Why do people vote? This seemingly simple question has intrigued social scientists for decades. The classical model of pivotal voting proposes that individuals vote because they may affect the outcome of the election (Downs 1957). But any single individual becomes very unlikely to affect the outcome of a large election, and yet people still vote in congressional and presidential elections. A second type of explanation is norm-based voting – individuals vote because they believe it is the right thing to do – even if their individual vote may not affect the outcome of the election (Riker and Ordeshook 1968, Harsanyi). This explanation has proven difficult to test or establish empirically.
We propose a model of voting due to social image motivations, which can be tested empirically. The intuition for the model is as follows. Imagine that, after the election, an individual is asked by a friend or colleague whether she voted. She may derive some pride from saying that she voted, or feel shame from admitting that she did not vote, similar in spirit to Harbaugh (1996). And suppose she dislikes lying, and incurs some disutility from it, consistent with introspection and laboratory evidence (Gneezy 2005). In this model, an individual would be motivated to vote simply because she anticipates being asked after the election! If she votes, she can later advertise her ‘good behaviour’ when asked. If she does not vote, she faces the choice of being truthful but incurring shame, or claiming that she voted but incurring the psychic cost of lying.
Our model of voting is stylised and does not capture myriad other motivations to vote. Yet, its simplicity allows us to estimate the model parameters using field data. We design a natural field experiment (Harrison and List 2004) that is tightly linked to the model, and estimate the value of voting that is due to the social image motivation described above.
The main experiment took place in the summer and fall of 2011 in towns around Chicago. We visited households and asked whether they were willing to answer a short survey, including a question on whether they voted in the previous year's (2010) Congressional election. In some cases, we posted a flyer on the doorknob a day in advance to announce the upcoming survey. Unbeknownst to the households, we accessed public voting records and thus knew whether the members of the household had truly voted or not.
The field experiment has three main sets of treatments. In the first set, we randomise the information on the flyer.
- In one group, the flyer informs households that the next day we will ask for their participation in a door-to-door survey at a specified time.
- In a second group, the flyer additionally specifies that the survey will be about "your voter participation in the 2010 congressional election."
Figure 1 below presents examples of one flyer of each kind. Changes in the share of households opening the door and completing the survey between the first and the second group reflect the value of being asked about voting. An increase in the participation of voting households indicates that voters feel pride from saying that they voted. A decrease among non-voting households indicates shame from admitting that they did not vote and the cost of lying.
Figure 1. Examples of Flyers used for the door-to-door treatments
Finding 1: Voters do not feel pride from saying they voted, but non-voters do feel shame
We find that, on average, voters do not sort in. In fact, voting households are slightly less likely to answer the door and do the survey when they are informed about the turnout question. However, non-voters sort out significantly, decreasing their survey participation by 20%.
The findings on sorting provide prima facie evidence of social-image utility. But in order to interpret the magnitude of the sorting response, we need to measure the cost of sorting in and out of the home to answer the survey. To do so, we introduce a second set of (crossed) treatments, in which we vary the promised payment for the survey ($10 versus $0) and the pre-announced duration (five minutes versus ten minutes).
- We find that the effect of reducing payment by $10 is comparable to the sorting response of non-voters to the election flyer. In other words, non-voters appear to dislike being asked whether they voted as much as they dislike being paid $10 less for completing the survey.
This suggests that the social image utilities and the lying cost are substantial, at least for non-voters.
These results may depend on the election considered. The 2010 elections were disappointing for Democrats and positive for Republicans, including in Illinois the loss of President Obama's prior seat in the Senate. The lack of pride among voters may reflect disappointment, given that the neighbourhoods visited were largely Democratic. Indeed, if we restrict the analysis to voters registered for the Republican primaries, we find evidence of sorting in.
Finding 2: Non-voters lie and claim they voted half the time, while voters tell the truth
Of course, we do more than just observe whether voters and non-voters become more or less likely to answer the door when warned about the survey content. We also know how they answered the question about voting (provided they answer the survey at all).
- We find that voters tell the truth and say they voted 90% of the time, while non-voters lie and claim to have voted 46% of the time (consistent with Silver, Anderson, and Abramson 1986). This provides another hint that non-voters prefer to be seen as voters, consistent with the assumptions of our social image model.
So far, we have shown that voters do not feel much pride from saying they voted, while non-voters seem to feel shame from admitting not voting, and lie half the time in order to appear to be voters. However, this does not mean that voters are not motivated by the anticipation of being asked by others. We need to know how much shame voters would feel were they to say they did not vote. This counterfactual is not provided by the sorting moments.
We thus introduce a third set of crossed treatments. We randomise incentives to provide a different response to the turnout question. Specifically, we inform half the respondents of the ten-minute survey that the survey will be eight minutes shorter if they state that they did not vote in the 2010 congressional election. For voters, this treatment amounts to an incentive to lie and permits us to quantify the disutility of voters were they to say (untruthfully) that they did not vote. For the 50% of non-voters who lie, this treatment provides an incentive to tell the truth. We provide a parallel $5 incentive in the five-minute survey to state that one did not vote.
