MEASURING VOTER DECISION STRATEGIES IN POLITICAL BEHAVIOR AND PUBLIC OPINION RESEARCH

Transcription

1 Public Opinion Quarterly, Vol. 82, Special Issue 2018, pp MEASURING VOTER DECISION STRATEGIES IN POLITICAL BEHAVIOR AND PUBLIC OPINION RESEARCH RICHARD R. LAU* MONA S. KLEINBERG TESSA M. DITONTO Abstract Although political science has advanced the study of voter decision-making, the discipline still understands very little about how citizens go about reaching those decisions. In this article, we introduce a five-factor self-report scale of political decision-making (PolDec-5) administered to six different samples with more than 6,500 respondents over the past four years. Analyses illustrate that our five subscales Rational Choice, Confirmatory, Fast and Frugal, Heuristic-Based, and Going with Your Gut have high internal consistency, relatively high discriminant validity (as they are largely distinct from existing measures of decision-making style), and significantly high predictive validity, as established by process tracing studies where actual decision strategies of voters can be observed directly. Finally, we discuss how these new measures can help predict important political outcomes. What strategies do voters employ to reach their voting decisions? Social scientists have utilized representative surveys to understand election outcomes since the 1950s, and research has shown that a number of considerations play Richard R. Lau is a Distinguished Professor in the Department of Political Science, Rutgers University, New Brunswick, NJ, USA. Mona S. Kleinberg is an assistant professor in the Department of Political Science, University of Massachusetts, Lowell, MA, USA. Tessa M. Ditonto is an assistant professor in the Department of Political Science, Iowa State University, Ames, IA, USA. The authors thank David Andersen, Larry Bartels, John Geer, Eric Johnson, Carolyn Lau, Douglas Pierce, David Redlawsk, and Mark Schlesinger for help at various earlier stages of this research. This research was supported by National Science Foundation grants [SES to R.R.L. and T.M.D. and SES to R.R.L. and M.S.K.] and research funds from the School of Arts and Sciences at Rutgers University to R.R.L. *Address correspondence to Richard R. Lau, Department of Political Science, Rutgers University, 89 George St, New Brunswick, NJ 08901, USA; ricklau@polisci.rutgers.edu. doi: /poq/nfy004 Advance Access publication March 22, 2018 The Author(s) Published by Oxford University Press on behalf of the American Association for Public Opinion Research. All rights reserved. For permissions, please journals.permissions@oup.com

2 326 Lau, Kleinberg, and Ditonto a role in shaping vote choice. Voters may prospectively consider agreement with the proposed policy stands of candidates or parties (Rabinowitz and MacDonald 1989; Jessee 2009), and on occasion their material self-interest (Sears et al. 1980; Sears and Funk 1991). Retrospective judgments about past performance in office can be influential, especially when incumbents seek reelection (Fiorina 1981). In candidate-centered electoral systems, the character of the competing politicians their competence and integrity is often important (Abelson et al. 1982; Rahn 1993). Stereotypes and prejudice can automatically (and unconsciously) influence decisions (Crawford et al. 2011; Lodge and Taber 2013), and social identifications, particularly with political parties, can determine vote choice (Campbell et al. 1960). If the primary objective is to identify factors associated with a candidate or party s victory, then aggregate-level analysis and common political science tools (e.g., well-specified multivariate models) may suffice. If the primary interest, however, is to understand why individual voters have reached the decisions they have, then the analyses we typically utilize in political science research are inadequate because they do not examine the process by which individuals actually make decisions. Humans are cognitively limited information processors (Simon 1979; Anderson 1983; Redlawsk and Lau 2013). Hence, they cannot consider all of the possible factors in a typical campaign and election simultaneously, nor would they want to even if they could. It takes a tremendous amount of effort to gather and understand the information necessary to employ many of the standard predictors in a voting model, far more effort than most people are willing to expend. Research suggests that most individuals adopt decision strategies that use only a very small number of all possible decision criteria when making their choice: Some individuals are issue voters, others are party voters, and yet another set of people focus on the characteristics of candidates as people. Clearly, the field would benefit from the identification of the specific decision strategies employed by individual voters, because such evidence would tell us not just more about who is likely to win an election, but also about how different people participate cognitively in the democratic process. The psychological concept of cognitive style captures this idea very well. Cognitive style refers to consistencies in an individual s manner of cognitive functions, particularly with respect to acquiring and processing information, and to fairly stable individual differences in the way people perceive, think, solve problems, learn, and relate to others (Kozhevnikov 2007, p. 464). Early work in psychology focused on individual differences in basic perception, developing dichotomies such as field independence versus field dependence (Witkin 1950). As this area of research matured, the concept was adopted by researchers in many applied fields, and utilized to understand much more complex tasks, such as decision-making, learning, and problem-solving. According to a recent review by Kozhevnikov (2007), many researchers have

