 -7.79 – 16.67 0.477

Thus, we did not find support for our main hypothesis, in that participants in the friction
condition did not decrease their app use, over the course of two weeks, more than participants in
the control condition. To further probe this effect, we conducted a multiverse analysis (Steegen
et al., 2016). Multiverse analyses involve running every theoretically sensible model
specification for a given research question. Multiverse analyses can address the garden of forking
paths, or the issue that the same general analysis could yield different results, depending on
HABITS AND FRICTION
95
analytic choices that may seem minor (Gelman & Loken, 2014). To that end, we fit every
possible combination of the main model, varying the parameters listed in Table 3 (1,008 models
overall). In support of the initial results, none of the main effects for condition were significant in
more than 10% of models. Thus, the lack of difference between conditions does not seem to be
due to particulars of the model specification used.

HABITS AND FRICTION
96
Table 3.3
Multiverse Analysis Parameters
Parameter Possible values
Dependent variable ● Minutes-on-app (difference between daily minutes-
on-app and average minutes-on-app in baseline week)
● # of times app was first to be used after unlocking
phone
Dependent variable
transformation
● None
● Log-transform
● Person-mean centering
Experimental condition
coding
● Dummy coding each cell
● Merging control and motivation conditions
● Collapsing friction into frequency conditions (daily /
weekly)
● Collapsing friction into dose conditions (same page /
different page)
Time variable ● Study week (1 / 2)
● Study day (1 - 13)
Covariate ● None
● Lagged daily intention to reduce app use
● Lagged daily expectation to reduce app use
● Lagged mean of daily intention and expectation to
reduce app use
Covariate transformation ● None
● Centering
● Person-mean centering
Data subset ● All two study weeks (no subsetting)
● Only first week
● Only first three days

HABITS AND FRICTION
97
Results - Perceived Intervention Efficacy from Intake Survey
In a second pre-registered model, a linear multilevel model (estimates nested within
participants) predicted perceived efficacy (0-100%; measured at the intake survey) from dummy-
coded intervention type (motivation, friction + same page, and friction + different page, with
willpower set as the reference level; see Table 4).
4
Results revealed that participants predicted
that both friction conditions would be less effective than willpower in reducing their app use. In
addition, participants predicted that motivation would be more effective than willpower in
reducing their app use.
Table 3.4
Results of Multilevel Linear Regression Predicting Initial Perceived Efficacy from Intervention
Predictors b ꞵ 95% CI p
Intercept 42.36 0.26 39.17 – 45.56
Motivation 10.14 0.39 6.40 – 13.89 <0.001
Same page -24.51 -0.95 -28.25 – -20.77 <0.001
Different page -11.99 -0.47 -15.73 – -8.25 <0.001

Exploratory Analysis - Condition Effects on Initial Intentions
To examine whether the motivation condition increased intentions to reduce app use on
day one of the intervention, a linear regression predicted intentions to reduce app use (measured
during the first evening diary) from experimental condition (see Table 5). None of the conditions

4
Note that participants only rated the perceived efficacy of relocating an app icon to the same page vs. different
page—we did not ask about daily vs. weekly, since a-priori we did not expect participants to perceive these
differently.
HABITS AND FRICTION
98
differed from control on these initial intentions to reduce app use. Furthermore, participants in
the motivation condition did not show stronger initial intentions on day one. The failure to find
any effects of the manipulation on participants’ intentions 24-36 hours post-intervention could
either reflect that (a) the motivation intervention effectively strengthened intentions, but its effect
weakened in the following 1-1.5 days, or (b) the motivation intervention was ineffective even
initially, or (c) our selection criteria created a ceiling effect, in that participants already had
strong intentions to reduce app use.
Table 3.5
Results of Multilevel Model Predicting Intentions to Limit App Use from Condition
Predictors b ꞵ 95% CI p
Intercept 6.04 0.23 5.63 – 6.45
Motivation -0.54 -0.50 -1.15 – 0.07 0.084
Friction (Weekly + Different Page) 0.17 0.16 -0.43 – 0.77 0.575
Friction (Daily + Same Page) -0.41 -0.38 -0.97 – 0.16 0.159
Friction (Weekly + Same Page) -0.30 -0.28 -0.88 – 0.28 0.311
Friction (Daily + Different Page) -0.36 -0.34 -0.92 – 0.20 0.205

Exploratory Analysis - Intention Dynamics over Time
An important question in behavior change is the dynamics of behavior change intentions
over time. One often-observed pattern consists of initially strong intentions to change behavior
that weaken as time goes by (as in the case of new year’s resolutions; Norcross & Vangarelli,
1988). To examine these dynamics in our data, I first confirmed that participants in the various
conditions did not differ in their initial (intake survey) intentions to reduce phone use (as would
HABITS AND FRICTION
99
be expected based on random assignment; see Table S1 in Appendix C). Then, to examine
whether conditions influenced intentions over time, I fit a multilevel linear regression predicted
daily intentions (days nested within participants) from study day (1-13) and experimental
condition (dummy-coded; see Table 6).
The main effect for study day was not significant, suggesting that participants’ intentions
did not weaken over time. Thus, as shown in Figure 4, participants’ intentions remained fairly
positive throughout the study, with no visible decline during the study’s 2 weeks. Interestingly,
main effects emerged for both daily friction conditions, suggesting that the daily interventions
depressed daily friction participants’ intentions. Additionally, a barely-significant main effect for
the motivation condition suggested that motivation participants held weaker intentions, although
it’s unclear to what extent this minor effect represents a replicable difference.

HABITS AND FRICTION
100
Figure 3.4
Intentions to Reduce Target App Use by Study Day

Note. Black horizontal bars represent mean intentions for that day. Light boxes represent 95%
confidence intervals for the mean. The confidence interval for day 14 is wider since the sample
for that day includes only the subset of participants who had their final session a day late.

HABITS AND FRICTION
101
Table 3.6
Results of Multilevel Linear Regression Predicting Intentions to Reduce App Use from Study
Day, Experimental Condition, and Day X Condition Interactions
Predictors b ꞵ 95% CI p
Intercept 6.17 0.32 5.78 – 6.55 <0.001
Day 0.01 0.02 -0.02 – 0.03 0.526
Motivation -0.58 -0.36 -1.15 – -0.00 0.049
Friction (Weekly + Different Page) -0.06 -0.06 -0.62 – 0.49 0.824
Friction (Daily + Same Page) -0.82 -0.70 -1.36 – -0.28 0.003
Friction (Weekly + Same Page) -0.47 -0.32 -1.01 – 0.07 0.089
Friction (Daily + Different Page) -0.67 -0.54 -1.20 – -0.15 0.012
Day X Motivation 0.02 0.05 -0.02 – 0.05 0.389
Day X Friction (Weekly + Different
Page)
-0.00 -0.01 -0.04 – 0.03 0.884
Day X (Daily + Same Page) -0.01 -0.04 -0.05 – 0.02 0.403
Day X (Weekly + Same Page) 0.01 0.02 -0.03 – 0.04 0.706
Day X (Daily + Different Page) -0.00 -0.01 -0.04 – 0.03 0.792

To examine whether friction burdened participants, I first examined perceived ease using
participants’ initial (intake survey) perception of the extent to which implementing each
intervention would be “a hassle”. A multilevel linear regression model (estimates nested within
participants) predicted perceived ease from strategy type. Participants rated both friction
conditions as far easier than willpower (see Table 7). Taken alone, this finding suggests that
HABITS AND FRICTION
102
participants may have overestimated how easy it would be to follow the friction manipulation.
To probe this possibility, I tested participants’ daily ease perceptions by fitting a multilevel linear
regression (days nested within participants) predicting perceived ease from study day and
experimental condition (dummy-coded; see results in Table S2 in Appendix C). Predicted values
based on this model are plotted in Figure 5. In contrast with the notion that the intervention
proved difficult as participants experienced it, the figure illustrates an opposite pattern: Perceived
ease increased over time, so that with repetition, the intervention seemed progressively easier.
Thus, the data does not support the notion that participants were initially optimistic about their
compliance, but then progressively became disillusioned as the intervention proved harder than
expected (see also Table S5 in Appendix C).
Exploratory Analyses - Intention-Behavior Gap
Finally, I examine the possibility that as the study progressed, participants’ behavior
became increasingly decoupled from their intentions, forming an intention-behavior gap (e.g.,
Sheeran & Webb, 2016). I fit a multilevel linear regression model predicting app use from study
day, intentions (measured in evening diaries), and experimental conditions (see Table S3 in
Appendix C). Results indicated at most only a marginal interaction between intentions and study
day; in other words, although intentions tended to more strongly predict behavior at the
beginning rather than end of the study, this interaction was not significant at the .05 level. Note,
however, that this analysis is predicated on the notion that intentions change over time during the
study period (because such variability is necessary for correlation). Therefore, given that
intentions were relatively static throughout the study period, it’s unclear whether the lack of a
significant interaction represents the lack of an intention-behavior gap in participants’
experience.
HABITS AND FRICTION
103
To further probe this explanation, I standardized both daily app use difference scores and
daily intentions by subtracting each daily score from its participant-mean, and then dividing the
resulting difference score by each participant’s standard deviation for that variable. I then
subtracted these scaled app use scores from scaled intentions, so that smaller difference scores
represent a larger intention-behavior gap (see Table S4 and Figure S1 in Appendix C). Results
did not show that this gap grew over time. Thus, the current data do not support an intention-
behavior gap that grew as the study progressed.

HABITS AND FRICTION
104
Figure 3.5
Plot of Multilevel Regression Model Predicting Daily Perceived Ease from Condition and Study
Day

HABITS AND FRICTION
105
Table 3.7
Results of Multilevel Linear Regression Predicting Perceived Ease from Condition
Predictors b ꞵ 95% CI p
Intercept 4.36 0.62 4.13 – 4.59 <0.001
Motivation -0.51 -0.28 -0.80 – -0.23 <0.001
Same page -2.09 -1.14 -2.38 – -1.80 <0.001
Different page -1.93 -1.05 -2.21 – -1.64 <0.001

