The More You Know: Voter Heuristics and the Information Search

Voters must also know how to effectively process and utilize this information, a skill lacking even amongst the knowledgeable, as argued by Cramer and Toff (2017).

Perhaps the clearest consequence of our focus on substantive representation is obscuring voters’ desire for descriptive representation—the congruence between the demographic characteristics of the representative and the represented (Pitkin 1967). Of course, given the underrepresentation of women and minorities, especially at higher offices, it would be reasonable for concerned voters to prioritize descriptive representation of such groups. Scholars typically argue that descriptive representation is valuable to the extent that someone like them is more likely to pursue their key substantive goals, or to the extent that having a fellow group member in office secures symbolic benefits for the group. A descriptively congruent representative, for instance, may be more likely to represent in the way Mansbridge (2003) refers to as gyroscopic—pursuing what constituents desire because it is what they themselves desire. In real life, however, voters may face tradeoffs between descriptive and substantive representation (Mansbridge 1999; Dovi 2002). Some evidence suggests in-group members choose substantive over descriptive representation, given a conflict between the two (Lerman and Sadin 2016).

The key question for the current project is whether voters know enough about even the functions of such offices to make retrospective considerations. Although we do not report the results in the main body of the paper, we show in SI Sect. 1.4 that, for all the offices we pose to respondents, a majority of respondents successfully identify key responsibilities of the office in question, suggesting that many voters can indeed engage in local retrospective voting, should they so choose. Nonetheless, due to considerations of length, we do not examine here how frequently voters engage in retrospective voting at the local level.

This scholarship faces a unique challenge: any consideration that seems to play a minor role in vote choice—for instance, issue opinions—may appear to be unimportant either because citizens are uninterested in evaluating their candidates on that basis, or because they lack the necessary information to do so. Despite this problem, most experiments examine voting behavior by manipulating either specific types of information about candidates—partisanship, policy stances, demographic characteristics, and so on—to assess their relative importance, or altering the presentation of those cues (such as with a prose vignette, photo, or video) to examine whether voters are sensitive to the medium.

We consider this heuristic conceptually distinct from other heuristic strategies, such as “take-the-best” or “fast and frugal” (Gigerenzer et al. 1999, 2008), for two reasons. First, take-the-best is meant for quickly choosing between two alternatives, while the deal-breaker heuristic we describe is often or even typically used in situations involving more than two alternatives. Second, take-the-best assumes the individual possesses prior knowledge about each alternative, then recalls from memory the most important information and looks for discriminating characteristics among them; we propose that voters employ the deal-breaker heuristic when they possess no prior knowledge on desired topics, and must instead choose the topics that they believe are both important and will allow them to eliminate at least some options. Recent work on information search and voting refers to what we call deal-breaking as “heuristic-based decision-making,” but we view this label as insufficiently precise, as distinct types of heuristics (minimalist, take-the-best, fast and frugal, etc.) would result in different types of search strategies--for instance, in shallow but comparable searches vs. non-comparable searches (see Lau et al. 2018, p. 4). For succinct conceptual definitions of take-the-best and fast and frugal heuristics, see Gigerenzer and Gaissmeier (2011).

We stress that this process applies only to candidates about whom the voters knows nothing before seeing their name on the ballot; the search process likely operates quite differently in high-profile races, where early information about candidates will not be self-sought but instead provided by the media.

We provide detailed information on the political knowledge batteries in SI Sect. 1.4. In the main paper, we use a five-item political knowledge battery per Delli Carpini and Keeter (1993), which is broadly understood to correlate with political interest (further mediated through media coverage), as well as education levels, which mediate uptake of the knowledge available (Jerit et al. 2006). While it was not possible to measure media coverage of a hypothetical candidate for office, we do also present evidence that using education levels (SI 3.1) or knowledge of local offices (SI 3.2) as the independent variable rather than political knowledge, provides similar results.

Our hope was that an open-ended text box most closely replicates the process of searching for information online, with the secondary advantage that we avoid putting ideas in the heads of our respondents. If asked to choose from a list of information about a candidate, the respondent might be confronted with many pieces of information (e.g., specific policy positions or candidate qualities) that they might otherwise never consider themselves.

We emphasize that we did not provide respondents with actual information regarding their request. Doing so would require us not only to anticipate all possible types of information requests, but also make assumptions about whether individual respondents would respond to a given revelation positively or negatively.

We assume that respondents would never react to disappointing information with increased likelihood of voting for the candidate, or vice versa.

Requests for political information are low for president because respondents were told this was a presidential primary candidate of their own party.

In SI Sect. 2.1, we also provide a breakdown of the content requests by both office and respondent political knowledge.

We acknowledge that citizens could potentially be justified in seeking ideological information about candidates for non-ideological offices. For instance, if voters see such offices as a springboard to higher office, some may want to eliminate ideologically divergent candidates early in their political careers. Other voters may be aware that some offices do touch on the ideological; Kentucky clerk Kim Davis’ refusal to provide marriage licenses to gay couples is one such example. Thus, we include all political requests as relevant for all offices. But we restrict the policy requests considered relevant based on each office (e.g., finance and economic policies would be considered relevant for a comptroller). See SI Sect. 1.3 for details.

We asked a battery of local political knowledge questions (e.g., does the mayor have the ability to hire city employees?) on Study 2 due to two concerns. First, we worried that typical political knowledge batteries, which focus on national offices and events, might overstate voters’ knowledge about local offices. Second, Ahler and Goggin (NP) report concerns that typical political knowledge batteries are too easy and too familiar (e.g., “who is president?), reducing variation in the variable of interest. As predicted, we find our results are even stronger with the local knowledge battery. We provide and interpret our results in SI Sect. 3.2.

Moreover, when aggregating across all requests made by an individual, we find that the base rates of each “unhelpful” type of search increase by 20–25%, with 80% of low knowledge and 65% of high knowledge respondents asking at least one unhelpful question (see SI Sect. 3.4). In SI Sect. 2.4, we show that less politically knowledgeable respondents are also less likely to report needing additional information to make a vote decision than high-knowledge respondents, and discuss the potential explanations for this finding.

This would create an information search asymmetry in most races: voters will know much less about candidates they quickly eliminate than candidates who meet that criteria. In other words, voters will learn much more about a candidate who exhausts their search preferences–the candidate they select–than their opponent(s). This prediction, though outside the scope of the present study, runs directly counter to most accounts of voting behavior, which presume that voters know the stances of all candidates on the criteria used to make a vote decision.

We show that the findings hold within each individual study in SI Sect. 4.2.