A Partisan Model of Electoral Reform

Introduction

Scholars of elections propose that the legitimacy of democratic regimes depends, in no small part, on public confidence in the integrity of the election process (Alvarez, Hall, and Llewellyn 2008; Birch 2008; Elkit and Reynolds 2005; Norris 2004; 2014). Democracy requires that both winners and losers view election outcomes as the result of a fair process (Anderson et al. 2005). Myriad aspects of the electoral process can affect perceptions that elections are fair; studies have found that electoral formula, bribery, vote buying, campaign finance rules, and quality of administration may affect trust and confidence in elections and support for democratic institutions (Anderson and Tverdova 2003; Birch 2008; Bowler, Brunell, Donovan, and Gronke 2015; McAllister and White 2011).

One paramount factor affecting perceptions of electoral integrity is confidence in the process of balloting and vote counting (Elkit and Reynolds 2002; 2005; Hall and Wang 2008). Indeed, the major motivation for electoral reform in the United States after the 2000 presidential election crisis in Florida was to assure public confidence in the process of casting and counting votes (e.g., Commission on Federal Election Reform 2005; National Commission on Federal Election Reform 2001). The 2002 Help America Vote Act (HAVA) was the primary legislative manifestation of this at the federal level. HAVA directed federal money to states to fund improvements in voter registration and ballot counting (Alvarez and Grofman 2014), while also opening the door for the dissemination of new voter identification laws (Stewart 2014).

Voter identification laws have proven to be the more controversial of these post-2000 reforms of election administration. Their adoption has been supported by claims that the laws are needed to improve public confidence in elections (e.g., Crawford v. Marion County Election Board 2008), but the politics of state adoption has been highly partisan. Most of the strict photo identification state laws were adopted by Republican-controlled legislatures. ¹ Some scholars describe state adoption of voter ID laws as “Jim Crow 2.0” (Bentele and O’Brien 2013) and see them as a vote suppression policy that reflects a Republican strategy to curtail Democratic electoral gains (Hicks et al. 2014).

This article focuses on how these laws might affect confidence in the integrity of elections; we set aside the better studied question about the effects of voter identification laws on turnout. Although the Court acknowledged in Crawford that the record contained no evidence of voter impersonation (see also Hasen 2012; Hood and Gillespie 2012; Minnite 2007; 2010; Wang 2012), ² the majority ruled that independent of any instances of fraud, photo identification laws were justifiable because of a state’s significant interest in “protecting public confidence in elections.” One reason for this, the Court opined, was that confidence in elections was a prerequisite for voter participation (for research on the relationship between identification laws and turnout, see Alvarez, Hall, and Llewellyn 2008; Ansolabehere 2009; Barreto, Nuno, and Sanchez 2009; Hood and Bullock 2012; Vercellotti and Anderson 2009).

We propose theory and hypotheses about the relationships between partisanship, election rules, and public attitudes relating to electoral integrity. We test our hypotheses with a 50-state (and Washington, DC) sample of opinion data that measured public confidence in how votes were cast and counted in a respondent’s state. Results confirm expectations derived from our theory. Rather than building broad-based confidence in how elections were conducted, voter identification laws corresponded with mass polarization in confidence about electoral integrity. Higher levels of confidence about the incidence of fraud that might be associated with strict identification laws were found among people who identified as Republican. We also find that Democrats living in states with strict identification laws, and Democrats asked for identification, were less confident in their state’s elections. Republicans in states with strict identification laws were substantially more confident. This implies that these laws may not have accomplished one of the key goals that was seen as essential in their justification.

A Partisan Model of Election Reform ³

Many rules governing the conduct of elections are the product of strategic interests of partisan incumbents who (largely) control the process of designing such rules. A wide variety of election rules can be seen as having a zero sum nature (Tsebelis 1990), such that change from a status quo arrangement may alter the balance of advantages and disadvantages between rival parties. Changes in election rules are often viewed through the lens of partisan self-interest; partisan incumbents often author the rules in their own self-interest (Bowler, Donovan, and Karp 2006).

