A few years ago,  Alford, Funk, and Hibbing challenged that environmental approach by showing that political orientations are genetically transmitted. Later work by Fowler and his colleagues has confirmed that our political leanings are genetically influenced. But although this genetic research has drawn our attention toward biological influences, it has not produced a theory that can explain why biology matters.

The goal of Mondak et al.’s recent APSR article is to develop a theory that can link these genetic studies with the more widespread environmental studies. The figure below (from the article) summarizes the theory. Note that they expect neither environmental factors nor personality traits to have much of a direct effect on political behavior. Instead, most of the effect is interactive. For example, if a person has an extroverted personality type, and if a form of political participation is social (e.g. a caucus as opposed to donating to a candidate via internet), then you will expect that person to participate. Here’s the figure:

Figure 1 from Mondak et al 2010

By “personality,” the authors refer to the “Big Five” personality index widely used within psychological circles. The Big Five traits include these:

• Openness to new experience. Folks who seek new experiences and information as opposed to folks content with their lot.
• Conscientiousness. Organized, hard-working folks as opposed to lazy or sloppy people.
• Extroversion vs introversion.
• Agreeableness. Warm, kind, sympathetic, generous people as opposed to unkind, distant, cold, miserly people.
• Emotional stability vs neuroticism. Calm, relaxed, stable as opposed to tense, nervous.

The authors stress a single main point: We cannot understand the effects of personality without accounting for the environment, and we cannot understand the effects of the environment without accounting for personality. They illustrate this argument by showing that certain types of political participation can be predicted well by interacting personality traits with environmental variables, but the empirical analysis seems peripheral here. As I understand it, the main goal of this paper is just to get political scientists thinking about the importance of personality.

They expect this personality research to supplant genetic research. The genetic research has shown an interesting relationship between biological factors and political behavior but without providing any sort of theoretical mechanism. By contrast, psychologists have shown that genes and other biological factors “account for most of the variance in personality traits” (p 89), but personality traits are the proximate cause of later behaviors.

Comments and Criticism

These are novel arguments, and I look forward to seeing how they influence future behavioral research. At the same time, I find myself wondering how much there is to gain by looking at personality. The authors have argued that personality can influence political behaviors (turnout and other political participation). But the genetic literature has shown that genetics influence political dispositions (liberal vs conservative, Republican vs Democratic). If Mondak et al. really want to show that personality is the real (proximate) cause of anything “caused” by genetics, then they need to show that personality influences political dispositions as well.

Update: See my review of a study that does just that–Gerber et al.’s “Personality and Political Attitudes.”

Given that this dispute revolves mostly around methodological (not theoretical) differences, one wonders why this article did not appear as a reply to Gerber et al, followed by a response from the original authors. As my critique at the end of this review will make clear, there are many holes that Gerber et al could poke in Denny and Doyle’s approach.

Details

Those familiar with Gerber et al’s 2003 article may be surprised to see methodological arguments leveled against it. After all, Gerber et al used sound experimental methods to neutralize methdological concerns. They selected a sample of voters who had not turned out in recent elections; they randomly assigned some of those non-voters to receive a mobilization treatment in 1998; of those who actually turned out in 1998 as a result of this mobilization effort, they then looked to see how many turned out again in 1999. At each stage, turnout information for each voter was gathered from public records–not from self-reported polling data.

The problem, Denny and Doyle claim, is not so much with the research design as with the statistical analysis. Gerber et al use instrumental variables in a lagged binary choice model, a procedure that does not yield consistent estimators. As a result, Denny and Doyle write, it appears that Gerber et al severely overestimated the effect that voting in previous elections has on voting in subsequent elections–that is, Gerber et al overestimated the extent to which voting is habit forming.

Denny and Doyle seek to estimate the strength of “habit” more accurately by using Britain’s National Child Development Study (NCDS), a six-stage panel that tracked children born in March of 1958 into the late 1990s. They use a large number of time-variant and time-invariant variables from this study to predict turnout in 1979, this cohort’s first opportunity to vote. For later elections, they use the same background variables, but insert a lag for whether the respondent had voted in previous elections. This procedure produces the estimate that voting in previous elections raises the probability of voting in subsequent elections by 13%.

Punchline: Yes, voting is habit-forming, but not as much as Gerber et al thought.

Critique

I’m surprised that AJPS did not provide Gerber et al the opportunity to reply to this article. I hope to see a reply from them in a future issue. Even if they accept Denny and Doyle’s econometric criticisms as valid (which they appear to be), I suspect that Gerber et al would find plenty of flaws in Denny and Doyle’s approach.

