Does descriptive representation lead to social desirability bias? Over-reporting of voting among blacks and Latinos in the United States

Abstract

This study explores whether black and/or Latino respondents are more likely to over-report voting in districts with US House candidates of the same race/ethnicity and whether the overstating of political participation is contingent on shared partisanship between the candidate and the respondent. We test these relationships using vote-validated data from the 2006 and 2010 Cooperative Congressional Election Study. We find that blacks and Latinos are not more likely to over-report voting in districts with descriptive candidates regardless of whether the candidate and respondent share the same party identification or not. The results of this study provide support for previous and future studies linking descriptive representation to higher levels of black and Latino turnout using non-voter validated data.

Introduction

In many surveys, it is not uncommon for those who failed to turnout on Election Day to report voting. Some estimates suggest that almost twenty percent of national survey respondents who report voting did not participate in the election (Bernstein et al., 2001). The most common explanation for respondents being deceitful about their electoral behavior is that they prefer to give the socially desirable response instead of admitting that they did not vote. Such an admission would violate societal expectations and lead to perceptions that the respondent is not a responsible citizen (Anderson and Silver, 1986; Burden, 2000; Silver et al., 1986). While these misleading responses inflate voting figures, they only bias the analysis of self-reported voting measures when the over-reporting is not random. Unfortunately, several studies demonstrate that many of the factors which predict turnout such as race, income, and education are also highly correlated with the over-reporting of voting (Bernstein et al., 2001; Silver et al., 1986). As a result, some have raised questions about the validity of previous research which makes the connection between these variables and turnout (Bernstein et al., 2001; Cassel, 2003).

While there are a number of studies which explore the over-reporting of voting, the effect of candidates who share the race/ethnicity of the respondent (i.e. descriptive candidates) on social desirability response bias in the United States has not yet been examined. This study addresses this gap in the literature by assessing whether over-reporting by blacks and Latinos in US House elections is greater when co-racial/ethnic candidates run for office. There is a plethora of research demonstrating that blacks and Latinos are more likely to vote when they have the opportunity to support a candidate who shares their race/ethnicity (Baretto et al., 2004; Bobo and Gilliam, 1990; Washington, 2006). However, many of these studies use self-reported measures of turnout (Bobo and Gilliam, 1990; Bullock and Gaddie, 2014; Clark, 2014; Griffin and Keane, 2006; Philpot et al., 2009; Tate, 2003) which could be problematic as several authors warn that those who have the opportunity to vote for descriptive candidates may feel the most pressure to over-report voting (Bernstein et al., 2001; Duefel and Kedar, 2010; Griffin and Keane, 2006; McKee et al. 2012). For example, Duefel and Kedar (2010: 308–309) note: “other things being equal, social desirability may be an even more powerful source of over-reporting among African-Americans who did not turn out in elections in which a black candidate competes”. If blacks and Latinos over-report voting more in districts with co-racial candidates, then previous and future research demonstrating a link between descriptive representation and turnout using non-voter validated data becomes questionable.

In order to investigate the effects of descriptive candidates on over-reporting, we utilize the 2006 and the 2010 Cooperative Congressional Election Studies which used political data vendor Catalist, LLC to validate voting responses for their nationwide sample. Controlling for a number of factors including age, income, education, racial district composition, and political interest, we find that blacks and Latinos are not more likely to over-report voting in districts with descriptive US House candidates. In fact, we find that blacks in districts with co-racial candidates are significantly less likely to overstate voting than blacks in districts without these candidates. We then assess whether blacks and/or Latinos are more likely to overstate their participation rates in districts where there is a candidate who both shares their race/ethnicity and partisanship. Here, we find that neither blacks nor Latinos are more likely to over-report voting in districts with candidates of the same race/ethnicity and party.

The over-reporting of voting

Previous research demonstrates that survey respondents provide misleading responses to surveyors to avoid embarrassment (Krysan, 1998; Kuran, 1995). For example, survey respondents are more likely to try to appear as being more racially tolerant, interested in politics, and more receptive to gender equality than they actually are (Krysan, 1998; Streb et al., 2008). Kuran (1995) suggests that when deciding whether to reveal their private preferences, people weigh the advantages of expressing these true opinions against the reputational damage caused by providing an untrue, but socially unacceptable response. When people feel that they will be severely judged for their views, they will be more likely to provide the untrue, but socially acceptable response to avoid humiliation (Kuran, 1995).

