Finds documents with both search terms in any word order, permitting "n" words as a maximum distance between them. Best choose between 15 and 30 (e.g. NEAR(recruit, professionals, 20)).
Finds documents with the search term in word versions or composites. The asterisk * marks whether you wish them BEFORE, BEHIND, or BEFORE and BEHIND the search term (e.g. lightweight*, *lightweight, *lightweight*).
The article delves into the representation of citizens' policy interests in direct democracy elections, focusing on spatial voting and the effects of information. It argues that citizens can choose ballot proposition alternatives that reflect their ideological positions and that political information enhances this ability. The study uses data from California ballot propositions to test a low-dimensional spatial model and the impact of party cues, policy information, and spatial maps on voting decisions. The findings highlight the significant role of ideological positions and political information in advancing citizens' policy interests in direct democracy elections.
AI Generated
This summary of the content was generated with the help of AI.
Abstract
In states and localities with direct democracy, citizens can advance their policy interests without the aid of elected officials. Research documenting citizens’ lack of political knowledge raises questions about their ability to do so. We conduct three studies during real-world direct democracy elections to determine whether citizens choose alternatives (the ballot proposal or status quo) that are closest to their ideological positions and whether political information improves this outcome. Using original surveys, our first two studies estimate citizens’ ideological positions and show that citizens regularly choose alternatives that are closest to these positions. Using a survey experiment, our third study indicates that political information (party cues, policy information, and spatial maps) further improves such spatial voting. These results demonstrate citizens’ capacity to advance their policy interests in direct democracy elections and identify conditions under which political information strengthens the relationship between citizens’ policy interests and choices about ballot propositions.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The extent to which citizens’ preferences are represented in the activities of government is a central question in the study of democratic politics. Regular elections offer citizens repeated opportunities to staff local, state, and federal offices with representatives who share their policy interests. In states with direct democracy, citizens can also advance their policy interests by supporting ballot propositions that reflect these interests and opposing those that do not. Today, citizens in these states decide many important issues. In November 2022, for example, citizens in California, Kentucky, Michigan, Montana, and Vermont turned to the ballot box to determine the fate of abortion policies following the Supreme Court’s Dobbs decision.
The frequent occurrence of such opportunities, however, does not guarantee that election outcomes will reflect citizens’ policy interests. As Downs (1957) emphasizes, citizens have weak incentives to acquire information that would help them make informed political decisions. Incentives are particularly weak in direct democracy elections, where citizens might see little benefit to learning about ballot propositions if they can simply allow a tolerable status quo to continue. Moreover, many ballot propositions involve complex issues and the costs of becoming informed can be high. Previous research indicates that citizens are often confused about ballot propositions (Bowler & Donovan, 1998; Cain & Miller, 2001; Cronin, 1989; Magleby, 1984). These deficiencies raise questions about citizens’ ability to advance their policy interests in direct democracy elections.
Advertisement
We argue that citizens can choose ballot proposition alternatives that reflect their policy interests and that political information helps them do so. Like other scholars (Lupia, 1992; Romer & Rosenthal, 1978), we describe citizens’ choices in direct democracy elections with a spatial model in which the alternatives (the ballot proposal and status quo) are compared against citizens’ ideal positions. Unlike these scholars, we hypothesize that citizens’ preferences for these alternatives are generated from a low-dimensional policy space that summarizes their positions across many issues. Rather than identify an ideal policy for every issue-specific space, citizens evaluate ballot propositions based on their own ideological position along a single low-dimensional policy space. That is, citizens will compare ballot proposition alternatives and choose the one closest to their own ideological position in this space. This implies that ballot propositions will map into a low-dimensional space and that this space is the same one used to evaluate other policies considered by state government. In addition, citizens’ positions in this space will accurately predict their preferences for many ballot propositions.
We test this hypothesis by conducting three studies during real-world direct democracy elections. In the first two studies, we analyze citizens’ decisions about California ballot propositions under active consideration in November 2016 and 2020. To this end, we administered original surveys before these elections asking respondents to express their opinions about policies that divided state lawmakers, as well as propositions that would appear on the ballot. This enables us to create objective estimates of respondents’ ideological positions (ideal points) on the same scale as the ballot propositions. We assess the extent to which respondents choose alternatives in direct democracy elections that are closest to their ideological position.
In the third study, we conduct survey experiments to assess whether and when political information improves respondents’ ability to pursue their policy interests in direct democracy elections. We randomly assign respondents to receive either (1) the Democratic and Republican parties’ official positions on ballot propositions (party cues), (2) policy information about each proposition’s likely consequences, (3) spatial maps that provide “complete information” about respondents’ own ideological position and the positions of ballot propositions, or (4) no additional information. Using this design, we examine two widely disseminated types of information (party cues and policy information) and compare their effects to a control group and to a real-world analogue of the complete information conditions of spatial voting models (spatial maps).
Our first two studies demonstrate that respondents’ choices about ballot propositions can be accurately predicted from their positions along the same liberal-conservative dimension that divides state lawmakers. A one-dimensional spatial model predicts 80.28 percent of respondents’ choices on 10 ballot propositions, reducing prediction errors by nearly 20 percent relative to the commonly used minority vote benchmark.1 Our third study shows that party cues and policy information improve respondents’ ability to choose the ballot proposition alternative closest to their own ideological position, relative to the control group. These effects are similar in size and significance to those of spatial maps. Our analyses also uncover variation in the strength of individual propositions’ connections to the ideological dimension. We consider several explanations for this variation and show that information is helpful when citizens are confused about ballot propositions or have weak attitudes about them. Together, these results shed light on citizens’ capacity to advance their policy interests in direct democracy elections and identify conditions under which political information improves this outcome.
Advertisement
Spatial Voting in Direct Democracy Elections
Citizens in representative democracies are responsible for choosing elected officials and, in direct democracy settings, public policies via elections. However, if citizens cannot identify candidates who share their policy interests and/or make decisions about ballot propositions that reflect these interests, then election outcomes will inaccurately convey citizens’ preferences. A growing body of research indicates that citizens can choose candidates whose ideological positions are closest to their own (i.e., engage in spatial voting) in presidential, congressional, and local elections (Boudreau et al., 2019; Jessee, 2012; Shor & Rogowski, 2016).
Can citizens similarly advance their policy interests in direct democracy elections? Citizens’ lack of substantive knowledge about ballot propositions raises questions about their ability to make informed choices (Bowler & Donovan, 1998; Cain & Miller, 2001). Nonetheless, citizens arguably need not know many details about the choice at hand if they have access to knowledgeable and trustworthy information providers (Lupia & McCubbins, 1998). Lupia (1994) shows that uninformed citizens who knew an information shortcut made choices about ballot propositions that resembled those of well-informed citizens. Other studies find that citizens’ preferences about ballot propositions are related to their partisanship, self-reported ideology, and other characteristics (Bowler & Donovan, 1998; Branton, 2003; Gerber & Lupia, 1995; Magleby, 1984). Finally, recent scholarship indicates that citizens can articulate reasons for their choices (Colombo, 2016) and use information about ballot propositions objectively (Boudreau & MacKenzie, 2014).
