Content uploaded by David W. Nickerson
Author content
All content in this area was uploaded by David W. Nickerson on Dec 20, 2013
Content may be subject to copyright.
Psychological Science
21(2) 194 –199
© The Author(s) 2010
Reprints and permission: http://www
.sagepub.com/journalsPermissions.nav
DOI: 10.1177/0956797609359326
http://pss.sagepub.com
Helping people to make plans can increase their likelihood of
following through on intentions. This finding has powerful
implications for interventions directed toward increasing
socially important behaviors that people intend to, but often
fail to, perform. In this study, we conducted a field experiment
during the 2008 U.S. presidential election, designed to increase
turnout by assisting voters in forming implementation inten-
tions. The results expanded our understanding of implementa-
tion intentions in two ways, one as hypothesized and another
unexpectedly. First, we found that implementation intentions
can be a potent addition to interventions aimed at increasing
intention fulfillment for a specific high-salience and socially
important behavior: voting. This turnout increase resulted
from concrete plan making, not from simply asking people if
they intended to vote. In fact, contrary to some past research,
self-prediction only marginally increased turnout. Second, an
unexpected heterogeneous treatment effect suggested a novel
moderator for interventions leveraging implementation inten-
tions. Participants in multiple-eligible-voter households were
unaffected by the intervention, whereas those in single-eligible-
voter households were strongly affected. We propose that
some previously unrecognized situational factors organically
facilitate implementation-intention formation more readily
than others and present data supporting this interpretation of
our differential treatment effect. Results provide insight into
when interventions leveraging implementation intentions will
be most potent for increasing voter turnout, and potentially
other intended behaviors.
Literature Review
Self-prediction
Predicting that one will follow through on a behavior may
increase one’s likelihood of doing so. This pattern of self-
fulfilling self-prediction has been studied and labeled in multi-
ple different areas of behavioral research (Greenwald, Carnot,
Beach, & Young, 1987; Greenwald, Klinger, Vande Kamp, &
Kerr, 1988; Morwitz, Johnson, & Schmittlein, 1993; Sherman,
1980). In fact, the original research on what was termed the
Corresponding Author:
Todd Rogers, Analyst Institute, 815 16th St., NW, Washington, DC 20006
E-mail: trogers@analystinstitute.org
Do You Have a Voting Plan?
Implementation Intentions, Voter
Turnout, and Organic Plan Making
David W. Nickerson1 and Todd Rogers2
1University of Notre Dame and 2Ideas42, Institute for Quantitative Social Science, Harvard University
Abstract
Phone calls encouraging citizens to vote are staples of modern campaigns. Insights from psychological science can make
these calls dramatically more potent while also generating opportunities to expand psychological theory. We present a field
experiment conducted during the 2008 presidential election (N = 287,228) showing that facilitating the formation of a voting
plan (i.e., implementation intentions) can increase turnout by 4.1 percentage points among those contacted, but a standard
encouragement call and self-prediction have no significant impact. Among single-eligible-voter households, the formation
of a voting plan increased turnout among persons contacted by 9.1 percentage points, whereas those in multiple-eligible-
voter households were unaffected by all scripts. Some situational factors may organically facilitate implementation-intentions
formation more readily than others; we present data suggesting that this could explain the differential treatment effect that
we found. We discuss implications for psychological and political science, and public interventions involving implementation-
intentions formation.
Keywords
implementation intentions, self-prediction, voting, nudges
Received 4/15/09; Revision accepted 7/8/09
Research Report
Do You Have a Voting Plan? 195
self-prophecy effect was conducted in the context of voter
turnout in the 1984 election (Greenwald et al., 1987). That
research reported a 23-percentage-point increase in turnout as
a result of people being asked if they intended to vote. A more
recent experiment failed to replicate those findings using iden-
tical procedures and a sample size nearly 10 times as large
(Smith, Gerber, & Orlich, 2003). Because the present experi-
ment focused on the novel translation of implementation-
intentions formation to voting, our design necessitated that we
initially ask respondents if they intended to vote. Therefore, a
self-prediction-only condition was included to attempt to rep-
licate the self-prophecy research with the largest sample yet.
