Vote Overreporting While Black: Identifying the Mechanism Behind Black Survey Respondents’ Vote Overreporting

Overreporting: Overview and Causes

Perhaps the biggest challenge for those who wish to use survey data to accurately understand the causes or effects of voter turnout is that individuals do not overreport voting at the same rate. Instead, the likelihood that someone overreports their turnout varies across a variety of demographic and psychological factors including income, satisfaction with the status-quo (Silver et al., 1986), education (e.g., Bernstein et al., 2001; Silver et al., 1986), strength of partisanship (e.g., Silver et al., 1986), church attendance (e.g., Bernstein et al., 2001), and race (e.g., Bernstein et al., 2001; Hill & Hurley, 1984), among other factors.

Overreporting presents difficulties to researchers, perhaps most importantly because the propensity to overreport is related to a number of popular covariates and variables of interest. Those attempting to test hypotheses involving voter turnout may encounter bias, leading to the potential that some hypotheses will be rejected in error and others accepted erroneously (e.g., Bernstein et al., 2001). Early work studying the effects of overreporting concluded that there was little to be concerned about (Katosh & Traugott, 1981; Sigelman, 1982). However, a renewed focus on how overreporting might bias results has suggested reasons to be concerned.

More recent work finds that results from various models of voting could change substantially depending on whether verified votes were used as a dependent variable or self-reports were used (Bernstein et al., 2001). Although, this may only be true for less than half of the variables common to voter behavior research (Cassel, 2003). More recently, Ansolabehere and Hersh (2012) suggest that accounting for overreporting changes results in some voting models significantly enough that they recommend, “the dramatic effect of misreporting on models of participation demands a renewed effort at theory-building.” Thus, developing a better understanding of the causes and consequences of overreporting is important to producing ways to account for overreporting in analyses of existing data, and discovering ways to reduce it in future studies.

There have been a number of explanations offered as to why survey respondents might report having voted when they did not, including misremembering, the desire to appear to have participated in a socially desirable activity, and internal pressures. Misremembering is thought to lead to overreporting because respondents, most often those individuals who are inconsistent voters, apply their experiences voting in earlier elections to the most recent one they are asked about, thus incorrectly “remembering” they voted (e.g., Abelson et al., 1992; Belli et al., 1999, who find both direct and conditional effects of memory on overreporting). Others (e.g., Bernstein et al., 2001; Silver et al., 1986), using voter validation designs from the ANES or other observational data sets, have suggested that external pressure to appear to have participated in a socially desirable activity, voting, is what leads many people to overreport because they don’t want to admit that they didn’t do so to another individual.

Work utilizing experimental designs have confirmed the role of social desirability bias as one of the factors responsible for overreporting. Holbrook and Krosnick (2010) use list experiments to demonstrate that social desirability bias is a factor causing overreporting. Duff et al. (2007), using an experimental design embedded in the 2002 ANES, demonstrate that a voter turnout question allowing respondents to give more socially desirable excuses for not voting in the last election reduces overreporting compared with questions that do not offer such response options. Further, Duff et al. (2007) find that questions offering socially desirable excuses for not voting are most successful at reducing overreporting among those least likely to vote due to low levels of income, education, and political efficacy. Similarly, while testing their “source-monitoring design,” Belli et al. (2006) demonstrate that longer questions about turning out to vote can help overcome overreporting due to memory problems and provide additional evidence of the efficacy of providing more socially acceptable response options at reducing overreporting.

Others have suggested that overreporting may be both a function of social desirability bias and misremembering. Much of this work suggests that not only does the desire to appear to act in socially desirable ways directly cause overreporting but that this desire can also influence the ways in which people remember things, causing individuals to incorrectly remember turning out to vote in an election when they did not. Belli et al. (1999) demonstrate that questions giving both socially acceptable response options for not voting in the past election and that encourage the respondent to think closely about the voting experience reduces overreporting by tackling both social desirability bias and misremembering.