Electoral rules provide many examples of these kinds of processes at work. Expansion of the franchise in the nineteenth and early twentieth centuries, for example, can be seen as a rule change that offered minority parties opportunities to challenge or surpass the dominance of a majority party of the day (Tsebelis 1990, 113). Elite support for changing to proportional electoral systems can also be understood in terms of strategic calculations by rival partisans. Changes to the electoral system in Australia (Farrell and McAllister 2005), British Columbia (Angus 1952), and New Zealand (Banducci and Karp 1999) have been documented in terms of strategic elites seeking partisan advantage (also see Bawn 1993 on Germany). Mass and elite support for election reform proposals in the United States such as legislative term limits (Karp 1995), changes in redistricting rules (Donovan 2011), primary election rules (Cain and Gerber 2001; Ware 2002), and direct election of the president (Karp and Tolbert 2010) can also be understood through strategic calculations of partisans. In many of these cases, the attitudes of partisans in the public can be seen as reflecting the positions of party elites.

A model of partisan self-interest applied to changes in electoral rules is not entirely novel. But we add to this the idea that, in democratic systems, partisan elites often need to generate popular support for proposals—or for challenges to proposals—that change rules about how elections are conducted. Given concern for procedural fairness as a widely held principle in mass attitudes (Tyler 1988), and given public concern over fairness in the process of democratic politics (Hibbing and Theiss-Morse 2001), it is not sufficient for a majority or minority party to promote a rule change with an argument along the lines of “this will advantage our party, we have the votes, so we will do it.” Rather, mass expectations over the integrity of the electoral process require that elites make normative arguments while advancing their partisan interests. Furthermore, research documents that in addition to their partisan interests, legislators evaluate rule changes in terms of their personal values about how democracy should work (Bowler, Donovan, and Karp 2006).

This means that election rules, particularly those that are newly adopted, exist as the result of an interaction between strategic calculations of rival partisan elites and political arguments those elites make to the public to promote or defeat a rule change (Bowler and Donovan 2013). New election rules are not simply the product of one dominant party being strong enough to push through any change it may see as an advantage. Legislators face the threat of reelection, or, in many states, institutions of direct democracy that may unravel rules baldy advanced by self-interested elites. As a result, when partisan elites advance or critique proposals to change election rules, they have strong incentives to make (biased) normative arguments that justify their position on a rule change in a manner that appeals to procedural concerns over fairness and electoral integrity.

Regardless of the veracity of the normative aspects of these arguments, the public is exposed to normative, partisan arguments about the pros and cons of a change to an election rule prior to the change taking effect and opinion is likely to divide along partisan lines, that is, people learn about political issues via cues and signals received from elites. Given the public has a rather limited grasp of the factual nuances of matters of policy and political institutions (Delli Carpini and Keeter 1996, 70–71), mass opinion on such topics can reflect positions taken by elites, or as Key (1966) put it, “the voice of the people is but an echo” of elite signals about such matters. Popular perceptions of a politicized election rule and its consequences, then, will reflect partisan polarization in mass opinion that mirrors signals from partisan elites (Lupia and McCubbins 1998; Zaller 1992).

It follows, then, that when prominent Democratic and Republican elites take divergent positions on arguments about an electoral reform, their partisan supporters in the public follow suit. Competing rules for redistricting practices provide an example. Democratic legislators who were in the minority in Florida and their allies, and Republican legislators and their allies in the minority in California both used similar arguments about the fairness and integrity of the electoral process while championing an electoral reform that served their respective party’s interest. ⁴ In this case, the minority parties advocated reforms that weakened a legislative majority’s control over districting (Bowler and Donovan 2013). Minority party incumbents in each state (Democrats in Florida, Republicans in California) promoted similar ballot measures designed to weaken the majority party’s control over redistricting, and their partisan supporters in the electorate responded in kind. These redistricting measures were more popular among Democratic voters in Florida and Republican voters in California (Donovan 2011). Term limits, likewise, were viewed more favorable by Republicans in states where Democrats were in the majority and relatively less favorably by Republican voters in states where Republicans held a majority in the legislature (Donovan and Snipp 1994). Wilson and Brewer (2013) likewise identified partisan polarization in opinion about voter identification laws in a national survey conducted in 2012.

Voter Identification Laws and Public Confidence in Elections

We suggest this partisan theory of electoral reform can be applied widely to evaluate public perceptions of many election reforms where parties are divided, and we apply it in this case to generate hypotheses about the effects of voter identification laws on public confidence in elections. The primary causal mechanism here is that reforms presented and promoted by rival partisan elites are likely to be seen as such by partisans in the public. Partisans in the public think about electoral laws in very similar ways to partisan elites, with voters often taking their cues about the reforms from elite rhetoric (Bowler and Donovan 2013). When election law reforms are implemented, then, we would expect voters’ responses to those laws to mirror themes from debates among elite opinion leaders.