Discussion
The present controlled trial tested the effects of subtle friction interventions on app use.
In addition, we orthogonally varied the frequency and dosage of friction to understand how these
intervention parameters influence app use over time. Participants’ phone use logs, as well as
daily intentions and attitudes, were tracked for the next 2 weeks, along with intake and final
surveys at the study’s outset and end. In addition to revealing the actual effect of friction and
motivation interventions on their behavior, participants reported their perceptions of each
intervention’s efficacy, both at the study’s outset as well as (for the friction conditions) every
evening during the study period.
In the friction conditions, app use decreased shortly after the study’s outset but quickly
reverted to baseline. A similar pattern was observed in the no-treatment control condition,
suggesting that even this temporary decrease in app use was not due to a unique benefit of the
friction intervention itself but reflected participants’ initial attempts to change their behavior.
HABITS AND FRICTION
106
That friction did not reduce app use (compared to both control and the enhanced motivation
conditions) seems to go against the notion that situational interventions generally outperform
motivation in influencing behavior (Duckworth et al., 2018), particularly when that behavior is
habitual. One caveat to this interpretation is that participants in the motivation condition did not
hold stronger intentions (as measured 1-1.5 after the intake survey), compared to participants in
the control condition. The current data does not speak to whether this lack of effect means that:
(a) the motivation intervention did not work, (b) the effect simply dissipated by the time
intentions were measured, or (c) there was a ceiling effect by which most participants held strong
intentions to begin with, making it difficult to increase intentions even further.
It may be that some features of phone app use make this behavior especially resistant to
intervention. Our participants initially reported very high initial levels of habit automaticity, with
a mean close to the extreme end of the Self-Report Habit Index. Accordingly, during the study
period, our participants unlocked their phones on average 138 times per day. Such frequency is
at least an order of magnitude larger than that of other undesirable habitual behaviors, such as
snacking a handful of times per day (e.g., Grenard et al., 2013). Such hyper-frequent repetition
may be sufficient to support habit formation even in the span of a single day post-intervention,
particularly when the new habit is highly similar to an established habit (the only difference
being having to tap on a different location on the same home screen, potentially only a
centimeter or two away from the original location).
There were several limitations to this study. Although I verified using screenshots that
participants in the friction conditions mostly adhered to the intervention (meaning that they
changed their app icon location when needed), we do not know whether the manipulation
successfully increased friction by making it more difficult for participants to access the app.
HABITS AND FRICTION
107
Thus, it is unknown whether the intervention made app use more difficult but participants
continued to use their phones despite that difficulty, or whether the intervention did not make
app use more difficult at all.
Knowing whether the intervention influenced effort is especially important because the
friction manipulation (moving an app icon around the same home page screen) was designed to
be extremely subtle. The rationale behind choosing such a minor intervention is that although
large amounts of friction (for example, completely deleting an app from your phone) can be
effective in principle, in practice people may fail to take up such measures, especially given that
smartphone apps are increasingly key to our day-to-day functioning. I hope that future research
helps elucidate the right balance to strike between intervention efficacy and acceptability.
Inference about the friction intervention’s success is further hampered by the lack of a
precise outcome measure. The friction intervention was designed to increase friction on opening
an app, rather than making it more difficult to use an app once it was already opened. As such,
the intervention should primarily have influenced the number of times that participants open
their app. However, because iOS does not track this metric, our main dependent variable was the
duration of time participants spent on their phone. However, it may be that the intervention
reduced the number of times participants opened their phone, but they compensated by spending
more time on their app each time that they opened it. If well-being is harmed by fragmentation of
phone use rather than overall duration (that is, how many times one opens an app rather than how
long one spends on it), then such an effect should be beneficial to well-being. In support, one
study that batched phone notifications to show up three times a day (instead of constantly
throughout the day) reduced stress and increased well-being (Fitz et al., 2019).

