More Candidates and Fewer Voters: How an Abundance of Choice Demobilizes the Electorate

Abstract

Voters, like consumers, experience choice overload when presented with too many similar options. The literature on consumer behavior finds overwhelmed consumers often opt out of making a decision altogether. Yet, how large candidate fields affect American voter behavior remains unexplored. I use an aggregate-level analysis leveraging exogenous changes to Louisiana’s electoral institutions and an individual-level analysis to investigate the effect of varying quantities of choice on ballot rolloff in U.S. House elections. Ballot rolloff is the phenomenon where voters turn out to vote, yet abstain in a given election. I find that an electoral institution that incentivizes candidate entry causes 5.7–7.1 percent more ballot rolloff, and also that individuals are more likely to rolloff when there are more candidates running in their House election. Louisiana is simply the best case to explore a phenomenon that is likely occurring in party primaries throughout the U.S.—choice overload at the ballot box.

Keywords

candidate choice choice overload ballot rolloff jungle primary voting elections

Voters experience choice overload at the ballot box when presented with too many similar candidates. Like shoppers choosing between gourmet jams (Iyengar & Lepper, 2000), voters become overwhelmed by choosing between more than two candidates because there is no obvious coordination strategy and there might not even be a preferred candidate. Consumers either opt out of making a decision entirely or make a decision and experience buyer’s remorse when facing a large choice set (Haynes, 2009; Jilke, Van Ryzin & Van de Walle, 2016). This phenomenon of becoming overwhelmed by too much choice is referred to as choice overload. Barry Schwartz popularized the phrase in his 2004 book, The Paradox of Choice: Why More Is Less. In it, he argues that an overabundance of options can lead to anxiety, decision fatigue, and dissatisfaction, challenging the notion that more choices inherently lead to greater happiness. Similarly, a large number of candidates in an election should cause voters to abstain. In this paper, I look at how the rate of voters opting out of making a decision increases as the burden of choice increases.

There are only two major-party candidates in most U.S. general elections, and U.S. primary elections—where the number of candidates varies—are low-salience competitions with much lower turnout. These two issues make it difficult to estimate the effect of crowded candidate fields on ballot rolloff. Additionally, there are no obvious top-of-the-ballot elections in primary elections to serve as a comparison for measuring ballot rolloff. I overcome these challenges by leveraging the state of Louisiana’s abandonment and readoption of the jungle primary to estimate the effect on ballot rolloff. Supreme Court cases led to Louisiana’s abandonment in 2008 and readoption in 2012 of the jungle primary for state and federal elections.

While primary rules differ across states, all states except Louisiana have a general election between either the party nominees, or the top two vote-getters from an open primary. Louisiana holds a jungle primary for its state and federal elections (with the exception of the presidential election) on November’s federal Election Day. Alaska, California, Washington, and Nebraska also use variants of the jungle primary, but these elections are not held on Election Day. Majority-vote primaries, as well as jungle primaries, incentivize candidate entry since candidates do not need a first-place finish to advance (Wright, 1989). This means that on presidential election years Louisiana has federal and state elections with many same-party candidates on the same ballot as the presidential race between only two major-party candidates. Thus, votes cast in the presidential election serve as a reliable baseline to votes cast in the crowded jungle primary House elections. I look at House elections since House elections are the most salient elections occurring each cycle. Louisiana is simply the best case to explore a phenomenon that is likely occurring in party primaries throughout the U.S.—candidate overload.

In this paper, I systematically explore the effect of large candidate fields on ballot rolloff. I argue that ballot rolloff will increase when the number of candidates in an election increases. To test my hypothesis, I use both aggregate- and individual-level analyses. For the aggregate analysis, I use a difference-in-differences design using the states of Louisiana and its similar neighbor, Mississippi, as the control (for robustness, I estimate a generalized synthetic control model where control counties are constructed using counties from Arkansas, Mississippi, and Texas). By leveraging a change in Louisiana’s electoral rules brought about by two Supreme Court decisions, the first research design is uniquely positioned to causally identify the effect of an institutional change on county-level ballot rolloff. Then, using Louisiana respondents to the Cooperative Election Study, I investigate whether voters in congressional districts with more candidates are more likely to abstain in their House election than voters with less crowded House elections. Both analyses lend support to my argument. Use of the so-called jungle primary—a majority-vote, nonpartisan blanket primary held on Election Day—led to a 5.7–7.1 percent increase in ballot rolloff at the aggregate level, and, on average, the addition of one more candidate in an election causes a one percent increase in the likelihood an individual rolls off in a given election. Thus, changes in the number of candidates affect voters’ decisions to vote in a given election after already deciding to turn out.

Although these results may be specific to the jungle primary in Louisiana, the implications should not be. Primary elections in the United States often feature more than two candidates from the same party—obscuring the usefulness of party labels and placing an informational burden on voters. Recent changes and reforms—such as ranked-choice voting and top-two primaries—to state and local electoral rules are spurring the growth of large candidate fields. Since 2000, Alaska, California, Maine, Nebraska, and Washington—along with many local or municipal governments—have adopted ranked-choice or top-two primaries.¹ There is evidence that ranked-choice and top-two primaries have normatively desirable effects on polarization and representation: these electoral systems can result in less ideologically extreme candidates (Crosson, 2021), or greater descriptive representation for marginalized groups (Terrell, Lamendola & Reilly, 2021). However, despite the benefits and increasing popularity of these reforms, they both affect candidate entry in a way that could be detrimental to voters.

John, Smith and Zack (2018) find cities that have switched to ranked-choice voting see a 24% increase in the average number of candidates. Similarly, top-two primaries incentivize the entry of challengers from the incumbent’s party (Sparks, 2020). The previous literature explores the effects of democratizing party primaries on the types of candidates that run and get elected, but it ignores the effect of large candidate fields on voter behavior. How the structure of our elections alleviates or exacerbates these burdens and the subsequent completion of ballots is less explored. My findings suggest there is a trade-off between choice in elections and voter participation. Knowledge of such a trade-off is especially pertinent as many states are adopting top-two primaries and ranked choice voting.