Finally, a fourth line of research suggests that there may be internal pressures within an individual that cause overreporting in addition to external pressures, such as social desirability bias (Hanmer et al., 2014). These internal pressures stem from voting being an expressive act for many people as a way to support their political party and/or participate in democracy and fulfill their civic duty. Thus, people adopt an attachment to the identity of being a voter even in instances when they don’t vote. Due to this attachment, respondents with high intrinsic value placed on the voter identity may be likely to report voting by interpreting the question about turnout in the recent election to be asking about more than the past election, and rather about their general voting behavior and place within the democratic system. Additionally, reporting to have voted while taking the survey, even when they didn’t actually vote, may help these individuals claim a bit of their voter identity. As evidence for the role of internal pressures, Hanmer et al. (2014) point to the fact that over 50% of validated non-voters in the 2008 CCES, an online only survey, reported voting. If only external pressures, such as social desirability bias, were at play this shouldn’t be observed. By experimentally manipulating whether or not respondents to an internet survey are told that their turnout behavior will be verified via official records, Hanmer et al. (2014) were able to reduce overreporting and improve the accuracy of responses, suggesting that internal pressures to vote can be somewhat overcome by encouraging respondents not to misreport for fear of being discovered. Thus, overreporting is suspected to be the result of a number of different processes including misremembering, external pressures to appear to behave in socially desirable ways, internal pressures, and a combination of the above.

Overreporting among African Americans

Perhaps one of the most consistently documented aspects of overreporting is that African Americans overreport at higher rates than whites (e.g., Abramson & Claggett, 1984, 1986, 1991; Anderson et al., 1988; Bernstein et al., 2001; Hill & Hurley, 1984; Katosh & Traugott, 1981; Presser et al., 1990; Sigelman, 1982; Silver et al., 1986; Traugott & Katosh, 1979). Considering the importance that race plays in our understanding of American politics (e.g., Lewis-Beck et al., 2008; Tesler, 2012; White, 2007), the necessity of understanding the nature and causes of such overreporting is evident. Despite proving a consistent source of vote overreporting, little attention has been given to fully understanding why it is that Black Americans are so much more likely to overreport their voting. Constrained by the small number of Black respondents in voter validated surveys and the lack of racial group-specific attitudinal measures in these surveys, researchers have been limited in their ability to investigate the mechanism underlying Black overreporting. Empirical limitations notwithstanding, many have speculated that Black overreporting likely stems from, among other things, group identity and a desire to improve the standing of the group (Belli et al., 2001; Deufel & Kedar, 2010) or poor voting record keeping in majority-black districts (Abramson & Claggett, 1992).

One of the more recent explanations for not just African American overreporting, but overreporting more generally, is from Bernstein et al. (2001) who argue that those who feel the most pressure to vote are the most likely to overreport. They argue that social pressure stems from the district in which one lives, and that minority respondents living in minority-dense districts should be more likely to overreport voting than minorities living in areas with a lower minority population. They demonstrate that African Americans living in more strongly African American districts are more likely to overreport than those African Americans living in areas with a smaller Black population. This is evidence, they claim, that due to both the increased salience of the minority group’s needs and pressures from group institutions in the community, minorities living in large minority-populated areas feel a stronger pressure to vote and support the group, leading them to feel more guilt when they don’t vote. And thus, also more likely to misreport their behavior to interviewers when asked.

We propose an additional explanation for African American overreporting—social pressure resulting from the social dynamics of the survey interview. In this paper, we investigate Black overreporting as the result of social pressure to conform to the norms of Black political behavior when respondents are faced with a same-race interviewer. We depart from prior work (e.g., Bernstein et al., 2001) by proposing an additional source of social pressure, that social pressure exerted by the dynamics of the survey interview. Further, we suggest this social pressure leads African American respondents to overreport. Thus, we examine the effects of social pressure to conform to racial group norms of political engagement, resulting from a unique survey interview context where a respondent is presented with a same-race interviewer as an explanation for Black American overreporting.

Racialized Norms of Voting and Survey Context

Building on the work on group norms in social psychology, we might expect that Black overreporting results from a need to conform to expectations of political behavior resulting from the racial context of the survey interview. When there are clear and common understandings and expectations of behavior within a group, often referred to as norms, along with equally clearly understood punishment for defecting from those norms, we should expect members of a group to act in line with those norms when other in-group members can perceive their behavior (White et al., 2014). Specifically, we argue that a particularly racialized context is created when a Black respondent is interviewed by a Black interviewer, resulting in social pressure that causes the respondent to be more likely to misreport their voting behavior so as to appear to align with the strongly held norms in the African American community of political participation.