Implementation intentions
Even more potent than self-prediction, assisting people in plan
making, or forming implementation intentions, can facilitate
the fulfillment of goals (Gollwitzer, 1999; Gollwitzer &
Sheeran, 2006). Articulating the when, where, and how of fol-
lowing through on an intention creates cognitive links between
an anticipated future situation and the intended behavior. These
pairs can be thought of as “if situation Y, then behavior X”
(Gollwitzer, Bayer, & McCulloch, 2005; Gollwitzer & Sheeran,
2006). Implementation intentions have been shown to affect
dozens of behaviors, including those that are repeated over
time, such as exercising (Lippke & Ziegelmann, 2002; Milne,
Orbell, & Sheeran, 2002), as well as onetime behaviors exe-
cuted within a finite window, such as picking up course reading
materials at an office at a certain time (Dholakia & Bagozzi,
2003). Nearly every context in which this cognitive process has
been studied has involved behaviors that were either new to the
participant (e.g., picking up books) or of a low-salience nature,
so that participants would be unlikely to organically discuss
them in detail with others (i.e., exercising). This article reports
the first well-powered experiment showing the impact of imple-
mentation intentions on a high-salience, socially important
behavior: voting. In the process, we discover a previously
unrecognized moderator of the power of implementation-
intentions formation to influence behavior.
“Get out the vote”
Voting is an important social behavior required by a function-
ing democratic society. In addition to its normative impor-
tance, tens of millions of dollars are spent by political
campaigns and organizations to increase turnout. Field experi-
mental research in political science has begun studying turn-
out interventions (e.g., Gerber & Green, 2000), most of which
has compared the relative impact of different modes of contact
including door to door, mail, e-mail, and phone (see Green &
Gerber, 2008, for review). The present experiment used phones
to encourage turnout. Research has shown that calls conducted
with a conversational tone (like those used in our standard
condition) can be more effective than calls made with a
scripted tone, even though they are not necessarily longer
(Nickerson, 2006, 2007). A meta-analysis of 28 experiments
found that a completed phone call from a commercial phone
bank increases turnout an average of 1.0 percentage point
(Green & Gerber, 2008). That is, for every 100 phone contacts,
one person votes who would have otherwise abstained. This
experiment introduces the psychological concept of imple-
mentation-intentions formation to voter-mobilization research.
Method
Our sample came from a list of registered Pennsylvanians eli-
gible to vote in the 2008 presidential primary, purchased from
a consumer data firm. Because only the Democratic primary
was competitive, the experiment was limited to registered
Democrats (N = 4,200,109). We selected the sample on the
basis of the following criteria. First, we selected people with
verified phone numbers that did not appear on any “do not
call” registry (n = 526,363). To avoid targeting citizens who
were extremely likely to vote, by definition unlikely to be
responsive to the experimental procedures, only those who
had voted in one or fewer primary elections since 2000 were
included in the experiment (excluding 155,669). To expedite
requesting a specific person while executing the experiment,
only households containing three or fewer registered Demo-
crats were included in the experiment (excluding another
14,578). Within the remaining households, one person was
randomly selected as a participant for the experiment (final n =
287,228), and households were randomly assigned to one of
the experimental conditions or the control condition (in which
there was no attempted phone call).
Six scripts were constructed to isolate the effect of self-
prediction and implementation intentions on turnout (see
Appendix S1 in the Supplemental Material available on-line).
The standard scripts encouraged participation by reminding
participants about the election and their duty to vote (Scripts A
and B). The self-prediction scripts were identical to the stan-
dard script but also asked whether the person intended to vote
(Scripts C and D). The implementation-intentions scripts were
identical to the self-prediction scripts but also asked three
follow-up questions designed to facilitate voting plan making:
what time they would vote, where they would be coming from,
and what they would be doing beforehand (Scripts E and F).
Two scripts were constructed for each experimental condition,
to test a different hypothesis regarding descriptive social
norms, which is not addressed in this article.1 Because assign-
ment to the scripts was random and our analysis was not
affected by which script of the two each participant received,
the discussion focuses only on the standard, self-prediction,
and implementation-intentions components of the scripts.