Choice Overload

It is well established in political science that the number of candidates (or parties) in an election is influenced by the design of the electoral institution (Duverger, 1954). Majority-vote elections lead to more candidates, because needing less than a plurality of the vote to make it to the runoff is much less costly than needing a plurality of the vote in the first round. Wright and Riker (1989) formally model the efficiency of plurality- and majority-vote elections at selecting Condorcet winners. They find that the most preferred candidate is less likely to win a majority-vote election than a plurality-vote election, because majority-vote elections feature more than two candidates on average. While Duverger and Wright and Riker investigate the effect of these electoral institutions on candidate entry or candidate selection, research ignores the effect on voter behavior. I contribute to this literature by showing that an electoral institution that positively incentivizes candidate entry negatively affects voters—the adoption of the jungle primary (in Louisiana) causes an increase in ballot rolloff.

In elections, voters assume the role of consumer—calculating first whether or not to make a purchase (vote) and, if purchasing (voting), which product (candidate) to purchase (vote for). The greater the informational burden placed on the voter, the more likely it should be that the voter abstains. Like shoppers choosing between gourmet jams (Iyengar & Lepper, 2000), voters become overwhelmed by choosing between more than two candidates. As I explain below, candidate overload occurs with a relatively small choice set because (1) voters struggle to distinguish between many similar candidates, and (2) coordinating with other voters becomes less clear as the number of candidates increases.

Voter Malaise

Informational burdens increase as the number of options in a choice set of limited range increases (Chernev, Böckenholt & Goodman, 2015). When these informational burdens increase, consumers become more reliant on cognitively cheap heuristics like brand-name recognition (Iyengar & Kamenica, 2010). The choice between foods at a grocery store is not as burdensome as the choice between many variations of one product. Because of the two-party system, expanded choice in the American political context mostly resembles the latter scenario—choosing between many variations of one product. An increase in candidates is more likely to be seen in Democratic or Republican party primaries. When consumers face a large choice set they either opt out of making a decision entirely or make a decision and experience buyer’s remorse (Haynes, 2009; Jilke, Van Ryzin & Van de Walle, 2016). Similarly, I expect an increase in the number of candidates to cause abstention among voters in the form of ballot rolloff.

With increasingly ideologically-homogenous parties (Webster & Abramowitz, 2017), a greater number of same-party candidates means less ideological distance and more issue overlap between each candidate. Voters are less capable of distinguishing between candidates as the number of candidates and ideological overlap increases (Kroh, 2009). Selecting between multiple candidates with overlapping issue positions requires additional thought or calculation from voters. This additional calculation is an informational burden or cost incurred by all voters. Candidate overload should be more pronounced in less salient elections where voters lack candidate-specific information. In House elections, for example, constituents struggle to recognize the names of challengers (Miller, 1990).

A Lack of Consensus

In addition to the difficulty of distinguishing between same-party candidates with issue overlap, voters must attempt to identify which one of their preferred candidates has the best chance of winning or making it to the second round. Voters are more likely to abstain or cast “wasted” votes—votes cast for a candidate with no chance of winning or advancing to the next round—when they feel less certain they can predict the decisions of other voters (Hall & Snyder Jr, 2015; Grosser & Schram, 2006). The jungle primary exacerbates coordination failures which can lead to two opposing-party candidates advancing to the second round (Alvarez & Sinclair, 2015). Coordination failures occur when votes are “wasted” on candidates without a real chance of winning. Voters should be less confident that they are casting a meaningful vote and subsequently more likely to abstain in jungle primaries featuring many same-party candidates.

Vote Choice, Abstention, and Ballot Rolloff

Despite the wide range of work done on consumer behavior and choice (Chernev, Böckenholt & Goodman, 2015), previous research has not investigated the effect of crowded candidate fields on voter behavior when party cues are constant across candidates. An emerging literature in political science investigates the paradox of choice in the context of Latin American elections, where there is an abundance of candidates and political parties. For example, recent research finds large fields of candidates in Brazilian elections make voters more dependent on cues such as race (Aguilar et al., 2015) and ballot position (Cunow et al., 2021). Additionally, Cohen (2018) finds that the effect of the number of candidates on abstention is moderated by the number of parties. An increase in the number of candidates causes more abstention when the number of parties is held constant. U.S. primary elections serve as a unique case where multiple candidates with the same party label run in the same election—by definition eliminating the usefulness of party cues. I expect voters, in the absence of differentiating party cues, will experience candidate overload as the number of same-party candidates exceeds one.

We understand from the literature that ballot rolloff is a product of informational demands placed upon or taken off voters. For example, Herrnson, Taylor and Curry (2015) find that, where multi-member districts are used in America, voters are more likely to abstain in their multi-member district elections. Multi-member district elections are much more cognitively demanding than their single-member counterparts. Similarly, the inclusion of complex language or ballot propositions causes an increase in rolloff (Bowler, Donovan & Happ, 1992; Reilly & Richey, 2011). Informational burdens do not need to be insurmountable to significantly affect rates of ballot rolloff. Regardless of the complexity, rolloff is more likely in elections near the bottom of the ballot. In contrast, technologies that make voting easier decrease ballot rolloff. The introduction of electronic voting machines led to an immediate decline in rolloff (Nichols & Strizek, 1995). While the ballot rolloff literature has widely explored the effects of various informational burdens (Damore, Waters & Bowler, 2012; Darcy & Schneider, 1989; Wattenberg, McAllister & Salvanto, 2000), previous work does not consider how variation in the burden of choice affects ballot rolloff. I contribute to the literature by showing an increase in the number of candidates causes an increase in rolloff.