While scholars have extensively studied citizens’ choices about ballot propositions, to our knowledge no existing study directly tests whether citizens choose ballot proposition alternatives that are closest to their ideological positions (i.e., engage in spatial voting) in direct democracy elections. This partly reflects the difficulty of developing comparable measures of citizens’ ideological positions and those of ballot propositions. There are also no experimental studies examining how political information affects spatial voting in direct democracy elections. Observational research on this topic relies on citizens’ observed levels of political knowledge or comparisons of uninformed citizens who know an information shortcut to well-informed citizens (Lupia, 1994; Shor & Rogowski, 2016). However, absent random assignment, respondents in these studies might differ in ways other than possessing political information that could explain differences in their choices (Arceneaux & Kolodny, 2009). Of course, many studies do manipulate political information (Bullock, 2011; Nicholson, 2011), but they rarely examine spatial voting as the outcome (see Sniderman & Stiglitz, 2012 and Boudreau et al., 2019 for exceptions).
We contribute to research on spatial voting, direct democracy, and political information in several ways. First, we apply a low-dimensional or basic space theory of spatial voting (Enelow & Hinich, 1984) to direct democracy elections. Whereas existing theoretical models characterize direct democracy elections as high-dimensional—with policy spaces specific to particular ballot propositions (Lupia, 1992; Romer & Rosenthal, 1978)—we hypothesize that most ballot propositions map into a low-dimensional space. We expect that citizens’ decisions about ballot propositions will be accurately predicted from their positions on the same liberal-conservative dimension that divides state lawmakers.
Second, we develop comparable measures of citizens’ ideological positions and those of ballot propositions. Specifically, we conduct original surveys that ask respondents to express their opinions about the same policy issues that divide state lawmakers, as well as their views about pending ballot propositions. We use this information to estimate ideal points for respondents and cut points for ballot propositions, which allows us to determine which alternative (the ballot proposal or status quo) is closest to respondents’ ideological positions. We assess the extent to which respondents’ choices are consistent with spatial voting theory.
Third, we use experiments to examine how political information affects spatial voting on ballot propositions. We assess whether two types of information (party cues and policy information) that are widely disseminated in direct democracy elections enhance spatial voting relative to respondents who receive no information. We also compare their effects to a second baseline (spatial maps) to assess whether they substitute for the “complete information” needed for perfect spatial voting.
Theory and Hypotheses
Methodological innovations that enable scholars to measure the ideology of candidates, citizens, and others have transformed the study of political representation. These methods draw on Converse (1964, p. 206), who defined a belief system as “a configuration of ideas and attitudes in which the elements are bound together by some form of constraint or functional interdependence.” Ideology captures this notion of constraint and implies that an individual’s positions across many issues can be predicted from her position on a small number of dimensions—the basic space. Enelow and Hinich (1984) expanded this notion of constraint by explaining how individuals’ positions in this low-dimensional basic space map onto the high-dimensional “action” space encompassing all political issues and government policies (Poole, 2005, pp. 1–18). As Downs (1957, p. 98) explains, individuals use ideologies to focus on the differences between alternatives, using this “short cut … to save himself the cost of being informed on a wider range of issues.”
Although Converse (1964) provides a well-articulated theory of constraint, his and other studies question whether most citizens hold stable ideological positions (Kinder & Kalmoe, 2017; Tausanovitch & Warshaw, 2018; Zaller & Feldman, 1992). Nonetheless, a large body of research argues that citizens do hold meaningful ideological positions. Citizens’ ideological positions predict their partisanship (Sniderman & Stiglitz, 2012), candidate evaluations (Carmines & Stimson, 1980), and vote choices in various settings (Jessee, 2012; Boudreau et al., 2019; Shor & Rogowski, 2016). Citizens’ issue positions are strongly related to their choices even after accounting for partisanship (Ansolabehere et al., 2008). Further, studies suggest that increasing elite polarization strengthens ideological constraint among citizens by communicating what issue positions partisans should hold (Barber & Pope, 2019). Whether derived from core values, economic circumstances, partisan or other attachments, there is ample evidence that citizens have a sense of what goes with what.
Direct democracy elections are a natural environment for spatial voting models. The universe of ballot propositions is high-dimensional, encompassing many different policies. It makes sense that citizens would rely on the same low-dimensional evaluative or ideological dimension(s) to inform their opinions about ballot propositions. While ballot propositions differ from the simple ideas and attitudes that comprise individuals’ belief systems (because they involve contests between concrete, often multi-faceted proposals for change and the status quo), the mapping from the ideological dimension(s) to alternatives in direct democracy elections is more straightforward than in candidate elections. This is because candidates are evaluated on many criteria (e.g., performance, likeability, race/ethnicity) besides their policy positions.
Previous research applying spatial models to direct democracy elections assumes the existence of issue-specific spaces where citizens have ideal policy positions. Romer and Rosenthal (1978), for example, cite packages of local school spending and expenditure proposals for new bridges as dimensions over which citizens have single-peaked preferences. Lupia (1992, p. 392) similarly imagines a “finite continuum of possible policy alternatives” over which citizens have symmetric and single-peaked utility functions. A world where citizens have well-defined preferences over each issue-specific space, while theoretically useful, is unrealistic. In the 2018 general election in California, citizens considered 11 ballot propositions; in the November 2016 election in San Francisco, citizens decided 25 local measures. Given mounting evidence that legislators’, candidates’, and citizens’ ideologies are low-dimensional, the policy space that theoretical models of direct democracy elections ought to be concerned with is the evaluative, or ideological one. The mismatch between what existing theoretical models require of citizens and what empirical studies indicate citizens know about ballot propositions helps explain many scholars’ pessimism about direct democracy elections.
We hypothesize that rather than evaluate ballot propositions based on issue-specific preferences, citizens’ opinions are generated from their positions in a single low-dimensional basic space. This distinction preserves the assumption that citizens are policy-seeking in direct democracy elections (they prefer alternatives closer to their own ideal policy) but focuses attention on ideology as key to advancing their policy interests in the issue-specific “action” spaces occupied by ballot propositions. Thus, we expect that citizens will compare ballot proposition alternatives (the ballot proposal and status quo) and choose the one closest to their own ideological position. Replacing issue-specific policy spaces with an evaluative space of one or two dimensions is more than a distinction without a difference. It implies, for example, that ballot propositions will map into a low-dimensional space and that this space is the same one used to evaluate other policies carried out by state government. Whereas other theories offer little reason to expect that preferences for local school spending and proposals for new bridges will be linked, we anticipate that measures of citizens’ positions in the evaluative space will accurately predict their preferences for many ballot propositions:
Hypothesis 1:
If citizens generate their opinions about ballot propositions from their positions in a low-dimensional basic space, then we will observe a strong relationship between citizens’ ideological positions and their decisions about ballot propositions.