Randomization was stratified by household size. The
experimental procedure was administered between Saturday
and Monday before Election Day by a professional firm that
delivered millions of “get out the vote” calls in 2008. Phone
numbers were provided to the firm in a random order and were
loaded into its computer-aided calling system gradually. At
196 Nickerson, Rogers
any given moment, the scripts were evenly distributed across
callers. Because the implementation-intentions script was lon-
ger than the standard and self-prediction scripts, fewer imple-
mentation-intentions calls were attempted and completed. The
randomized ordering of the phone numbers ensured that the
imbalance only reduced efficiency and did not bias the analy-
sis. Appendix S2 in the Supplemental Material available on-
line shows that the covariates did not differ significantly across
condition or across contacted participants.
After the election, participants in the experimental and con-
trol conditions were matched to official voter-turnout records to
determine who voted, so the dependent variable was measured
identically for both groups and not subject to self-reporting
biases. When not every participant receives the assigned treat-
ment, experiments commonly provide two estimates of treat-
ment effect. The intent-to-treat analysis compared rates of voter
turnout across conditions, independent of contact rates. How-
ever, most participants were not actually contacted because they
were unavailable, and even fewer participants in the implemen-
tation-intentions condition were contacted because of the con-
tract with the phone vendor, as described in the previous
paragraph. Thus, we also adjusted for the contact rate using the
random assignment as an instrument for actual contact (Angrist,
Imbens, & Rubin, 1996; Gerber & Green, 2000), to estimate the
average treatment effect among the treated (ATT).
Results
Table 1 contains turnout and contact rates for the four condi-
tions. Although very similar to scripts that have been success-
ful in other settings, the standard script did not have any
mobilizing impact (ATT = –0.4). Targets who heard the self-
prediction script were 0.5 percentage points (ATT = 2.0) more
likely to vote than were targets in the control group, which was
marginally significant. Those assigned to the implementation-
intentions condition were 0.9 percentage points (ATT = 4.1)
more likely to vote compared with the control group, which
was highly significant. Although the experiment involved
287,228 targeted voters, the results were not precise enough to
disentangle the unique contribution of self-prediction to the
overall impact of the implementation-intentions condition.
Analyzing the experimental effect by number of eligible
voters in a household helps to explain the differential impact of
the self-prediction and implementation-intentions scripts and
suggests, post hoc, a moderator for the impact of the imple-
mentation-intentions script. Table 2 shows that targets in
Table 1. Contact Rate and Impact of Condition on Voter Turnout Across All Household Sizes
ITT analysis ATT analysis
Condition Turnout Contact rate
Difference from the
control condition SE prep ATT SE n
Control 42.9% 0.0% — — — — — 228,995
Standard GOTV 42.8% 26.3% –0.1% 0.4 .46 –0.2% 1.5 19,411
Self-prediction 43.4% 25.8% 0.5% 0.4 .83 2.0% 1.5 19,411
Implementation
intentions 43.8% 23.0% 0.9% 0.4 .95 4.1% 1.7 19,411
Note: GOTV = “get out the vote.” Turnout was verified using the official Democrat voter file for Pennsylvania. The intent-to-
treat (ITT) analysis compared the rate of voter turnout in each experimental condition with the rate in the control condition,
independent of contact rates. The average treatment effect among the treated (ATT) estimates the impact on turnout among
individuals who were contacted by adjusting the ITT effect for the contact rate (Angrist, Imbens, & Rubin, 1996). Standard
errors reflect a fixed-effects estimator to control for strata of randomization. All preps are based on one-tailed ps.
Table 2. Contact Rate and Impact on Voter Turnout for Households With One Eligible Voter
ITT analysis ATT analysis
Condition Turnout Contact rate
Difference from the
control condition SE prep ATT SE n
Control 40.7% 0.0% — — — — — 199,131
Standard GOTV 40.5% 25.0% –0.2% 0.5 .41 –0.7% 2.1 9,487
Self-prediction 40.9% 24.3% 0.3% 0.5 .64 1.1% 2.1 9,474
Implementation
intentions 42.7% 22.1% 2.0% 0.5 .997 9.1% 2.3 9,484
Note: GOTV = “get out the vote.” Turnout was verified using the official Democrat voter file for Pennsylvania. The intent-
to-treat (ITT) analysis compared the rate of voter turnout in each experimental condition with the rate in the control
condition, independent of contact rates. The average treatment effect among the treated (ATT) estimates impact on turnout
among individuals who were contacted by adjusting the ITT effect for the contact rate (Angrist, Imbens, & Rubin, 1996).