Why Louisiana?

Different primary rules may lead to crowded candidate fields in the primaries, but general elections are effectively limited to the two major-party candidates and one or two third party candidates. The jungle primary in Louisiana, which occurs on federal Election Day, is both a primary and a general election rolled into one. If no candidate gets a majority of the vote, then the top two vote-getters move on to the runoff 32 days later in December. If a candidate receives a majority of the vote, then there is no runoff election.² Louisiana’s jungle primary is the only electoral system in the U.S. that allows for multiple candidates from the same party to compete on Election Day.

The state of Louisiana and its election of House and Senate candidates provides an interesting case to investigate how electoral changes can affect ballot rolloff. Since the 1970s, Louisiana had the majority-vote primary (known as the jungle primary) in October concurrently paired with a plurality vote general election. However, in 1997, the Supreme Court ruled in Foster v. Love that the jungle primary was unconstitutional since the winner could be selected before the federal Election Day if any one candidate received a majority of the vote in the first round. Louisiana adhered to the ruling by moving the jungle primary from October to November’s federal Election Day. Then, in California Democratic Party v. Jones, the Supreme Court ruled that states could not control party nomination processes—affecting all states with blanket primary systems. In compliance with the United States Justice Department, the Louisiana legislature adopted the closed primary system for the 2008 and 2010 federal elections. Following a 2008 Supreme Court ruling upholding Washington’s electoral system (Washington State Grange v. Washington State Republican Party) and effectively reversing their prior ruling, Louisiana readopted the jungle primary on Election Day.

Does the jungle primary incentivize candidate entry? The 2008 election is the only congressional election, with a concurrent presidential election, in which Louisiana does not use the jungle primary. Instead, only party nominees compete in the election on Election Day. Unsurprisingly, when the jungle primary is in use there are, on average, two more candidates in each congressional election. The wide variation in the number of candidates in each congressional election justifies my mixed-methods approach to identifying the effect of large candidate fields on ballot rolloff.

In my first set of analyses, I use generalized difference-in-differences and synthetic control models. I compare ballot rolloff in Louisiana to ballot rolloff in Mississippi and ballot rolloff in a synthetic state before and after Louisiana switched back to the jungle primary. Mississippi serves as the control in the difference-in-differences model for two reasons: first, Mississippi is demographically and economically similar to Louisiana—both of which affect voter turnout—and, second, Mississippi neighbors Louisiana, thus, it is geographically proximate and similar. Because there’s only one presidential election (2008) in which Mississippi and Louisiana have similar electoral designs, I cannot be certain Mississippi satisfies the parallel trends assumption necessary to use a difference-in-differences design. I include a synthetic control model to alleviate these concerns (Abadie, Diamond & Hainmueller, 2010, 2015). Figure 1 shows that the differences in ballot rolloff between Louisiana and Mississippi, and Louisiana and the synthetic control are smallest in 2008. Ballot rolloff in the synthetic control is very similar to ballot rolloff in Mississippi, although it is slightly less volatile across elections in the synthetic control.

Figure 1.

Average level of ballot rolloff, 2004–2020.

Expectations

For the aggregate-level study, I hypothesize that the use of the jungle primary causes an increase in ballot rolloff.

Hypothesis 1

Use of the jungle primary in Louisiana House elections will cause an increase in ballot rolloff.

In my second analysis, I see whether voters in Louisiana congressional districts with more candidates are more likely to rolloff. This second analysis is not as strong of a causal test, but it provides evidence consistent with my general theory about choice and abstention. For the individual-level study, I hypothesize that voters are more likely to rolloff at the ballot box when there are more candidates in their House election.

Hypothesis 2

Louisiana voters will be more likely to rolloff when there are more House candidates in their congressional district.

Aggregate-Level Analysis

I look at U.S. House of Representatives elections in Louisiana, Mississippi, and a synthetic control state to estimate the effect of Louisiana’s jungle primary on parish-level ballot rolloff. To test my theory, I use a generalized difference-in-differences model with cluster-robust standard errors at the county level,

Y_{i t} = β_{1} + β_{2} (T_{t} * D_{i}) + β_{3} X_{i t} + λ_{1} Z_{i} + ω_{1} Q_{t} + ϵ_{i t} .

where Y_it is ballot rolloff, D_i is the treatment indicator, T_t is the time indicator, X_it are the controls, and Z_i and Q_t are the county and election fixed effects. This model is estimated on two different sets of controls: counties in Mississippi, and synthetic counties. I use generalized synthetic control to estimate a synthetic control model on multiple treated units. Functionally, generalized synthetic control is a difference-in-differences model where the control units are synthetic controls (Xu, 2017). The synthetic counties are constructed using counties in states that border Louisiana: Texas, Arkansas, and Mississippi. Synthetic counties are weighted on pretreatment (2008) levels of poverty, unemployment, median household income, and racial and gender composition.³

Following the generalized difference-in-differences framework, my explanatory variable of interest is an interaction between two dummy variables indicating the treatment condition and the period in which the jungle primary is in effect, (T_t∗D_i). The treatment dummy variable, D_i, takes on the value of 1 for Louisiana parishes and 0 for Mississippi or synthetic counties. The time dummy variable, T_t, takes the value of 0 for the 2008 election and 1 for the 2004, 2012, 2016 and 2020 elections. The constitutive terms from this interaction are dropped from the models since two-way fixed effects are included. The interaction between these two dummy variables captures the average treatment effect on the treated—the effect of Louisiana’s jungle primary on ballot rolloff in Louisiana parishes. I include controls, X_it, in my generalized difference-in-differences model, but not the generalized synthetic control model, to reduce the error variance in the regression and to obtain more precise estimates.