Randomization vs Control Variables

For example, I find it surprising that Denny and Doyle think their statistical control variables can remove individual-level background conditions (fixed effects) as well as Gerber et al’s randomization can. As Denny and Doyle acknowledge near the beginning of their article, the greatest empirical obstacle to identifying an effect of habit is that you must first control for anything that might influence an individual voter’s propensity to vote. If you omit a single variable that influences propensity to vote, then that variable’s effects will be swallowed up in the lagged turnout measure–inflating the estimated effect of habit.

Gerber et al use randomized experimentation to completely remove concerns about omitted variables; Denny and Doyle by contrast, use a (large) vector of control variables taken from the NCDS. They make a valiant effort to argue that these variables (and their sophisticated methods) can account for this problem, but let’s face it–no matter how many control variables you use, control variables can never improve on randomized experimentation.

Is Britain comparable to America?

More broadly, I wonder whether this study is even comparable to Gerber et al’s. Gerber et al studied American elections, and argued that habit raised your probability of turning out by 47%; Denny and Doyle study British elections and find that habit raises your probability of turning out by 13%. But do we really expect habit to have the same effect in such different contexts? Elections are far more frequent in the United States than in Britain–so much so that some worry about “voter fatigue” hurting American turnout. This difference alone implies two things relevant to habit formation. First: Since American elections are more frequent, voters have more opportunity to develop a strong habit. Second: Since American elections happen in more rapid succession, there is less time for the habit to “wear off” before the next election occurs. (Recall that Gerber et all compared turnout in 1998 to 1999–only one year apart; is it any surprise that this would have a stronger effect than comparing British turnout in 1979, 1987, and 1997?)

In sum: An interesting article, but I’d like to see more. I guess that’s how I end most of my reviews, though.

Briefly: The authors argue that mobilization has the strongest effects on voters who are indifferent about turning out. Efficient campaign managers should identify these fence-sitters and mobilize only them; money spent mobilizing those who are likely to turn out (or stay home) regardless of the campaign’s efforts is money wasted. Crucially, however, the authors demonstrate that these indifferent voters are not the same from one election to the next. In highly visible elections (like presidential elections), mobilization efforts should target those who rarely vote; in obscure elections (like legislative primaries), mobilization efforts should target those who regularly vote; and in mid-level elections (like Congressional or mayoral races), mobilization efforts should target those who vote occasionally.

Contribution to the Literature

This argument resolves a conflict in the literature between four different models of mobilization, which the authors summarize in their figure 1 (at right; click to enlarge). In panel A, mobilization influences all voters equally; in panel B, mobilization influences those who would be least likely to turn out otherwise (that is, mobilization has the strongest effect on “low propensity” voters); in panel C, mobilization has the strongest effect on high-propensity voters; and in panel D, mobilization has the strongest effect on voters with a moderate propensity to vote.

Although the authors reject panel A, their theory can produce a theory that can lead to either B, C, or D. In high-salience elections, panel B is accurate–since in high-salience elections, it is low-propensity voters who are debating whether to turn out. In low-salience elections, panel C is accurate–since it is the regular voters who are debating whether to turn out.

Empirical Test

For empirical evidence, the authors re-evaluate the results of 11 previous experiments.² They use turnout among each study’s control group as a proxy for salience. They estimate each voter’s “propensity to vote” using a bunch of demographic variables (mostly) and past turnout data. Sure enough, they confirm their theory.

Critique

The authors characterize a voter’s propensity to vote as an “enduring, individual-level trait.” I find this puzzling. It is well-known, for example, that turnout rises with age (to a point). We also know that voting can be habit-forming; a voter mobilized in one election becomes more likely to turn out in subsequent elections (Green and Shachar 2000; Gerber, Green, and Shachar 2003; Fowler 2006). Of course, I doubt that this measurement choice undermines their results.

On the whole, though, a welcome contribution. Campaign consultants should read this closely. When political scientists spend NSF money on mobilization experiments, they can use a blanket strategy. But when campaigns spend hard-earned dollars on mobilization efforts, they need to know exactly which voters to target.

Update: Research published by Catalist, a Democratic group, seems to support this paper’s conclusions. Read more here.