HABITS AND FRICTION
108
Mechanisms of Friction
To understand the mechanisms underlying friction effects, we independently varied two
features of the friction manipulation. We varied frequency, so that participants changed the target
app icon location either daily or weekly. We also varied dose, so that participants relocated the
target app icon location into either a different location on the same page (disrupting the habit
without increasing objective effort) or into a folder (which, in addition to disrupting habits,
slightly increases objective effort). Surprisingly, neither feature influenced app use. However,
participants in the daily friction conditions held weaker daily intentions to reduce app use
(despite showing similarly strong intentions to reduce their app use at intake).
In contrast with the hypothesis that participants in the daily friction conditions were
burdened by the intervention, data from both intake and daily diaries suggested that participants
generally perceived the friction intervention as easy, and that ease in fact increased rather than
decreased as the study progressed. Therefore, rebounding to baseline was not driven by a process
by which an intervention that seemed easy initially proved difficult to comply with when
implemented repeatedly.
Perceived Intervention Effects
Participants overwhelmingly ranked willpower and motivation as more effective than
either friction intervention (as measured during the intake survey). Therefore, although
motivation and friction did not differ in their actual effects, participants overestimated the
relative efficacy of motivation and willpower compared with friction. This pattern fits previous
findings that people underestimate the impact of friction on behavior (Mazar et al., in press), as
well as the general finding that westerners discount the influence of situations on behavior (Choi
et al., 1999). Furthermore, it may help explain why students under-utilize situational strategies
HABITS AND FRICTION
109
when regulating their studying behavior (Duckworth, White, et al., 2016). This is an important
finding, given that public support for nudges increases when nudges are perceived as effective
(Petrescu et al., 2016). It may similarly be that up-shifting the perceived efficacy of friction
could encourage people to make use of it in their daily lives to change their own behavior.
Temporal Dynamics of Behavior Change
In this study, participants’ intentions did not significantly shift over time. In contrast with
an account that would posit rosy initial intentions that quickly sour (as when new year’s
enthusiasm is replaced with February’s disenchantment), participants’ intentions remained fairly
stable throughout. In other words, participants’ behavior returned to baseline even as intentions
remained strong. Such a pattern suggests an intention-behavior gap, such that behavior takes
place counter to people’s intentions (with the caveat that regression analyses revealed only a
marginal intention-behavior interaction over time, so that the link between intentions and app use
marginally weakened as the study progressed).
Conversely, intentions were not informed by behavior, so that decreasing success in
changing app use did not weaken intentions. These findings call into question common self-
regulation theories such as Social-Cognitive Theory and the Theory of Planned Behavior, which
posit that behavior is largely purposive and intentional (Ajzen, 1991; Bandura, 1991). In
contrast, such findings do align with the notion that app use is highly habitual, and so operates in
relative independence from intentions and attitudes (Wood et al., 2021). As a result, these results
reinforce that changing intentions often does not suffice to change habits (Mazar et al., 2021).
General Discussion
Human behavior is often steered by learned cue-response habit associations as well as the
situational ease of enacting a behavior. In Chapter 1, I tracked the actual and perceived effects of
HABITS AND FRICTION
110
habit on behavior. In a naturalistic study, participants’ coffee drinking was about equally driven
by habit and fatigue. However, participants chiefly attributed their coffee drinking to fatigue,
overestimating fatigue’s impact while overlooking habit. This pattern was replicated in a
controlled experiment that independently manipulated habit strength and mood. In Chapter 2, I
examine the effects of friction and beliefs on voter turnout. As in the initial research, I found that
friction exerted a similar or slightly smaller effect on behavior as did beliefs. In contrast,
participants overwhelmingly attributed their turnout to beliefs over friction. Furthermore, this
bias was modestly but robustly correlated with support for policies that increase friction on
voting as well as opposition to policies that make voting easier. Finally, in Chapter 3, I tested the
effects of a minor friction intervention on habitual app use in a randomized controlled trial. To
probe the mechanism by which friction might exert its effects, I varied the frequency and dosage
of friction, and tracked on a daily basis both objective phone use as well as participants’
experience. Although most participants reduced their app use for up to a couple of days, this
pattern did not significantly differ by condition, and friction did not reduce app use more than a
no-treatment control. In contrast with an account suggesting that behavior change attempts falter
once initial optimistic intentions turn dour (Polivy & Herman, 1999), participants continued to
hold strong intentions to reduce their app use throughout the study, even as their app use returned
to baseline. This suggests that this rebounding is driven by processes independent of attitudinal
and intentional mechanisms.
The findings in this manuscript carry implications for the future of the field of behavior
change. First, the gaps between the perceived and actual effects of habit (in Chapter 1) and
friction (in Chapter 2) highlight that humans often have relatively little insight as to the causes of
their own behavior. That introspection is limited is hardly a novel research finding (Pronin, 2009;
HABITS AND FRICTION
111
Watson, 1913; Wilson, 2002), and yet research continues to rely heavily on self-reports, perhaps
because of the relative ease of collecting self-report data. Our findings provide further
documentation that relying on self-reported causes of behavior often fails to shed light on actual
causes of behavior. For example, people who score high on self-report emotional eating scales do
not necessarily eat more in response to negative feelings; instead, such surveys may tap a
tendency to explain one’s eating as emotionally-driven, regardless of whether that inference is
valid (Adriaanse et al., 2016; Adriaanse, Ridder, et al., 2011).
Chapter 3 tweaked features of the friction intervention (frequency, dose) to test the
effects of different variations of the same treatment. Often, the specifics of intervention delivery
are determined heuristically prior to data collection. Although the different variations did not
differ in their effects on app use duration, results did indicate that participants in the daily
conditions had lower intentions to reduce their app use. I hope that this assay encourages
researchers to test variations on an intervention, with the aim of developing evidence-based
guidance for fleshing out theoretical ideas into on-the-ground interventions. Such an approach is
valuable regardless of whether the variations differ in their effects. If they do, such differences
could inform theory. On the other hand, if differences are not found, this indicates that an
intervention is robust to such changes, which could help boost intervention generalizability
(Yarkoni, 2019).
Together, these findings of these three chapters shed light on the actual and perceived
roles that habit and friction play in our lives. People are often unaware of the antecedents of their
behavior, and this bias carries a wide gamut of consequences, from failing to recognize the
habitual component in one’s morning brew routine to supporting policies that maintain friction
and prevent others from realizing their franchise.
HABITS AND FRICTION
112
References
Adolph, K. E., Eppler, M. A., & Gibson, E. J. (1993). Crawling versus Walking Infants’
Perception of Affordances for Locomotion over Sloping Surfaces. Child Development,
64(4), 1158–1174. https://doi.org/10.2307/1131332
Adriaanse, M. A., Prinsen, S., de Witt Huberts, J. C., de Ridder, D. T. D., & Evers, C. (2016). ‘I
ate too much so I must have been sad’: Emotions as a confabulated reason for overeating.
Appetite, 103, 318–323. https://doi.org/10.1016/j.appet.2016.04.028
Adriaanse, M. A., Ridder, D. T. D. de, & Evers, C. (2011). Emotional eating: Eating when
emotional or emotional about eating? Psychology & Health, 26(1), 23–39.
https://doi.org/10.1080/08870440903207627
Adriaanse, M. A., Vinkers, C. D. W., De Ridder, D. T. D., Hox, J. J., & De Wit, J. B. F. (2011).
Do implementation intentions help to eat a healthy diet? A systematic review and meta-
analysis of the empirical evidence. Appetite, 56(1), 183–193.
https://doi.org/10.1016/j.appet.2010.10.012
Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision
Processes, 50(2), 179–211. https://doi.org/10.1016/0749-5978(91)90020-T
Ajzen, I. (2002). Constructing a TpB Questionnaire: Conceptual and Methodological
Considerations. Undefined. https://www.semanticscholar.org/paper/Constructing-a-TpB-
Questionnaire%3A-Conceptual-and-
Ajzen/6074b33b529ea56c175095872fa40798f8141867?p2df
Allcott, H., Gentzkow, M., & Song, L. (2021). Digital Addiction (Working Paper No. 28936).
National Bureau of Economic Research. https://doi.org/10.3386/w28936
Allcott, H., & Rogers, T. (2014). The Short-Run and Long-Run Effects of Behavioral
Interventions: Experimental Evidence from Energy Conservation. American Economic
Review, 104(10), 3003–3037. https://doi.org/10.1257/aer.104.10.3003
American Psychological Association. (2012). Stress in America: Our Health at Risk.
Anderson, I. A., & Wood, W. (2021). Habits and the electronic herd: The psychology behind
social media’s successes and failures. Consumer Psychology Review, 4(1), 83–99.
https://doi.org/10.1002/arcp.1063
Ansolabehere, S., & Hersh, E. (2012). Validation: What Big Data Reveal About Survey
Misreporting and the Real Electorate. Political Analysis, 20(4), 437–459.
HABITS AND FRICTION
113
Appelhans, B. M., French, S. A., Olinger, T., Bogucki, M., Janssen, I., Avery-Mamer, E. F., &
Powell, L. M. (2018). Leveraging delay discounting for health: Can time delays influence
food choice? Appetite, 126, 16–25. https://doi.org/10.1016/j.appet.2018.03.010
Avni‐Babad, D. (2011). Routine and feelings of safety, confidence, and well-being. British
Journal of Psychology, 102(2), 223–244. https://doi.org/10.1348/000712610X513617
Avni-Babad, D., & Ritov, I. (2003). Routine and the perception of time. Journal of Experimental
Psychology. General, 132(4), 543–550.
Bandura, A. (1991). Social cognitive theory of self-regulation. Organizational Behavior and
Human Decision Processes, 50(2), 248–287. https://doi.org/10.1016/0749-
5978(91)90022-L
BARC, & Nielsen. (2020). India: COVID-19 impact on daily smartphone usage 2020. Statista.
https://www-statista-com.libproxy1.usc.edu/statistics/1114423/india-coronavirus-impact-
on-smartphone-usage-time/
Baumeister, R. F., Vohs, K. D., & Tice, D. M. (2007). The Strength Model of Self-Control.
Current Directions in Psychological Science, 16(6), 351–355.
https://doi.org/10.1111/j.1467-8721.2007.00534.x
BCT Taxonomy Interventions. (2016, December 8). BCT Taxonomy Interventions.
https://www.bct-taxonomy.com/PHP-4.1.1/htdocsNew/interventions
Black, R. E. G. (2014). From Charioteer Myth to Shoulder Angel: A Rhetorical Look at Our
Divided Soul. Colloquy, 10, 36–49.
Blumenstock, J., Callen, M., & Ghani, T. (2018). Why Do Defaults Affect Behavior?
Experimental Evidence from Afghanistan. American Economic Review, 108(10), 2868–
2901. https://doi.org/10.1257/aer.20171676
Boltz, M. G., Kupperman, C., & Dunne, J. (1998). The role of learning in remembered duration.
Memory & Cognition, 26(5), 903–921. https://doi.org/10.3758/BF03201172
Borenstein, M., Hedges, L. V., Higgins, J. P. T., & Rothstein, H. R. (2021). Introduction to
Meta-Analysis. John Wiley & Sons.
Bouton, M. E., Todd, T. P., Vurbic, D., & Winterbauer, N. E. (2011). Renewal after the
extinction of free operant behavior. Learning & Behavior, 39(1), 57–67.
https://doi.org/10.3758/s13420-011-0018-6
Brennan Center for Justice. (2019). AVR Impact on State Voter Registration.
HABITS AND FRICTION
114
https://www.brennancenter.org/our-work/research-reports/avr-impact-state-voter-
registration
Brennan Center for Justice. (2021). Voting Laws Roundup. https://www.brennancenter.org/our-
work/research-reports/voting-laws-roundup-march-2021
Brijs, K., Daniels, S., Brijs, T., & Wets, G. (2011). An experimental approach towards the
evaluation of a seat belt campaign with an inside view on the psychology behind seat belt
use. Transportation Research Part F: Traffic Psychology and Behaviour, 14(6), 600–613.
https://doi.org/10.1016/j.trf.2011.07.003
Brooklyn Historic Railway Association. (n.d.). Science of Railway Locomotion. Retrieved March
21, 2021, from http://www.brooklynrail.net/science_of_railway_locomotion.html
Cantoni, E. (2020). A Precinct Too Far: Turnout and Voting Costs. American Economic Journal:
Applied Economics, 12(1), 61–85. https://doi.org/10.1257/app.20180306
Carroll, K. M. (1998). A Cognitive-behavioral Approach: Treating Cocaine Addiction. U.S.
Department of Health and Human Services, National Institutes of Health, National
Institute on Drug Abuse.
Choi, I., Nisbett, R. E., & Norenzayan, A. (1999). Causal attribution across cultures: Variation
and universality. Psychological Bulletin, 125(1), 47–63.
http://dx.doi.org.libproxy1.usc.edu/10.1037/0033-2909.125.1.47
Cohen, G. L., Garcia, J., Apfel, N., & Master, A. (2006). Reducing the Racial Achievement Gap:
A Social-Psychological Intervention. Science, 313(5791), 1307–1310.
https://doi.org/10.1126/science.1128317
Crandall, C. S., D’Anello, S., Sakalli, N., Lazarus, E., Nejtardt, G. W., & Feather, N. T. (2001).
An Attribution-Value Model of Prejudice: Anti-Fat Attitudes in Six Nations. Personality
and Social Psychology Bulletin, 27(1), 30–37.
https://doi.org/10.1177/0146167201271003
Danner, U. N., Aarts, H., & Vries, N. K. de. (2008). Habit vs. intention in the prediction of future
behaviour: The role of frequency, context stability and mental accessibility of past
behaviour. British Journal of Social Psychology, 47(2), 245–265.
https://doi.org/10.1348/014466607X230876
De Camp, J. E. (1920). Relative distance as a factor in the white rat’s selection of a path.
Psychobiology, 2(3), 245–253. http://dx.doi.org.libproxy2.usc.edu/10.1037/h0075411
Delgado, M. M., Han, B. S. G., & Bain, M. J. (2022). Domestic cats (Felis catus) prefer freely
HABITS AND FRICTION
115
available food over food that requires effort. Animal Cognition, 25(1), 95–102.
https://doi.org/10.1007/s10071-021-01530-3
digitaltrends. (2020, October 25). How Frictionless Technology Is Making Us Miserable. Digital
Trends. https://www.digitaltrends.com/web/the-frictionless-internet/
Duckworth, A. L., Gendler, T. S., & Gross, J. J. (2016). Situational Strategies for Self-Control.
Perspectives on Psychological Science, 11(1), 35–55.
https://doi.org/10.1177/1745691615623247
Duckworth, A. L., & Kern, M. L. (2011). A meta-analysis of the convergent validity of self-
control measures. Journal of Research in Personality, 45(3), 259–268.
https://doi.org/10.1016/j.jrp.2011.02.004
Duckworth, A. L., Milkman, K. L., & Laibson, D. (2018). Beyond Willpower: Strategies for
Reducing Failures of Self-Control. Psychological Science in the Public Interest, 19(3),
102–129. https://doi.org/10.1177/1529100618821893
Duckworth, A. L., White, R. E., Matteucci, A. J., Shearer, A., & Gross, J. J. (2016). A Stitch in
Time: Strategic Self-Control in High School and College Students. Journal of
Educational Psychology, 108(3), 329–341. https://doi.org/10.1037/edu0000062
Eagly, A. H., & Chaiken, S. (1993). The psychology of attitudes (pp. xxii, 794). Harcourt Brace
Jovanovich College Publishers.
Ebeling, F., & Lotz, S. (2015). Domestic uptake of green energy promoted by opt-out tariffs.
Nature Climate Change, 5(9), 868–871. https://doi.org/10.1038/nclimate2681
Ersche, K. D., Gillan, C. M., Jones, P. S., Williams, G. B., Ward, L. H. E., Luijten, M., de Wit,
S., Sahakian, B. J., Bullmore, E. T., & Robbins, T. W. (2016). Carrots and sticks fail to
change behavior in cocaine addiction. Science, 352(6292), 1468–1471.
https://doi.org/10.1126/science.aaf3700
Eyal, N. (2014). Hooked: How to Build Habit-Forming Products (R. Hoover, Ed.; Illustrated
edition). Portfolio.
Fitz, N., Kushlev, K., Jagannathan, R., Lewis, T., Paliwal, D., & Ariely, D. (2019). Batching
smartphone notifications can improve well-being. Computers in Human Behavior, 101,
84–94. https://doi.org/10.1016/j.chb.2019.07.016
Fujiwara, T., Meng, K., & Vogl, T. (2016). Habit Formation in Voting: Evidence from Rainy
Elections. American Economic Journal: Applied Economics, 8(4), 160–188.
HABITS AND FRICTION
116
Funder, D. C. (2016). Taking Situations Seriously: The Situation Construal Model and the
Riverside Situational Q-Sort. Current Directions in Psychological Science, 25(3), 203–
208. https://doi.org/10.1177/0963721416635552
Galla, B. M., & Duckworth, A. L. (2015). More than resisting temptation: Beneficial habits
mediate the relationship between self-control and positive life outcomes. Journal of
Personality and Social Psychology, 109(3), 508–525.
http://dx.doi.org.libproxy1.usc.edu/10.1037/pspp0000026
Garber, M. (2013, August 26). 10 Things You Probably Did Not Know About Eraser
Technology. The Atlantic. https://www.theatlantic.com/technology/archive/2013/08/10-
things-you-probably-did-not-know-about-eraser-technology/279028/
Gardner, B. (2015). A review and analysis of the use of ‘habit’ in understanding, predicting and
influencing health-related behaviour. Health Psychology Review, 9(3), 277–295.
https://doi.org/10.1080/17437199.2013.876238
Gardner, B., Abraham, C., Lally, P., & Bruijn, G.-J. de. (2012). Towards parsimony in habit
measurement: Testing the convergent and predictive validity of an automaticity subscale
of the Self-Report Habit Index. International Journal of Behavioral Nutrition and
Physical Activity, 9(1), 102. https://doi.org/10.1186/1479-5868-9-102
Gelman, A., & Loken, E. (2014). The statistical crisis in science: Data-dependent analysis—A
“garden of forking paths”—Explains why many statistically significant comparisons
don’t hold up. Scientific American, 02(6), 6.
Gershman, S. J., Gerstenberg, T., Baker, C. L., & Cushman, F. A. (2016). Plans, Habits, and
Theory of Mind. PLoS ONE, 11(9). https://doi.org/10.1371/journal.pone.0162246
Gibson, J. J. (1966). The Senses Considered as Perceptual Systems (Vol. 44). Houghton Mifflin.
https://www.journals.uchicago.edu/doi/10.1086/406033
Godin, G., & Kok, G. (1996). The Theory of Planned Behavior: A Review of its Applications to
Health-Related Behaviors. American Journal of Health Promotion, 11(2), 87–98.
https://doi.org/10.4278/0890-1171-11.2.87
Google. (2018, February). Find Out How You Stack Up to New Industry Benchmarks for Mobile
Page Speed. Think with Google. https://www.thinkwithgoogle.com/marketing-
strategies/app-and-mobile/mobile-page-speed-new-industry-benchmarks/
Grenard, J. L., Stacy, A. W., Shiffman, S., Baraldi, A. N., MacKinnon, D. P., Lockhart, G.,
Kisbu-Sakarya, Y., Boyle, S., Beleva, Y., Koprowski, C., Ames, S. L., & Reynolds, K. D.
(2013). Sweetened drink and snacking cues in adolescents. A study using ecological
HABITS AND FRICTION
117
momentary assessment. Appetite, 67, 61–73. https://doi.org/10.1016/j.appet.2013.03.016
Gross, L. S., Li, L., Ford, E. S., & Liu, S. (2004). Increased consumption of refined
carbohydrates and the epidemic of type 2 diabetes in the United States: An ecologic
assessment. The American Journal of Clinical Nutrition, 79(5), 774–779.
https://doi.org/10.1093/ajcn/79.5.774
Hackbarth, G., Grover, V., & Yi, M. Y. (2003). Computer playfulness and anxiety: Positive and
negative mediators of the system experience effect on perceived ease of use. Information
& Management, 40(3), 221–232. https://doi.org/10.1016/S0378-7206(02)00006-X
Hagger, M. S., Rebar, A. L., Mullan, B., Lipp, O. V., & Chatzisarantis, N. L. D. (2015). The
subjective experience of habit captured by self-report indexes may lead to inaccuracies in
the measurement of habitual action. Health Psychology Review, 9(3), 296–302.
https://doi.org/10.1080/17437199.2014.959728
Hardwick, R. M., Forrence, A. D., Krakauer, J. W., & Haith, A. M. (2019). Time-dependent
competition between goal-directed and habitual response preparation. Nature Human
Behaviour, 3(12), 1252–1262. https://doi.org/10.1038/s41562-019-0725-0
Heintzelman, S. J., & King, L. A. (2019). Routines and Meaning in Life. Personality and Social
Psychology Bulletin, 45(5), 688–699. https://doi.org/10.1177/0146167218795133
Hofmann, W., Vohs, K. D., & Baumeister, R. F. (2012). What People Desire, Feel Conflicted
About, and Try to Resist in Everyday Life. Psychological Science, 23(6), 582–588.
https://doi.org/10.1177/0956797612437426
Hollands, G. J., Carter, P., Anwer, S., King, S. E., Jebb, S. A., Ogilvie, D., Shemilt, I., Higgins,
J. P. T., & Marteau, T. M. (2019). Altering the availability or proximity of food, alcohol,