Ballot rolloff is calculated as one minus the proportion of parish/county votes in the House election to votes in the presidential election.

rolloff = 1 - \frac{votes cast for House}{votes cast for president}

The proportion is subtracted from one so the measure indicates the proportion of presidential election voters that did not vote in their respective House election. Higher values of ballot rolloff indicate a larger proportion of people voted in the presidential election than the House election. Whereas, smaller values indicate that the gap between presidential and House votes is narrow. The parish/county vote totals are from Louisiana and Mississippi’s secretaries of state, as well as Dave Leip’s Atlas of U.S. Presidential Elections and the MIT Election Data and Science Lab (for Arkansas and Texas counties). I use parish and county vote so the analysis is not affected by redistricting. There are 64 parishes in Louisiana, 82 counties in Mississippi, and 16 synthetic counties.⁴ To measure county-level ballot rolloff there must be a concurrent House and presidential election. Thus, the data are limited to the five presidential elections from 2004 to 2020.⁵ Additionally, House elections with only one candidate are dropped from the analysis because there are no votes cast in single-candidate races in Louisiana, Mississippi, or Texas. As a result, there are 655 observations in the generalized difference-in-differences model and 160 observations in the synthetic control model.

I include four controls in the generalized difference-in-differences model. First, I control for whether there is a major-party challenge in the House election.⁶ Ballot rolloff should be higher when there is only one viable option presented to voters, since voters are more likely to rolloff when an election is not contested by the opposing major party (Bonneau & Zaleski, 2021; Fisk, 2020; Patterson Jr., 2020; Streb, Frederick & LaFrance, 2009). Second, I control for whether there is a concurrent Senate election.⁷ The salience of an election on the ballot affects the rate of ballot rolloff (Milton, 1983). If the presence of a concurrent Senate election attracts more voters, then these additional voters should be more likely to rolloff in the House election. Third, I control for the competitiveness of the presidential election in the House district.⁸ Voters are less likely to cast a vote in noncompetitive elections (Hall, 2007). Fourth, I control for whether an incumbent is running for reelection.⁹ The presence of a partisan incumbent running for reelection should cause more ballot rolloff as incumbency also serves as a signal of election competitiveness.

Findings

The difference between ballot rolloff in Louisiana and Mississippi when Louisiana uses the jungle primary is much larger than the difference between ballot rolloff in the two states when Louisiana temporarily abandons the jungle primary. Taking the averages for both states in both time periods, the difference in means is 7.92% when the jungle primary is in use, and 1.67% when the jungle primary is abandoned. Similarly, the difference in ballot rolloff between Louisiana parishes and the synthetic control counties is 6.30 when the jungle primary is in use, and 0.58 when the jungle primary is abandoned. Thus, with no controls or fixed effects, the difference in differences is between 5.72 and 6.26%—ballot rolloff is approximately six percent higher in Louisiana when the jungle primary is in use.

The coefficients from the generalized difference-in-differences and generalized synthetic control models are presented in Figure 2. The coefficient for the difference-in-differences estimator, “Jungle Primary,” is statistically significant (p < .00) and substantively large in both models—providing support for Hypothesis 1. There is a 5.7 to 7.1% increase in ballot rolloff in Louisiana when Louisiana uses the jungle primary. The effect is larger in the differences-in-differences model than in the synthetic control model. On average, Louisiana House elections feature two more candidates when the jungle primary is in use (4.7 candidates vs. 2.6 candidates). Therefore, the entry of one additional candidate causes a 2.8 to 3.5% increase in ballot rolloff.

Figure 2.

The effect of the jungle primary on ballot rolloff. The full regression results from these two models are included in the first and third columns in Table A1 of the Appendix.

There are no coefficients on the treatment or time dummy variables since they are collinear with the fixed effects for Louisiana parishes and election years. Three of the four controls have significant effects in the direction I expected. When there is only one party in a House election, the parishes/counties within that district have 5.3% more ballot rolloff. Concurrent Senate elections cause ballot rolloff to increase by 0.9%, and an incumbent running for reelection causes an increase of 3.1%. Interestingly, the competitiveness of the district in the presidential election has no effect on ballot rolloff. Perhaps incumbency better serves as a proxy for the current election’s competitiveness. Compared to the other coefficients, the magnitude of the effect of the jungle primary on ballot rolloff is the largest. The jungle primary causes more ballot rolloff than elections with candidates from only one of the major parties. The full regression results from these two models are included in the first and third columns, and a placebo test using synthetic control and Mississippi counties is included in the fourth column of Table A1 of the Appendix. The placebo test provides further evidence that the parallel trends assumption is satisfied in the synthetic control model.

A change in electoral rules to the jungle primary, which leads to a greater number of candidates, causes an increase in ballot rolloff. However, by using the electoral change and a difference-in-differences design, I do not directly test whether the number candidates in a House election causes greater rates of ballot rolloff. In the following section, I use survey data to show that the number of House candidates predicts the probability an individual rolls off.

Individual-Level Analysis

For the individual-level analysis, I use Louisiana respondents in the Cooperative Election Study (CES) time series to investigate how individuals are affected by the number of candidates in their respective House election. Consistent with the previous analysis, the dependent variable is ballot rolloff. Ballot rolloff is coded 1 if the respondent reports that they voted in the presidential election but not their district’s House election, and 0 otherwise.¹⁰ I estimate a logistic regression with standard errors clustered at the congressional district since the dependent variable is dichotomous and respondents are nested in their congressional district.¹¹ The sample is limited to the 1043 validated self-reported voters.