Standard errors reflect a fixed-effects estimator to control for strata of randomization. All preps are based on one-tailed ps.
Do You Have a Voting Plan? 197
one-eligible-voter households were unaffected by the standard
and self-prediction scripts. However, targets in these house-
holds in the implementation-intentions condition were 2.0 per-
centage points (ATT = 9.1) more likely to vote than those in the
control group.2 Table 3 shows that in multiple-eligible-voter
households, none of the scripts significantly affected turnout
relative to the control group. A significant interaction between
household size and condition (implementation intentions vs.
control) confirmed that turnout was increased only among tar-
gets in the implementation-intentions condition who came
from one-eligible-voter households (prep = .98; see Fig. 1).
Why would targets in the one-eligible-voter households
respond to the implementation-intentions intervention while
those in multiple-eligible-voter households remained unaf-
fected? One possible explanation is that citizens in the two
types of households differed in their responsiveness. Those in
multiple-eligible-voter households voted at the same rates in
the 2006 general election as those in one-eligible-voter house-
holds (31.0%), were marginally younger (mean age = 42.6
years vs. 43.1 years, prep > .99), were more likely to be men
(52.6% vs. 56.7%, prep > .99), were more likely to be African
American (26.9% vs. 22.6%, prep > .99), and were slightly
more likely have been in the implementation-interventions
group (23.8% vs. 26.2%, prep > .99). The heterogeneous effect of
the implementation-intentions intervention was unaffected by
whether we controlled for these attributes. To our knowledge,
no research has determined that certain attributes systemati-
cally affect people’s responsiveness to an implementation-
intentions intervention. Nonetheless, the previous comparisons
show that eligible voters who live together are different from
those who live alone. Although there is no supportive evi-
dence, it is conceivable that these differences could explain
why eligible voters who live together are impervious to the
implementation-intentions intervention.
However, we found supportive evidence for a plausible
explanation that could partly explain our results. Eligible
voters who live together may be more likely to organically
make voting plans than those who live alone. Casting a vote can
be time-consuming, personally important, and deliberative and
can require transportation and child care—characteristics that
might make a topic more likely to be the subject of intrahouse-
hold conversation. Perhaps targets in one-eligible-voter house-
holds were less likely to have made voting plans before our
intervention than were those in multiple-eligible-voter house-
holds. This hypothesis was confirmed in the data. Targets in
one-eligible-voter households were more likely to answer “do
not know” than were those in multiple-eligible-voter house-
holds to each of the three plan-making questions (see Table 4).
Discussion
Phone calls encouraging citizens to vote are staples of modern
campaigns. Insights from psychological science can make these
Table 3. Contact Rate and Impact on Voter Turnout for Households With Two or Three Eligible Voters
ITT analysis ATT analysis
Condition Turnout Contact rate
Difference from the
control condition SE prep ATT SE n
Control 44.2% — — — — — — 29,864
Standard GOTV 44.3% 27.4% 0.1% 0.6 .26 0.3% 2.1 9,924
Self-prediction 45.0% 27.2% 0.8% 0.6 .84 3.0% 2.1 9,927
Implementation
intentions 43.8% 23.8% –0.4% 0.6 .21 –1.5% 2.4 9,927
Note: GOTV = “get out the vote.” Turnout was verified using the official Democrat voter file for Pennsylvania. The intent-
to-treat (ITT) analysis compared the rate of voter turnout in each experimental condition with the rate in the control
condition, independent of contact rates. The average treatment effect among the treated (ATT) estimates impact on
turnout among individuals who were contacted by adjusting the ITT effect for the contact rate (Angrist, Imbens, & Rubin,
1996). Standard errors reflect a fixed-effects estimator to control for strata of randomization. All preps are based on
one-tailed ps.