and tobacco products to change their selection and consumption. Cochrane Database of
Systematic Reviews, 9. https://doi.org/10.1002/14651858.CD012573.pub3
Houten, R. V., Nau, P. A., & Merrigan, M. (1981). Reducing elevator energy use: A comparison
of posted feedback and reduced elevator convenience. Journal of Applied Behavior
Analysis, 14(4), 377–387. https://doi.org/10.1901/jaba.1981.14-377
Hull, C. L. (1943). Principles of behavior: An introduction to behavior theory (Vol. 422).
Appleton-century-crofts.
Ickes, W., Snyder, M., & Garcia, S. (1997). Personality Influences on the Choice of Situations.
In Handbook of Personality Psychology (pp. 165–195). Academic Press. https://www-
sciencedirect-com.libproxy1.usc.edu/science/article/pii/B9780121346454500081
HABITS AND FRICTION
118
Inzlicht, M., Shenhav, A., & Olivola, C. Y. (2018). The Effort Paradox: Effort Is Both Costly
and Valued. Trends in Cognitive Sciences, 22(4), 337–349.
https://doi.org/10.1016/j.tics.2018.01.007
Itzchakov, G., Uziel, L., & Wood, W. (2018). When attitudes and habits don’t correspond: Self-
control depletion increases persuasion but not behavior. Journal of Experimental Social
Psychology, 75, 1–10. https://doi.org/10.1016/j.jesp.2017.10.011
Ji, M. F., & Wood, W. (2007). Purchase and Consumption Habits: Not Necessarily What You
Intend. Journal of Consumer Psychology (Taylor & Francis Ltd), 17(4), 261–276.
https://doi.org/10.1080/10577400701542403
Johnson, E. J., & Goldstein, D. (2003). Do Defaults Save Lives? Science, 302(5649), 1338–
1339. https://doi.org/10.1126/science.1091721
Jones, E. E., & Nisbett, R. E. (1971). The actor and the observer: Divergent perceptions of the
causes of behavior. In E. E. Jones, D. E. Kanouse, H. H. Kelly, R. E. Nisbett, S. Valins,
& B. Weiner (Eds.), Attribution: Perceiving the causes of behavior (pp. 77–94). General
Learning Press.
Jung, D.-J. (2017). Irrationalizing the rational choice model of voting: The moderating effects of
partisanship on turnout decisions in Western and postcommunist democracies. Electoral
Studies, 50, 26–38. https://doi.org/10.1016/j.electstud.2017.09.005
Kenny, D. A., Mohr, C. D., & Levesque, M. J. (2001). A social relations variance partitioning of
dyadic behavior. Psychological Bulletin, 127(1), 128–141.
http://dx.doi.org.libproxy1.usc.edu/10.1037/0033-2909.127.1.128
Knowlton, B. J., & Diedrichsen, J. (2018). Editorial overview: Habits and skills. Current
Opinion in Behavioral Sciences, 20, iv–vi. https://doi.org/10.1016/j.cobeha.2018.02.009
Koehler, D. J., & Poon, C. S. K. (2006). Self-predictions overweight strength of current
intentions. Journal of Experimental Social Psychology, 42(4), 517–524.
https://doi.org/10.1016/j.jesp.2005.08.003
Kool, W., McGuire, J. T., Rosen, Z. B., & Botvinick, M. M. (2010). Decision Making and the
Avoidance of Cognitive Demand. Journal of Experimental Psychology. General, 139(4),
665–682. https://doi.org/10.1037/a0020198
Labrecque, J. S., & Wood, W. (2015). What measures of habit strength to use? Comment on
Gardner (2015). Health Psychology Review, 9(3), 303–310.
https://doi.org/10.1080/17437199.2014.992030
HABITS AND FRICTION
119
Lewin, K. (1936). Principles of topological psychology. (F. Heider & G. M. Heider, Trans.).
McGraw-Hill. https://doi.org/10.1037/10019-000
Lewin, K. (1947). Frontiers in Group Dynamics: II. Channels of Group Life; Social Planning and
Action Research. Human Relations, 1(2), 143–153.
https://doi.org/10.1177/001872674700100201
Loewenstein, G. (1996). Out of Control: Visceral Influences on Behavior. Organizational
Behavior and Human Decision Processes, 65(3), 272–292.
https://doi.org/10.1006/obhd.1996.0028
Luque, D., Molinero, S., Watson, P., López, F. J., & Le Pelley, M. E. (2020). Measuring habit
formation through goal-directed response switching. Journal of Experimental
Psychology: General, 149(8), 1449–1459.
http://dx.doi.org.libproxy1.usc.edu/10.1037/xge0000722
MacLeod, C. M. (1992). The Stroop task: The “gold standard” of attentional measures. Journal
of Experimental Psychology: General, 121(1), 12–14.
http://dx.doi.org.libproxy1.usc.edu/10.1037/0096-3445.121.1.12
Madrian, B. C., & Shea, D. F. (2001). The power of suggestion: Inertia in 401(k) participation
and savings behavior. The Quarterly Journal of Economics, 116(4).
Malle, B. F. (2006). The actor-observer asymmetry in attribution: A (surprising) meta-analysis.
Psychological Bulletin, 132(6), 895–919.
http://dx.doi.org.libproxy1.usc.edu/10.1037/0033-2909.132.6.895
Mantzari, E., Vogt, F., Shemilt, I., Wei, Y., Higgins, J. P. T., & Marteau, T. M. (2015). Personal
financial incentives for changing habitual health-related behaviors: A systematic review
and meta-analysis. Preventive Medicine, 75, 75–85.
https://doi.org/10.1016/j.ypmed.2015.03.001
Mazar, A., Tomaino, G., Carmon, Z., & Wood, W. (2021). Habits to Save our Habitat: Using the
Psychology of Habits to Promote Sustainability. Behavioral Science and Policy, 7(2).
https://doi.org/10.31234/osf.io/kwdh9
Mazar, A., & Wood, W. (2018). Defining Habit in Psychology. In B. Verplanken (Ed.), The
Psychology of Habit: Theory, Mechanisms, Change, and Contexts (pp. 13–29). Springer
International Publishing. https://doi.org/10.1007/978-3-319-97529-0_2
McFerran, B., & Mukhopadhyay, A. (2013). Lay Theories of Obesity Predict Actual Body Mass.
Psychological Science, 24(8), 1428–1436. https://doi.org/10.1177/0956797612473121
HABITS AND FRICTION
120
McQueen, A., & Klein, W. M. P. (2006). Experimental manipulations of self-affirmation: A
systematic review. Self and Identity, 5(4), 289–354.
https://doi.org/10.1080/15298860600805325
Metcalfe, J., & Mischel, W. (1999). A hot/cool-system analysis of delay of gratification:
Dynamics of willpower. Psychological Review, 106(1), 3–19.
http://dx.doi.org.libproxy2.usc.edu/10.1037/0033-295X.106.1.3
Michie, S., Richardson, M., Johnston, M., Abraham, C., Francis, J., Hardeman, W., Eccles, M.
P., Cane, J., & Wood, C. E. (2013). The behavior change technique taxonomy (v1) of 93
hierarchically clustered techniques: Building an international consensus for the reporting
of behavior change interventions. Annals of Behavioral Medicine: A Publication of the
Society of Behavioral Medicine, 46(1), 81–95. https://doi.org/10.1007/s12160-013-9486-
6
Mills, C., & D’Mello, S. (2014). On the Validity of the Autobiographical Emotional Memory
Task for Emotion Induction. PLOS ONE, 9(4), e95837.
https://doi.org/10.1371/journal.pone.0095837
Milyavskaya, M., & Inzlicht, M. (2017). What’s So Great About Self-Control? Examining the
Importance of Effortful Self-Control and Temptation in Predicting Real-Life Depletion
and Goal Attainment. Social Psychological and Personality Science, 8(6), 603–611.
https://doi.org/10.1177/1948550616679237
Mitchell, T. R., Thompson, L., Peterson, E., & Cronk, R. (1997). Temporal Adjustments in the
Evaluation of Events: The “Rosy View.” Journal of Experimental Social Psychology,
33(4), 421–448. https://doi.org/10.1006/jesp.1997.1333
MMD Labo. (2020). Japan: Smartphone usage duration 2019. Statista. https://www-statista-
com.libproxy1.usc.edu/statistics/1028679/japan-duration-use-smartphone/
Morris, A., Phillips, J., Huang, K., & Cushman, F. (2021). Generating Options and Choosing
Between Them Depend on Distinct Forms of Value Representation: Psychological
Science. https://doi.org/10.1177/09567976211005702
Murray, H. A. (1938). Explorations in personality: A clinical and experimental study of fifty men
of college age (pp. xiv, 761). Oxford Univ. Press.
Neal, D. T., Wood, W., & Drolet, A. (2013). How do people adhere to goals when willpower is
low? The profits (and pitfalls) of strong habits. Journal of Personality and Social
Psychology, 104(6), 959–975. http://dx.doi.org.libproxy1.usc.edu/10.1037/a0032626
Neal, D. T., Wood, W., Wu, M., & Kurlander, D. (2011). The Pull of the Past: When Do Habits
HABITS AND FRICTION
121
Persist Despite Conflict With Motives? Personality and Social Psychology Bulletin,
37(11), 1428–1437. https://doi.org/10.1177/0146167211419863
Nickerson, D. W., & Rogers, T. (2010). Do You Have a Voting Plan?: Implementation
Intentions, Voter Turnout, and Organic Plan Making. Psychological Science, 21(2), 194–
199. https://doi.org/10.1177/0956797609359326
Nielsen. (2020). Australia—Average total time spent online by active internet users on
smartphones by age group December 2019. Statista. https://www-statista-
com.libproxy1.usc.edu/statistics/1112228/australia-average-total-time-spent-online-by-
active-internet-users-smartphones-by-age-group/
Norcross, J. C., & Vangarelli, D. J. (1988). The resolution solution: Longitudinal examination of
New Year’s change attempts. Journal of Substance Abuse, 1(2), 127–134.
Norman, D. (2013). The Design of Everyday Things: Revised and Expanded Edition (Revised
edition). Basic Books.
Open Secrets. (2021a, February 1). Expenditures Breakdown, Donald Trump, 2020 cycle.
OpenSecrets. https://www.opensecrets.org/2020-presidential-race/donald-
trump/expenditures?id=N00023864
Open Secrets. (2021b, February 1). Expenditures Breakdown, Joe Biden, 2020 cycle.
OpenSecrets. https://www.opensecrets.org/2020-presidential-race/joe-
biden/expenditures?id=N00001669
Orbell, S., & Verplanken, B. (2010). The automatic component of habit in health behavior: Habit
as cue-contingent automaticity. Health Psychology, 29(4), 374–383.
http://dx.doi.org.libproxy1.usc.edu/10.1037/a0019596
Orbit Media Studios. (2015, August 26). Web Design Standards: 10 Best Practices on the Top 50
Websites. https://www.orbitmedia.com/blog/web-design-standards/
Ouellette, J. A., & Wood, W. (1998). Habit and intention in everyday life: The multiple
processes by which past behavior predicts future behavior. Psychological Bulletin,
124(1), 54–74. http://dx.doi.org.libproxy1.usc.edu/10.1037/0033-2909.124.1.54
Oulasvirta, A., Rattenbury, T., Ma, L., & Raita, E. (2012). Habits make smartphone use more
pervasive. Personal and Ubiquitous Computing, 16(1), 105–114.
https://doi.org/10.1007/s00779-011-0412-2
Petrescu, D. C., Hollands, G. J., Couturier, D.-L., Ng, Y.-L., & Marteau, T. M. (2016). Public
Acceptability in the UK and USA of Nudging to Reduce Obesity: The Example of
HABITS AND FRICTION
122
Reducing Sugar-Sweetened Beverages Consumption. PLOS ONE, 11(6), e0155995.
https://doi.org/10.1371/journal.pone.0155995
Pew Research Center. (2014). Emerging and Developing Economies Much More Optimistic than
Rich Countries about the Future.
https://www.pewresearch.org/global/2014/10/09/emerging-and-developing-economies-
much-more-optimistic-than-rich-countries-about-the-future/
Pew Research Center. (2021). Share of Republicans saying ‘everything possible’ should be done
to make voting easy declines sharply. https://www.pewresearch.org/fact-
tank/2021/04/01/share-of-republicans-saying-everything-possible-should-be-done-to-
make-voting-easy-declines-sharply/
Plato. (1967). Laws (R. G. Bury, Trans.). Harvard University Press. (Original work published
347 B.C.E.)
Polivy, J., & Herman, C. P. (1999). The effects of resolving to diet on restrained and
unrestrained eaters: The “false hope syndrome.” International Journal of Eating
Disorders, 26(4), 434–447. https://doi.org/10.1002/(SICI)1098-
108X(199912)26:4<434::AID-EAT9>3.0.CO;2-0
Potrafke, N., & Roesel, F. (2020). Opening hours of polling stations and voter turnout: Evidence
from a natural experiment. The Review of International Organizations, 15(1), 133–163.
https://doi.org/10.1007/s11558-018-9305-8
Preischl, W., & Hellmich, M. (2013). Human error probabilities from operational experience of
German nuclear power plants. Reliability Engineering & System Safety, 109, 150–159.
https://doi.org/10.1016/j.ress.2012.08.004
Pronin, E. (2009). The Introspection Illusion. In M. P. Zanna (Ed.), Advances in experimental
social psychology. (Vol. 41). Academic Press.
Quinn, J. M., Pascoe, A., Wood, W., & Neal, D. T. (2010). Can’t Control Yourself? Monitor
Those Bad Habits. Personality and Social Psychology Bulletin, 36(4), 499–511.
https://doi.org/10.1177/0146167209360665
Rauthmann, J. F., Gallardo-Pujol, D., Link to external site, this link will open in a new window,
Guillaume, E. M., Todd, E., Nave, C. S., Sherman, R. A., Ziegler, M., Jones, A. B., &
Funder, D. C. (2014). The Situational Eight DIAMONDS: A taxonomy of major
dimensions of situation characteristics. Journal of Personality and Social Psychology,
107(4), 677–718. http://dx.doi.org.libproxy1.usc.edu/10.1037/a0037250
Rebar, A. L., Elavsky, S., Maher, J. P., Doerksen, S. E., & Conroy, D. E. (2014). Habits Predict
HABITS AND FRICTION
123
Physical Activity on Days When Intentions Are Weak. Journal of Sport & Exercise
Psychology, 36(2), 157–165.
Reeves, B., Ram, N., Robinson, T. N., Cummings, J. J., Giles, C. L., Pan, J., Chiatti, A., Cho,
M., Roehrick, K., Yang, X., Gagneja, A., Brinberg, M., Muise, D., Lu, Y., Luo, M.,
Fitzgerald, A., & Yeykelis, L. (2021). Screenomics: A Framework to Capture and
Analyze Personal Life Experiences and the Ways that Technology Shapes Them.
Human–Computer Interaction, 36(2), 150–201.
https://doi.org/10.1080/07370024.2019.1578652
Reis, H. T. (2008). Reinvigorating the Concept of Situation in Social Psychology. Personality
and Social Psychology Review, 12(4), 311–329.
https://doi.org/10.1177/1088868308321721
Rosset, E. (2008). It’s no accident: Our bias for intentional explanations. Cognition, 108(3), 771–
780. https://doi.org/10.1016/j.cognition.2008.07.001
Rozin, P. (2011). Nudge to Nobesity I: Minor Changes in Accessibility Decrease Food Intake.
Judgment and Decision Making, 6(4), 12.
Ruschendorf, L. (1995). Convergence of the Iterative Proportional Fitting Procedure. The Annals
of Statistics, 23(4), 1160–1174.
Sassenberg, K., & Ditrich, L. (2019). Research in Social Psychology Changed Between 2011 and
2016: Larger Sample Sizes, More Self-Report Measures, and More Online Studies.
Advances in Methods and Practices in Psychological Science, 2(2), 107–114.
https://doi.org/10.1177/2515245919838781
Schultz, P. W., Nolan, J. M., Cialdini, R. B., Goldstein, N. J., & Griskevicius, V. (2007). The
Constructive, Destructive, and Reconstructive Power of Social Norms. Psychological
Science, 18(5), 429–434. https://doi.org/10.1111/j.1467-9280.2007.01917.x
Schwabe, L., & Wolf, O. T. (2009). Stress Prompts Habit Behavior in Humans. Journal of
Neuroscience, 29(22), 7191–7198. https://doi.org/10.1523/JNEUROSCI.0979-09.2009
Service, O., Hallsworth, M., Halpern, D., Algate, F., Gallagher, R., Nguyen, S., Ruda, S., &
Sanders, M. (2015). EAST - Four simple ways to apply behavioral insights.
https://www.behaviouralinsights.co.uk/wp-content/uploads/2015/07/BIT-Publication-
EAST_FA_WEB.pdf
Shane, F. (2005). Cognitive Reflection and Decision Making. The Journal of Economic
Perspectives, 19(4), 25.
HABITS AND FRICTION
124
Sheeran, P., & Webb, T. L. (2016). The Intention–Behavior Gap. Social and Personality
Psychology Compass, 10(9), 503–518. https://doi.org/10.1111/spc3.12265
Shiffman, S., Hufford, M., Hickcox, M., Paty, J. A., Gnys, M., & Kassel, J. D. (1997).
Remember that? A comparison of real-time versus retrospective recall of smoking lapses.
Journal of Consulting and Clinical Psychology, 65(2), 292–300.
http://dx.doi.org.libproxy2.usc.edu/10.1037/0022-006X.65.2.292.a
Shiffman, S., Stone, A. A., & Hufford, M. R. (2008). Ecological Momentary Assessment.
Annual Review of Clinical Psychology, 4(1), 1–32.
https://doi.org/10.1146/annurev.clinpsy.3.022806.091415
Simons, D. J., Shoda, Y., & Lindsay, D. S. (2017). Constraints on Generality (COG): A
Proposed Addition to All Empirical Papers. Perspectives on Psychological Science,
12(6), 1123–1128. https://doi.org/10.1177/1745691617708630
Statista. (2021). Mobile phone daily usage time in U.S. 2023. https://www-statista-
com.libproxy2.usc.edu/statistics/1045353/mobile-device-daily-usage-time-in-the-us/
Steegen, S., Tuerlinckx, F., Gelman, A., & Vanpaemel, W. (2016). Increasing Transparency
Through a Multiverse Analysis. Perspectives on Psychological Science, 11(5), 702–712.
https://doi.org/10.1177/1745691616658637
Stevens, D., Sullivan, J., Allen, B., & Alger, D. (2008). What’s Good for the Goose is Bad for
the Gander: Negative Political Advertising, Partisanship, and Turnout. The Journal of
Politics, 70(2), 527–541. https://doi.org/10.1017/S0022381608080481
Stewart, C., & Ansolabehere, S. (2015). Waiting to Vote. Election Law Journal: Rules, Politics,
and Policy, 14(1), 47–53. https://doi.org/10.1089/elj.2014.0292
Stone, A., Shiffman, S., Atienza, A., & Nebeling, L. (2007). Historical Roots and Rationale of
Ecological Momentary Assessment (EMA). In The Science of Real-Time Data Capture:
Self-Reports in Health Research (pp. 3–10). Oxford University Press.
Tangney, J. P., Baumeister, R. F., & Boone, A. L. (2004). High Self-Control Predicts Good
Adjustment, Less Pathology, Better Grades, and Interpersonal Success. Journal of
Personality, 72(2), 271–324. https://doi.org/10.1111/j.0022-3506.2004.00263.x
Thaler, R. H., & Sunstein, C. R. (2009). Nudge: Improving Decisions About Health, Wealth, and
Happiness. Penguin.
Thøgersen, J., & Møller, B. (2008). Breaking car use habits: The effectiveness of a free one-
month travelcard. Transportation, 35(3), 329–345. https://doi.org/10.1007/s11116-008-
HABITS AND FRICTION
125
9160-1
Thompson, E. R. (2007). Development and Validation of an Internationally Reliable Short-Form
of the Positive and Negative Affect Schedule (PANAS). Journal of Cross-Cultural
Psychology, 38(2), 227–242. https://doi.org/10.1177/0022022106297301
U.S. Census Bureau. (2019). American Community Survey 1-year estimates (2019).
https://www.census.gov/newsroom/press-kits/2020/acs-1year.html
Verplanken, B., & Orbell, S. (2003). Reflections on Past Behavior: A Self-Report Index of Habit
Strength. Journal of Applied Social Psychology, 33(6), 1313–1330.
https://doi.org/10.1111/j.1559-1816.2003.tb01951.x
Verplanken, B., & Orbell, S. (2022). Attitudes, habits, and behavior change. Annual Review of
Psychology, 73, 327–352. https://doi.org/10.1146/annurev-psych-020821-011744
Verplanken, B., & Roy, D. (2016). Empowering interventions to promote sustainable lifestyles:
Testing the habit discontinuity hypothesis in a field experiment. Journal of
Environmental Psychology, 45, 127–134. https://doi.org/10.1016/j.jenvp.2015.11.008
Vivo, & Cybermedia Research. (2019). Time spent by Indians on smartphones as of December
2019, by age group. Statista. https://www-statista-
com.libproxy1.usc.edu/statistics/1116572/india-time-spent-on-smartphones-by-age-
group/
Watson, J. B. (1913). Psychology as the behaviorist views it. Psychological Review, 20(2), 158–
177. https://doi.org/10.1037/h0074428
Webb, T. L., Sheeran, P., & Luszczynska, A. (2009). Planning to break unwanted habits: Habit
strength moderates implementation intention effects on behaviour change. British Journal
of Social Psychology, 48(3), 507–523. https://doi.org/10.1348/014466608X370591
White, K., Hardisty, D. J., & Habib, R. (2019). The Elusive Green Consumer. Harvard Business
Review, 97(4), 124–133.
Wilson, T. D. (2002). Strangers to Ourselves. Harvard University Press.
Wilson, T. D., Laser, P. S., & Stone, J. I. (1982). Judging the predictors of one’s own mood:
Accuracy and the use of shared theories. Journal of Experimental Social Psychology,
18(6), 537–556. https://doi.org/10.1016/0022-1031(82)90072-5
Wood, W. (2017). Habit in Personality and Social Psychology. Personality and Social
Psychology Review, 21(4), 389–403. https://doi.org/10.1177/1088868317720362
HABITS AND FRICTION
126
Wood, W., Mazar, A., & Neal, D. (2021). Habits and Goals in Human Behavior: Separate but
Interacting Systems. PsyArXiv. https://doi.org/10.31234/osf.io/qvrby
Wood, W., & Neal, D. T. (2016). Healthy through habit: Interventions for initiating &
maintaining health behavior change. Behavioral Science & Policy; Washington, 2(1), 71–
83.
Wood, W., Quinn, J. M., & Kashy, D. A. (2002). Habits in everyday life: Thought, emotion, and
action. Journal of Personality and Social Psychology, 83(6), 1281–1297.
http://dx.doi.org.libproxy2.usc.edu/10.1037/0022-3514.83.6.1281
Wood, W., & Rünger, D. (2016). Psychology of Habit. Annual Review of Psychology, 67(1),
289–314. https://doi.org/10.1146/annurev-psych-122414-033417
Wood, W., Tam, L., Link to external site, this link will open in a new window, & Witt, M. G.
(2005). Changing circumstances, disrupting habits. Journal of Personality and Social
Psychology, 88(6), 918–933. http://dx.doi.org.libproxy1.usc.edu/10.1037/0022-
3514.88.6.918
Yablonski, J. (2021). Jakob’s Law. Laws of UX. /jakobs-law/
Yang, Y., Read, S. J., & Miller, L. C. (2006). A taxonomy of situations from Chinese idioms.
Journal of Research in Personality, 40(5), 750–778.
https://doi.org/10.1016/j.jrp.2005.09.007
Yang, Y., Read, S. J., & Miller, L. C. (2009). The Concept of Situations. Social and Personality
Psychology Compass, 3(6), 1018–1037. https://doi.org/10.1111/j.1751-
9004.2009.00236.x
Yarkoni, T. (2019). The Generalizability Crisis [Preprint]. PsyArXiv.
https://doi.org/10.31234/osf.io/jqw35
Zhang, Z., & Yuan, K.-H. (2018). Practical Statistical Power Analysis Using Webpower and R.
Zimmerman, B. J. (2000). Attaining Self-Regulation: A Social Cognitive Perspective. In
Handbook of Self-Regulation (pp. 13–39). Academic Press. https://www-sciencedirect-
com.libproxy1.usc.edu/science/article/pii/B9780121098902500317
Zimmermann, L., & Sobolev, M. (2020). Digital Nudges for Screen Time Reduction: A
Randomized Control Trial with Performance and Wellbeing Outcomes. PsyArXiv.
https://doi.org/10.31234/osf.io/nmgdz