The explanatory variable of interest in this model is the number candidates in a respondent’s congressional district. The number of candidates in each election are taken from the Louisiana secretary of state’s election report. I include three of the same district-level dummy controls from the aggregate-level model, and slightly change one to make it more applicable to the individual level. The three that are the same are the two dummy variables indicating whether there is only one major party running in the House election and whether there is a concurrent Senate election, and the continuous measure of district competitiveness. I do control for incumbency, but, with individual-level data, I only control for whether the incumbent is from the respondent’s party or the opposing party. This variable is coded 1 if there is an incumbent from the respondent’s party, −1 if there is an incumbent from the opposing party running for office, and 0 otherwise. I include election year fixed effects since there could be election-specific factors influencing an individual’s propensity to rolloff. For example, more salient or competitive presidential elections might turn out voters that are less likely to vote in down ballot elections regardless of the number of candidates. I cannot include unit fixed effects since the respondents vary in each wave of the survey. Since I cannot include time-invariant fixed effects, I include a series of individual-level controls that could affect the likelihood a respondent rolls off. These controls include respondent age, party identification, education, employment status, marital status, race, and gender.¹²

Findings

Table 1 shows the average marginal effects from two logistic regressions. The marginal effects in the first column are from the model that only includes election-specifc controls, and the marginal effects in the second column are from the model that includes both election-specific and individual-level controls. The full regression results from these models are presented in Table A2 of the Appendix. Supporting my hypothesis, the number of candidates has a positive and statistically significant effect on ballot rolloff under both specifications. I find the presence of one additional candidate in an election causes a 1.0% increase in ballot rolloff. Looking at the controls with significant effects, I find the absence of a major-party challenger causes an 10.0% increase, and the presence of a same-party incumbent causes a 1.9% decrease. The controls for a concurrent Senate election and district competitiveness have a null effect on rolloff. Interestingly, none of the individual-level controls have a significant effect on ballot rolloff.

Table 1.

The Average Marginal Effect of a Unit Increase in Each Explanatory Variable.

	(1)	(2)
	W/o Ind.-Level	W/Ind.-Level
Number of candidates	0.010^∗	0.010^∗
Number of candidates	(0.004)	(0.004)
Election-Specific Controls
Single-party	0.100^∗∗∗	0.099^∗∗∗
Single-party	(0.024)	(0.023)
Senate	0.004	−0.002
Senate	(0.037)	(0.034)
Competitive	−0.000	−0.000
Competitive	(0.001)	(0.001)
Incumbent	−0.026^∗∗∗	−0.019^∗∗
Incumbent	(0.005)	(0.007)
Individual-Level Controls
Age		0.000
Age		(0.000)
Republican		−0.020
Republican		(0.019)
Democratic		0.010
Democratic		(0.015)
Education		−0.006
Education		(0.004)
Employment status		0.018
Employment status		(0.015)
Marital status		−0.007
Marital status		(0.015)
Race		−0.012
Race		(0.017)
Gender		0.013
Gender		(0.010)
N	1043	1043

Standard errors in parentheses.

^∗p < .05, ^∗∗p < .01, ∗∗∗p < .001.

The results in Table 1 show that, on average, each additional candidate causes an increase in ballot rolloff, but the jungle primary often includes more than just one additional candidate. In Figure 3, I present the predicted probability of rolling off across the range of candidate field sizes. Underlaying the plot of predicted probabilities, the histogram depicts the percent of respondents in elections of each size. Approximately half of the respondents voted in House elections with two or three candidates, while the other half voted in elections with four or more candidates. When there are twelve candidates, a respondent is more than six times as likely to roll off than when there are two candidates—the likelihood of rolling off increases from 2.9% to 19.1%. But there is only one election in 2016 with 12 candidates, and such large candidate fields are rare. However, ballot rolloff becomes significantly more likely as the number increases by only three candidates. As a respondent moves from an election with two candidates to five they become approximately 2.7% (p = .016) more likely to roll off. These findings provide support for my theory that choice becomes burdensome when there are more than two candidates in an election, and that an individual’s likelihood of rolling off is affected by the number of candidates.

Figure 3.

The probability of rolling off and the number of candidates.

Limitations

There are several limitations to my analyses that must be acknowledged. First, the study does not fully explore the linkages between consumer psychology, particularly Schwartz’s (2004) theory of choice overload, and voter behavior. Schwartz’s work suggests that an overabundance of choices can lead to anxiety and decision paralysis. We suspect these forces are at play in crowded elections, but the dependent variable in these analyses is ballot rolloff. Therefore, these analyses cannot reveal whether or not an increase in candidates causes more anxiety or buyer’s remorse.

Second, Louisiana is a unique state, and its electoral system within this context may burden voters in a novel way. For example, Louisiana has a parish-based government system as opposed to the county-based system used in other states. The administrative differences between parishes and counties are non-existent (Benton, 2003), so the presence of parish-based government should not impose any greater level of voter burden. Louisiana’s reputation for political corruption could also play a role in discouraging voter participation. Corruption can lead to a lack of trust in the electoral process (Richey, 2010), which may result in lower rates of political engagement (Uslaner & Brown, 2005). Yet, Mississippi—which I compare Louisiana to—is identified as the most corrupt state (Cordis & Milyo, 2016). Regardless of Louisiana’s unique context, the difference-in-differences design employed in this study should help alleviate concerns. By comparing ballot rolloff rates in Louisiana to ballot rolloff rates in both Mississippi and a synthetic control before and after the adoption of the jungle primary, the study controls for time-invariant factors that could influence voter behavior. If factors such as the parish-based government structure and political corruption affect Louisiana, they would likely have consistent effects across all election years examined (2004, 2008, 2012, 2016, and 2020). This design ensures that the observed differences in ballot rolloff can be confidently attributed to the changes in the primary system rather than to other unobserved variables.

Third, the potential staying power of vote-by-mail may prove these warnings of choice overload in politics to be contextually dependent on in-person voting. Recalling Figure 1, Louisiana witnessed a sharp decline in ballot rolloff in 2020 despite continuing to use the jungle primary. The use of vote-by-mail during the Coronavirus pandemic potentially explains this dip by providing informationally burdened voters with more time. Previous research finds vote-by-mail led to greater levels of ballot completion in Colorado (Menger, Stein & Vonnahme, 2018). One of the conditions for experiencing choice overload is a limited amount of time, that is, shopping in a grocery store (Chernev, Böckenholt & Goodman, 2015). This would suggest there are interactions between different electoral institutions and methods of voting. If states are to adopt institutions that increase the amount of choice facing voters, then they should be cognizant of the predominant voting strategies and methods.

Conclusion

Electoral rules matter, and rational politicians respond to rules that incentivize candidate entry (Wright & Riker, 1989). Recently adopted reforms such as ranked-choice and top-two primaries democratize the primary process and lower the cost of entry for candidates. When the cost of entry is lower, more candidates will enter an election. Normatively, democratizing the process may seem like a good thing; however, democratizing the primaries may come at a cost to political participation as voters struggle to navigate large candidate fields. In this paper, I argue that electoral rules that incentivize candidate entry depress voter participation. My analyses find that the entry of an additional candidate causes a 2.8–3.5% increase in aggregate-level ballot rollof or makes an individual one percent more likely to rolloff. While these findings may be specific to Louisiana, the implications are not. Larger candidate fields in primaries are becoming increasingly more common as the cost of candidate entry cheapens.

In 2020, Alaska voted to adopt a top-four non-partisan primary election and ranked-choice general election. Less than two years after the adoption of the new reform, the House special election taking place on June 11, 2022 features 48 candidates who will all compete on the same ballot. One of the 48 candidates, John Coghill, noted that for voters “the overload is going to be very difficult to navigate through. It’s not impossible — it’s difficult”.¹³ Similarly, California had five top-two House primaries featuring eight or more candidates in 2020, and four in 2018. Democratic activists in California have expressed concern over the perpetually crowded top-two primaries.¹⁴ Lawmakers and voters in more states and cities are voting to adopt these democratizing electoral systems. In 2021, bills were proposed in 29 states to adopt ranked-choice voting in some capacity.¹⁵

Voters already struggle to navigate low-information elections in a nationalized media environment with declining local news (Chapp & Aehl, 2021; Darr, Hitt & Dunaway, 2018). What happens when these conditions are paired with electoral systems that overwhelm voters with choice? When the media are nationalized, voters are more likely to rely on party cues in these less salient elections (Trussler, 2021). But, how do voters distinguish between candidates when multiple candidates are from the same party? Future research should continue to investigate how electoral designs that incentivize candidate entry affect the informational demands placed upon voters.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Spencer Goidel

Notes

Appendix

Use of the jungle primary in Louisiana leads to an approximately 5.7–7.5% increase in ballot rolloff. The identified effect is robust to the three model specifications in the first three columns of Table A1. The model with Mississippi as the control and the model with synthetic control—both presented in the body of this article—are shown in the first and third columns. The model in the second column only includes Louisiana parishes and Mississippi counties with House elections in all five election cycles. Table A1.

Effect of Jungle Primary on Ballot Rolloff.

	(1)	(2)	(3)	(4)
	Full	Balanced	Synthetic	Placebo
Jungle primary	7.109^∗∗∗	7.474^∗∗∗	5.720^∗∗	1.115
Jungle primary	(1.352)	(1.386)	(1.601)	(0.745)
Single-party	5.252^∗∗∗	6.346^∗∗∗
Single-party	(0.469)	(0.858)
Senate	0.904^∗	1.398
Senate	(0.414)	(0.818)
Competitive	−0.118	−0.142
Competitive	(0.0763)	(0.0968)
Incumbent	3.095^∗∗∗	3.452^∗∗∗
Incumbent	(0.443)	(0.832)
Constant	−0.746	−2.229	3.764^∗∗	2.633^∗∗∗
Constant	(1.315)	(1.581)	(1.180)	(0.500)
Time FEs	✓	✓	✓	✓
Unit FEs	✓	✓	✓	✓
R²	0.68	0.41	0.33	0.00
N	655	420	160	680

Standard errors in parentheses.

^∗p < .05, ^∗∗p < .01, ^∗∗∗p < .001.

I estimate the same model on a placebo state (Mississippi) and its synthetic control. The placebo test is shown in column four of Table A1. The difference-in-differences estimator has a null effect when the model is estimated on Mississippi counties. Table A2.

Logistic Regressions Predicting Ballot Rolloff.

	(1)	(2)
	W/o Ind.-Level	W/ Ind.-Level
Number of candidates	0.249^∗∗	0.256^∗∗
Number of candidates	(0.086)	(0.099)
Constant	−5.327^∗∗∗	−5.106^∗∗∗
Constant	(0.941)	(1.335)
Election-specific controls
Single-Party	2.588^∗∗∗	2.632^∗∗∗
Single-Party	(0.436)	(0.446)
Senate	0.096	−0.054
Senate	(0.954)	(0.902)
Competitive	−0.003	−0.006
Competitive	(0.024)	(0.025)
Incumbent	−0.664^∗∗∗	−0.494^∗∗
Incumbent	(0.143)	(0.155)
Individual-level controls
Age		0.008
Age		(0.011)
Republican		−0.517
Republican		(0.478)
Democratic		0.262
Democratic		(0.422)
Education		−0.149
Education		(0.100)
Employment Status		0.472
Employment Status		(0.404)
Marital status		−0.192
Marital status		(0.400)
Race		−0.312
Race		(0.461)
Gender		0.335
Gender		(0.260)
Year FEs	✓	✓
N	1,043	1,043
pseudo R²	0.143	0.166
Wald χ²	94.6	1155.1

Standard errors in parentheses.

^∗p < .05, ^∗∗p < .01, ^∗∗∗p < .001.

Table A3.

Multilevel Logistic Regressions Predicting Ballot Rolloff.

	(1)	(2)
	W/o Ind.-Level	W/ Ind.-Level
Number of candidates	0.249^∗∗	0.256^∗∗
Number of candidates	(0.0916)	(0.0948)
Constant	−5.326^∗∗∗	−5.106^∗∗∗
Constant	(1.173)	(1.440)
Election-specific controls
Single-party	2.587^∗∗∗	2.632^∗∗∗
Single-party	(0.614)	(0.630)
Senate	0.0972	−0.0536
Senate	(1.137)	(1.146)
Competitive	−0.00288	−0.00630
Competitive	(0.0355)	(0.0363)
Incumbent	−0.664^∗∗∗	−0.494^∗
Incumbent	(0.192)	(0.234)
Individual-level controls
Age		0.00837
Age		(0.0121)
Republican		−0.517
Republican		(0.496)
Democratic		0.262
Democratic		(0.418)
Education		−0.149
Education		(0.113)
Employment Status		0.472
Employment Status		(0.371)
Marital status		−0.192
Marital status		(0.346)
Race		−0.311
Race		(0.373)
Gender		0.335
Gender		(0.340)
Year FEs	✓	✓
N	1043	1043
Log-likelihood	−158.88	−154.71
Wald χ²	44.79	50.78

Standard errors in parentheses.

^∗p < .05, ^∗∗p < .01, ^∗∗∗p < .001.

Author Biography

Spencer Goidel is an Assistant Professor in the Department of Political Science at Auburn University. He received his PhD from the Department of Political Science at Texas A&M University, and BA from the University of Louisiana at Lafayette. Spencer studies public opinion, voting behavior, and elections. His research investigates how political institutions, objective conditions, subjective evaluations, and the media shape the behaviors and attitudes of American voters.

References

Abadie

Diamond

Hainmueller

(2010). Synthetic control methods for comparative case studies: Estimating the effect of California’s tobacco control program. Journal of the American Statistical Association, 105(490), 493–505. https://doi.org/10.1198/jasa.2009.ap08746

Abadie

Diamond

Hainmueller

(2015). Comparative politics and the synthetic control method. American Journal of Political Science, 59(2), 495–510. https://doi.org/10.1111/ajps.12116

Aguilar

Cunow

Desposato

Barone

L. S.

(2015). Ballot structure, candidate race, and vote choice in Brazil. Latin American Research Review, 50(3), 175–202. https://doi.org/10.1353/lar.2015.0044

Alvarez

R. M.

Sinclair

J. A.

(2015). Nonpartisan primary election reform. Cambridge University Press.

Benton

J. E.

(2003). County government (pp. 261–266). Encyclopedia of Public Administration and Public Policy.

Bonneau

D. D.

Zaleski

(2021). The effect of California’s top-two primary system on voter turnout in US House Elections. Economics of Governance, 22(1), 1–21. https://doi.org/10.1007/s10101-021-00249-8

Bowler

Donovan

Happ

(1992). Ballot propositions and information costs: Direct democracy and the fatigued voter. The Western Political Quarterly, 45(2), 559–568. https://doi.org/10.1177/106591299204500216

Chapp

Aehl

(2021). Newspapers and political participation: The relationship between ballot rolloff and local newspaper circulation. Newspaper Research Journal, 42(2), 235–252. https://doi.org/10.1177/07395329211014968

Chernev

Böckenholt

Goodman

(2015). Choice overload: A conceptual review and meta-analysis. Journal of Consumer Psychology, 25(2), 333–358. https://doi.org/10.1016/j.jcps.2014.08.002

10.

Cohen

M. J.

(2018). A dynamic model of the invalid vote: How a changing candidate menu shapes null voting behavior. Electoral Studies, 53(2), 111–121. https://doi.org/10.1016/j.electstud.2018.04.015

11.

Cordis

A. S.

Milyo

(2016). Measuring public corruption in the United States: Evidence from administrative records of federal prosecutions. Public Integrity, 18(2), 127–148. https://doi.org/10.1080/10999922.2015.1111748

12.

Crosson

(2021). Extreme districts, moderate winners: Same-party challenges, and deterrence in top-two primaries. Political science research and methods, 9(3), 532–548. https://doi.org/10.1017/psrm.2020.7

13.

Cunow

Desposato

Janusz

Sells

(2021). Less is more: The paradox of choice in voting behavior. Electoral Studies, 69(1), 102230. https://doi.org/10.1016/j.electstud.2020.102230

14.

Damore

D. F.

Waters

M. M.

Bowler

(2012). Unhappy, uninformed, or uninterested? Understanding “none of the above” voting. Political Research Quarterly, 65(4), 895–907. https://doi.org/10.1177/1065912911424286

15.

Darcy

Schneider

(1989). Confusing ballots, roll-off, and the black vote. The Western Political Quarterly, 42(3), 347–364. https://doi.org/10.2307/448432

16.

Darr

J. P.

Hitt

M. P.

Dunaway

J. L.

(2018). Newspaper closures polarize voting behavior. Journal of Communication, 68(6), 1007–1028. https://doi.org/10.1093/joc/jqy051

17.

Duverger

(1954). Political parties: Their organization and activity in the modern state. University Printing House, Cambridge (Methuen & Co Ltd).

18.

Fisk

C. A.

(2020). No republican, no vote: Undervoting and consequences of the top-two primary system. State Politics and Policy Quarterly, 20(3), 292–312. https://doi.org/10.1177/1532440019893688

19.

Grosser

Arthur

(2006). Neighborhood information exchange and voter participation: An experimental study. American Political Science Review, 100(2), 235–248.

20.

Hall

A. B.

Snyder Jr

J. M.

(2015). Information and wasted votes: A study of US primary elections. Quarterly Journal of Political Science, 10(4), 433–459. https://doi.org/10.1561/100.00015027

21.

Hall

M. G.

(2007). Voting in state supreme court elections: Competition and context as democratic incentives. The Journal of Politics, 69(4), 1147–1159. https://doi.org/10.1111/j.1468-2508.2007.00614.x

22.

Haynes

G. A.

(2009). Testing the boundaries of the choice overload phenomenon: The effect of number of options and time pressure on decision difficulty and satisfaction. Psychology and Marketing, 26(3), 204–212. https://doi.org/10.1002/mar.20269

23.

Herrnson

P. S.

Taylor

J. A.

Curry

J. M.

(2015). The impact of district magnitude on voter drop-off and roll-off in American elections. Legislative Studies Quarterly, 40(4), 627–650. https://doi.org/10.1111/lsq.12091

24.

Holman

M. R.

Lay

J. C.

(2021). Are you picking up what I Am laying down? Ideology in low-information elections. Urban Affairs Review, 57(2), 315–341. https://doi.org/10.1177/1078087420908933

25.

Iyengar

S. S.

Kamenica

(2010). Choice proliferation, simplicity seeking, and asset allocation. Journal of Public Economics, 94(7-8), 530–539. https://doi.org/10.1016/j.jpubeco.2010.03.006

26.

Iyengar

S. S.

Lepper

M. R.

(2000). When choice is demotivating: Can one desire too much of a good thing? Journal of Personality and Social Psychology, 79(6), 995–1006. https://doi.org/10.1037/0022-3514.79.6.995

27.

Jilke

Van Ryzin

G. G.

Van de Walle

(2016). Responses to decline in marketized public services: An experimental evaluation of choice overload. Journal of Public Administration Research and Theory, 26(3), 421–432. https://doi.org/10.1093/jopart/muv021

28.

John

Smith

Zack

(2018). The alternative vote: Do changes in single-member voting systems affect descriptive representation of women and minorities? Electoral Studies, 54(2), 90–102. https://doi.org/10.1016/j.electstud.2018.05.009

29.

Keele

L. J.

Shah

P. R.

White

Kay

(2017). Black candidates and black turnout: A study of viability in Louisiana mayoral elections. The Journal of Politics, 79(3), 780–791. https://doi.org/10.1086/690302

30.

Kroh

(2009). The ease of ideological voting: Voter sophistication and party system complexity. Oxford University Press.

31.

Menger

Stein

R. M.

Vonnahme

(2018). Reducing the undervote with vote by mail. American Politics Research, 46(6), 1039–1064. https://doi.org/10.1177/1532673x17737059

32.

Miller

A. H.

(1990). Public judgments of senate and House candidates (pp. 525–542). Legislative Studies Quarterly.

33.

Milton

(1983). A cross-sectional analysis of the roll-off vote in New York state, 1948-1974. Polity, 15(4), 613–629. https://doi.org/10.2307/3234459

34.

Nichols

S. M.

Strizek

G. A.

(1995). Electronic voting machines and ballot roll-off. American Politics Quarterly, 23(3), 300–318. https://doi.org/10.1177/1532673x9502300303

35.

Patterson

Jr. (2020). Estimating the unintended participation penalty under top-two primaries with a discontinuity design. Electoral Studies, 68(1), 102231. https://doi.org/10.1016/j.electstud.2020.102231

36.

Reilly

Richey

(2011). Ballot question readability and roll-off: The impact of language complexity. Political Research Quarterly, 64(1), 59–67. https://doi.org/10.1177/1065912909349629

37.

Richey

(2010). The impact of corruption on social trust. American Politics Research, 38(4), 676–690. https://doi.org/10.1177/1532673x09341531

38.

Richie Terrell

Lamendola

Reilly

(2021). Election reform and women’s representation: Ranked choice voting in the US. Politics and Governance, 9(2), 332–343. https://doi.org/10.17645/pag.v9i2.3924

39.

Schwartz

(2004). The paradox of choice: Why more is less. ECCO.

40.

Sparks

(2020). Quality challenger emergence under the top-two primary: Comparing one-party and two-party general election contests. Electoral Studies, 65(5), 102136. https://doi.org/10.1016/j.electstud.2020.102136

41.

Streb

M. J.

Frederick

LaFrance

(2009). Voter rolloff in a low-information context: Evidence from intermediate appellate court elections. American Politics Research, 37(4), 644–669. https://doi.org/10.1177/1532673x08326045

42.

Trussler

(2021). Get information or get in formation: The effects of high-information environments on legislative elections. British Journal of Political Science, 51(4), 1529–1549. https://doi.org/10.1017/s0007123419000577

43.

Uslaner

E. M.

Brown

(2005). Inequality, trust, and civic engagement. American Politics Research, 33(6), 868–894. https://doi.org/10.1177/1532673x04271903

44.

Wattenberg

M. P.

McAllister

Salvanto

(2000). How voting is like taking an SAT test: An analysis of American voter rolloff. American Politics Quarterly, 28(2), 234–250. https://doi.org/10.1177/1532673x00028002005

45.

Webster

S. W.

Abramowitz

A. I.

(2017). The ideological foundations of affective polarization in the US electorate. American Politics Research, 45(4), 621–647. https://doi.org/10.1177/1532673x17703132

46.

Wright

G. C.

(1989). Policy voting in the U.S. Senate: Who is represented? Legislative Studies Quarterly, 14(4), 465–486. https://doi.org/10.2307/439954

47.

Wright

S. G.

Riker

W. H.

(1989). Plurality and runoff systems and numbers of candidates. Public Choice, 60(2), 155–175. https://doi.org/10.1007/bf00149243

48.

(2017). Generalized synthetic control method: Causal inference with interactive fixed effects models. Political Analysis, 25(1), 57–76. https://doi.org/10.1017/pan.2016.2