A potential barrier to spatial voting in direct democracy elections is citizens’ lack of information about how their policy interests relate to their choices about ballot propositions. Previous research identifies different types of political information that might help citizens in this regard. Here, we focus on two widely available information sources, party cues and policy information, and a third, spatial maps, that provides the information needed for perfect spatial voting (the locations of the ballot proposal and status quo relative to their own ideological position).
Party cues are among the most important information sources in direct democracy elections. The Democratic and Republican parties regularly contribute to the campaigns for or against ballot propositions and advertise their positions. Because the parties are perceived as knowledgeable about political matters and have well-known ideological reputations, their endorsements can help citizens determine where their own interests lie (Lupia & McCubbins, 1998; Sniderman & Stiglitz, 2012). The two parties take opposing positions on most ballot propositions, thereby providing signals about the relative ideological positions of the ballot proposal and status quo. That is, a party’s support for (opposition to) a ballot proposition communicates to citizens that the proposed policy change (status quo) is among the set of policies preferred by party members and consistent with its ideological reputation.
Hypothesis 2:
Citizens who receive party cues are more likely to choose ballot proposition alternatives that are closest to their own ideological position than citizens who do not receive this information.
Policy information, which we conceive of as clarifying the substance and likely consequences of a ballot proposition, is often circulated by nonpartisan experts seeking to educate policymakers and citizens. In states like California, government agencies analyze proposed ballot propositions and report their findings to the public. We contend that such information can help citizens determine the direction of the proposed policy change relative to the status quo and, as such, improve their ability to relate ballot proposal and status quo alternatives to their own position along the ideological dimension. When such policy information comes from a nonpartisan expert, citizens are likely to trust its characterization of a proposition’s substance and likely consequences:
Hypothesis 3:
Citizens who receive policy information from a credible source are more likely to choose ballot proposition alternatives that are closest to their own ideological position than citizens who do not receive this information.
Spatial maps, which are based on legislators’, candidates’, and/or citizens’ responses to a set of roll calls or policy questions, offer a visual summary of these actors’ positions along the ideological dimension(s). They are increasingly used by civic organizations to educate voters about which candidates hold policy views closest to their own (Boudreau et al., 2018; Garzia et al., 2017). In direct democracy elections, spatial maps similarly convey which alternative (the ballot proposal or status quo) is closest to a citizen’s ideological position. To the extent that citizens can interpret such spatial maps, they can strengthen spatial voting in direct democracy elections:
Hypothesis 4:
Citizens who receive spatial maps depicting their own ideological position and cut points for individual ballot propositions are more likely to choose ballot proposition alternatives that are closest to their own ideological position than citizens who do not receive this information.
Our predictions about low-dimensional spatial voting and political information are neither obvious nor empirically settled. Scholars’ skepticism about citizen competence (Converse, 1964), especially in direct democracy elections (Cain & Miller, 2001; Cronin, 1989; Magleby, 1984), weighs against observing strong spatial voting on ballot propositions. Moreover, if citizens cannot connect their policy interests to ballot proposition alternatives, they might also have trouble using political information. Studies of candidate elections find that partisanship and party cues weaken spatial voting, as citizens react by choosing party-endorsed candidates over more ideologically-similar alternatives (Jessee, 2012; Boudreau, Elmendorf, and MacKenzie 2015). With respect to policy information, citizens might ignore expert advice because they cannot relate the source’s interests to their own (Calvert, 1985). Studies of policy information find that it moves Democrats’ and Republicans’ opinions in the same direction (Boudreau & MacKenzie, 2014), indicative, perhaps, of a quality rather than ideological signal. To the extent that political information triggers non-ideological reactions, it will not increase spatial voting in direct democracy elections.
Our focus on ideology as the linchpin for policy-seeking behavior in direct democracy elections also raises questions about how well particular ballot propositions will map onto the liberal-conservative dimension and whether this will condition the effects of information. Although we predict that most ballot propositions will map onto the same space used to evaluate other state government policies, the strength of this relationship likely varies. Some propositions address issues that are obviously related to the liberal-conservative dimension, while others involve esoteric policies with weak connections to this dimension. Given our hypotheses that political information will convey ideological content, we expect it to be especially helpful on propositions with weaker connections to the ideological dimension (where citizens might be confused and/or lack strong attitudes).
Study 1: Spatial Voting in Five 2016 Direct Democracy Elections
Our first study examines whether citizens choose ballot proposition alternatives that are closest to their own ideological position in real-world direct democracy elections. We begin by estimating citizens’ ideological positions and cut points for five 2016 ballot propositions contested in California. First, we scaled roll call votes in the California Assembly between 2013 and 2016. These analyses indicated that a dominant first (liberal-conservative) dimension explains a large share of assemblymembers’ votes. We selected 34 votes that ranked high in their ability to discriminate California legislators along the liberal-conservative dimension.
Next, we measured citizen ideology on the same liberal-conservative dimension that explains voting in the California Assembly. To do so, we recruited 3,040 Californians from the Survey Sampling International (SSI) panel.2 We administered our survey online using Qualtrics software from August 5 to August 11, 2016, three months before the 2016 general election. We asked respondents to express their opinions about the 34 policy proposals that successfully distinguish California legislators’ ideological positions. Table A1 in the Online Appendix (OA) summarizes these questions and respondents’ answers. Using these answers, we estimated each respondent’s position along the dominant liberal-conservative ideological dimension.
To determine the cut points of ballot propositions on the same ideological dimension, we also asked respondents to express their opinions about five citizen-initiated propositions under active consideration.3 These included 1) a referendum on California’s law that prohibits grocery stores from providing single-use plastic bags, as well as initiatives that would 2) require background checks before individuals can purchase ammunition, 3) increase the cigarette tax by $2 per pack, 4) allow inmates convicted of nonviolent crimes to receive early parole consideration, and 5) require a two-thirds vote in the state legislature to change how fees that hospitals pay to Medi-Cal (California’s health-care program for low-income patients) are used. While time/attention considerations prevented us from examining all 17 propositions on the ballot, these five represent a range of important issues and offer a suitable test of Hypothesis 1. Table 1 summarizes the propositions, as well as respondents’ opinions.
