Verified by 
The paradox of voter turnout – why do millions of people bother to vote in large national elections when they have virtually no chance of affecting the outcome? – is one of the major theoretical puzzles in the rational choice theory of politics. It’s a theoretical puzzle that I attempted to tackle when I was a rational choice theorist.
In my 1998 article in the Journal of Politics, I rely on the stochastic learning model (which is a variant of operant conditioning in behaviorism) to try to solve the paradox of voter turnout. I suggest that maybe voters are not trying to affect the electoral outcome in the future, but instead responding to the past electoral outcomes. Maybe they are using the combination of what they did in the last election and how the election turned out as a clue to what they should do now in their process of learning. So if they voted in the last election and their favorite candidate won, then their voting was reinforced and, following the logic of reinforcement, they should become more likely than before to vote in the next election. If, on the other hand, they voted in the last election and their favorite candidate lost, then their voting was punished and they should become less likely than before to vote again in the next election.
The same logic applies if they abstained in the last election. If their favorite candidate won, then their abstention was reinforced and, once again following the logic of reinforcement, they should become even more likely to abstain (or less likely to vote) in the next election. If their favorite candidate lost, then their abstention was punished and they should become less likely to abstain (or more likely to vote) in the next election.
I test this stochastic learning model of voter turnout with the American National Election Study data. The study examines people’s voting pattern in three consecutive national elections (1972 Presidential election, 1974 midterm Congressional election, and 1976 Presidential election). One unique feature of the American National Election Study is that the respondent’s vote is validated. In other words, after asking respondents whether or not they voted in an election, the survey taker went to the respondents’ precinct and validated their voting. Because voting is considered to be so normative (an important part of carrying out one’s civic duty), a lot of people lie about their voting. (But, as you can imagine, they only lie in one direction. Non-voters often claim that they voted, but no voters ever claim that they didn’t vote.) The American National Election Study allows the researchers to cut through the respondents’ misrepresentation and observe whether or not they truly voted, regardless of what they claim to have done.
My analysis of the data supports the stochastic learning model of voter turnout perfectly. Those who voted for the winner in the 1972 Presidential election (Richard M. Nixon) become significantly more likely to vote again both in the 1974 midterm Congressional election and the 1976 Presidential election. Those who voted for the loser in the 1972 Presidential election (George McGovern, John G. Schmitz, and other minor candidates) become significantly less likely to vote both in the 1974 midterm Congressional election and the 1976 Presidential election.
The same process is operative among nonvoters as well. Those who abstained in the 1972 Presidential election but supported Nixon become even less likely to vote in subsequent elections (because they got what they wanted without bothering to vote). In contrast, the 1972 abstainers who supported McGovern or Schmitz become more likely to vote in subsequent elections. These initial results from the 1972-1974-1976 elections from the American National Election Study are later replicated with the General Social Survey data on a larger number of Presidential elections from 1972 to 1992. These results are published in my 2000 American Sociological Review article.
In most Presidential elections, roughly half the people vote for the winning candidate, and roughly the other half vote for the losing candidate. In other words, about half of the voters have their voting reinforced in any given election, and the other half have their voting punished. So the stochastic learning model of voter turnout can explain, among other things, why roughly half the eligible voters choose to vote in each Presidential election.
So it does appear that people may be backward-looking adaptive learners (who respond to the past reinforcement contingencies and become more likely to repeat the reinforced response and less likely to repeat the punished response), rather than forward-looking utility maximizers (who engage in rational behavior designed to bring about the desired outcome in the future). It’s not the prospect of influencing the future electoral outcome that affects their likelihood of voting; it’s the association of their behavior with the desired outcome in the past.
But this conclusion simply raises more questions. Why are people backward-looking adaptive learners? Why do people respond to the outcomes of past elections as if their vote made a difference? If it is irrational to believe that their vote will make a difference in the outcome of the next election (which it is), why is it rational to believe that their vote made a difference in the past election (it is not)? Why do people who voted for the winner feel like their behavior was rewarded? Why do people who voted for the loser feel like their behavior was punished? I will explain in my next post.