HABITS AND FRICTION
127
Appendices
Appendix A - Supplementary Materials for Chapter 1
Addendum to Study 1
Table S1 gives the results of the main analyses without participant exclusions.
Comprehension Check
Due to a technical issue, the comprehension check was not displayed for some participants.
Thus, participants who did not answer it (183, or 18%) were included in analyses. Excluding
these participants did not notably alter the results.
Secondary Analyses
To compare the size of the mood-habit difference in attribution to that in actual behavior, we
created attribution difference scores by subtracting each participant’s habit attribution from their
mood attribution. A one-sample t-test compared these scores to the actual difference in Tjur’s R
2
between a habit-only and mood-only model (a small .57% difference in favor of habit; R
2
mood =
.03%, R
2
habit = .60%). Demonstrating the bias to overlook habits, the perceived difference in
favor of mood in attributions, M = 17.10%, 95% CI [15.66, 18.53], was significantly larger than
the actual difference of .57% in favor of habit, t(800) = 24.09, p < .001, Cohen’s d = 1.21, 95%
CI [1.10, 1.31]. Although the results of these pre-registered analyses are qualified by the
difference in orders of magnitude between actual and perceived effects, this secondary analysis
may provide some additional evidence of the bias to overlook habits in favor of inner states.
In testing the habit measure (SRBAI), helping behavior (yes/no) was predicted from perceived
automaticity (SRBAI), positive affect (positive PANAS items) and negative affect (negative
PANAS items). SRBAI had no effect, OR = 0.97, 95% CI [0.89, 1.06], p = .537. There was a
non-significant trend in the direction of a higher likelihood of helping in the positive affect
HABITS AND FRICTION
128
condition, OR = 1.09, 95% CI [0.99, 1.20], p = .067. Negative affect had no effect, OR = 0.91,
95% CI [0.80, 1.03], p = .140.
Exploratory analyses predicted separately the attributions to mood and habit from the mood and
habit manipulations. A linear regression model predicting attribution to habit revealed no
significant effects. An additional linear regression model predicting attribution to mood revealed
only that participants in the strong habit conditions tended to attribute less importance to mood,
compared with participants in the weak habit condition, b = -1.76, 95% CI [-2.94, -0.58], p =
.003.
Table S1.1
Study 1: Results of Main Regression Analysis Without Exclusions (N = 1015)
Predictors Odds Ratios 95% CI p
(Intercept) 0.92 0.74 – 1.15 0.463
Habit strength 0.87 0.76 – 0.99 0.033
Negative mood 1.03 0.76 – 1.39 0.862
Positive mood 1.07 0.79 – 1.44 0.684
Note. Linear regression predicting likelihood of helping (yes/no) from manipulated habit
strength (weak vs. strong) and mood (positive, negative, neutral)..