Table 1
2016 ballot propositions with respondents’ answers and classification metrics
Ballot proposition
Respondents
Item parameters
Correct
classif. (%)
PRE
Y-N-DK (%)
γj
αj
Prop. 52. Permanently extend state fee on private hospitals and require a 2/3 vote in the state legislature or a statewide vote to change it
66-12-22
− 0.785
− 0.924
84.80
.009
Prop. 56. Increase cigarette tax by $2 per pack to fund healthcare and tobacco use prevention
77-18-5
− 0.942
− 0.754
83.13
.101
Prop. 57. Allow inmates convicted of non-violent crimes to receive a parole hearing upon completing their primary sentence and let prison officials award credits toward early release for good behavior
71-20-9
− 0.933
− 0.733
80.07
.055
Prop. 63. Require background checks for ammunition purchases and prohibit ownership of large-capacity magazines
81-15-4
− 1.345
− 0.987
88.57
.260
Prop. 67. Support state law prohibiting single-use plastic bags and requiring retailers to charge 10 cents for paper bags
63-32-5
− 1.538
− 0.190
79.52
.373
Total
83.19
.198
γj and αj are policy proposal parameters in equation (1)
Proportional reduction in error (PRE) for each ballot proposition is calculated as (Minority votes − Classification errors)/Minority votes
For all ballot propositions, the aggregate proportional reduction in error is Σnj=1(Minority votesj − Classification errorsj)/Σnj=1Minority votesj
Respondents’ opinions about the ballot propositions, as well as their answers to the policy questions, enable us to identify a cut point for each proposition. In a one-dimensional spatial model, a ballot proposition’s cut point is the point equidistant between the ballot proposal and status quo. It separates citizens with ideal points closer to the ballot proposal from those with ideal points closer to the status quo. In Fig. 1, for example, the dashed line shows the cut point of a hypothetical ballot proposition (BP) that seeks to move policy leftward, relative to the status quo (SQ). Respondents with ideological positions to the left of the line are, according to the model, likely to support this proposition. Respondents with ideological positions to the right of the line, like Respondent X, are likely to oppose it. Respondents with ideological positions close to or the same as the ballot proposition’s cut point are equally likely to support or oppose it. Together, our measures of respondents’ ideological positions and the cut point for each ballot proposition enable us to assess whether respondents’ opinions about the ballot propositions accord with their policy interests.
Fig. 1
Spatial map with respondent ideal point and ballot proposition cut point
To estimate respondents’ ideological positions and the ballot proposition cut points, we scaled respondents’ answers to the 34 policy and five ballot proposition questions together using the item-response model developed by Clinton et al. (2004).4 The model assumes that each respondent i’s utility from a policy proposal j’s yea and nay outcomes (BPj and SQj) declines with its squared distance from the respondent’s ideal point, xi. The statistical model implied by this Euclidean spatial voting model is equivalent to the following two-parameter item-response model used in education testing applications (Jackman, 2001: 228–229):
where yij = 1 if y*ij > 0 and 0 otherwise. The additional assumption, εij ~ N(0, 1), implies a probit model with respondents’ ideal points, xi, and policy proposal parameters, γj and αj, as predictors to be estimated. Because the policy proposal parameters, γj and αj, are functions of the positions of the yea and nay alternatives, BPj and SQj, the probit model recovers cut points for the policy proposals rather than the exact positions of the alternatives.
While most studies of spatial voting focus on measuring candidate and citizen ideal points, we also examine the policy proposal parameters. The item difficulty parameter, αj, is related to a policy proposal’s general level of support. Holding ideology constant, higher values of αj reduce the probability that a respondent will support the proposal. The item discrimination parameter, γj, indicates how strongly a proposal distinguishes respondents along the different dimensions of the ideological space (Jackman, 2001). In a one-dimensional model, γj measures the extent to which a respondent’s ideal point, xi, translates into support for policy proposal j. Large and significant γj indicate that support for the proposal j and ideology are strongly related.
To further investigate the influence of citizens’ ideological positions on their choices about ballot propositions, we also estimate models of support for each proposition using respondents’ ideal points and partisanship as predictors. To ensure that our measure of respondents’ ideology is independent of their opinions about ballot propositions, we re-estimated respondents’ ideal points by scaling only their answers to the 34 policy questions. Our dependent variables in these models indicate whether a respondent “strongly supports,” “somewhat supports,” “somewhat opposes,” or “strongly opposes” a ballot proposition (rescaled to range from 0 [least supportive] to 1 [most supportive]). For ease of presentation, we estimate a separate OLS model for each proposition and plot first differences (changing ideology and partisanship from their 25th [relatively liberal/Democratic] to 75th [relatively conservative/Republican] percentile values). Given that the five 2016 ballot propositions we examine sought to move policy leftward, we expect to observe negative first differences.
Results
Our results provide evidence of significant spatial voting in direct democracy elections. Table 1 contains the item parameters for the five 2016 ballot propositions. Each of the discrimination parameters, γj, is significantly different from zero, which indicates that all are substantively related to the liberal-conservative ideological dimension. There are differences in the strength of this relationship, with the ammunition limits and plastic bag ban propositions having values of – 1.345 and – 1.538, respectively, and the Medi-Cal fees proposition having a value of – 0.785. The discrimination parameter, γj, is also significantly different from zero for each policy question that respondents answered (see Table A1 in the OA). This affirms that the same policy disputes that divide state lawmakers along liberal-conservative lines also divide our respondents.
With both the policy questions and ballot propositions related to the liberal-conservative dimension, we have a solid basis for expecting spatial voting in these elections. The success of the one-dimensional spatial model in predicting respondents’ ballot proposition choices underscores this point. A one-dimensional spatial model correctly classifies 83.19 percent of respondents’ choices on the five ballot propositions. Table 1 reports the proportional reduction in error (PRE) for each proposition and the aggregate proportional reduction in error (APRE) for all five propositions.5 A one-dimensional model reduces classification errors by 19.8 percent above the commonly used minority vote benchmark.
Consistent with our first hypothesis, respondents’ ideal points have large effects on their support for the five ballot propositions. Figure 2 plots first differences from our models of support. The right-hand panel, for example, indicates that changing a respondent’s ideal point from its 25th to 75th percentile value reduces support for the plastic bag ban referendum by 0.23, a significant difference. The effects of ideology on the other propositions are also significant. We observe the smallest effects for the Medi-Cal fees proposition, which Table 1 indicates is weakly-related to the liberal-conservative dimension. The effects of ideology are comparable to those of respondents’ partisanship.
