Menu Adam R Brown

Notes navigation: Browse by titleBrowse by authorSubject index

Kanazawa: A new solution to the collective action problem

Disclaimer. Don't rely on these old notes in lieu of reading the literature, but they can jog your memory. As a grad student long ago, my peers and I collaborated to write and exchange summaries of political science research. I posted them to a wiki-style website. "Wikisum" is now dead but archived here. I cannot vouch for these notes' accuracy, nor can I say who wrote them.

Kanazawa. 2000. A new solution to the collective action problem: The paradox of voter turnout. American Sociological Review 65 (June): 433-442.

In Brief

Voters are backward-looking adaptive learners. Election results act as rewards and/or punishments that influence voting behavior.


Utility Maximizers?

Kanazawa builds on the work of Macy, who challenges the application of subjective expected utility maximization theory to human behavior. He claims that the calculations involved in prospective utility maximization are, in terms of cognition, simply too demanding. People use a simpler "win-stay-lose-shift" rubric. (This concept is basically Pavlovian conditioning.)

The Stochastic Learning Model (SLM):

The key here is correlation rather than causation. People base their behavioral decisions on whether acts have been "successful" or not in the past. If a behavior and a desired outcome are positively correlated, then the behavior is reinforced. If there is a negative correlation, then the behavior is punished.

Voting Calculus and the SLM:

Kanazawa applies the SLM to voting, and explains how this type of model would alter the logic of the standard calculus of voting model (pB + D > C). He claims that the SLM causes the "p" term to average out to .500, since, under this model, it "represents the probability that one's vote was associated with a win in the past" (982). The "D" term is endogenized under the SLM, since attachment to norms is also affected by reinforcement/punishment over time. Instrumental learning (the "p" term) is immediately responsive to the presence or absence of a correlation between behavior and outcome, while normative learning (the "D" term) will be somewhat insulated from this relationship (i.e. it may take many punishments/reinforcements to significantly affect "D").


Note that the actual chart on page 439 (Kanazawa 2000) is incorrect!

An Intuitive Explanation: If I vote in an election and my favored candidate wins, I'll be really likely to vote in the next election. If I vote in an election and my favored candidate loses, it makes me less likely to vote than I was before, but I might still have a strong sense of duty that brings me back to the polls next time. If I vote in 3 consecutive elections and my favored candidate loses each time, then perhaps my sense of duty will be "worn down," and I may stop voting. If I stop voting and my candidate is successful, then I realize that I can get away with not voting and I won't vote next time. If I don't vote and my favored candidate loses, then I feel guilty and may vote next time.


Empirical Support Part I:

In his first article (1998) Kanazawa uses NES data from 1972-1974-1976 to test his model. These are the only years that he can use because respondents' answers about whether they had voted or not were checked using voting records. His hypotheses were supported, but he notes that this result would NOT have been the same if he would have taken respondents at their word (i.e. many people lied about their voting behavior)! The problem with this data is that the years 1972-1976 were very exceptional for American politics. Think Vietnam, Nixon, etc...

Empirical Support Part II:

In his second article (2000) Kanazawa uses General Social Survey (GSS) data from 1973-1993. This dataset allows him to remedy the historical problem with his earlier test, but it does NOT allow him to check whether respondents were truthful or not (which, remember, he identified as an extremely important step in his first article). His hypotheses are again supported.

Comments and Criticisms

The SLM has some advantages. I especially like that it doesn't require people to do lots of math when deciding whether to vote or not. However, there are a number of questions that go unaddressed in these two essays. Do people only care about presidential elections, or are they punished/reinforced for each candidate/proposition/ bond measure that they vote for? Are people really conditioned by prior correlations between voting and outcome? While sophisticated probability calculation is difficult, it seems almost demeaning to suggest that we are backward-looking lab rats who are merely "trained" by circumstances to vote or not vote. Finally, the empirical results are quite shaky, since each article's test is susceptible to either historical bias or false reporting by respondents.

What about evidence that the same voters turn out year after year? Kanazawa's model goes against findings that voting is habitual (see Gerber et al). See also Fowler's (2006) critique. [[]]

Research on similar subjects


Kanazawa, Satoshi (author)Political TheoryVotingTurnout

Wikisum home: Index of all summaries by title, by author, or by subject.