This logic helps explain which survey respondents are the most likely to overstate their participation in elections. While all individuals have incentives to overstate their voting habits, some individuals may perceive that they have more to lose by acknowledging their abstention from voting (Bernstein et al., 2001; Kuran, 1995). The pressure to participate in elections is generally greater among citizens who are perceived as having some stake in the election result (Bernstein et al., 2001; Silver et al., 1986). Based on this research, there is reason to believe that blacks and Latinos in districts with co-racial candidates will feel more pressure to overstate their levels of electoral participation than co-racial/ethnic respondents in districts without these candidates. Blacks in the United States generally have high levels of group consciousness because of their socio-historical experiences with discrimination and their continued underrepresentation in government (Dawson, 1995; Tate, 1994). This sense of group consciousness is hypothesized to be so strong that blacks tend to believe that what happens to other African-Americans plays a significant role in predicting their own opportunities (Dawson, 1995). While Latinos generally have lower levels of linked fate than African- Americans (Masuoka, 2006), Latinos too have a strong sense of community. In a recent study, 68% of Latinos believed their success was tied to other Latinos (Sanchez and Masuoka, 2010). Given blacks’ and Latinos’ strong sense of group consciousness, there may be the expectation that blacks and Latinos in districts with co-racial candidates should be especially active.

The social pressure to support a descriptive candidate may feel particularly intense for both of these racial/ethnic groups as they have faced strong formal and informal barriers to government offices for most of American history (Bernstein et al., 2001). Moreover, Casellas and Wallace (2015) show that both Latinos and blacks have strong preferences for more descriptive representation in Congress. The elevated social pressure to turnout to increase diversity in government may lead blacks and Latinos who have the potential to be descriptively represented to feel ashamed for staying home on Election Day. Those who fail to vote in an election with a co-racial candidate may also feel that they have let down members of their racial/ethnic group and feel embarrassed for their abstention. In combination, blacks and Latinos in districts with co-racial candidates may be the most likely to over-report voting. As a result, researchers using non-voter validated data to explore the effects of descriptive candidates on turnout may arrive at biased conclusions.

However, there is also good reason to believe that blacks and Latinos in districts with co-racial/ethnic candidates will not over-report voting significantly more than other co-racial/ethnic respondents. The increased social pressure for African-Americans to vote that comes from the sacrifices of the Civil Rights Movement may make it so that even blacks in districts without black candidates feel pressure to overstate their voting record (Bernstein et al., 2001). Similarly, given pressure on Latinos to assimilate into the American polity, Latinos as a group may have a greater incentive to report voting even when they did not (Johnson, 1997). Consistent with these arguments, Fraga (2015) demonstrates that blacks and Latinos are the most likely of any racial/ethnic group to overstate participating in elections. If the pressure to over-report voting is higher for all blacks and Latinos, then the baseline comparison (i.e. blacks and Latinos without descriptive candidates) may not leave enough room for significant differences to occur.

Moreover, recent research demonstrates that individuals with strong political identities are the most likely to over-report voting (Brenner, 2012). As a result, we expect that blacks and Latinos who identify most with their racial group should feel the most pressure to overstate their participation rates when there is a co-racial/ethnic candidate on the ballot. However, it is also these individuals who feel the most pressure to overstate voting whether a co-racial/ethnic candidate is on the ballot or not. As a result, we may find that co-racial/ethnic candidates have little influence on increasing the number of blacks and Latinos who over-report voting.

Data and methods

The 2006 and 2010 CCES

The 2006 and 2010 Cooperative Congressional Election Studies (CCES) surveyed a nationwide sample of Americans on their attitudes about government and voting. In both years, the CCES was administered by YouGov/Polimetrix and partnered with political data firm Catalist, LLC to validate voting records. Using its’ nationwide database of voters, Catalist was able to accurately match respondents to their voting records provided by state governments (Ansolabehere and Hersh, 2012). The CCES is suitable for this study because it includes a national sample of respondents, a large number of black and Latino respondents, and was collected in years where a large number of black and Latino candidates sought federal office. ¹

Our dependent variable for this study is dichotomous and measures whether an individual reported voting in the 2006 or 2010 general election when actuality they did not vote based on the validated data (i.e. they over-reported voting). Similar to other studies, we confine the sample to individuals who stated that they voted, as only they can misreport voting (Bernstein et al., 2001; Silver et al., 1986). We dropped respondents from the analysis who said they were not citizens, said that they did not vote, said they were not registered to vote, or were not matched by Catalist, LLC indicating that it was not possible to know whether they voted or not. Finally, our analysis only includes respondents who identify as African-American or Latino. ²

Our independent variable of interest measures whether there was at least one co-racial/ethnic US House candidate in the black or Latino respondent’s congressional district. ³ To isolate this relationship, we control for several factors shown to increase rates of over-reporting on surveys. First, we control for socio-demographic factors such as gender, age, income and education. Previous research shows that older, more educated, and wealthier respondents are more likely to mislead surveyors about voting (Bernstein et al., 2001; Cassel, 2003). We also control for whether the respondent identifies as a Democrat or a Republican. Partisans, because they are perceived as having more of a stake in the election, may feel an expectation to vote and thus may feel more pressure to overstate voting (Bernstein et al., 2001). Additionally, we account for whether respondents were contacted by a political party during the election cycle and their levels of political interest. Both of these variables are associated with respondents over-reporting voting (Presser, 1990; Silver et al., 1986).

We also control for the percent of co-racial/ethnic individuals who reside in the respondent’s congressional district based on estimates from the American Community Survey. Bernstein et al. (2001) show that blacks and Latinos are more likely to over-report voting in districts where their racial group makes up a sizable portion of the population. Given that this is also where black and Latino candidates are most likely to arise, it is important to control for this variable.

Results

Descriptive representation and over-reporting of voting

Figure 1(a) presents the results of a logit regression model predicting whether black respondents who said that they voted, over-reported turning out in the 2006 and 2010 midterm elections. Figure 1(b) presents the same analysis for Latinos. The results in Figure 1 present a marker for the regression coefficient and 90 to 95 percent confidence intervals. To make binary and continuous variables more comparable, the coefficients in Figure 1 have been standardized by dividing continuous variables by two standard deviations (see Gelman, 2008).

Figure 1.

Logit regression predicting the over-reporting of voting for (a) black respondents in districts with black US House candidates and (b) Latino respondents in districts with Latino US House candidates with 90% and 95% confidence intervals.

While there should be increased social pressures for members of underrepresented groups to overstate their political participation rates in certain electoral contexts, the results in Figure 1 demonstrate that neither blacks nor Latinos were more likely to over-report voting in districts with US House candidates of the same race/ethnicity once we account for several variables. In fact, blacks were significantly (p<0.05) less likely to over-report voting in districts with black US House candidates than blacks in districts without these candidates. This provides confidence that previous studies which examine the relationship between descriptive representation and turnout using non-voter validated data are not biased by the over-reporting of voting once accounting for other factors. Moreover, they may actually be understating candidate effects for African-Americans.

While the opportunity for descriptive representation did not lead blacks or Latinos to significantly overstate their voter participation rates, there were several variables related to greater levels of social desirability response bias. Younger blacks and Latinos and black men were significantly more likely to over-report voting. Given that these individuals have lower rates of turnout compared to older racial/ethnic minorities and black women, they may have greater opportunities to mislead surveyors when they fail to vote. Blacks who were partisans were also more likely to over-report voting. Perceptions that partisans should have a greater stake in the election outcome may be driving this result. Contrary to previous studies, we find that blacks and Latinos who were contacted by a campaign were less likely to over-report voting. This result may be driven by the factors specific to midterm elections, such as fewer people being contacted during these years. Previous studies on this topic focus predominately on presidential elections (Bernstein et al., 2001). Finally, in line with previous research, blacks in majority black congressional districts were much more likely to over-report voting (Bernstein et al., 2001).

Not all candidates are equal: Partisanship and over-reporting of voting

While the above analysis demonstrated that blacks and Latinos were not more likely to overstate their voting record in districts with descriptive candidates, there may not be uniform pressure to turnout to support any candidate of the same race/ethnicity. For example, a black Democrat may not be perceived as having a special stake in an election where the only black candidate on the ballot is a Republican. Thus, black Democrats in this scenario may only feel increased pressure to vote and in turn overstate voting when a candidate who shares both their race and party is on the ballot.

To assess this possibility, we reanalyze the model presented in Figure 1. However, instead of examining the relationship between all descriptive candidates and black’s and Latino’s propensity to over-report voting, we only examine the effect of co-partisan descriptive candidates on social desirability response bias; thus, we create a dummy variable in which respondents who live in a congressional district with a US House candidate who shares their race and partisanship are given a score of one and all others receive a score of zero. For example, a black Democratic respondent who has the opportunity to vote for a black Democratic congressional candidate is given a score of one. Conversely, a black Republican in a district where the only black candidate on the ballot is a Democrat is given a score of zero. All individuals who do not have the opportunity to support a descriptive candidate are also given a zero. To assess the effect of co-partisan descriptive candidates on the over-reporting of voting, we again estimate a logit regression for blacks and Latinos predicting whether individuals who were not validated as voting reported turning out in the 2006 or 2010 midterm elections. These regression models control for the same factors used in the previous analysis such as age, income, education and campaign contact.

Figure 2 presents the results from two logit regression models predicting the over-reporting of turnout for blacks (Figure 2(a)) and Latinos (Figure 2(b)). Again, only those who stated that they voted were included in the analysis; Figure 2 presents coefficient estimates of each variable with 90 and 95 percent confidence intervals. Again, the continuous variables have been divided by two standard deviations to make them comparable to the dichotomous variables. While blacks and Latinos should feel more pressure to turnout to support candidates who share both their party and race/ethnicity, we again find that blacks and Latinos are not more likely to overstate voting when there is a co-partisan descriptive candidate on the ballot. Again, there does not appear to be evidence that opportunities for descriptive representation lead to greater levels of social desirability response bias.

Figure 2.

Logit regression predicting the over-reporting of voting for (a) black respondents in districts with black US House candidates of the same party and (b) Latino respondents in districts with Latino US House candidates of the same party with 90% and 95% confidence intervals.

Conclusion

This study provides comfort to individuals who examine the link between descriptive representation and turnout for blacks and Latinos using non-voter validated surveys such as the American National Election Study (ANES) (which has not validated turnout since 1988) or the Current Population Survey. The mere presence of candidates of the same race/ethnicity does not lead blacks or Latinos to overstate their voting records to a greater degree than non-descriptively represented minorities when controlling for several variables. Instead, the results in Figure 1 suggest that non-vote validated data may actually provide conservative estimates of the link between descriptive representation and voting for blacks. All else being equal, blacks in districts with co-racial candidates are about 7% less likely to overstate voting than other blacks. ⁴ If blacks face universal pressure to overstate turnout, it is possible that this result could be attributed to the fact that blacks in districts with co-racial candidates actually vote more. As a result, all else being equal, blacks in districts with co-racial candidates should have less opportunity to over-report voting than blacks without co-racial candidates. ⁵

In line with previous research on the topic (Bernstein et al., 2001), we also find that blacks in districts with a large percentage of co-racial individuals were much more likely to over-report voting. Our results suggest that the overstating of voting when blacks live around co-racial individuals is not simply driven by the desire to increase descriptive representation. Thus, the overstating of turnout in majority–minority districts may be most tied to increased pressure to engage in the act of voting among co-racial/ethnic individuals given the sacrifices of the Civil Rights Movement, rather than pressure to alter the election outcome.

While this study provides new information about the role of candidates from underrepresented groups on social desirability response bias, future work should build on this research with the following considerations. First, given concerns about space, we did not explore internal differences within the black and Latino community. It is possible, that some racial/ethnic minorities may be more likely to overstate voting when there is a descriptive candidate on the ballot. For example, minorities with high levels of linked fate may feel more pressure to overstate voting when they have the opportunity to support a co-racial candidate.

Second, the CCES has many advantages, including having a large number of black and Latino respondents and vote validation measures; however, there is research suggesting that the CCES contains a higher percentage of respondents who over-report voting than surveys such as the ANES during the 1980s (Ansolabahere and Hersh, 2012). As a result, it is possible that the format in which the survey was collected could have led all black and Latino voters to overstate their participation, thus obscuring differences between those who have the opportunity to be descriptively represented and those who do not. Third, future research should examine whether other underrepresented groups such as Asian-Americans (there were too few Asian-American respondents in districts with Asian-American candidates in both the 2006 and 2010 CCES to draw meaningful results) are more likely to overstate or understate their voting levels when a candidate sharing their race is on the ballot. While this study is an important first step in exploring the relationship between descriptive representation and the over-reporting of voting, more work is necessary.