3 Measuring Voter Decision Strategies 327 concluded that cognitive style is often a better predictor of success in particular fields than general intelligence or prominent situational factors. Our goal in this article is to introduce a practical self-report tool that can reliably measure different styles of voter decision-making that should be of particular interest to political psychologists and survey researchers. We readily acknowledge that there are other ways one could address the question of how voters decide, for example with agent-based (Laver 2005) or even more complex computational models (Kollman, Miller, and Page 1992; Taber and Steenbergen 1995; Kim, Taber, and Lodge 2010), but we focus on methods that are more readily accessible and easier to implement on a large scale for public opinion researchers. We build on prior theoretical work of Lau and Redlawsk (2006) and their four styles of political decision-making, and present a 13-item Political Decision-Making scale that includes multi-item subscales designed to measure Lau and Redlawsk s four different models, plus a fifth going with your gut decision style. After reviewing their theory, we briefly describe six different datasets in which these new items have been administered, present the measurement properties of our new scales, and establish their discriminant and predictive validity. We conclude by suggesting how these new measures of voters decision styles could provide fresh insights into perennial questions of political behavior. Political Decision Strategies Broadly speaking, a decision strategy is a set of mental and physical operations that an individual uses to reach a decision (Lau and Redlawsk 2006, p. 30; see also Payne, Bettman, and Johnson [1993]; Lau [2003]; Redlawsk and Lau [2013]). At the very least, decision strategies involve plans for gathering relevant information (from the external environment, and/or by search through memory), evaluating that information, and choosing among alternative courses of action. Lau and Redlawsk described four broad types (or, in their words, models ) of decision strategies that are employed by citizens in making vote decisions. These four strategies differ in how much information is gathered (depth of search), and how evenly that search is distributed across alternatives (comparability of search) the two major dimensions identified by psychologists across which various decision strategies differ (e.g., Jacoby et al. 1987; Ford et al. 1989; Payne, Bettman, and Johnson 1993). Strategy 1, Classic Rational Choice, involves actively (though dispassionately) gathering as much information as one possibly can, about every candidate/party on the ballot in other words, deep, comparable search. It is cognitively difficult and more time consuming than other strategies, as it involves carefully weighing the positive and negative attributes associated with each alternative and balancing between the two but has the benefit

4 328 Lau, Kleinberg, and Ditonto of promising the highest probability of finding a value-maximizing outcome (Enelow and Hinich 1984; Hastie and Dawes 2009; Chong 2013). Strategy 2, Confirmatory decision-making, is based on early affectivebased socialization toward or against prominent symbols such as political parties, and a subsequent motivation to maintain cognitive consistency with the early learned affect. Party identification is generally the lens through which political information is selectively perceived. Information search is more often passive than active, although in the right circumstances, such as a high-profile national election, could involve selectively gathering and learning a lot of information. In other circumstances, however (e.g., low-profile elections), confirmatory search could consist of little more than learning the candidates party affiliations. To the extent that search is consciously guided, it should be biased in favor of the in-party candidate, which translates to shallow to deep but clearly unequal search (Sears 1975; Kunda 1990; Lodge and Taber 2013). Strategy 3, Fast and Frugal decision-making, assumes that voters are motivated primarily by efficiency, actively seeking only the most diagnostic information that will allow them to quickly make the correct choice. Information-seeking should be limited to the one or two most important/ diagnostic criteria, but those few criteria should be evenhandedly applied to every alternative in the choice set that is, shallow but comparable search (Gigerenzer and Goldstein 1996; Gigerenzer and Todd 1999). Strategy 4, Heuristic-Based decision-making, views voters as cognitively limited information processors who are generally motivated to make good (although not necessarily best ) decisions, as easily as possible. Information search is generally quite limited. Various cognitive shortcuts and heuristics are heavily utilized, particularly those that avoid having to make cognitively difficult value trade-offs. This is usually accomplished by limiting search to a single satisfactory alternative, or eliminating alternatives as soon as any negative information about them is encountered, thus usually resulting in noncomparable search across alternatives (Kahneman, Slovic, and Tversky 1982; Lau and Redlawsk 2001). How do we know which strategy a decision-maker is following when they are making a decision? As Payne, Bettman, and Johnson (1993) note, observing decision-makers information search behavior while they are making the decision provides valuable insights into the decision strategies they are utilizing. As noted above, different combinations of shallow or deep, and comparable or noncomparable information search, uniquely identify each one of Lau and Redlawsk s (2006) four decision-making styles. We also propose a fifth possible type of decision-making, one that gets more attention in the popular press than among psychologists, and is colloquially referred to as going with your gut. Keeping this common colloquial label, strategy 5, Gut decision-making, is strictly affective, usually unconscious, and involves no deliberate external searching for information. It should surely be