Or maybe–just maybe–voting is “contagious.”¹

The trouble is, it is difficult to prove empirically whether voting is contagious. To do so requires somehow controlling for selection effects and other outside factors. In the article reviewed here, Nickerson reports on an innovative experiment that does just that. Ahead of the September 10th, 2002, Congressional primaries in Denver and Minneapolis, Nickerson sent canvassers door to door to households that had (exactly) two registered voters. At each household, the canvasser gave a brief message to whomever answered the door. Based on random assignment, each household received either a “get out the vote” (GOTV) appeal or a placebo message about the importance of recycling.²

There are actually two experiments going on here. The first tests the direct effect of delivering a GOTV appeal to whomever answered the door. Consistent with previous experiments, Nickerson finds that this group voted at a rate 9.8 percentage points higher than the placebo group.

The second experiment is far more interesting. For it, Nickerson looks at turnout rates among those who did not answer the door. Those who did not answer the door at houses receiving the GOTV message voted at a rate 6.0 percentage points higher than those who did not answer the door in the placebo group. This is the contagion effect.

In other words, 61.2% of the treatment effect was passed on from the mobilized voters to their housemates.³ This is huge. Nickerson points out that this “contagion effect” has a stronger effect on turnout than education, income, or age.

Of course, every study has its shortcomings. Experiments like this are the gold standard for making a causal claim. However, experiments also suffer from external validity problems–that is, this study’s conclusions may not apply in every circumstance. Nickerson points out several potential concerns. Among them:

• This study looks only at two-voter households. But what about college roommates, multi-generational households, and other living arrangements?
• Nickerson conducted this study during a primary. But what about during higher-salience elections?
• Turnout might be contagious, but what about other behaviors, like volunteering, donating, or even vote choice?

Nickerson’s paper uses the best of methods to make a fascinating contribution to the literature. Still, I would ask just a couple more questions.

First, Nickerson does not randomly select which member of each household receives the message; instead, the message goes to whomever answers the door. It seems plausible that the person more likely to answer the door might differ from the person who does not. Might the effect have been (slightly) different if the message were delivered to the second person instead of the first?

Second, what causes this contagion? Nickerson mentions two possible reasons, the lowered costs of voting (sharing a ride to the polls) and the social pressure to conform (see Gerber, Green, and Larimer 2008). My hunch is that it’s the latter; it’s easy to ignore your civic duty to vote when your spouse also ignores it, but if your spouse bothers to vote, the situation changes. It might be nice to see further research looking specifically at how social pressures and civic duty operate within the family.

In an article by Gerber, Green, and Larimer, these three political scientists report the result of a large-scale randomized experiment involving 180,000 Michigan voters.¹ You may not know it, but several states (including Michigan) make it a matter of public record whether or not you vote. Anybody willing to pay a small access fee can have a list of registered voters, including names, addresses, and turnout information.²

Once you’ve got this information, designing an experiment is easy. First, find out who has voted in recent elections and who hasn’t. Then, apply some sort of treatment to a randomly selected group of these people. Finally, observe turnout in a future election and see whether the treated group increased its turnout.

What the Researchers Did

That’s the pattern the authors followed. They acquired this turnout information for 180,000 voters, then divided them into a control group (with 100,000 members) and four treatment groups (with 20,000 voters each). These five groups were all similar in terms of household size, gender balance, average age, and average turnout over the previous few elections. The general question was this: Can we come up with a “treatment” that will cause a treated group to vote at significantly higher levels than the untreated control group?

The four treatments were designed to play on our sense that voting is a civic duty. There’s an internal and an external side to civic duty. The internal side represents the positive “warm fuzzies” that you might feel inside if you fulfill your duty by voting. The external side represents the negative social shame you might feel from others if you fail to vote.

Treatment #1 was a simple postcard with a gentle reminder to recipients that voting is a duty, so they should participate in the upcoming primary election. Treatment #2 added a note that researchers were watching this election to learn about turnout (to control for any Hawthorne effects). Treatment #3 removed this research note, but added a reminder “that who votes is a matter of public record,” with the recipient’s turnout behavior over the past couple elections written below; following was an ominous reminder that “we intend to mail you an updated chart” after the primary, indicating whether you voted.

Treatment #4 was the doozy. At the top, in capital letters, was the question, “What if your neighbors knew whether you voted?” After a brief explanation, the mailer displayed the turnout record for the recipient and all the recipient’s neighbors. This is the sort of mailer that causes new cases of paranoid schizophrenia. If you’re not picturing what this mailer would have looked like, take a look at this sample (from the article’s appendix; click on the image to enlarge it):

What the Researchers Found

Treatment #1 yielded a turnout rate 1.8 percentage points higher than the control group’s. Previous researchers have tried things similar to treatment #1 and found similar results. Treatment #2 raised turnout 2.5 points relative to the control. Treatment #3 raised it 4.9 relative to the control. These are large changes.