Fig. 2
Effects of ideology and partisanship on support for 2016 ballot propositions
Note: Circles (triangles) are predicted first differences [with 95 percent critical intervals] of the effects of ideology (partisanship) on support for each ballot proposition generated from the models in Tables A2 and A3 of the OA
Study 2: Spatial Voting in Five 2020 Direct Democracy Elections
Our second study replicates Study 1 by estimating citizens’ ideological positions on the same scale as five new ballot propositions contested in 2020. As in Study 1, we first scaled roll call votes taken by members of the California Assembly, this time between 2017 and 2020. We selected 20 votes that ranked high in their ability to discriminate California legislators along a dominant liberal-conservative dimension. We then recruited 645 Californians from the Lucid panel and asked them to express their opinions about the 20 policy proposals, which we used to determine their position along the liberal-conservative dimension. We administered our survey online using Qualtrics software from October 2 to October 22, 2020, just before the 2020 general election.
We also asked respondents to express their opinions about five 2020 citizen-initiated ballot propositions. These included (1) a referendum on California’s law that would replace money bail with a system for pretrial release based on public safety, as well as initiatives that would (2) tax commercial properties based on current market value rather than purchase price, (3) authorize felony sentences for certain crimes defined as misdemeanors and restrict eligibility for a state parole program, (4) define app-based drivers as “independent contractors” and restrict local regulation of them, and (5) allow local governments to establish rent control on properties over 15 years old. As in 2016, time/attention considerations prevented us from examining all 12 propositions on the ballot, but these five encompass a range of important issues. Table 2 summarizes these propositions and respondents’ opinions.
Table 2
2020 ballot propositions with respondents’ answers and classification metrics
Ballot proposition
Respondents
Item parameters
Correct
classif. (%)
PRE
Y-N-DK (%)
γj
αj
Prop. 15. Tax commercial and industrial properties based on current market value rather than their purchase price
59-27-14
− 0.968
− 1.005
75.97
.184
Prop. 20. Authorize felony sentences for certain offenses and restrict eligibility for state parole program for non-violent offenders
55-32-13
− 0.365
− 0.592
68.10
.024
Prop. 21. Allow local governments to establish rent control on properties over 15 years old
60-28-12
− 1.262
− 1.153
78.69
.302
Prop. 22. Define app-based drivers as “independent contractors” and restrict local regulation of them
64-26-10
− 0.273
− 0.744
73.43
− .007
Prop. 25. Replace money bail with a system for pretrial release based on public safety and flight risk
56-29-15
− 1.411
− 1.177
80.52
.391
Total
75.33
.183
Data Analysis and Results
We use the same procedures described for Study 1 to estimate respondents’ ideological positions and cut points for the 2020 ballot propositions on the same scale. Our model also estimates item parameters for the 20 policy proposals and five propositions. Table 2 contains the item parameters for the five 2020 ballot propositions. Each of the discrimination parameters is significantly different from zero. As in Study 1, we find variation in the connections between the ballot propositions and the liberal-conservative dimension. Three propositions (split roll tax, rent control, bail reform) are strongly related to this dimension while two others (felony charges, app-based drivers) appear disconnected. All the discrimination parameters for the 20 policy proposals are significantly different from zero (see Table A4 in the OA).
The predictive validity of the spatial model offers further evidence for hypothesis 1. A one-dimensional spatial model correctly classifies 75.33 percent of respondents’ choices on the five 2020 ballot propositions. Table 2 reports the PRE for each proposition and the APRE for all five propositions. The APRE is similar to what we observe in 2016; that is, a one-dimensional model reduces classification errors by 18.3 percent above the minority vote benchmark.
As in Study 1, we estimated models of respondents’ support for each 2020 ballot proposition with ideology and partisanship as predictors. Because respondents expressed opinions about all five propositions, we were also able to pool these responses and use factor analysis to calculate an overall support score. Ansolabehere et al. (2008) show that combining multiple survey items into a scale factor or simple average reduces measurement error. We re-estimated respondents’ ideal points by scaling only their answers to the 20 policy questions and rescaled their support for the ballot propositions, separately and combined, to range from 0 (least supportive) to 1 (most supportive).
Figure 3 plots first differences for the five 2020 ballot propositions. As in Study 1, ideology is a significant predictor of support for the five propositions (separately and combined), with effects comparable to respondents’ partisanship. We observe the largest effects of ideology on respondents’ support for the three propositions that Table 2 indicates are strongly related to the liberal-conservative dimension. Ideology has modest effects on support for the app-based drivers and felony charges propositions. Nonetheless, the pattern of first differences resembles what we observe in Study 1. The left panel of Fig. 3 plots first differences for both ideology and partisanship on the five-item support score. Changing ideology from its 25th to 75th percentile value reduces support by 0.19, a large and statistically significant difference.6
Fig. 3
Effects of ideology and partisanship on support for 2020 ballot propositions
Note: Circles (triangles) are predicted first differences [with 95 percent critical intervals] of the effects of ideology (partisanship) on support for each ballot proposition generated from the models in Tables A5 and A6 of the OA
Collectively, these results from two studies conducted four years apart and examining 10 ballot propositions that vary in substance, complexity, and salience, provide strong evidence that citizens can advance their policy interests in direct democracy elections. To be sure, we find variation in the strength of the relationship between citizens’ ideological positions and their opinions about ballot propositions. Although spatial voting theory is agnostic as to why this variation occurs, one explanation we can rule out is that citizens’ decision-making reflects more than one ideological dimension. In the OA, we show that adding dimensions to our spatial model seldom improves our ability to predict citizens’ choices about the 10 ballot propositions.
Study 3: The Effects of Political Information on Spatial Voting
In light of the significant spatial voting we observed in Study 1, we conducted a follow-up survey with an embedded experiment to test our hypotheses about how political information affects this outcome. Specifically, we recruited an additional 998 Californians, none of whom participated in Study 1, from the SSI panel. We administered this survey online using Qualtrics from October 1 to October 8, 2016, one month before the 2016 general election. To place these respondents’ ideological positions on the same scale as the five ballot propositions from Study 1, we asked them 18 of the policy questions that respondents in Study 1 answered. Based on their answers, we were able to estimate their ideal points.
We then randomly assigned respondents to either a control group or one of three treatment groups. All respondents were asked to express their opinions about the five 2016 ballot propositions. In the control group, respondents receive only the brief description of each proposition that respondents in Study 1 received. For example, on the parole credits ballot proposition, control group respondents read the following:
This November, Californians will be asked to vote on a ballot measure that would allow inmates convicted of nonviolent crimes to be given parole consideration upon completion of their primary sentence. Currently, many prisoners receive both a primary sentence for a crime and “enhancements” or extra time if there are multiple victims or if they previously were in prison. This measure would allow prison officials to award credits toward early release to prisoners who demonstrate good behavior, efforts to rehabilitate themselves, or participate in prison education programs.