HABITS AND FRICTION
129
Addendum to Study 2
Table S1.2
Study 2: Main Analysis Results Using SRHI
Predictors OR 95% CI
Intercept 0.18 0.11 – 0.28
Habit strength 1.27 1.09 – 1.49
Fatigue (within) 1.22 1.08 – 1.39
Fatigue (between) 0.92 0.81 – 1.07
F. attribution 1.00 0.99 – 1.01
Habit strength X fatigue (within) 0.90 0.80 – 1.01
F. attribution X fatigue (within) 0.97 0.86 – 1.09

Analyses Using Soft Drink Consumption and Attributions
A multilevel model tested soft drink consumption in a given prompt from soft drink consumption
habit strength, fatigue, and fatigue attribution. The results revealed more consumption among
those with stronger habits, OR = 1.97, 95% CI [1.54, 2.51], and less consumption given greater
fatigue, OR = 0.80, 95% CI = [0.65, 0.98]. No other effects were significant.

HABITS AND FRICTION
130
Table S1.3
Study 2: Main Analysis with Intention and Attitudes as Covariates
Predictors OR 95% CI
Intercept 0.17 0.14 – 0.19
Habit strength 1.30 1.09 – 1.55
Fatigue (within) 1.20 1.06 – 1.37
Fatigue (between) 0.94 0.82 – 1.09
Fatigue attribution 0.99 0.84 – 1.15
Intention 0.98 0.84 – 1.14
Liking 1.13 0.96 – 1.35
Habit strength X Fatigue (within) 0.92 0.81 – 1.04
Fatigue attribution X Fatigue (within) 0.99 0.87 – 1.12

HABITS AND FRICTION
131
Supplemental Materials
Study 1: Additional Measures
Attribution (Open Ended)
In an open-text response, participants indicated: “Why did you [decline/agree] to do additional
trials? You may answer in 3-15 words.”
Suspicion Probe
In an open-text response, participants answered the question: “What do you think was the
purpose of the study?”
Study 2: Additional Measures
Context-Specific Habit Measure (EMA and Follow-Up Survey)
In the final survey, a script was used to randomly select up to seven coffee-drinking prompts and
seven non-coffee drinking prompts. Participants were then shown each of their individually
generated context descriptions, and rated for each context:
Behavioral Frequency. To mirror the BFCS, participants indicated, “How often do you get (or
make) coffee in this situation?” With responses “… % of the times when I’m in that
situation,”(0-20%, 20-40%, 40-60%, 60-80%, 80-100%).
Automaticity. To mirror the SRBAI, participants indicated, “In this situation, getting (or
making) coffee is something that I would have to think about,”With responses ranging from 1
(strongly disagree) to 7 (strongly agree).
Intentions. Participants indicated, “I intend to get (or make) coffee in this situation,” With
responses ranging from 1(strongly disagree) to 7 (strongly agree).
HABITS AND FRICTION
132
Baseline frequency. Participants indicated, “How often are you in this situation?” with
responses: Less than once a week, 1-2 times a week, 3-5 times a week, 6-7 times a week, more
than 7 times a week (that is, more than once a day).
After data collection, situation descriptions were excluded from analysis if they were highly
unlikely to be ones in which coffee drinking occurs, such as sleeping or showering (166 prompts,
or 13%).
Social Norms
On a scale from 1 (Strongly disagree) to 7 (Strongly agree), participants rated descriptive
norms ,“Most people who are like me drink coffee regularly”, and injunctive norms, “Most
people who are important to me expect me to drink coffee”; “My friends do not expect me to
drink coffee” (reverse-coded).
Exploratory Open-Text Self-Regulation Measure (Final Survey)
Participants completed the following: “I drink coffee (more often than / less often than /
as much as) I would like.” If they indicated that they drank coffee more than they’d like, they
responded to the prompt: “If you wanted to reduce the frequency of your coffee drinking, how
would you do that?” (open-text).
Conditional Probability Estimates (Follow-Up Survey)
In addition, participants estimated “how often you drank coffee following each tiredness
score. For each tiredness score, estimate what was the likelihood of you drinking coffee in the 2
hours after surveys in which you reported the corresponding tiredness score.” Participants rated,
for each score on the fatigue scale from 1-6, their conditional coffee drinking likelihood from
0%-100%. Participants then estimated how many times they reported each tiredness score during
HABITS AND FRICTION
133
the study, from 0 to 40 (the maximum number of prompts that could be answered during the
study).
Studies not Included in Main Manuscript
Pilot Study: Effects of Mood and Habit on Coffee Purchase Expectations
We examined the effects of mood and habit strength on coffee purchase intentions in a
hypothetical job interview. Participants were 150 mTurkers who imagined getting a phone call a
day after being interviewed for a job and being offered the position (success condition), not
offered it (failure condition), or notified in two weeks (control condition). Participants rated their
intentions to buy coffee,completed an attribution measure, and habit strength to purchase coffee
(BFCS and SRBAI).
Linear regression predicted purchase intentions from habit strength, interview outcome as
indicator of mood, and interactions. Stronger habit participants reported stronger intentions, b =
2.41, 95% CI [0.52, 4.29], β = .31, p = 0.125, but interview outcome had little effect.
Attributions were analyzed with a one sample t-test test comparing with 50% (representing equal
attribution to mood and habit). Participants in the success and failure conditions assigned greater
weight to affect compared with habit in attributions for their purchase intentions, t(100) = 2.42, p
= .017.
In summary, regardless of the interview’s outcome, participants with stronger habits
reported stronger purchase intentions. Nevertheless, when explaining their intended behavior,
participants in the positive and negative outcome conditions assigned equal or greater weight to
mood than to habit. These findings provided initial support for our primary hypothesis of
overlooking habit.