But treatment #4 raised turnout by a whopping 8.1 percentage points from the control.

Let’s put that in perspective. In other experiments, the greatest boost in turnout came when experimenters sent door-to-door canvassers. The effect of these “social pressure” mailings was roughly the same as sending somebody to the voter’s door. However, it was far more cost effective. Canvassing costs roughly $20 per vote, while these mailers cost roughly $1.93 per vote.³ Of course, that doesn’t mean political consultants should jump on this new tactic; many recipients of treatment #4 called the number listed on the mailer to complain.

So what does this all mean?

It means two things. First, it means that there is a substantial internal “civic duty” motivation that gets people to vote. Sure, they might vote because they care about the outcome, or because they think their vote matters for the outcome. But (as treatment #1 shows) they also vote because they want the internal “warm fuzzies” of voting.

Second, there is an even stronger external “civic duty” motivation. Social shame matters, and in a big way. This sense of (external) civic duty gets activated when people know that others can observe their behavior. Shame works. The authors seem almost surprised by this finding. The psychological and sociological literatures warn of “reactance” when heavy-handed tactics are used to enforce norm compliance–so the authors seem to have half expected turnout to fall among group #4. But instead of reactance, they saw a large rise in compliance.

Most precincts give out those little “I voted” stickers to encourage people to vote, presumably to tap into these social rewards. If we really want to improve turnout, maybe the government should start mailing out postcards similar to treatment #4 before (and after) every election!

As Nichter points out, these two situations are far from hypothetical. These efforts to buy the vote happen occasionally in the United States,¹ but frequently in other countries–such as Argentina, the focus of this study.

But Nichter asks us to reconsider what’s happening here. When we observe these behaviors, are we observing “vote buying” (as we usually assume) or “turnout buying”? The question isn’t merely academic; “vote buying” smacks of corruption, but “turnout buying” looks more like mobilization, a (usually) laudable activity.

Nichter’s central claim is that parties engage primarily in turnout buying, not vote buying (although they may engage in both, to some extent). Rather than try to purchase support from moderately opposed voters, parties try to encourage non-voting supporters to turn out. This is a major shift from previous work, which has focused on vote buying. In making this claim, Nichter responds directly to Stokes (2005), even using her own data against her.

Stokes, like almost all other political scientists, assumed that when the Argentine Peronist party gave out financial rewards to voters that it was engaging in “vote buying,” not “turnout buying.” This led her down two avenues. First, she expected that money was flowing to moderately opposed voters, the group that could be persuaded to change sides most cheaply. Second, she sought to explain how the Peronists monitored these voters; given that ballots are secret, how do the Peronists know that people aren’t taking their money and then voting against them?

By shifting the focus to “turnout buying,” Nichter doesn’t have to worry about the secret ballot. He expects that parties will give money to their strongest supporters–albeit the ones that are not inclined to vote. If this is true, then the Peronists don’t need to monitor secret ballots; if such a voter bothers to turn out, the Personists can be reasonably sure that the voter will vote for them.

Using the same data (and regressions, more or less) that Stokes used, Nichter shows that Peronist money flowed predominantly toward strong Peronist supporters. Stokes noticed this too, of course, but argued it away, saying that voters had already received the Peronist money (possibly over several election cycles) and may have inflating their claims of support to exit pollsters as a result. Still, Nichter correctly points out that “the most straightforward interpretation” is that Peronists were buying turnout, not votes.

However, the same set of regressions also showed that past voting behavior had no bearing on reward distribution. That is, even if a voter told pollsters that he did not vote in the previous election (1999), that didn’t make him more (or less) likely to have received Peronist rewards during the present election. This finding runs counter to Nichter’s expectations; nonvoters should have been more likely to receive incentives. Nichter argues the point away, but he does so using similar logic as what Stokes used to argue away the relationship between Peronist support and receiving rewards. And if you read the preceding paragraph of this summary, you’ll see that Nichter rejected the logic that time around. To use Nichter’s words (about Stokes) against him, “the most straightforward interpretation” is that the Peronists weren’t targeting non-voters.

All in all, an interesting paper. His point that the parties are supporting their own partisans is well made. His point that they are targeting those that otherwise would not vote is not. Further research is needed. As Nichter points out in his article, a panel study would be ideal to parse out exactly what is going on.