As for that debate, Stewart (2014, 97) noted that starting in 2002, HAVA fostered a political process that “stoked paranoia about elections being stolen, either because of hacked voting machines or fraudulent voting.” If there was initial bipartisan concern over the need for strict voter ID laws (see Commission on Federal Election Reform 2005), it faded quickly. Partisan arguments over the adoption of photo identification laws can be simplified as one where Republicans contended that voter impersonation presented risks of elections being “stolen” (e.g., Fund 2004; von Spakovsky 2011), whereas Democrats contended that requirements for photo identification were attempts at voter suppression (e.g., Overton 2006). By 2003, six states had adopted “nonstrict, nonphoto ID” rules, with five of these states having Republican legislatures (the sixth being Alabama). The first strict photo identification laws were adopted by Republican legislatures in Georgia and Indiana in 2005, and by Republicans in Missouri in 2006. ⁵ As of 2014, 31 states had voter identification laws, with strict photo identification laws in effect in Kansas, Georgia, Indiana, Mississippi, Tennessee, Texas, and Virginia. Photo identification laws have often been adopted on near-pure party-line floor votes, with nearly all Republicans in favor and nearly all Democrats in opposition (Bentele and O’Brien 2013; Bowler and Donovan 2013, 30; Davidson 2009). Hicks et al. (2015) found that adoption was most likely in electorally competitive states where Republicans controlled government.

Mass opinion about the need for voter identification reflects these elite divisions. Multiple opinion polls conducted since the adoption of HAVA have shown that while voter identification requirements were popular, Republicans were much more concerned about the threat of voter impersonation than Democrats. A 2007 Cooperative Congressional Election Study (CCES) survey found Republican identifiers more than twice as likely as Democrats to respond that voter fraud was “very common” (35% vs. 15%; Ansolabehere and Persily 2008). A 2008 YouGov/Polimetrix survey found 21% of Republicans responding that voter fraud occurred “very often,” compared with just 7% of Democrats who had such perceptions (Ansolabehere and Persily 2008). Table 1 reports opinion data measured in a 2014 Survey on the Performance of American Elections (SPAE) and illustrates that these partisan differences in perceptions of elections persisted into 2014.

OPEN IN VIEWER

Table 1.

Percent Saying They Think Various Forms of Illegal Voting Are “Very Common” and “Occasional” in Respondent’s City or County, and Percent Who Were “Very Confident” in Their State’s Vote Count, 2014.

	Democrat (%)	Republican (%)	Independent (%)
People voting more than once in an election	11	27	23
People pretending to be someone else when voting	13	28	24
People voting who are not U.S. citizens	13	39	30
Very confident in state’s vote count	45	50	39

Source. 2014 Survey of the Performance of American Elections.

Despite the importance that the Court assigned to the presumed effects of voter identification laws on confidence in elections in Crawford, there is relatively little research into how these laws are associated with perceptions of the conduct of elections. Of the studies we have identified, there is little to suggest that voter identification laws affect confidence in elections. Atkeson’s (2014, 117) study of voter confidence in New Mexico concluded, “voter identification policies appear to have little effect.” Ansolabehere (2009), likewise, concluded from surveys of national opinion that those living in states with strict identification laws were no more confident about elections than people in states with the weakest identification rules, and that people asked to show identification at the polls were no more confident about electoral integrity than those who were not asked (Ansolabehere and Persily 2008).

But by neglecting to examine partisan differences in perceptions of elections, these studies may be missing the potential polarizing effects that voter identification laws have on perceptions of elections. Moreover, the laws were relatively new when those studies were conducted. We expect that because partisan elites have provided clear, divergent signals about their positions on the consequences of having or not having voter identification rules that public perceptions of the application of voter identification laws will reflect these partisan divisions. As Atkeson’s (2014, 118) study demonstrates, most people do not experience election fraud personally; they live it vicariously through media (see also Gronke and Hicks 2009).

For approximately a decade since the first states began adopting these laws, rival partisan narratives about the laws’ effects on the conduct of elections have been disseminated. News media have directed attention to state legislative battles over adopting voter identification laws, to several state and federal court challenges to the laws, to Justice Department actions associated with the laws, to various conspiracy “theories,” to stories about the Association of Community Organizations for Reform Now (ACORN), and given coverage to the arguments of rival party elites. Partisan signals about voter identification laws, then, have been available to the public. Selective exposure to partisan-themed media outlets by partisans in the public (Stroud 2008; 2010) could further increase the likelihood that partisans absorbed their respective party’s narrative on the consequences of strict voter identification laws. ⁶ There is some evidence consistent with this. Wilson and Brewer (2013) found that other things held constant, Fox News viewers were more supportive of voter identification laws.

Hypotheses

How then, might we expect people to respond to these laws where they are in effect? That is, what does our partisan model of election reform suggest we should expect about how people evaluate the integrity of elections where these laws are in effect? Given signals from Republican elites, we assume Republicans in the public were more likely to absorb the narrative that strict voter identification laws are an improvement to the conduct of elections because, without the laws, fraudulent voting will occur. Republicans may thus see the laws as good, effective election practice. As such, Republicans living in states that used strict ID laws should be more confident that fraud was uncommon and that elections in their state were conducted properly.

One might expect that states with higher instances of fraud were more likely to adopt strict identification laws. But, as noted above and as recognized by the Court in Crawford, there is little or no evidence of voter impersonation occurring (see also Minnite 2007; 2010). Strict identification laws thus seem to exist almost entirely independent from actual levels of voter impersonation in a state. Nevertheless, if it were true that stricter ID laws responded to actual problems of fraud then we would expect all voters in strict ID states—including Republicans—to be more likely to think fraud was common and be less confident in elections.

Democrats, however, are assumed to have absorbed the narrative of Democratic elites that strict voter identification is an attempt at vote suppression aimed at diluting the Democratic vote. Democrats living in states with strict ID laws, then, should be less confident that elections in their state were conducted properly, and adoption of strict ID laws should have had no effect on Democrats’ perceptions of the instances of voter fraud. That is, while a partisan model would lead Republican voters to see ID laws as effective, the laws should have no effect on Democrats.

We also expect that there should be partisan differences in how the application of voter identification rules affects perceptions of electoral integrity. Studies have documented that the implementation of voter identification rules by poll workers was uneven, and that voter identification laws were not applied in a race-neutral manner. Cobb, Greiner, and Quinn (2010) demonstrated that black voters and Latino voters, and Atkeson et al. (2010) demonstrated that Latino voters were more likely than white voters to be asked for identification when voting. This suggests that the election day experience for some voters involves witnessing people of color being asked for identification—an experience that is consistent with, and could reinforce, the Democratic narrative that voter identification laws degrade the integrity of elections by disproportionately affecting groups of voters known to affiliate with the Democratic Party. ⁷ Given this, and given the Democratic narrative about vote suppression, we expect that partisans interpret the experience of being asked for identification differently. Democrats who were asked for identification, then, should be more cynical about the conduct of elections. Given the prevailing Republican narrative about voter fraud, a Republican experiencing a request for identification while voting may view the request as a process that increases electoral security. Republicans who were asked for identification, then, should be less cynical about the conduct of elections.

Previous research suggests a number of additional factors that should be included in models estimating attitudes about the conduct of elections and confidence in elections. Research has documented that Latinos in Colorado and New Mexico have been no less confident in elections than others (Atkeson et al. 2011; Atkeson et al. 2010; Atkeson and Saunders 2007). However, given the discussion above about the uneven application of voter identification laws, and given the history of disenfranchisement motivated by race and ethnicity (Grofman and Davidson 1992; Keyssar 2000), we expect that Latinos and African Americans (e.g., Alvarez, Hall, and Llewellyn 2008; Bullock, Hood, and Clark 2005) could have less confidence in the conduct of elections. ⁸ Education has also been found to be associated with voter confidence (Alvarez, Hall, and Llewellyn 2008), as has gender, with women found to be less confident than men (Alvarez, Hall, and Llewellyn 2008; although see Atkeson and Saunders 2007; Hall, Monson, and Patterson 2009).

We also include age as a control, and we account for whether or not a person voted, and whether or not one voted in person or by mail. Voters can be expected to be more confidence in elections than nonvoters, and previous research suggests that there are more counting errors associated with voting by mail (Stewart 2010) and that people are less confident in votes cast by mail (Alvarez, Hall, and Llewellyn 2008; Atkeson and Saunders 2007). Partisanship is included in the models with terms for Republican and Democratic identifiers, respectively, interacted with experiences with voter identification, to test our primary hypotheses. ⁹

Estimating Perceptions of Elections

Our hypotheses are tested with data from a postelection survey of 10,200 American adults conducted in Fall of 2014. The 2014 SPAE included samples of 200 registered voters from all 50 states and the District of Columbia (Stewart 2015). The survey was directed by Charles Stewart of the Caltech/MIT Voting Technology Project and was conducted online by Polimetrix during the 2014 midterm election. Data and documentation are publicly available via the Harvard Dataverse. ¹⁰ Of respondents who reported voting (more than 80% of the sample), 64% voted in person on election day, 16% voted in person before election day, and 20% voted by mail. The sample was 54% female, 81% white, with a partisan distribution of 34% Democrat, 27% Republican, 30% independent, and 9% “other” and “not sure.” ¹¹ The data are weighted in the analysis to be representative of respondents nationwide.

The first dependent variable for this study is constructed from three SPAE questions that asked about perceptions of illegal voting that voter identification laws might be expected to police: “people voting more than once,” “people pretending to be someone else when going to vote,” and “people voting who are not U.S. citizens.” Response options for these items were “very common,” “occurs occasionally,” “occurs infrequently, “it almost never occurs,” and “I’m not sure.” In the analysis reported here, responses are rescaled, so high scores reflect perceptions that the illegal act was very common, with “I’m not sure” responses at the center of a five-point scale. These three items are combined into an index that reflects perceptions of fraud. A factor analysis found that each of the three items load on a single tab at .79 or higher. ¹²

Our second dependent variable is an item asking respondents to “think about vote counting throughout [respondent’s state]. How confident are you that votes in [respondent’s state] were counted as voters intended?” Response options were rescaled to range from “not at all confident,” “not too confident,” “I don’t know,” “somewhat confident,” to “very confident.” With all analysis reported here, results are largely unaffected by whether or not the “I don’t know” and “I’m not sure” responses are omitted or included as a midpoint on a scale (see Table A1 of the appendix). This voter confidence item is similar to measures that were included on the 2006 CCES and 2008 SPAE asking voters about their own vote being counted. Atkeson and Saunders (2007) and Alvarez, Hall, and Llewellyn (2008) suggested a link between the individualistic version of this confidence item and attitudes about democracy generally (where the former causes the latter). Given that the state-centric version of the voter confidence question asked people to think about counting generically in their state, we expect the item reflects attitudes not just about the personal experience of voting but a broad view of a person’s confidence in the conduct of elections in their own state. ¹³

Our main independent variable is a measure of each state’s voter identification law that was in effect as of October 31, 2014. We use the National Conference of State Legislatures’ (NCSL) five-point scale for this, with states coded 1 if they had no documentation required for voting, 2 if they had nonstrict, nonphoto identification rules, 3 if they had nonstrict photo identification rules, 4 if the state had strict nonphoto identification requirements, and 5 if a state had strict photo identification requirements. ¹⁴ Our main hypotheses are tested by interacting a respondent’s partisanship with the status of her state’s voter identification rule, and by interacting partisanship with whether or not the respondent was asked for identification when voting. Standard estimates of individual opinions conducted with large sample surveys collected across a relatively small cluster of higher level (state) cases could bias estimates of standard errors such that the likelihood of Type I errors would increase (Steenbergen and Jones 2002). Given the multilevel nature of our data, our hypotheses were tested with multilevel mixed models with fixed effects, and for comparison, with single-level models specified with robust standard errors clustered by state. Multilevel linear regression (column 1, Table 2) and ordinary least squares (OLS) with robust standard errors clustered by state (column 2, Table 2) are used to model the three-item index of perceptions of voter fraud. Mixed effects ordered logit (column 1, Table 3) and ordered logit with robust standard errors clustered by state (column 2, Table 3) are used to estimate confidence in a state’s vote counting.

OPEN IN VIEWER

Table 2.

Partisan Models of Perceptions of Voter Fraud: Three-Item Index, Higher Values = More Fraud.

Variables	Mixed effects linear regressions	Ordinary least squares
Democrat	−1.86**	−1.84**
	(.19)	(.18)
Republican	0.90**	0.86**
	(.19)	(.20)
Black	0.97**	1.19**
	(.16)	(.15)
Latino	0.40	0.80*
	(.24)	(.27)
Female	−0.08	−0.09
	(.07)	(.06)
Age	−0.012**	−0.011**
	(.002)	(.002)
Education	−0.23**	−0.23**
	(.03)	(.03)
Asked ID	0.09	0.02
	(.16)	(.18)
Voted	−1.04**	−1.05**
	(.12)	(.12)
Vote by mail	0.47**	0.54**
	(.10)	(.15)
State photo ID score	0.02	0.03
	(.06)	(.06)
Republican × State ID Score	−0.07	−0.06
	(.07)	(.07)
Democrat × State ID Score	0.05	0.04
	(.06)	(.07)
Republican × Asked ID	−0.38	−0.42
	(.21)	(.23)
Democrat × Asked ID	0.23	0.27
	(.25)	(.25)
Constant	9.6**	9.57**
	(.25)	(.25)
Variance (individual-level)	11.60
Variance (state-level)	.30
Wald χ2/R2	1,308**	.10
Level 1 N	10,146	10,146
Level 2 N	51

Source. 2014 Survey on the Performance of American Elections.

Note. Robust standard errors clustered by state are in parentheses.

*

p < .05. **p < .01 (two-tailed).

OPEN IN VIEWER

Table 3.

Partisan Models of Confidence in State’s Elections (Confident that Votes in State Counted as Intended, Higher Scores = More Confident).

Variables	Mixed effects ordered logit estimates	ordered logit estimates
Democrat	0.74**	0.74**
	(.16)	(.15)
Republican	0.16	0.17
	(.10)	(.13)
Black	−0.29**	−0.40**
	(.10)	(.11)
Latino	0.04	−0.16
	(.08)	(.09)
Female	−0.08*	−0.07
	(.04)	(.04)
Age	0.0003	−0.0004
	(.0014)	(.0013)
Education	0.06**	0.06**
	(.02)	(.02)
Asked ID	0.17	0.19
	(.10)	(.11)
Voted	0.83**	0.80**
	(.07)	(.07)
Vote by mail	−0.07	−0.09
	(.06)	(.08)
State photo ID score	−0.04	−0.03
	(.04)	(.04)
Republican × State ID Score	0.14**	0.13**
	(.03)	(.04)
Democrat × State ID Score	−0.16**	−0.16**
	(.06)	(.06)
Democrat × Asked ID	−0.27	−0.26
	(.15)	(.15)
Republican × Asked ID	0.01	0.02
	(.11)	(.11)
Constant 1	−2.47**	−2.45**
Constant 2	−1.09**	−1.09**
Constant 3	−0.68**	−0.68**
Constant 4	1.25**	1.20**
Variance (cons)	.144
Wald χ2	608.9*	692.3**
Level 1 N	10,196	10,196
Level 2 N	51

Source. 2014 Survey on the Performance of American Elections.

Note. Robust standard errors clustered by state are in parentheses.

*

p < .05. **p < .01 (two-tailed).

Results

Table 2 displays results of estimates of our voter fraud perception index—note that the higher a score on the dependent variables in Table 2, the more a person perceived fraud as common. The partisan differences in perceptions of voter fraud illustrated in Table 1 persist here when demographic traits, mode of voting, and other factors are accounted for. Given the partisan battles over the need for voter identification laws, it may come as no surprise that Republicans were significantly more likely to perceive that fraud was common. Democrats, in contrast, were significantly less likely to see fraud as common. We also find people of color and people who voted by mail were more likely to have perceived that fraud was common. As expected, voters were less skeptical about fraud than nonvoters, and age and education correspond with being less likely to perceive voter fraud.

As for the potential effects of strict voter identification laws on confidence in state elections, we find evidence that the relationship between the laws and public perceptions of elections was polarized on party lines. None of our model specifications yielded evidence that strict photo identification laws had any broad, unconditioned effects on voter confidence. The likelihood that a person perceived voter fraud as common was unrelated to how strict her state’s voter identification laws were. However, there was a conditional relationship associated with partisanship. Although Republicans were significantly more suspicious of voter fraud, Republicans who were asked for ID were less so (Table 2), this relationship is only marginally significant (p < .05, one-tailed).

As for public confidence in a state’s election process (Table 3), we find that partisans generally, those with more education, and (not surprisingly) those who voted, were more confident that votes were counted in their state as they were supposed to be. African Americans and women, in contrast, had less confidence in their state’s elections. Again, we find no unconditional relationship between the strictness of a state’s voter identification law and public confidence in elections. However, the polarized, conditional relationship between voter identification laws and attitudes about the conduct of elections is quite evident when we examine a person’s confidence in their state’s elections. Republicans living in states with stricter identification laws were significantly and substantially more confident in their state’s elections, whereas Democrats were significantly and substantially less confident. ¹⁵ Democrats were slightly less confident their state’s elections if they were asked for identification (p < .05, one-tailed) when they voted. This study is not equipped to assess causality or change in perceptions across time, but these results do suggest that if voter identification laws had any positive effects on voter confidence, they were limited to Republicans, and any potential positive effects among Republicans may have been offset by depressed confidence among Democrats.

The magnitude of partisan differences in voter confidence associated with voter identification laws is substantial—larger than the presumed effects of race and ethnicity. Table 4 displays Clarify simulations (Tomz, Wittenberg, and King 2003) generated from ordered logit estimates of confidence in state elections (from column 2, Table 3), where we display the predicted probability that a respondent reported being “very confident” in her state’s vote count. Our baseline respondent, a white, political independent, female voter of median age and education, who resided in a state with the mean level of voter identification strictness, is estimated to have a .42 probability (SE = .02) of being very confident in her state’s elections. As shown in Table 3, Democratic partisans were generally more confident in elections, all else equal. But the predicted probabilities in Table 4 demonstrate how much this confidence was conditional on whether a state was using strict voter identification procedures, and, to a lesser extent, on whether or not someone was asked for identification at the polls.

OPEN IN VIEWER

Table 4.

Predicted Probability of Being Very Confident in State’s Vote Count.

Baseline	0.42 (.02)	Δ, from no ID state to strict photo ID
Democrat in non-ID state	0.57 (.03)
Democrat in strict photo ID state	0.39 (.03)	−.18
Democrat, asked for ID in strict ID state	0.37 (.03)	−.20
Black Democrat, non-ID state	0.50 (.05)
Black Democrat, strict photo ID state	0.30 (.03)	−.21
Black Democrat, asked for ID, strict ID state	0.28 (.03)	−.23
Republican in non-ID state	0.47 (.04)
Republican in strict photo ID state	0.60 (.03)	+.13
Republican asked for ID, strict ID state	0.65 (.03)	+.18

Note. Estimated from ordered logistic regression with robust standard errors clustered by state. Baseline respondent is a white, female, independent who voted at the polls. Standard errors in parentheses.

A Democrat had a rather high probability of being very confident in her state’s elections (.57, SE = .03) if her state had no voter identification requirements. A Republican, in contrast, had nearly the same probability (.60, SE = .03) of being very confident if her state had the strictest photo identification rules. This probability of being very confident melts away for Democrats in states with the strictest photo identification ( p = .39), and it drops a bit further for Democrats who were asked for identification in the strictest identification states ( p = .37). The relationship is reversed for Republicans. A Republican in a state without voter identification had an almost even probability (.47) of being very confident in her state’s vote counts. This increased (p = .65) for Republicans in strict identification states who were asked for identification.

We should note that these results are quite robust; they are largely insensitive to how the models are estimated. Hypothesis tests conducted with estimates of our five-point opinion measure produce the same substantive results when modeled as regular ordered logit with standard errors clustered by state, and when modeled with multilevel mixed effects linear regression. Likewise, our hypothesis tests are consistent regardless of whether or not “don’t knows” are omitted or included in the analyses, and when state-level controls for income and a Pew Center measure of the quality of the states election performance (circa 2012) are included (not reported). Finally, when these models are estimated with voters only, and with three additional controls that account for a respondent’s election day experiences, we find results identical to those reported here. Table A1 of the appendix reports some of these models, estimated to include controls for time spent in line, experience with problem associated with voting equipment, and experience with registration problems—each of which have significant relationship confidence in a state’s election count. Again, there was no generalized effect of voter identification laws on voter confidence, but we do find voter identification laws had a significant relationship with confidence that was conditioned by partisanship. Republican voters in states with strict identification laws were more confident in their state’s elections, whereas Democratic voters in strict identification states, and Democrats asked for identification, were less confident. In short, we consistently find that the relationship between photo identification laws and confidence in state elections was polarized and heavily conditioned by party identification.

Discussion

Results from our study reinforce something well established—that the manner in which the process of voting is conducted has major effects on how people perceive the integrity of elections (Alvarez, Hall, and Llewellyn 2008; Atkeson 2014; Atkeson and Saunders 2007; Hall, Monson, and Patterson 2007). Our major contribution is to highlight the consequences of politicized reforms on public confidence in elections. We demonstrate that when a feature of election administration is highly partisan, there are reasons to expect that the potential effect of that reform on public confidence in elections will be conditioned by partisanship.

We offer a model of election reform where public response to a reform reflects the partisan environment from which the rule change emerged. In this model, rival partisan elites send signals on the need for and consequences of a proposed reform, with partisans in public adopting those positions. After the law is passed and implemented, we anticipate partisan reactions among voters to mirror those of the party debate that around the change itself. We applied this model to examine how state use of voter identification laws might have affected voter confidence in elections, and found polarization in partisans’ confidence that corresponded with use of the laws. These results are consistent with a process where (1) polarized elites send divergent narratives to the public about the needs for and consequences of an election reform, (2) partisans in the electorate adopt their party’s narrative about the benefit (or harm) of the reform, and (3) this conditions how partisans in the public subsequently evaluate and experience the reform when it is applied. Republicans were told by Republican elites that voter identification would improve the integrity of elections, whereas Democrats were told by Democratic elites that the reform would erode the legitimacy of elections. It follows, then, that when the laws were implemented, Democrats and Republicans would experience them differently. We find evidence that was the case. Republicans living in strict photo identification states were more confident in their elections, where Democrats were less confident.

The results raise broader questions about the conditions necessary and sufficient for an election reform to improve voter confidence in elections. With reforms that are the product of a relatively consensual adoption process, or the product of a process where party divisions are less overt, we might expect some broad, generalized effect of the reform on voter confidence—particularly where there is high public support for a new election rule and support is not divided along party lines. But this raises questions of how often such conditions can be met. Given the partisan stakes of so many election rules, there is a possibility that many rule changes that do result from elite consensus may only be those so banal that the public may be unaware of them and so have little effect on citizen attitudes toward electoral practices.

It is the case that at least some aspects of the conduct of elections are not highly politicized. Apart from matters of fiscal implementation costs, for the most part rival partisan elites do not launch highly divisive contests over reforms such as those designed to improve the training of polling place workers—at least not as loudly as they contested voter identification laws. Similarly, rival elites do not (overtly) contest policies aimed at improving the effectiveness of statewide registration databases. And, for the most part, they do not construct rival narratives about the pros and cons of having effective technology for counting ballots. As with nearly all election laws, there are potentially partisan stakes even with these matters (e.g., discussions of registration databases beget discussions of purging voter rolls, and discussions of electronic voting systems beget discussions of vote theft). But these may be areas of reform where there is some partisan consensus (or much less division), where the interests of rival party elites do not reach a level of such salience that public attitudes about the consequences of the reform polarize along party lines. Elites from different parts of the political spectrum could communicate a uniform message about the positive consequences of a reform. ¹⁶ In such cases, we might expect reform to generally have positive effects on confidence in elections, provided the reform is something the public can actually notice.

But again, how often do such conditions apply? In many cases, a proposal for a change in status quo election rules—particularly those that are highly visible—is viewed differently by different partisans. One party may see a gain where the other sees a loss. Examples of such cases include proposals for changes in redistricting practices, changes in electoral formula, various campaign finance reforms, changing local elections from at-large to district, adopting term limits, voting by mail, election day registration, lowering the voting age, and allowing noncitizens the vote in nonfederal elections. In these cases, partisan elites tend to see what is at stake in a reform, and if change is imminent they provide their supporters with rival narratives about how the change might harm or help the democratic process. When this is the case, we suggest the reform will be perceived in partisan terms, and its effects on how people view the integrity of elections and representation may be polarized.

The application of this partisan model of election reform to voter identification laws illustrates some manner of irony. To a degree, the Democrats’ narrative about photo identification laws matched the Court’s Crawford decision: both recognized that fraud was largely undocumented and rare. The Court, however, took the potential for even rare, undocumented instances of voter fraud as justifying the need for strict voter identification laws. This justification rested, in part, on the idea that voter identification laws could “protect” public confidence in elections. Or, as federal reform commissions have put it, election reforms generally are said to be needed to “build confidence” and to “assure pride and confidence” in elections.

Yet the highly politicized, acrimonious nature of legislative adoption, litigation, and implementation of strict voter identification laws, and the partisan stakes, we contend, likely determined that public evaluation of and public response to these laws would be defined on party lines. Republicans argued that there would be fraud without photo identification laws. Democrats argued (and some empirical studies report) that voter identification laws would be applied in ways that disadvantage groups of voters who tend to vote for Democrats. It makes some sense, then, that despite their proponent’s stated goal of building confidence in elections, and despite the Court’s rationalization of the laws as needed for building voter confidence, state identification laws were associated with Republicans being more confident and Democrats being less confident in their states’ elections. We suggest that the prospects for such polarized effects of reform on public confidence in elections are not limited to voter identification laws.