HABITS AND FRICTION
134
Additional Study: Effects of Mood and Habit on Food/Drink Purchase after Exams
In a study measuring attributions and actual effects of habit and mood on students’
purchases of food or drink after exams, attributions reflected mood as much as habit, and both
influenced behavior. We speculate that this departure from our hypothesis is due to our definition
of habit as tied to this emotion-inducing event, which may have inadvertently confounded
measurement of habit and mood.
Participants were 173 undergraduate students. Several days before an exam (Time 1),
they reported habit strength for purchasing food or drink after exams. Immediately after the
exam (Time 2), they reported their current mood on a scale ranging from 1 (sad face) to 100
(happy face). Finally, between 6 and 48 hours after the exam (Time 3), they indicated (yes/no)
whether they had purchased food or drink immediately after the exam and rated purchase
attributions on scales ranging from 0% (no effect of mood/habit) to 100% (behavior solely due to
mood/habit).
A logistic regression predicted actual food or drink purchase (yes/no) from habit strength
(using the BFCS), post-exam mood, and their interaction. Habit strength significantly predicted
purchases, OR = 1.97, 95% CI [1.26, 3.08], p = .003, but the linear predictor of mood did not,
OR = 1.00, 95% CI [0.99, 1.01], p = .817. In addition, no interaction was evident between mood
and habit strength, OR = 1.00, 95% CI [0.99, 1.02], p = .767. However, a mood intensity term
(coded as absolute deviation of mood from 50, so that 0 = neutral mood and 50 = intense
negative or positive mood) revealed both an effect of habit strength and an effect for mood
intensity, OR = 1.04, 95% CI [1.02, 1.06], p = .001.
A paired-samples t-test compared each participant’s mood attribution rating with their
habit attribution rating (for the 89 participants who reported a purchase). No significant
HABITS AND FRICTION
135
difference emerged between the two, Mdifference = 0.05, t(88) = 0.01, p = .989. Sensitivity
analyses revealed a similar pattern when all participants were included, Mdifference = -3.67, t(172)
= -1.14, p = .255.

HABITS AND FRICTION
136
Appendix B - Supplementary Materials for Chapter 2
Section S2.1: Presidential Campaign Video Coding
We defined a campaign video as being two minute or less in length, focusing on the 2020
presidential election, and being posted between each party’s national convention and election
day. Overall, 131 such videos were posted by Joe Biden’s campaign, and 74 videos were posted
by Donald Trump’s campaign. Videos were coded as mentioning beliefs and/or friction by a
research assistant who was blind to the hypotheses.

HABITS AND FRICTION
137
Table S2.1
Perceived Effects of Beliefs and Friction on Turnout only Among Intended In-person Voters
Measure type Variable Target Count (%) /
Mean (SD)
Beliefs - friction
difference (Cohen’s
h / Cohen’s d)
Open text (% of participants
who noted at least one response
for each factor)
Beliefs Self 908 (92.00%) 1.88
Friction Self 111 (11.25%)
Beliefs Other 909 (92.10%) 1.58
Friction Other 223 (22.59%)
Open text (mean # of responses
listed for each factor)
Beliefs Self 1.88 (1.22) 1.27
Friction Self 0.15 (0.45)
Beliefs Other 2.47 (1.36) 1.25
Friction Other 0.31 (0.69)
Open text (mean importance
attributed to each factor)
Beliefs Self 8.59 (2.88) 1.68
Friction Self 0.94 (2.71)
Beliefs Other 7.99 (2.90) 1.29
Friction Other 1.79 (3.53)
Composite Beliefs Self 4.93 (1.13) 0.89
Friction Self 3.80 (1.43)
Beliefs Other 5.21 (1.04) 0.81
Friction Other 4.32 (1.28)
HABITS AND FRICTION
138
Single-item Beliefs Self 80.70 (21.89) 0.35
Friction Self 69.91 (33.03)
Beliefs Other 77.37 (20.63) 0.41
Friction Other 64.57 (30.22)
Note. Effect sizes are in Cohen’s d for continuous measures and Cohen’s h for proportions (% of
participants who noted at least one response for a factor).

HABITS AND FRICTION
139
Table S2.2
Actual Effects of Beliefs and Friction on Turnout – Unweighted Sample
Variable Tjur’s R
2
R
2
Beliefs - friction
difference (R
2
)
Beliefs - friction
difference (d)
Beliefs .18 .15 - -
Friction (full) .20 .17 -0.02 -0.06
Friction (narrow
model)
.09 .09 0.06 0.17

HABITS AND FRICTION
140
Table S2.3
Perceived Effects of Beliefs and Friction on Turnout – Unweighted Sample
Measure type Variable Target Count (%) /
Mean (SD)
Beliefs -
friction
difference
(Cohen’s h /
Cohen’s d)
Open text (% of participants
who noted at least one response
for each factor)
Beliefs Self 1,168 (91.25%) 1.84
Friction Self 152 (11.88%)
Beliefs Other 1,170 (91.41%) 1.54
Friction Other 297 (23.20%)
Open text (# of responses listed
for each factor)
Beliefs Self 1.89 (1.23) 1.23
Friction Self 0.15 (0.46)
Beliefs Other 2.46 (1.34) 1.23
Friction Other 0.34 (0.73)
Open text (mean importance
attributed to each factor)
Beliefs Self 8.59 (2.89) 1.59
Friction Self 1.01 (2.82)
Beliefs Other 7.99 (2.91) 1.25
Friction Other 1.91 (3.61)
Composite Beliefs Self 4.84 (1.19) 0.80
Friction Self 3.78 (1.46)
Beliefs Other 5.19 (1.04) 0.74
Friction Other 4.36 (1.25)
HABITS AND FRICTION
141
Single-item Beliefs Self 79.35 (23.72) 0.29
Friction Self 69.33 (32.76)
Beliefs Other 76.17 (21.44) 0.38
Friction Other 64.34 (29.88)

HABITS AND FRICTION
142
Table S2.4
Perceived Effects of Beliefs and Friction on Turnout – Post Election Survey
Measure type Variable Target Count (%) /
Mean (SD)
Beliefs - friction
difference
(Cohen’s h /
Cohen’s d)
Open text (% of participants
who noted at least one response
for each factor)
Beliefs Self 1,205 (94.14%) 2.00
Friction Self 130 (10.16%)
Beliefs Other 1,191 (93.05%) 1.89
Friction Other 156 (12.19%)
Open text (# of responses listed
for each factor)
Beliefs Self 2.03 (1.16) 1.36
Friction Self 0.15 (0.50)
Beliefs Other 1.64 (1.09) 1.14
Friction Other 0.16 (0.52)
Open text (mean importance
attributed to each factor)
Beliefs Self 8.84 (2.58) 1.75
Friction Self 0.91 (2.70)
Beliefs Other 8.31 (2.74) 1.61
Friction Other 0.96 (2.74)

HABITS AND FRICTION
143
Table S2.5
Perceived Effects of Beliefs and Friction on Turnout among those who Reported Not Voting in
Post Election Survey
Measure type Variable Target Count (%) /
Mean (SD)
Beliefs - friction
difference
(Cohen’s h /
Cohen’s d)
Open text (% of participants
who noted at least one response
for each factor)
Beliefs Self 96 (77.42%) .94
Friction Self 40 (32.26%)
Beliefs Other 21 (16.94%) 1.44
Friction Other 103 (83.06%)
Open text (# of responses listed
for each factor)
Beliefs Self 1.30 (1.04) 0.58
Friction Self 0.43 (0.71)
Beliefs Other 1.43 (1.12) 0.82
Friction Other 0.27 (0.69)
Open text (mean importance
attributed to each factor)
Beliefs Self 6.92 (4.11) 0.52
Friction Self 2.99 (4.40)
Beliefs Other 7.14 (3.60) 1.02
Friction Other 1.45 (3.30)

HABITS AND FRICTION
144
Table S2.6
Regressions Predicting Support for Policies Affecting Friction from Belief vs. Friction
Importance Ratings
Measure
Automatic Voter
Registration
Name
Match
Signature
Match
More Poll
Workers
More Polling
Places
Fewer
Polling
Places
No Controls -.18*** .12*** .12*** -.12*** -.12*** -.13***
With
Controls
-.09*** .04 .03 -.10** -.08** -.17***
Note. Numbers represent standardized coefficients for regressions predicting policy support from
friction discounting (belief-friction difference scores) for others. The sample size for the simple
regressions was N = 1,280, whereas that of the control regressions was N = 1,229 (since some
participants declined to respond to certain measures or gave an “unsure” response).
** p < .01 *** p < .001

HABITS AND FRICTION
145
Section S2.2: Exploratory Follow-up Studies
Readers interested in further understanding the dynamics of friction discounting may be
interested in additional preliminary follow-up studies we had conducted. We present these two
exploratory studies below.
Method
In two studies, participants (Study S1a: n = 333; Study S1b: n = 341) read about our main
study and tried to predict the factors that most impacted turnout in our prior sample. To
incentivize participants for accuracy, they were informed that the three most accurate
participants (as compared against the actual study results) would receive $50 Amazon gift cards.
Participants first listed up to six pieces of information that they would most like to know
about an individual in order to estimate if that individual voted or not. Participants then rated, on
a 7-point scale, each factor’s perceived importance in influencing turnout).
Next, participants saw a more detailed description of our main study (e.g., sample size,
states of interest, timing of study, specific predictors). In Study S1a, these included seven
variables, with three belief variables, three friction variables, and one variable that may reflect
both beliefs and friction (voter registration status). In Study S1b, each participant saw a different
set of four variables that were chosen at random from a list of 17 (the variables included in each
study are shown in Table S7). Participants then completed five rounds of the prediction task. In
each round, participants saw the actual responses of a randomly-selected participant from the
main study for the relevant variables. Based on these responses, participants estimated the
likelihood that this person voted, both in terms of a binary item (voted / did not vote) and as a
probability judgment (0-100%). Participants then answered a series of demographic questions,
were thanked, debriefed, and paid.

HABITS AND FRICTION
146
Results and Discussion
Participants’ open-ended responses were coded by coders blind to the research
hypotheses, as in the main study. Free responses in Study S1a revealed friction discounting: 91%
of participants mentioned at least one belief, as opposed to 38% who mentioned at least one
source of friction. Participants' importance weights revealed a similar pattern, with beliefs (M =
14.05, SD = 8.00) rated as more important than friction (M = 3.89, SD = 6.25), t(318) = 16.21, p
< .001, d = 0.91. Free-response answers in Study S1b similarly featured beliefs much more often
than friction: 91% of participants mentioned at least one belief, whereas only 33% mentioned
any form of friction. Once again, importance weights on these free-response factors revealed M =
7.50 (SD = 7.58), M = 2.18 (SD = 4.76), t(340) = 10.47, p < .001, d = 0.57.
In each study, to compare participants’ predictions to the actual main study results, we computed
two logistic regression models for each variable: One model used the main study data, predicting
actual turnout (voted / did not vote) from that variable. A second multilevel model (with
predictions nested within participants) used the prediction study data, with the binary turnout
prediction (voted / did not vote) as the dependent variable instead of actual turnout. Then, for
each variable, we compared how well a variable explained actual turnout (using Tjur’s R
2
) with
how well it explained predicted turnout. If a variable’s predicted R
2
> actual R
2
, this suggests
that that variable’s effect is over-estimated, whereas a predicted R
2
< actual R
2
, suggests that a
variable’s contribution is under-estimated.
Actual and predicted R
2
values are shown in Table S7. In Study S1a, participants’ predictions
were remarkably close to the actual results, with predicted values being mostly within ±5% of
the actual values. Nevertheless, the belief variables tended to be overestimated whereas the
friction variables tended to be under-estimated. In Study S1b using all 17 variables, participants
HABITS AND FRICTION
147
overestimated the effects of all variables compared with these variables’ actual contributions, but
much more so for beliefs than for frictions: The effect of belief variables was over-estimated on
average by 8%, compared with 2% for friction effects. In illustration, splitting the variable list in
half into the six most over-estimated variables and six least over-estimated ones, the most over-
estimated variables include five beliefs and only one friction (travel mode), whereas the least
over-estimated variables include four frictions and two beliefs. Although we did not pre-register
quantitative comparison of the regression coefficients, an exploratory analysis using a Wilcoxon
signed-rank test revealed that the rank distribution of perceived predictors differs from that of
actual predictors, V = 91, p < .001.