Finding 3: Voters strongly dislike saying they did not vote, while non-voters are easily swayed
The results, displayed in Figure 2, indicate that non-voters are significantly more sensitive to these incentives than voters. When incentivised, the share of voters who lie increases only insignificantly – by two percentage points – from a low baseline of 10 percentage points. These results reveal a strong preference of voters for saying that they voted. Instead, the truth-telling incentive has sizeable effects for non-voters. The share of non-voters who lie decreases significantly – by 12 percentage points – from a baseline of, on average, 45 percentage points. Hence, the incentives to tell the truth reduce by about one quarter the share lying. The results are similar for time and monetary incentives.
Figure 2. Response to Incentives to Respond “Did not vote”
We combine the data from the three sets of treatments to estimate the parameters of our model. The estimates provide no evidence of pride in voting. On average, voters get negative utility even from saying that they voted. However, they obtain an even lower utility, by $15, from untruthfully saying that they did not vote. Non-voters are estimated to be on average indifferent between saying truthfully that they did not vote or lying and saying that they voted, with negative average utility from either option. We estimate substantial heterogeneity in social image utility, especially among voters.
These estimates identify the key parameters up to an additive lying cost, which remains unidentified. Since the lying cost is an integral part of the social-image value of voting, we adopt two approaches to address this limitation.
- First, we compute the value of voting for a range of values of the lying cost, including one we estimate from the laboratory evidence in Erat and Gneezy (2012).
In this range, the implied value of voting ‘because of being asked once’ is in the range of $1-$3 for voters, as the bottom line in Figure 3 shows. The estimated value is smaller for non-voters, which suggests that differences in such social image motivations can explain some of the differences in turnout.
- Second, we identify the subsample of households with similar turnout histories prior to 2010, but different turnout in 2010.
Voters and non-voters (in 2010) in this subsample are likely to be similar, and we assume that they have the same social-image and lying parameters. Under this assumption, we estimate lying costs of $5, leading to a value of being asked once of $1.50. Hence, the estimates are quite similar under both approaches.
To compute the overall value of voting due to being asked, we scale up the estimated value of being asked once by the average number of times asked. Our survey respondents report being asked, on average, five times whether they voted in the 2010 congressional election, implying an estimated value of voting ‘because others will ask’ in the range of $5-$15, a sizeable magnitude (dotted line in Figure 3). Given that our survey respondents report being asked about Presidential elections twice as often as Congressional elections, the social-image value of voting in presidential elections could be about twice as high, in the range of $10-$30, consistent with the higher turnout in presidential elections.
Figure 3. Overall value of being asked
The main field experiment was designed to measure the value of voting without affecting voting itself. Instead, we rely on sorting, survey completion, and survey responses. Yet, the model suggests an intervention to increase turnout – individuals with social-image motives are more likely to vote the more they expect to be asked. Experimentally increasing such expectation should thus lead to an increase in turnout.
In November of 2010 and of 2012, we did just that. A few days before the election, a flyer on the doorknob of treatment households informed them that “researchers will contact you within three weeks of the election [...] to conduct a survey on your voter participation.”
A control group received a flyer with a mere reminder of the upcoming election. The results are consistent with the model, though statistically imprecise. In 2010, the turnout of the treatment group is 1.3 percentage points higher than the control group (with a one-sided p-value of 0.06). In 2012, the turnout difference is just 0.1 percentage points (not significant).
Our results suggest that the anticipation of being asked whether one voted provides a substantial motivation for individuals to vote, even in large elections where the probability of influencing the result is miniscule. Indeed, the effect of being asked is as large as being paid $5-$15 to vote for a Congressional election. This paper thus provides rigorous evidence from a field experiment of a social image motivation to vote.
We end by mentioning a caveat. The results are specific to their time and location – the 2010 congressional elections in Illinois. The lack of estimated pride in voting could be related to the disappointing results for Democrats in 2010. It would be interesting to apply this methodology to other elections.
Downs, Anthony (1957), An economic theory of democracy, New York, Harper.
Erat, Sanjiv and Uri Gneezy (2012), “White lies”, Management sciences, 58(4): 723-733
Harsanyi, John (1977), “Morality and the theory of rational behaviour”, Social Research 44(4): 623-569
Harbaugh, William T (1996), “If people vote because they like to, then why do so many of them lie?”, Public Choice 89(1-2): 63-76
Harisson, Glen and John List (2004), “Field experiments”, Journal of Economic Literature, 42: 1009-1055
Gneezy, Uri (2005), “Deception: The role of consequences”, The American Economic Review, 95(1): 384-394
Riker, W H and Ordershook, P C (1968), “A theory of the calculus of voting”, American Political Science Review, 62: 25-42
Silver, Brian D, Barbara A Anderson, and Paul R. Abramson (1986), “Who overreports voting?”, American Political Science Review, 80: 613-624