5 Measuring Voter Decision Strategies 329 associated with shallow information search, with no effort whatsoever to compare alternatives on anything other than how they make you feel (Dane, Rockmann, and Pratt 2012). Allegedly, it often provides very good decisions or at least choices that, retrospectively, decision-makers feel good about. Table 1 presents a set of self-report items we have developed to measure the extent to which respondents follow each of these five different strategies when making their vote choices ( PolDec-5 for short), while table 2 summarizes the expected relationship of each subscale with depth and comparability of decision-relevant information search. Method BRIEF STUDY DESCRIPTIONS The new decision-making items were administered in six different studies: three nationally representative surveys, and three experimental studies of mock election campaigns. While the survey data are representative, they are limited because they do not allow us to peer inside the black box of decisionmaking. To actually observe subjects engaging with information as they make their vote decisions, we employ experimental methods to simulate the information environments of multiple campaign scenarios. More information can be found about each study in the associated publications, as listed below. Study 1 and study 2 were both mock election experiments using the Dynamic Process Tracing Environment (DPTE) experimental platform, run more or less simultaneously during the late spring and summer of About a third of the subjects in each study were paid $20 to come into our laboratory at the university to complete the study. The remaining subjects were recruited from Amazon s Mechanical Turk (MTurk), and were paid $5 to complete the study online. Each of these experiments took about 45 minutes to complete. DPTE simulates the ongoing flow of information during an election campaign, by presenting so-called information boxes to subjects that scroll down a computer screen at a steady pace. Each information box displays a candidate s name, a small picture of the candidate s face (sufficiently large to perceive gender and race), a colored (blue or red) border indicating the candidate s party, and a short label indicating what a subject might learn by clicking on and thereby opening the box (e.g., Clark s Stand on Abortion). The purpose of the DPTE program is to give subjects considerable discretion over the type and amount of information they access in order to learn about competing candidates as is the case in actual political campaigns. 1. Interested readers can go to to learn more about the DPTE program.

6 330 Lau, Kleinberg, and Ditonto Table 1. POLDEC-5: a political decision-making scale Strategy 1, Rational Choice Decision-Making RC1. When I have an important choice to make, I like to gather as much information as I possibly can. RC2. If I learn something about one candidate running for office, I try to find out the same information about other candidates. RC3. I find it important to carefully consider all likely alternatives whenever I am making a decision. RC4. When I have to make a quick decision, I try to be as objective and balanced as I possibly can. (Added to revised scale) Strategy 2, Confirmatory Decision-Making PC1. All I need to know when making a tough political decision is what party a candidate belongs to. PC2. The parties are so polarized and distinct today that it is hard for me to imagine ever voting for a candidate from another party. a PC3. I usually see mostly good things about the candidates from my party and many bad things about the candidates from other parties. (Revised scale only) b Strategy 3, Fast and Frugal Decision-Making FF1. There are only one or two issues I really care about in most elections. I make my decision by comparing the candidates on those one or two issues. FF2. Whenever I have to make a tough choice, I focus on the most important aspects of the decision and leave it at that. Strategy 4, Heuristic-Based Decision-Making Heur1. Choosing a familiar candidate is an easy way for me to make a reasonably good vote choice. Heur2. If one option meets my needs I will save time and go with it without really looking at others. Heur3. In deciding how to vote, I often follow the recommendations of people or groups I trust. Strategy 5, Gut Decision-Making Gut1. When making decisions, I usually just go with my gut. c Note. We have typically utilized a seven-point response scale running from Strongly Disagree, Disagree, Slightly Disagree, Neither Agree nor Disagree, Slightly Agree, Agree, or Strongly Agree. We would recommend a shorter five-point scale on a phone survey (dropping the two slightly alternatives) that requires respondents to keep fewer alternatives in mind as they answer each item. a In the original version of this scale, this item read from the opposite party. We revised this item slightly so that it would work better in a multiparty electoral system. We would argue (and our data suggest) that in a two-party system, with either wording the item is interpreted the same way. b In the original version of the PolDec-5 scale, this item read When deciding how to vote, I usually end up learning more about the candidate from my own party than candidates from other parties. c In the original version of the PolDec-5 scale, this subscale included a second item that proved to be problematic: I never sweat even big decisions. The best choice is usually pretty obvious to me. If researchers are looking for additional measures of Gut decision-making, borrowing from Scott and Bruce (1985), we would suggest, When I make vote decisions, I tend to rely on my intuition and I generally make political decisions that feel right to me. In study 6, this threeitem subscale has a reliability (coefficient alpha) of.80.

7 Measuring Voter Decision Strategies 331 Table 2. Predicted relationship between each decision-making scale and the depth and comparability of decision-relevant information search PolDec-5 subscale Depth of search Comparability of search Rational choice Positive Positive Confirmatory Positive Positive/Negative a Fast and frugal Negative Positive Heuristic-based Negative Negative Go with gut Negative Negative a Confirmatory decision-making should be positively associated with comparability of information search in a within-party election such as a primary election in the United States. Confirmatory decision-making should be negatively associated with the comparability of search in any multiparty election, with greater search directed at the in-party candidate. Participants in study 1 experienced a mock presidential primary contested by two candidates in each of the Democratic and Republican parties, followed by a general election campaign where the winning Democratic candidate faced off against the winning Republican (see Ditonto [2013, 2017] for more detail). In study 2, subjects chose to vote in either a Democratic or Republican presidential primary election (see Kleinberg [2014] for more detail). Both studies began with subjects completing a pre-experiment questionnaire. In study 1, this pretest survey included the 13 items of the original version of our PolDec-5 scale. In study 2, the PolDec-5 items were included in the post-experiment questionnaire. Both the lab and MTurk subjects are nonrepresentative convenience samples (details available in table 3). This article will focus on the unique data provided by these experiments, a record of the actual information search behavior of the subjects in these two experiments. Study 3 and study 4 were modules purchased from the 2012 Cooperative Campaign Analysis Project (CCAP), a nationally representative online panel run by YouGov/Polimetrix during the 2012 US presidential election campaign. The first module (study 3), with 1,500 respondents, was run in June 2012; it included the 13 items of the original PolDec-5 scale. The second module (study 4), with 2,000 respondents, was run in September 2012, and included the 13 items of the slightly revised scale. 2 (See Geer, Lau, and Nickerson [2013] 2. The 2012 CCAP study began in December 2011 when 45,000 adult American citizens were initially interviewed. These respondents were drawn from YouGov s panel of online respondents who had agreed to take occasional surveys. In June 2012, our study 3 invited a stratified sample of over 1,800 of those respondents to complete the 13 items from the original PolDec-5 scale. In September 2012, our study 4 invited a separate but similarly stratified sample of over 2,400 respondents to complete the 13 questions from the revised PolDec-5 scale. YouGov then employs a matching procedure, which begins by creating a synthetic sampling frame from nationally representative Census Bureau data. Our data comprise 1,500 respondents from the YouGov sample (for study 3; 2,000 respondents in study 4) who best matched the synthetic sampling frame.

8 332 Lau, Kleinberg, and Ditonto for more detail about these two studies.) For each of these modules, a common core of background and general political beliefs was available for all of these respondents, along with current evaluations of the major presidential candidates, Barack Obama and Mitt Romney. The data also include respondents self-reported 2012 vote choice. Study 5 was a nationally representative panel survey conducted by the Eagleton Center for Public Interest Polling at Rutgers University. A total of 1,531 respondents completed a minute phone interview in early October 2012, which included the 13 items of our revised PolDec-5 scale. 3 (See Lau et al. [2017] for a description of the larger project from which these data are drawn.) At the end of the October survey, respondents were asked if we could call them again for a brief survey after the November election. A total of 719 (47 percent) of the original respondents were successfully contacted after the election, and reported their vote choice. As with the CCAP surveys, we have current evaluations from a representative sample of the actual major presidential candidates during the 2012 presidential election, and eventually reported vote choice from about half of them. Study 6 was a 10-wave experimental study of a presidential primary election campaign, conducted four times between February 2014 and April A total of 609 subjects completed the DPTE study across these four replications. The subjects were a mix of undergraduates enrolled in research methods classes in two large universities who participated in the study as part of a class assignment (44 percent), parents of some of those undergraduates who volunteered to take part in the study at the same time as their offspring (10 percent), and MTurkers who were paid up to $12 for their participation in the study (46 percent). 4 The 13 items of the revised PolDec-5 scale were administered in the first wave of the panel study, during which the two candidates running in the subject s chosen primary were also introduced. In waves 2 through 9 of the study, subjects had the opportunity to view new items about the candidates, a mix of policy stands taken from the candidates web pages, endorsements of several interest groups, and media articles about the campaign. In wave 10 of the study, subjects voted, evaluated the two competing candidates, and answered a series of questions about them. (See Kleinberg and Lau [2017] for a thorough description of this study.) SUBJECTS Table 3 provides a summary of the demographic characteristics of the six different samples used in this article. A total of just over 6,500 respondents provided 3. The phone numbers, which included both landlines and cell phones, were provided by Survey Sampling International. The sampling frame was registered voters, and it excluded Alaska and Hawaii. The response rate, using AAPOR s method 1, was 2 percent. 4. There were no important differences between samples for any of the analyses in studies 1, 2, or 6.

9 Measuring Voter Decision Strategies 333 Table 3. Study characteristics Study 1 Summer 2012 Study 2 Summer 2012 Study 3 June 2012 Study 4 Sept Study 5 Oct Study 6 DPTE Feb Apr Sample description Convenience sample of local subjects run in laboratory (24%) and recruited thru MTurk (76%) Convenience sample of local subjects run in laboratory (33%) and recruited thru MTurk (67%) Nationally representative online sample provided by Cooperative Campaign Analysis Project Nationally representative online sample provided by Cooperative Campaign Analysis Project Nationally representative phone survey conducted by the Center for Public Interest Polling at Rutgers Scale version Original Original Original Revised Revised Revised % Female 58 % NA 51% 53% 56% 48% % Nonwhite 24% 36% 28% 27% 15% 37% Median edu. 15 years 14 Years 13 Years 13 Years 14 Years 14 Years Median age Partisanship 51% Dems, 58% Dems, 34% Dems, 35% Dems, 34% Dems, 44% Dems, 32% Ind, 25% Ind, 40% Ind, 41% Ind, 32% Ind, 39% Ind, 17% Reps 18% Reps 26% Reps 24% Reps 34% Reps 17% Reps Sample size ,500 2,000 1, Convenience sample of students (44%), parents (10%), and MTurkers (46%)

10 334 Lau, Kleinberg, and Ditonto answers to the decision-making questions, with over 5,000 coming from different nationally representative surveys. A little over half of the respondents were female, and about a quarter of them nonwhite. Results SCALE DEVELOPMENT The purpose of this research is to develop a concise scale of different decisionmaking strategies that is short enough to be utilized in standard survey research. Respondents were asked to indicate their level of agreement with statements that describe how they typically make political decisions, as opposed to how they believe they should be made (see question wording in table 1). It was our goal to devise two to four items that could reliably measure each of these different decision strategies. The three items measuring Rational Choice each focus on crucial aspects of that mode of decision-making gathering a lot of information about all viable alternatives, and trying to gather comparable information about each of those alternatives. The first two items in the Confirmatory subscale emphasize party as the primary consideration in making a vote decision. The third item to represent this mode of decision-making measured biased information search (a focus on the in-party candidate), but was replaced in the revised scale by an item that measures biased perception of the information you learn about the in-party candidate and candidates from other parties. The two items representing Fast and Frugal decision-making quantify the extent to which a respondent exclusively focuses on a very limited number of decision criteria (without specifying what they are), applied to all alternatives under consideration. The three items representing Heuristic-Based decision-making each mention a particular shortcut for making a decision: familiarity, satisficing, and following the recommendations of trusted experts. We consciously measure this mode of decision-making broadly, trading greater face validity and wider applicability for the higher internal consistency that a more narrowly focused set of items might presumably have obtained. Finally, although they do not explicitly use these words, the two items proposed to measure Gut decision-making clearly endorse making decisions quickly without much deliberation. We administered the initial version of the scale in two different mock election experiments, and one large nationally representative online survey (studies 1, 2, and 3). We first conducted a confirmatory factor analysis on our proposed scale using the computer program EQS 6.3 for Windows (Bentler 2006). There is no single way to evaluate the fit of a confirmatory factor analysis. With relatively large sample sizes, as we have in all six studies, everything is statistically significant. The EQS program provides seven different fit indices to judge how well a hypothesized model fits any particular sample. Across the

11 Measuring Voter Decision Strategies 335 first three studies that administered the original version of our scale, these fit indices varied between a low of.813 and a high of.949. The median fit indices in our first three samples were.867,.897, and.895, respectively. These numbers are in the lower range of most published work. Table A.1 in the online appendix shows the results of the confirmatory factor analysis from study 3, with its large nationally representative survey, but the CFAs from the first two studies with their convenience samples do not differ meaningfully. A preliminary examination of the measurement data from these first three studies suggests that we successfully developed strong measures of Rational Choice and of Heuristic-Based decision-making, and a good measure of Fast and Frugal decision-making. However, one of the three items designed to measure the Confirmatory subscale had strong correlations with other factors, while the two items designed to measure Gut decision-making never correlated more strongly with each other than they did with items from some other factor. This preliminary analysis also suggested that Rational Choice stands in stark contrast to the other four factors, which all have positive correlations with each other, but negative correlations with Rational Choice. We therefore revised our original scale by replacing the problematic item from the Confirmatory subscale with a new item that focused more on motivated evaluation rather than biased search. Because Rational Choice is empirically distinct from the other four subscales, we dropped the problematic measure of Gut decision-making, and replaced it with a fourth item hypothesized to measure Rational Choice, focusing on unbiased ( objective and balanced ) processing of information. This left us with a single item (albeit one with high face validity) measuring Gut decision-making. Table 4 displays the key results from a confirmatory factor analysis of the revised decision-making scale from study 5, the most recently fielded nationally representative survey we have in which this new scale was administered. (Table A.2 in the online appendix reports a similar analysis from study 4, with another large nationally representative sample that was in the field a month earlier.) As can be seen in table 4, the data fit the hypothesized factor structure extremely well. Each of the 13 items has a positive and highly significant loading on its hypothesized factor, of course, and all of the various fit indices are above.95. The median fit index from this study was.965. The median fit indices from studies 4 and 6 were even higher:.994 and.974, respectively. We then computed summary measures of each subscale by averaging together the items that were hypothesized to measure each decision-making style. Descriptive statistics for all of these summary measures, including their internal consistency (coefficient alpha), are shown in table 5, while table 6 presents subscale correlations. Although this is primarily a scale development article and therefore largely a descriptive study, we do hold some expectations about the data. For example, although our items ask respondents to report on what they do, rather than what they believe they should do when making

12 336 Lau, Kleinberg, and Ditonto Table 4. Confirmatory factor analysis, revised scale, study 5 measurement equations Rational choice Confirmatory Fast and frugal Heuristic-based Go with gut RC1.311 (.021) RC2.328 (.029) RC3.470 (.026) RC4.277 (.021) PC1.863 (.034) PC2.597 (.032) PC3.622 (.036) FF1.716 (.037) FF2.506 (.033) Heu1.612 (.032) Heu2.598 (.027) Heu3.342 (.035) Gut (.035) Note. Table entries are unstandardized factor loadings, with standard errors in parentheses. These coefficients are all many times larger than their standard errors, and thus highly statistically significant. Measures of goodness of fit Chi-square (57 df) Bentler-Bonett normed fit index.955 Bentler-Bonett non-normed fit index.953 Comparative fit index.967 Bollen s fit index.967 McDonald s fit index.951 Joreskog-Sorbom s GFI fit index.978 Joreskog-Sorbom s AGFI fit index.965 Standardized root mean-square residual.033

13 Measuring Voter Decision Strategies 337 Table 5. Subscale characteristics, by sample Study 1 a Study 2 a Study 3 b Study 4 b Study 5 b Study 6 a Rational choice (Original 3 items, Revised 4 items) Confirmatory decisionmaking (3 items) Fast and frugal decisionmaking (2 items) Heuristic-based decision-making (3 items) M = 5.72 c M = 5.85 c M = 4.14 c M = 4.08 d M = 4.22 d M = 5.81 d sd = 0.86 sd = 0.79 sd = 0.59 sd = 0.61 sd = 0.50 sd = 0.75 α =.67 α =.68 α =.66 α =.72 α =.60 α =.68 M = 3.50 c M = 3.63 c M = 2.84 c M = 2.72 d M = 2.51 d M = 3.64 d sd = 1.34 sd = 1.32 sd = 0.89 sd = 0.83 sd = 0.89 sd = 1.32 α =.62 α =.65 α =.68 α =.69 α =.61 α =.67 M = 4.03 e M = 4.60 e M = 3.31 e M = 3.11 e M = 2.98 e M = 4.10 e Sd = 1.30 sd = 1.20 sd = 0.82 sd = 0.78 sd = 0.93 sd = 1.20 α =.39 α =.40 α =.43 α =.41 α =.40 α =.39 M = 3.31 e M = 3.58 e M = 2.70 e M = 2.55 e M = 2.57 e M = 3.69 e Sd = 1.23 sd = 1.11 sd = 0.78 sd = 0.77 sd = 0.80 sd = 1.10 α =.59 α =.51 α =.56 α =.58 α =.47 α =.51 Combined low information M = 3.60 M = 3.99 M = 2.94 M = 2.78 M = 2.74 M = 3.85 rationality (5 items) f Sd = 1.06 sd = 0.94 sd = 0.67 sd = 0.67 sd = 0.68 sd = 0.95 α =.64 α =.58 α =.65 α =.67 α =.61 b α =.61 Gut decision-making (original 2 items) M = 3.56 c M = 3.81 c M = 3.02 c M = 3.15 d M = 2.79 d M = 4.09 d sd = 1.34 sd = 1.32 sd = 0.86 sd = 1.02 sd = 1.18 sd = 1.57 α =.46 α =.40 α =.37 a Strongly Disagree, Disagree, Slightly Disagree, Neither Agree nor Disagree, Slightly Agree, Agree, or Strongly Agree seven-point response scale. b Strongly Disagree, Disagree, Neither Agree nor Disagree, Agree, to Strongly Agree five-point response scale. We believe a shorter response scale works better with a phone interview, such as was used in study 5. c Original scale. d Revised scale. e Same items in original and revised scales. fthis row shows the measurement properties of a combined five-item scale that includes all of the items from both the Fast and frugal and Heuristic-based decision strategies.

14 338 Lau, Kleinberg, and Ditonto Table 6. Subscale correlations Rational choice political decisions, we expect self-presentation to play some role in how respondents answer these questions, particularly when it comes to Rational Choice. Rational Choice describes decision-making in a way that is typically assumed to result in high-quality decisions. Thus, we expect mean levels of agreement with the Rational Choice subscale to be higher than levels of agreement with items representing the other subscales. Table 5 shows that this expectation is clearly met. Table 6 reports the subscale correlations from study 5. The pattern shown here replicates across all six studies (see table A.3 in the online appendix). Rational Choice is the most distinctive decision-making style. It has small negative correlations with each of the other subscales, the strongest of which (r =.21) is with Heuristic-Based decision-making, its opposite (deep comparable search vs shallow noncomparable search). But the remaining decision strategies all have somewhat stronger positive correlations with each other, including the Confirmatory and Fast and Frugal styles, which also are polar opposites. This makes sense given that these four decision strategies are similar in that they are not formally rational and are characterized by less information and less effort. DISCRIMINANT VALIDITY Confirmatory Fast and frugal Heuristicbased Go with gut Rational choice 1.00 Confirmatory Fast and frugal Heuristic-based Go with gut Note. Data come from study 5. All observed correlations are significantly different from 0, p <.01 or better. N varies between 1,496 and 1,521. Having established the measurement properties of the revised PolDec-5 subscales, we now demonstrate their distinctiveness, not only from previous and more general measures of decision-making styles, but also from various common demographic variables, and from familiar indicators of political interest and partisanship. On day one of study 6, subjects responded not only to the 13 items from the revised PolDec-5 scale, but also to 15 additional items that are commonly used as indicators of different decision-making styles: Rational, Impulsive, Intuitive (Scott and Bruce 1995), Maximization, and Avoiding Regret (Schwartz et al. 2002). The correlations between the PolDec-5 subscales and these more general measures of decision-making styles are shown

15 Measuring Voter Decision Strategies 339 in table 7. With an N of over 600, most of the pairwise correlations are reliably different from zero, but few of them are particularly large. Indeed, in only three instances did our new political decision-making subscales share as much as 12 percent of the variance with any of these prior measures of decision-making style (each highlighted in bold type in table 7). As would be expected, our new political measure of Rational Choice has a strong positive correlation (r =.62) with Scott and Bruce s measure of a Rational decision-making style. And the single item measuring Gut decision-making has positive correlations with both Scott and Bruce s measure of an Impulsive decision-making style (r =.399) and their measure of Intuition (r =.654). With these three very understandable exceptions, our new measures of political decision-making styles are quite distinct from existing more general measures of decision-making. To demonstrate that these new measures of political decision-making styles have different etiologies, and are not simply summaries of commonly available measures of demographic differences and more general measures of political interest and partisanship, we regressed summary measures of each Table 7. Correlations of political decision-making subscales with general measures of decision-making style (study 6) Rational a Impulsive b Intuition c Maximization d Avoiding regrets e Rational choice.620**.289** **.309** Confirmatory **.140**.086*.000 Fast and frugal.131**.233**.312**.152**.044 Heuristic-based.147**.322**.240**.133**.053 Go with gut.224**.399**.654**.098*.051 a The three Rational items were: I double-check my information sources to be sure I have the right facts before making decisions. I make decisions in a logical and systematic way. My decision-making requires careful thought. b The three Impulsive items were: I generally make snap decisions. I often make decisions on the spur of the moment. When making decisions, I do what seems natural at the moment. c The three Intuition items were: When I make decision, I tend to rely on my intuition. I generally make decisions that feel right to me. When I make a decision, I trust my inner feelings and reactions. d The three Maximization items were: When I watch TV, I channel-surf, often scanning through the available options even while attempting to watch one program. When I m in the car listening to the radio, I often check other stations to see in something better is playing, even if I m relatively satisfied with what I m listening to. No matter how satisfied I am with my job, it s only right for me to be on the lookout for better opportunities. e The three Avoiding regrets items were: Whenever I make a choice, I m curious about what would have happened if I had chosen differently. Whenever I make a choice, I try to get information about how the other alternatives turned out. When I think about how I m doing in life, I often assess opportunities I have passed up. *p <.05, **p <.001

16 340 Lau, Kleinberg, and Ditonto of the subscales on a common set of 11 background variables and indicators of political interest and partisanship in studies 3, 4, and 5 (the three studies with nationally representative samples). We present these regressions as nothing more than descriptive analyses, although we had a few strong a priori expectations for example, that individuals with strong party identification would strongly endorse Confirmatory decision-making. We also expected Gut decision-making to correlate negatively with political interest and political knowledge. The complete results of these 15 OLS regressions are presented in the online appendix, and are summarized in table 8. Barring a few minor inconsistencies, the regression results are consistent across three different large national samples. Rational Choice is particularly high among women, young people, and respondents with high levels of political interest. Confirmatory decision-making is particularly strong among men, blacks, Latinos, and as predicted among those with strong party identifications. Fast and Frugal Table 8. Predictors of political decision-making subscales across studies 3, 4, and 5 Rational choice Confirmatory Fast and frugal Heuristic based Go with gut Female Black Latino Education + + Family income Age Ideology (Conservative high) + Party ID (Republican high) Strength of party ID Political interest Political knowledge + + Average R-squared Note. Table summarizes the results of a series of OLS regressions where each of the decision- making subscales are regressed on a common series of demographic and general political attitudes and knowledge. These regressions were conducted in studies 3, 4, and 5 the three studies with nationally representative samples. A + sign means that the predictor had a positive significant effect (p <.05 or better) on the decision-making subscale heading each column, while a sign means the predictor had a significant negative effect on the particular decision-making subscale, controlling for every other predictor in the equation. The location of the + or sign within each cell of the table refers to the particular study in which the significant relationship occurred. For example, the + + entry in the first cell of the table indicates that Female was positively associated with the Rational choice decision-making scale in studies 3 and 4 but not study 5.

17 Measuring Voter Decision Strategies 341 decision-making is utilized by African Americans, older people, Republicans, and respondents with lower levels of political interest and political knowledge. Heuristic-Based decision-making is endorsed by those with strong partisan identification (itself an important political heuristic), low levels of political interest, and low levels of political knowledge. Finally, Gut decision-making is utilized more frequently by women but not Latinos, people with less education, low levels of political interest, and less political knowledge. PREDICTIVE VALIDITY These new measures of decision-making can provide a powerful tool for social scientists to explain important political behaviors. Before they can be applied, however, we have to examine their predictive validity. As noted earlier, DPTE allows us to observe actual information gathering during mock election campaigns, and to measure what decision strategy a subject employs. We hypothesize that people s self-reported decision-making style corresponds to their observed information search (e.g., people high in Rational decision-making will in fact engage in deep search across alternatives). While this prediction may seem obvious, significant research in social psychology suggests that most people have very little insight into their own mental processes and the causes of their own behavior (see, e.g., Nisbett and Wilson [1977]). It is therefore not clear what the relationship between self-reported and actual behavior is. A mismatch would render our research moot. If, however, a reasonably strong correspondence emerges between self-reported and actual decisionmaking (information search) behavior, then these scales can be used to reliably predict decision strategies. The best data to examine this hypothesis come from our first study. Recall that study 1 featured mock Democratic and Republican primary elections followed by a general election campaign. We examine two validity coefficients: depth of search and comparability of search. Both are relevant to all five decision strategies. Because these data include two primaries and one general election campaign, six validity coefficients are available for each of the five decision-making subscales. How these two indicators of search were actually operationalized varies a bit across these different elections, and is reserved for the online appendix. We regress each of the available validity indicators on the summary measures of the five PolDec subscales, plus a small set of covariates (age, political interest, and liberal-conservative identification) that, in preliminary analysis, had significant effects in one or more equations. We certainly do not expect our new measures of decision-making styles to be the sole predictors of actual behavior. Decision-making is far too complex and situation specific for that. Nor do we expect the correlations to be the same across the primary and general election campaigns, as these two elections have many obvious differences. We do, however, expect these new measures to be

18 342 Lau, Kleinberg, and Ditonto partial explanations of actual decision-making behavior, and to help predict what voters actually do. The full regression results are reserved for the online appendix. Table 9 summarizes the results, presenting both the bivariate correlations, and then, to give a better sense of the relative magnitudes of the effects, OLS regression coefficients. Overall, the data provide strong evidence ( passing 5/6 tests) for the predictive validity of the Rational Choice, Confirmatory, and Gut PolDec-5 subscales. The validity coefficients for the Heuristic-Based subscale had the wrong signs in the primary election, but were noticeably larger in the general election campaign. The data do not, however, provide evidence for the predictive validity of the Fast and Frugal decision style. POLITICAL BEHAVIOR Questionnaire measures that can reliably predict actual information search are not needed by experimentalists who can directly observe that behavior. These scales should find much broader application among survey researchers who are interested in predicting common measures of political behavior such as candidate evaluation and the vote choice, but who have no direct means of observing information search or decision strategies. We have room in this article to illustrate only one such application using affective polarization the extent to which a voter views the out-party as a distrusted and disliked outgroup compared to their in-party (e.g., Iyengar, Sood, and Lelkes 2012; Kleinberg and Lau 2016; Lau et al. 2017; Lau and Pierce 2017). We hypothesize that strategy 2 (Confirmatory) should lead to high levels of affective polarization, as this is exactly the type of motivated reasoning this decision strategy describes. In contrast, strategy 3 (Fast and Frugal) and strategy 4 (Heuristic-Based) should both be negatively related to affective polarization. These two strategies are meant to produce quick and easy, satisfactory but not necessarily extreme, decisions. We have no clear expectation about strategy 5 (Gut), but little reason to think it should lead to extreme evaluations. The possible influence of strategy 1 (Rational Choice) is the most difficult to predict a priori, as we see good reasons that its influence could be either positive or negative. On the one hand, it is hard to imagine that anyone save those who are most engaged with and interested in politics would even contemplate spending the time necessary for strategy 1 decision-making but those are exactly the people who tend to have the strongest and most extreme political opinions (Abelson 1988; Delli Carpini and Keeter 1996; Holbrook et al. 2005). On the other hand, Rational Choice demands balanced search and objective information processing, which should force decision-makers to confront and have to reconcile good and bad points about all candidates resulting in, we would expect, more moderate (and therefore less polarized) evaluations.

19 Measuring Voter Decision Strategies 343 Table 9. Political decision-making subscales predicting actual information search (study 1) Bivariate correlations Multivariate regression weights Primary election campaign Depth of search Comparability Depth of search Comparability Rational choice a (6.185).071 (.079) Confirmatory b.169**.108** **(3.639).063 (.056) Fast and frugal c (2.744).026 (.056) Heuristic-based d (3.171).114 (.064) Go with gut e.085* * (2.005).068* (.040) General election campaign Bivariate correlations Multivariate regression weights Depth of search Comparability Depth of search Comparability Rational choice.115**.134** * (8.891).225* (.105) Confirmatory.118** * (2.518).043 (.069) Fast and frugal (3.914).011 (.068) Heuristic-based * (4.486).019 (.078) Go with gut.089*.093* 5.137* (2.817).081* (.049) Combined election campaigns Bivariate correlations Multivariate regression weights Depth of search Comparability Depth of search Comparability Rational choice.100*.098* * (11.216).164* (.079) confirmatory.223***.121** **(4.020).075* (.041) Continued

20 344 Lau, Kleinberg, and Ditonto Table 9. Continued Combined election campaigns Bivariate correlations Multivariate regression weights Depth of search Comparability Depth of search Comparability Fast and frugal (6.285).018 (.044) Heuristic-based * (7.185).075 (.051) Go with gut.094*.101* 9.884* (4.543).077* (.032) Note. All independent variables have been recoded to have a one-point range, as do the various measures of comparability of search. Numbers in parentheses are standard errors associated with the reported regression weights. The depth of search dependent variables have a 91-point range in the primary election, a 79-point range in the general election, and a 175-point range overall. Because as indicated in table 2 all of our hypothesis tests are directional, one-tailed significance levels are reported. a We expect positive correlations with total information search across all candidates from both parties, and positive correlations with comparability of search across the two candidates on the ballot, in both the primary and general election campaigns, and therefore overall. b Expect positive correlations with differential search favoring in-party over out-party candidates. Expect positive correlations with comparable search toward the in-party candidates in the primary election, but negative correlations with comparable search in the general election, and overall. c Expect negative correlations with depth of search, but positive correlations with comparability of search, directed toward the two candidates on the ballot in both the primary and general election. d Expect negative correlations with depth of search, and negative correlations with comparability of search, directed toward the two candidates on the ballot in both the primary and general election. e Expect negative correlations with depth of search, and negative correlations with comparability of search, directed toward the two candidates on the ballot in both the primary and general election. *p <.05; **p <.01; ***p <.001