Respondents were then asked whether they strongly support, somewhat support, somewhat oppose, or strongly oppose the proposition, or whether they don’t know. The exact wording of the ballot proposition questions and treatments is provided in the OA.
In the “party cues” treatment group, respondents also received the Democratic and Republican parties’ official positions on each ballot proposition. On the parole credits proposition, for example, respondents were told, “The Democratic Party supports this measure. The Republican Party opposes it.” In this example, the party cues imply that the proposition would replace the status quo with a more liberal policy.
In the “policy information” treatment group, respondents received information about the likely consequences of passing each ballot proposition. This information clarifies the direction of the proposed policy change, relative to the status quo. The policy information is drawn from materials produced by California’s nonpartisan Legislative Analyst’s Office (which estimates the fiscal and other impacts of ballot propositions). For example, on the parole credits ballot proposition, respondents in this treatment group received the following information:
This initiative would help reduce significant overcrowding problems in state prisons by increasing the number of non-violent inmates eligible for parole consideration. California’s nonpartisan Legislative Analyst’s Office estimates that this initiative could save the state tens of millions of dollars each year in correctional and other costs.
In this example, the policy information indicates that the early parole measure would move policy in a more liberal direction (because it would reduce correctional costs and allow more inmates to be eligible for early release).
Respondents assigned to the “spatial map” treatment group were shown a visual depiction of their own ideological position relative to each ballot proposition’s cut point. That is, respondents learn whether they should support or oppose each measure based on their actual ideological position. To create these spatial maps, we used nine policy questions representing a range of issues that respondents in Study 3 answered to create 512 “citizen profiles,” one for every combination of yes/no answers to these questions (e.g., nine “yes,” nine “no,” “yes” to the first five and “no” to the last four questions, etc.). We obtained an estimated ideal point for each profile by scaling the 512 profiles along with the survey responses of Study 1 respondents.7 We then drew spatial maps that depict the estimated ideal point for each profile, as well as the cut point for each ballot proposition. Respondents were shown the spatial map that corresponds to their answers to the nine policy questions. Figure 4 provides an example of a spatial map that a respondent in this group might view before expressing an opinion about the parole credits ballot proposition.
Fig. 4
Example of spatial map treatment for parole credits proposition
To measure Study 3 respondents’ ideological positions on the same scale as the ballot proposition cut points, we scaled their responses to the 18 policy questions together with Study 1 respondents’ answers to the 34 policy and five ballot proposition questions. This yielded estimated ideal points for respondents in Study 3 and new cut points for the five ballot propositions. Importantly, Study 3 respondents’ opinions about the ballot propositions did not influence our estimates of their ideal points or the ballot proposition cut points. This ensures that our measure of these respondents’ ideal points and the positions of the five ballot propositions remain independent of the political information manipulated in our experiments. It also reduces the accuracy of respondents’ estimated ideal points (by ignoring their opinions about ballot propositions), thus biasing us against finding effects for information.
To assess whether respondents’ choices about ballot propositions are consistent with spatial voting theory (i.e., preferring the alternative closest to one’s ideal point), we calculated the distance between each respondent’s estimated ideal point and the cut point for each ballot proposition. Recall that the cut point is the position at which a respondent is indifferent between the ballot proposal and status quo. In a one-dimensional spatial model, each ballot proposition’s cut point is given by the ratio (see Clinton & Jackman, 2009):
For each ballot proposition, subtracting the cut point from the estimated ideal point, i.e., (xi – τj), provides a measure of how far away the cut point is from respondents’ ideal policy positions. Because γj < 0 (the position of the ballot proposal is to the left of the status quo) for all five 2016 ballot propositions, the spatial model predicts that a respondent will support (oppose) the ballot proposition when this distance is negative (positive).
To capture this intuition, our dependent variable, Vote_Spatialij, is coded as 1 when (xi – τj) < 0 and the respondent strongly or somewhat supports the proposition or when (xi – τj) > 0 and the respondent strongly or somewhat opposes the proposition, and zero otherwise. We calculated the percentage of opinions in each treatment group and the control group that are consistent with spatial voting theory. We conducted difference-of-means tests to examine whether more respondents choose the ballot proposition alternative closest to their ideal point when they receive party cues, policy information, or spatial maps, relative to the control group. We also examined how well party cues and policy information approximate the “complete information” baseline that spatial maps provide. We report the results of these analyses having pooled responses to the five proposition questions and then separately for each proposition.
Results
Our results indicate that political information significantly strengthens spatial voting in direct democracy elections. The large effects of political information are apparent in Fig. 5, which plots for each group the percentage of choices about the five ballot propositions where respondents choose the alternative closest to their ideal point. In the control group, respondents choose the alternative closest to their own ideal point 71.2 percent of the time. Respondents in the party cues treatment group do so 74.7 percent of the time.8 The difference between these groups is significant and supports our second hypothesis. The effects of party cues are indistinguishable from those of spatial maps. Indeed, in the spatial map treatment group, respondents choose the alternative closest to their ideal point 74.3 percent of the time. In this way, our results demonstrate that a widely disseminated “information shortcut” can substitute for “complete” spatial information about ballot propositions.
Fig. 5
Spatial voting by control and treatment groups
Note: Numbers are percentages of respondents who chose the alternative closest to their own ideal point (see Table A10 of the OA). *Difference with control is significant (p< 0.05, one-tailed)
Consistent with our third hypothesis, we find that policy information also strengthens spatial voting. As Fig. 5 shows, respondents in the policy information treatment group choose the ballot proposition alternative closest to their own ideal point 75.4 percent of the time. This is a significant increase in spatial voting relative to the control group. It is also comparable to the effects of party cues and spatial maps. The sizable effects of policy information confirm that respondents can connect substantive information about ballot proposals to their policy interests and testify to the value of real-world efforts to disseminate expert advice.
As expected, the extent to which political information improves spatial voting varies across the five propositions. Figure 6 plots the difference between each treatment group and the control group. All three types of information increase the percentage of respondents who choose the alternative closest to their ideal point on the plastic bag ban and ammunition limits propositions, which Table 1 suggests have the strongest relationships with the liberal-conservative dimension. For example, spatial maps increase spatial voting on the plastic bag ban referendum by 7.6 percent. Party cues (6.5 percent) and policy information (11.0 percent) have comparable effects on this proposition.