HABITS AND FRICTION
148
Table S2.7
Variables’ Predicted vs. Actual Explained Variance in Studies S1a and S1b
Variable S1a
Actual
S1a
Predicted
S1a
Difference
S1b
Actual
S1b
Predicted
S1b
Difference
Voting as Duty 0.234 0.258 0.024 0.234 0.367 0.133
Candidate
Support
— — — 0.015 0.131 0.116
Trump
Attitude
0.018 0.027 0.009 0.018 0.112 0.094
Vote matters — — — 0.192 0.274 0.082
Biden attitude — — — 0.021 0.082 0.061
Travel mode 0.002 0.01 0.008 0.002 0.058 0.056
Registration 0.279 0.204 -0.075 0.279 0.334 0.055
Ideology — — — 0.023 0.066 0.043
Childcare 0.018 0.002 -0.016 0.018 0.035 0.017
Min. to poll — — — 0.014 0.029 0.015
Car ownership 0.023 0 -0.023 0.023 0.037 0.014
Partisanship 0.046 0.058 0.012 0.046 0.055 0.009
Time off work — — — 0.01 0.018 0.008
Note. Min. to poll = Expected time it would take to travel to the nearest polling place (in
minutes). Variables are ordered by their level of overestimation in Study S1b, running from most
overestimated (belief that voting is one’s duty) to most under-estimated (requiring to take time
off of work to vote).

HABITS AND FRICTION
149
Appendix C - Supplementary Materials for Chapter 3
Table S3.1
Results of Linear Regression Model Predicting Initial (Intake Survey) Intentions to Reduce
Phone Use from Condition
Predictors b ꞵ 95% CI p
Intercept 6.14 -0.07 5.90 – 6.37
Motivation 0.06 0.09 -0.29 – 0.42 0.717
Friction (Weekly + Different Page) 0.06 0.09 -0.27 – 0.40 0.706
Friction (Daily + Same Page) 0.22 0.30 -0.12 – 0.56 0.209
Friction (Weekly + Same Page) -0.00 -0.00 -0.33 – 0.33 0.983
Friction (Daily + Different Page) -0.02 -0.02 -0.34 – 0.31 0.922

HABITS AND FRICTION
150

Table S3.2
Results of Multilevel Linear Regression Model Predicting Daily Perceived Ease from Study day
and Experimental Condition
Predictors b ꞵ 95% CI p
Intercept 3.88 -0.08 3.36 – 4.39
Day 0.10 0.22 0.04 – 0.15 <0.001
Motivation 0.22 -0.05 -0.56 – 1.00 0.585
Friction (Weekly + Different Page) 0.19 0.03 -0.58 – 0.95 0.631
Friction (Daily + Same Page) 0.58 0.19 -0.15 – 1.32 0.121
Friction (Weekly + Same Page) 0.27 -0.10 -0.48 – 1.01 0.482
Friction (Daily + Different Page) 0.86 0.25 0.15 – 1.58 0.018
Day X Motivation -0.04 -0.10 -0.12 – 0.03 0.271
Day X Friction (Weekly + Different
Page)
-0.02 -0.04 -0.09 – 0.06 0.622
Day X (Daily + Same Page) -0.04 -0.09 -0.11 – 0.03 0.306
Day X (Weekly + Same Page) -0.06 -0.14 -0.14 – 0.01 0.093
Day X (Daily + Different Page) -0.06 -0.15 -0.13 – 0.00 0.069

HABITS AND FRICTION
151
Table S3.3
Results of Multilevel Linear Regression Model Predicting Daily App Use from Daily Intentions
(person-mean-centered; measured in evening diaries), Study Day, and Condition
Predictors b ꞵ 95% CI p
Intercept -15.26 0.00 -26.49 – -4.02
Intentions 1.95 -0.03 -5.01 – 8.91 0.582
Motivation 2.31 0.05 -13.86 – 18.48 0.779
Friction (Weekly + Different Page) -10.69 -0.22 -26.12 – 4.75 0.175
Friction (Daily + Same Page) -7.07 -0.14 -22.57 – 8.43 0.371
Friction (Weekly + Same Page) 8.15 0.16 -6.92 – 23.23 0.289
Friction (Daily + Different Page) 5.55 0.11 -9.29 – 20.40 0.463
Study Day 0.54 0.04 0.10 – 0.97 0.017
Intentions X Motivation -0.64 -0.01 -7.66 – 6.38 0.858
Intentions X Friction (Weekly +
Different Page)
2.35 0.04 -5.07 – 9.78 0.534
Intentions X (Daily + Same Page) 0.57 0.01 -6.12 – 7.26 0.868
Intentions X (Weekly + Same Page) 5.19 0.08 -2.17 – 12.55 0.167
Intentions X (Daily + Different Page) -2.13 -0.03 -9.05 – 4.79 0.546
Intentions X Study Day -0.49 -0.03 -1.05 – 0.06 0.083
HABITS AND FRICTION
152
Table S3.4
Results of Multilevel Linear Regression Model Predicting Daily Intention - App Use Difference
Score (scaled by person-mean and person-standard-deviation) from Study Day

Predictors b 95% CI
Intercept -0.04 -0.18 – 0.10
Study Day -0.01 -0.03 – 0.01

HABITS AND FRICTION
153
Table S3.5
Results of Multilevel Linear Regression Model Predicting Daily App Use in the Friction
Conditions from Perceived Efficacy and Experimental Condition
Predictors b ꞵ CI p
Intercept -18.85 -0.15 -30.64 – -7.06
Friction (Daily + Same Page) 0.38 0.01 -16.35 – 17.10 0.965
Friction (Weekly + Same Page) 14.00 0.28 -2.28 – 30.27 0.092
Friction (Daily + Different Page) 14.56 0.29 -1.44 – 30.56 0.074
Efficacy -7.85 -0.30 -10.92 – -4.78 <0.001
Efficacy X (Daily + Same Page) 3.74 0.14 -0.53 – 8.01 0.086
Efficacy X (Weekly + Same Page) 1.65 0.06 -2.68 – 5.99 0.454
Efficacy X (Daily + Different
Page)
5.49 0.21 1.05 – 9.93 0.015

HABITS AND FRICTION
154
Figure S3.1

Daily Intention - App Use Difference Scores (scaled by person-mean and person-standard-
deviation) by Study Day

Note. Black horizontal bars represent mean intentions for that day. Light boxes represent 95%
confidence intervals for the mean.

Abstract (if available)

Abstract People tend to do (a) what they’ve done in similar situations in the past, and (b) what’s easiest. In other words, human behavior is strongly shaped by both habit and friction. However, people’s perceptions may underestimate the pervasive pull of both forces on behavior, with far-reaching consequences. In Chapter 1, I measured the actual and perceived impact of habits on behavior. In two studies—a naturalistic study using Ecological Momentary Assessment (EMA) and an online controlled experiment—I found that participants’ coffee drinking and helping behavior was substantially driven by habit, but participants underestimated habits’ influence, implicating inner states (like fatigue and mood) instead. In Chapter 2, I tracked the effects of friction and beliefs on voter turnout in the 2020 United States election. I found that friction depressed turnout, but participants underappreciated its effects. Furthermore, those participants who neglected friction (vs. beliefs) also tended to support policies that make voting more difficult. In both Chapters 1 and 2, underestimation of habit and friction was remarkably robust, persisting regardless of financial incentives as well as when making attributions for others (vs. self). Finally, Chapter 3 compared perceptions and actual effects of a subtle friction intervention on habitual phone app use. In estimating intervention effects, participants once again overlooked the potential impact of friction and instead emphasized willpower and motivation. In contrast, I anticipated that imposition of minor friction may suffice to disrupt habitual phone use. However, results do not indicate that friction reduced app use more than a control condition. Participants across conditions exhibited an initial reduction in app use quickly followed by a rapid rebounding to baseline—after several days or even a single day. Collectively, these accumulated findings underscore the omnipresent yet overlooked power of habits and friction, while highlighting the difficulties of creating and maintaining behavior change in field interventions.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Illusory motives, elusive habits: lay theories of everyday behavior

PDF

Habits as friend and foe to self-control: the dual roles of routinized behavior

PDF

The role of good habits in facilitating long-term goals

PDF

Actual and perceived social reinforcements of weight-related cognitions and behaviors in adolescent peer groups

PDF

Performance and attention novelty slows hedonic adaptation during habit formation

PDF

Examining the effects of personalized feedback about ALDH2*2, alcohol use, and associated health risks on drinking intentions and consumption: the role of self-efficacy and perceived threat

PDF

Facebook habits: rewards, cues, and automaticity

PDF

Beyond active and passive social media use: habit mechanisms are behind frequent posting and scrolling on Twitter/X

PDF

Habits and policy: the socio-cognitive foundations of foreign policy stability

PDF

The effects of dual coding theory on memory retention

PDF

The interplay between social connection and substance use

PDF

Links between discrimination and sleep difficulties: romantic relationships as risk and resilience factors

PDF

Individual differences in heart rate response and expressive behavior during social emotions: effects of resting cardiac vagal tone and culture, and relation to the default mode network

PDF

Internet-delivered eating disorder prevention: a randomized controlled trial of dissonance-based and cognitive-behavioral treatments

PDF

Bound in hatred: a multi-methodological investigation of morally motivated acts of hate

PDF

Perceived social consensus: a metacognitive perspective

PDF

Expressing values and group identity through behavior and language

PDF

Impulsivity interferes with emotion regulation strategy prioritization in everyday life

PDF

Effects of mint, menthol, and tobacco-flavored e-cigarettes on appeal and sensory effects, tobacco withdrawal…

PDF

How misinformation exploits moral values and framing: insights from social media platforms and behavioral experiments

Asset Metadata

Creator Mazar, Asaf (author)

Core Title Habits and Friction: Actual and Perceived Effects on Behavior

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Psychology

Degree Conferral Date 2023-08

Publication Date 05/26/2023

Defense Date 07/26/2022

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

Tag field study,friction,habit,lay theories,OAI-PMH Harvest

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Wood, Wendy (committee chair), Beam, Chris (committee member), Dunton, Genevieve (committee member), Stone, Arthur (committee member)

Creator Email amazar@usc.edu,mazaras17@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113138314

Unique identifier UC113138314

Identifier etd-MazarAsaf-11896.pdf (filename)

Legacy Identifier etd-MazarAsaf-11896

Document Type Dissertation

Format theses (aat)

Rights Mazar, Asaf

Internet Media Type application/pdf

Type texts

Source 20230530-usctheses-batch-1049 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu