Theories

Proximity theory is the best-known of these three theories. It makes a basic claim: If you line up all the candidates from most liberal to most conservative, voters will pick the candidate whose ideology is most similar to their own. This theory serves as a basic assumption of the median voter theorem and other spatial models.

Discounting theory is similar to proximity voting, but with a recognition that winning candidates will have to battle other elected officials to get anything done. Imagine a genuinely moderate voter who desires middle-of-the-road policies. If the president is a staunch conservative, then that voter might prefer an extremely liberal Congress over a moderate Congress, given that actual policy outcomes will lie somewhere between what Congress and the president wants. Thus, a discounting voter weights candidate proximity based on the status quo.

The third theory, directional theory, argues that voters view the world in black and white. They want to vote for whichever candidate is on their side of the issues. For example, a moderate Republican would rather vote for an extremely conservative Republican than for a moderate Democrat, because the Republican is on the same side of the issues as the voter is–even if the moderate Democrat is closer to the voter ideologically.

Test

Previous attempts to test these three theories against one another empirically have run into several problems. For one thing, candidate positioning is endogenous–candidates try to place themselves ideologically in a winning position. There have also been a variety of measurement problems.

The main problem, though, has been a lack of critical tests. Previous research has not spelled out the exact circumstances under which these three theories yield different empirical results. The authors begin by filling that gap with a formal model. They show that only for voters within a narrow ideological range do the proximity and discounting rules produce divergent predictions; likewise, the proximity and directional rules produce divergent predictions only within another narrow ideological range. (See Figure 1, Table 1, and Figure 2 in the article for summaries of these scenarios.)

Knowledge of these ranges enables the authors to design critical tests of the theories with a simple experimental survey. Three survey questions provide most of the necessary data. First, they ask voters to place themselves on an 11-point ideological scale dealing with health care reform. Second, they ask them to choose one of two hypothetical candidates based only on each candidate’s position on this 11-point scale. And third, they ask voters to place current government policy along this 11-point scale.

Results

Using sophisticated statistical analysis of these data, the authors find that proximity voting is by far most common, followed by discounting and (distantly) directional voting. There is considerable heterogeneity within the sample; 57.7% are proximity voters, 27.6% are discounters, and only 14.7% are directional voters.

Demographic factors help explain some of this heterogeneity. For example, directional voter was twice as common among less educated respondents as among more educated ones. Most interestingly, discounting was much more common among moderates and independents,¹ a finding that may help explain why candidates polarize rather than converging to the ideological center.²

Critique

The authors provide a rigorous answer to a long-unanswered question. Their formal model is thoroughly persuasive, and it shows exactly where the three theories diverge. Thanks in large measure to this formal model, the authors have a compelling paper.

The empirical work is interesting, but I question whether they have underestimated discounting (and overestimated proximity voting). They do acknowledge a potential bias in this direction (p 310), but only in passing. My concern: When voters choose a hypothetical candidate in step two of the questionnaire, how do we know they are not taking account of the status quo? If they are, then they are making a discounted decision–not a proximity decision.

The authors attempt to control for this through question ordering–that is, by placing the status quo question last, not first, so as not to prime respondents toward discounting. But even if the status quo question were absent altogether, you still might have respondents considering the status quo when choosing a candidate. After all, when voters show up on election day, there is not a question on the ballot asking them to place status quo policies on an ideological scale–yet it appears that at least 27.6% of voters do so.

This article did set up a critical test between directional voting and the other two theories, but it did not set up a critical test between proximity and discounting. Granted, it showed that discounting is more common among moderate/independent voters than among others, but that’s only a marginal effect–a worthwhile finding, but not an answer to the puzzle.

Punchline: A significant contribution to this literature, but not a final answer.

In the absence of direct democracy, political parties might not have much reason to promote moderate policies. Republican legislators would generally prefer policies to the right of the median, while Democratic legislators would prefer policies to the left. Whichever party has the legislative majority has a variety of tools at its disposal to help it push policy away from the median and towards its own ideal point. These tools include control of the legislative agenda, control of committee chairmanships, and so on. In addition, the need to appease major campaign donors and partisan activists increases the incentive to use these tools. As a result, it is not the legislative majority that governs; it is the majority of the majority party that governs. When legislative control shifts from Republicans to Democrats, you might see a large shift in policy outcomes–even if only a couple of legislative seats changed hands.

But this is exactly the sort of thing that the Progressive movement sought to end. A purpose of direct democracy was to force legislators to produce policy outcomes closer to what the median voter would want–regardless of which party has the legislative majority. The purpose of Phillips’s article is to ask whether they succeeded.

There are two main mechanisms by which direct democracy can be a median-enhancing institution (from Gerber 1996). The first mechanism is direct; voters can impose specific legislation on the legislature, as happened with California’s famous Prop 13. The second is indirect; even if voters never use the initiative, its presence acts as a deterrent against extreme behaviors by the legislature.

Phillips finds evidence that direct democracy does matter. Party government is weaker in states with the citizen initiative than in states without. Here’s how he did it:

• Dependent variable: The state tax burden.¹
• Main independent variables: First, a few dummies to measure partisan control.² Second, a dummy indicating whether a state has direct democracy. Third, interactions between these two sets of variables.

As expected, he finds that the partisan dummies have strong relationships with the state tax burden. Democratic governments tax the most; Republicans tax the least; divided governments are in the middle. But here’s the rub: these relationships disappear in states with direct democracy.

The existing literature on the relationship between partisan control and size of government has had mixed results. Phillips contends that these mixed results can be explained, at least in part, by looking at direct democracy.

I have only one complaint with this article. I’m not convinced that a simple dummy variable can adequately measure direct democracy. From state to state, there are huge variations in how easy it is to use the initiative process, leading to huge differences in how frequently the process is used. This article could be more persuasive if it discussed this problem, which Phillips does not even mention. In fact, this problem is severe enough that, to me at least, it undermines the entire argument.