Fig. 6
Average treatment effects for 2016 ballot propositions
Note: Symbols are differences [with 90 percent confidence intervals] in percentage of respondents who chose the alternative closest to their own ideal point for all five ballot propositions and each proposition individually, generated from Table A10 of the OA
In contrast, Fig. 6 shows that political information does not significantly increase the percentage of respondents who choose the alternative closest to their ideal point on either the cigarette tax or parole credits propositions, which have weaker connections to the liberal-conservative dimension. However, both party cues and spatial maps increase spatial voting on the Medi-Cal fees proposition. This is somewhat surprising as this proposition appears disconnected from the ideological dimension. In what follows, we explore different explanations for the variation in the ballot propositions’ connections to the liberal-conservative dimension and the effects of political information.
Explaining Variation in Spatial Voting and the Effects of Political Information
We consider two potential explanations for the variation in the 2016 and 2020 ballot propositions’ connections to the liberal-conservative dimension. One possibility is that a proposition’s weak connection stems from confusion about its substance. Alternatively, a proposition might be weakly related to the ideological dimension because its popularity transcends ideological divisions. To assess these explanations, we analyzed the amount of time respondents take to express their opinions, their propensity to respond “don’t know,” and the strength of their support or opposition on each proposition. If respondents are confused about a proposition’s substance, then we should observe longer response times and more “don’t know” responses. If a proposition is overwhelmingly popular, then we should observe shorter response times, fewer “don’t knows,” and stronger attitudes. We should also observe high levels of overall support.
The top two panels of Fig. 7 plot the absolute value of each proposition’s discrimination parameter against response times and attitude strength. As Fig. 7a shows, two propositions (Medi-Cal fees, felony charges) with the smallest discrimination parameters (i.e., the weakest connections to the liberal-conservative dimension) exhibit the longest response times. These propositions also generated more “don’t know” responses (see the OA). These results are consistent with respondent confusion. A third proposition (app-based drivers) that appears disconnected from the liberal-conservative dimension was overwhelmingly popular. Respondents spent less time making their choice, were more likely to offer “strong” support or opposition, and were less likely to say “don’t know.” As Table 2 conveys, this proposition also registered the highest support among the 2020 ballot propositions we examined.
Fig. 7
Discrimination and treatment effects by ballot proposition
Note: Points in panels a and b indicate absolute value of discrimination parameters from Tables 1 and 2. Points in panel c and d indicate effects of party cues from Tables A10 and A17. See Tables A15 and A16 for average response times and measures of attitude strength for individual ballot propositions
Can political information improve respondents’ ability to relate their policy interests to their choices about ballot propositions that appear disconnected from the liberal-conservative dimension? While we did not conduct a follow-up experimental study in 2020, we utilize data from 278 additional respondents from the Lucid panel who participated in a different study, but who answered the same policy questions as respondents in Study 2 and received party cues before registering their opinions about the 2020 ballot propositions (see the OA for details).
The bottom two panels of Fig. 7 plot our estimates of the effects of party cues on spatial voting against response time and attitude strength for both 2016 and 2020. As Fig. 7c indicates, party cues tend to have stronger effects on ballot propositions characterized by longer response times. Indeed, the two propositions with the longest response times (Medi-Cal fees and felony charges) exhibit the largest increases in spatial voting. Figure 7d shows that the effects of party cues also tend to be largest on propositions where respondents’ attitudes are relatively weak. Overall, these analyses suggest that party cues tend to be particularly effective at increasing spatial voting on ballot propositions when respondents are either confused or lack strong attitudes.
Conclusion
Our results provide three new types of evidence of citizens’ ability to advance their policy interests in direct democracy elections. First, our surveys of Californians making choices about 10 ballot propositions across four years revealed significant spatial voting. We compiled detailed measures of ideology by asking citizens over four dozen policy questions that divided state lawmakers along a dominant liberal-conservative dimension. Using a one-dimensional spatial model, we found that citizens’ ideological positions accurately predict 80.28 percent of their choices. Second, our experimental analyses of citizens’ choices on these ballot propositions indicate that party cues and policy information strengthen spatial voting. Indeed, citizens who receive this information behave as if they possess the “complete” information (spatial maps) needed for spatial voting. Third, the few ballot propositions that appear disconnected from the ideological dimension are those that respondents either find confusing or overwhelmingly support. While the small number of such propositions in our study does not permit us to be conclusive, political information appears to help citizens connect their policy interests to their choices on these propositions.
The fact that we conduct our three studies in California, where Democrats predominate in the state legislature and electorate, might be viewed as a limitation. Nonetheless, with the nation’s largest, most diverse population, and the world’s fifth largest economy, direct democracy elections in California address myriad issues and are consequential in their own right. Although legislators and voters trend liberal, low signature thresholds enable groups aligned with both parties to put propositions on the ballot (see Tables A21 and A22 in the OA). Several ballot propositions we examine were sponsored by right-leaning groups or endorsed by the Republican Party. Beyond these propositions, right-leaning groups historically have used direct democracy to lock in low property tax rates, voice opposition to illegal immigration and affirmative action, and adopt sentence enhancements for repeat criminal offenders. California citizens might have distinct preferences, but they resemble citizens in other direct democracy settings in having policy interests motivating their behavior.
For scholars, our findings demonstrate the benefits of focusing on ideology as key to understanding how citizens pursue their policy interests in direct democracy settings. In doing so, our study advances empirical assessments of the quality of citizens’ decisions by facilitating individual-level measures of “improvement” in citizens’ decisions—as opposed to relying on group-level comparisons of informed and uninformed citizens. It also offers a more thorough explanation of how and when political information will be helpful. Finally, it raises questions about how citizens form such evaluative dimensions in the first place and how spatial voting in direct democracy elections compares to candidate elections.
Our findings also have implications for practical efforts to inform citizens about their choices in direct democracy elections. The improved spatial voting we observe in response to policy information highlights the salutary effects of state laws that provide for nonpartisan expert evaluation of pending ballot propositions. Nonetheless, the significant effects of political information we observe also indicate that many citizens lack information about ballot propositions near Election Day. Increasing citizens’ access to the types of information we examine (e.g., party cues), either on the ballot as in many candidate elections or via official ballot pamphlets or websites, can improve citizen competence in these settings. Indeed, the success of the spatial maps we examined suggests a heretofore unstudied tool for helping citizens bring their policy interests to bear in direct democracy elections. Such interactive spatial maps are often provided on the Internet by civic groups and public agencies for citizens seeking information about candidates. Our study demonstrates that such efforts could be extended to direct democracy elections with potentially powerful results. In Europe, voter advice applications (VAAs) help citizens apply their issue preferences to ballot propositions, albeit with different information than spatial maps (Stadelmann-Steffen et al., 2023). Future research can examine whether VAAs or other information, such as about donors to campaigns for and against ballot propositions, similarly improve spatial voting.