Standard
GOTV
−4
−2
0
0.3
−0.7
−1.5
3.0
1.1
9.1
Average Treatment Effect
Among the Treated
2
4
2+-Voter Household
1-Voter Household
6
8
10
Self-Prediction
Condition
Implementation
Intentions
Fig. 1. Average treatment effect among the treated as a function of condition
and number of eligible voters in the household. In all three conditions,
participants were reminded by phone about the election and their duty to vote.
Targets in the standard “get out the vote” (GOTV) condition received this
reminder only. Those in the self-prediction condition were also asked if they
intended to vote, and those in the implementation-intentions condition were
asked three follow-up questions designed to facilitate making plans for voting.
198 Nickerson, Rogers
calls dramatically more potent while also generating opportuni-
ties to expand psychological theory. In our experiment, a stan-
dard phone call had no impact on overall turnout, a call eliciting
a vote-intention self-prediction produced a marginally signifi-
cant turnout increase, and a script that incorporated both
self-prediction and implementation intentions resulted in a
4.1-percentage-point increase in turnout among those in the
experimental conditions. The weak self-prediction impact sup-
ports recent contentions (Smith et al., 2003) that the much-cited
large self-prediction effect detected in Greenwald et al. (1987)
is no longer replicable. Unexpectedly, we found that among
single-eligible-voter households, targets were 9.1 percentage
points more likely to vote when they received a script guiding
them to make a plan, whereas targets in multiple-eligible-voter
households were unaffected by the same script. We propose that
one source of this heterogeneity could be the social context in
which targets reside. Targets living in multiple-eligible-voter
households were much less likely to have a preexisting voting
plan than were targets living in single-eligible-voter households.
This suggests that targets living with others who might share an
interest in the focal behavior are more likely to engage in plan
making on their own, which might explain the impotence of the
plan-making intervention when directed at them. This rationale
seems most plausible for behaviors about which intrahousehold
conversation is greatest (i.e., high-salience, personally impor-
tant behaviors). In addition to studying the conditions under
which implementation intentions naturally arise in the world,
future research should examine the types of behaviors for which
these intentions occur on their own.
By introducing the psychological construct of implementation
intentions to this political topic, we broaden traditional con-
ceptualizations of the cost of voting, which usually relate to
the cost of the time required to go to one’s polling place (Blais,
2000). Guiding the formation of implementation intentions
does not reduce the time required to cast a vote; rather, it
increases the likelihood that one will plan time for it.
This research contributes to a growing body of work using
behavioral science to facilitate socially important behaviors
(Thaler & Sunstein, 2008). Campaign professionals can use psy-
chological science more widely to help citizens follow through
on their intentions to vote. In addition to showing that implemen-
tation intentions can be a valuable addition to “get out the vote”
efforts, we provide prescriptive guidance for other public
interventions. For prosocial behaviors that are highly salient and
widely discussed, our results suggest interventions will have the
greatest impact on individuals least likely to have engaged in
those discussions. For example, after television reporter Tim
Russert died of a heart attack, many health organizations used the
high-salience event to encourage people to get medical examina-
tions (Johnson, 2008). This was likely a topic of conversation
within many households, especially those with multiple adults.
Under these conditions, our results suggest that single-adult
households might have benefited most from an implementation-
intentions intervention for the goal of getting an exam.
Acknowledgments
We thank Alan Gerber, Peter Gollwitzer, Mike Norton, and the
Department of Social and Decision Sciences, Carnegie Mellon
University, for comments. The authors are listed in alphabetical order
and contributed equally to this research.
Declaration of Conflicting Interests
The authors declared that they had no conflicts of interests with
respect to their authorship and/or the publication of this article.
Funding
This research was funded by the Harvard Business School and the
Yale Institution for Social and Policy Studies.
Supplemental Material
Additional supporting information may be found at http://pss.sagepub
.com/content/by/supplemental-data
Notes
1. We also studied whether a script emphasizing high turnout would
be more effective than one emphasizing low turnout. The results are
to be reported elsewhere.
2. Contrasts showed that this increase was significantly greater
than in the standard “get out the vote” condition (SE = .007, p = .002,
prep > .98) and in the self-prediction condition (SE = .007, p < .007,
prep > .96).