Finally, our results place recent efforts to restrict citizens’ use of direct democracy (by increasing filing fees, tinkering with signature-gathering rules, or requiring voter supermajorities) in a different light. Proponents of such measures frequently cite concerns over citizen competence as justification for using direct democracy sparingly. Our results suggest efforts to manipulate the process might say more about the intentions of groups seeking particular outcomes, as in Ohio where abortion opponents sought to raise the threshold for passing constitutional amendments in advance of a November 2022 pro-abortion rights measure. For practitioners with sincere concerns about citizens’ understanding of ballot propositions, our methods could be used to identify propositions that are likely to be confusing to voters. This might inform legal challenges to ballot propositions, the drafting of ballot proposition titles and descriptions, or voter education efforts to improve citizen decision-making.
Declarations
Conflict of interest
The authors have no relevant financial or non-financial interests to disclose.
Ethical Approval
The questionnaires and methodology for these studies were reviewed by the Institutional Review Board at UC Davis, which deemed the studies to be Exempt.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This benchmark model predicts that all respondents take the majority position on each proposition, thereby making classification errors equal to the number of minority votes. The relevant comparison is how much a spatial model reduces classification errors relative to this benchmark.
Our respondents resemble California’s population in many respects, though they are better educated than the general population. Our samples more closely resemble voters, the population that decides direct democracy elections (see the OA).
To minimize fatigue, respondents answered a random subset of 23 to 26 of the 34 policy questions and two to three of the five ballot proposition questions.
We used the pscl R package to estimate a one-dimensional model with uninformative priors for all model parameters with 200,000 iterations after discarding the first 10,000 and thinning by 100. The first dimension correctly classifies 79.67 percent of responses.
Bridging the profiles with respondents from Study 1 enhances the precision of the estimated ideal points. To assist respondents’ interpretation of the spatial maps, we converted the cut points and ideal points to a 1–7 scale and added “most liberal,” “in the middle,” and “most conservative” labels.
Because respondents in Study 3 answered fewer policy questions and our estimates of their ideal points are unaffected by their views on pending ballot propositions, these percentages understate the extent of spatial voting. Similarly, Vote_Spatialij takes the value 0 for “don’t know” responses. This explains why the percentages in Fig. 3 are lower than the percentages in Table 1.
Ansolabehere, S., Rodden, J., & Snyder, J. M. (2008). The strength of issues. American Political Science Review,102(2), 215–232.CrossRef
Arceneaux, K., & Kolodny, R. (2009). Educating the least informed. American Journal of Political Science,53(4), 755–770.CrossRef
Barber, M., & Pope, J. C. (2019). Does party trump ideology? American Political Science Review,113(1), 38–54.CrossRef
Boudreau, C., Elmendorf, C. S., & MacKenzie, S. A. (2018). Roadmaps to representation. Political Behavior,41, 1001–1024.CrossRef
Boudreau, C., Elmendorf, C. S., & MacKenzie, S. A. (2019). Racial or spatialvoting? American Journal of Political Science,63(1), 5–20.CrossRef
Boudreau, C., & MacKenzie, S. A. (2014). Informing the electorate? American Journal of Political Science,58(1), 48–62.CrossRef
Bowler, S., & Donovan, T. (1998). Demanding choices. University of Michigan Press.CrossRef
Branton, R. (2003). Examining individual-level voting behavior on state ballot propositions. Political Research Quarterly,56(3), 367–377.CrossRef
Bullock, J. (2011). Elite influence on public opinion in an informed electorate. American Political Science Review,105(3), 496–515.CrossRef
Cain, B. E., & Miller, K. P., et al. (2001). The populist legacy. In L. J. Sabato (Ed.), Dangerous democracy. Rowman and Littlefield.
Calvert, R. (1985). The value of biased information. Journal of Politics,47, 530–555.CrossRef
Carmines, E. G., & Stimson, J. A. (1980). The two faces of issue voting. American Political Science Review,74, 78–91.CrossRef
Clinton, J. D., & Jackman, S. (2009). To simulate or nominate. Legislative Studies Quarterly,34(4), 593–621.CrossRef
Clinton, J. D., Jackman, S., & Rivers, D. (2004). The statistical analysis of roll call data. American Political Science Review,98, 355–370.CrossRef
Colombo, C. (2016). Justifications and citizen competence in direct democracy. British Journal of Political Science,48, 787–806.CrossRef
Converse, P. E. (1964). The nature of belief systems in mass publics. In D. E. Apter (Ed.), Ideology and discontent. Free Press.
Cronin, T. E. (1989). Direct democracy. Harvard University Press.CrossRef
Downs, A. (1957). An economic theory of democracy. HarperCollins.
Enelow, J. M., & Hinich, M. J. (1984). The spatial theory of voting. Cambridge University Press.
Garzia, D., Trechsel, A. H., & De Angelis, A. (2017). Voting advice applications and electoral participation. Political Communication,34, 424–443.CrossRef
Gerber, E. R., & Lupia, A. (1995). Campaign competition and policy responsiveness in direct legislation elections. Political Behavior,17(3), 287–306.CrossRef
Jackman, S. (2001). Multidimensional analysis of roll call data via bayesian simulation. Political Analysis,9(3), 227–241.CrossRef
Jessee, S. A. (2012). Ideology and spatial voting in american elections. Cambridge University Press.CrossRef
Kinder, D. R., & Kalmoe, N. P. (2017). Neither liberal nor conservative. University of Chicago Press.CrossRef
Lupia, A. (1992). Busy voters, agenda control, and the power of information. American Political Science Review,86(2), 390–403.CrossRef
Lupia, A. (1994). Shortcuts versus encyclopedias. American Political Science Review,88, 63–76.CrossRef
Lupia, A., & McCubbins, M. D. (1998). The democratic dilemma. Cambridge University Press.
Magleby, D. B. (1984). Direct legislation. Johns Hopkins University Press.CrossRef
Nicholson, S. P. (2011). Dominating cues and the limits of elite influence. Journal of Politics,73(4), 1165–1177.CrossRef
Poole, K. T. (2005). Spatial models of parliamentary voting. Cambridge University Press.CrossRef
Romer, T., & Rosenthal, H. (1978). Political resource allocation, controlled agendas, and the status quo. Public Choice,33, 27–44.CrossRef
Sniderman, P. M., & Stiglitz, E. H. (2012). The reputational premium. Princeton University Press.CrossRef
Stadelmann-Steffen, I., Rajski, H., & Ruprecht, S. (2023). The role of vote advice application in direct-democratic opinion formation. Acta Politica,58, 792–818.CrossRef
Tausanovitch, C., & Warshaw, C. (2018). Does the ideological proximity between candidates and voters affect voting in U.S. house elections. Political Behavior,40, 223–245.CrossRef
Zaller, J., & Feldman, S. (1992). A simple theory of the survey response. American Journal of Political Science,36(3), 579–616.CrossRef