References
Angrist, J., Imbens, G., & Rubin, D. (1996). Identification of causal
effects using instrumental variables. Journal of the American Sta-
tistical Association, 91, 444–456.
Table 4. Percentage of Targets Who Responded With “Do Not Know” to Each Question in the
Implementation-Intentions Battery and Tests of the Difference Between 1-Voter and 2+-Voter
Households
Household size
Difference between types of
households
Question 1 voter 2+ voters χ2(1) n p prep
What time will you vote? 22.4% 18.0% 9.70 3,252 < .01 .98
Where will you be coming from? 10.4% 7.9% 5.66 2,941 .02 .93
What will you be doing beforehand? 10.8% 7.9% 7.55 2,885 < .01 .96
Do You Have a Voting Plan? 199
Blais, A. (2000). To vote or not to vote: The merits and limits of ratio-
nal choice theory. Pittsburgh, PA: University of Pittsburgh Press.
Dholakia, U.M., & Bagozzi, R.P. (2003). As time goes by: How goal
and implementation intentions influence enactment of short-fuse
behaviors. Journal of Applied Social Psychology, 33, 889–922.
Gerber, A.S., & Green, D.P. (2000). The effects of canvassing, tele-
phone calls, and direct mail on voter turnout: A field experiment.
American Political Science Review, 94, 653.
Gollwitzer, P.M. (1999). Implementation intentions: Strong effects of
simple plans. American Psychologist, 54, 493–503.
Gollwitzer, P.M., Bayer, U., & McCulloch, K. (2005). The control
of the unwanted. In R. Hassin, J. Uleman, & J.A. Bargh (Eds.),
The new unconscious (pp. 485–515). Oxford, England: Oxford
University Press.
Gollwitzer, P.M., & Sheeran, P. (2006). Implementation intentions
and goal achievement: A meta-analysis of effects and processes.
Advances in Experimental Social Psychology, 38, 69–119.
Green, D.P., & Gerber, A.S. (2008). Get out the vote! Washington,
DC: Brookings Institution Press.
Greenwald, A.G., Carnot, C.G., Beach, R., & Young, B. (1987).
Increasing voting-behavior by asking people if they expect to
vote. Journal of Applied Psychology, 72, 315–318.
Greenwald, A.G., Klinger, M.R., Vande Kamp, M.E., & Kerr, K.L.
(1988). The self-prophecy effect: Increasing voter turnout by
vanity-assisted consciousness raising. Unpublished manuscript,
University of Washington, Seattle.
Johnson, B. (2008, June 24). Doctors renew health advice. Washington
Post. Retrieved from http://www.washingtonpost .com/wp-dyn/
content/article/2008/06/20/AR2008062002570 .html
Lippke, S., & Ziegelmann, J.P. (2002). Self-regulation and exercise:
A study on stages of change and successful aging. Unpublished
manuscript, Free University of Berlin, Berlin, Germany.
Milne, S., Orbell, S., & Sheeran, P. (2002). Combining motivational
and volitional interventions to promote exercise participation:
Protection motivation theory and implementation intentions.
British Journal of Health Psychology, 7, 163.
Morwitz, V.G., Johnson, E., & Schmittlein, D. (1993). Does measur-
ing intent change behavior? Journal of Consumer Research, 20,
46–61.
Nickerson, D.W. (2006). Volunteer phone calls can increase turnout.
American Politics Research, 34, 271–292.
Nickerson, D.W. (2007). Quality is job one: Professional and volun-
teer voter mobilization calls. American Journal of Political Sci-
ence, 51, 269–282.
Sherman, S.J. (1980). On the self-erasing nature of errors of predic-
tion. Journal of Personality and Social Psychology, 39, 211–221.
Smith, J.K., Gerber, A.S., & Orlich, A. (2003). Self-prophecy effects
and voter turnout: An experimental replication. Political Psychol-
ogy, 24, 593–604.
Thaler, R.H., & Sunstein, C.R. (2008). Nudge: Improving decisions
about health, wealth, and happiness. New Haven, CT: Yale Uni-
versity Press.