Recently, the role of social pressure in influencing the behavior of individuals in the political realm has received interest. For example, the norm or belief in the virtue of political participation that exists within American society allows threats to make an individual’s turnout record public effective at increasing turnout (e.g., Gerber et al., 2008). This line of work often identifies fear of being shamed or sanctioned for violating norms as the reason that social pressure induces people to engage in norm-conforming behavior (e.g., Gerber et al., 2010).

Not all norms are equal in influencing member behavior. For norms, and the threat of sanction for deviating from them, to influence an individual’s behavior they must be crystalized and have intensity (e.g., Jackson, 1965). Crystallization of a norm is the extent to which a norm is thought of as being understood and agreed upon by group members. The higher the level of norm crystallization the more generally the norm is agreed upon by members throughout the group. The lower the level of norm crystallization, the less enforceable it is by social pressure (Cialdini et al., 1990). Norm intensity is how important a norm is to group members. When intensity of a norm is low the threat of sanctions from other members might not constrain behavior because the deviating member may think their behavior will go unnoticed due to the low importance, or salience, of that norm to group members. On the other hand, when a norm has high intensity, it is considered more important to group members and member violations of that norm are more likely to be noticed, and thus the norm will more strongly constrain group member behavior.

Within the African American community there are a number of norms regarding political behavior that have both high intensity and crystallization. That is, they are widely held by group members and considered important to members. One of the best documented of these is the norm of Black support for the Democratic party (e.g., White & Laird, 2020; White et al., 2014). We suggest that electoral participation is another norm of Black political behavior that meets the above criteria. As noted earlier, African Americans’ unique political history, for example, the Civil Rights Movement’s focus on electoral rights, has led to the development of both highly crystallized and intensely held norms of the belief in the importance of political participation among group members. Thus, we suggest that it is social pressure to appear to comply with the norm of participation and avoid social sanction that, at least partially, accounts for vote overreporting among Black respondents, and that this social pressure is uniquely present when in the presence of another Black individual.

That is, given the high intensity and crystallization of the electoral participation norm of Black political behavior, and evidence of strong sanctioning for deviation from norms of political behavior within the Black community (e.g., Chong, 1991; Starkey, 2012; White & Laird, 2020; White et al., 2014), we argue that overreporting by Black respondents interviewed by a member of the same race should be greater than when interviewed by a member of another race. Black respondents who did not vote in the recent election should be more likely to report having voted in the presence of an interviewer of their own race relative to an interviewer of another race, out of a desire of acceptance from other group members and fear of sanctioning for defecting from that group norm. ¹ If this is the case, we should expect that the race of the interviewer affects overreporting, with Black respondents overreporting at much higher rates in the presence of a Black interviewer than an interviewer of a different race. Additionally, we expect that this effect should exist only among Black respondents and not those of other races or ethnicities, because this norm is believed to be unique to Black Americans.

Therefore, we suggest:

Linked Fate

While previous research has offered little insight into the nature of Black respondents’ overreporting we do know that Blacks’ psychological attachment to their racial group tends to be very important to understanding how Black Americans make political decisions. Scholars of Black political behavior have demonstrated that the degree to which Black Americans see what happens to the racial group as linked to what happens to them as individuals is strongly related to a number of political behaviors, including turnout (see Dawson, 1994; Tate, 1993). This may mean that instead of social pressure resulting from a racialized survey context, feelings of linked fate might explain Black overreporting. Thus, in addition to our theoretical expectations above, we also investigate whether feelings of linked fate explain Black respondents’ overreporting.

Black Americans feel a sense of “linked fate” with one another, often expressing feelings that their interests are the same or similar as those of the larger group (e.g., Dawson, 1994; Tate, 1991). As a likely result of their unique history of slavery, segregation, and discrimination, Black Americans typically express higher levels of linked fate with each other than most other racial or ethnic groups in America, although recent work suggests Latinos, particularly younger Latinos, may exhibit levels of linked fate nearing that of Black Americans (e.g., Dawson, 1994, 2003; Hurwitz et al., 2015). Linked fate represents the idea that support for policies, candidates, or parties that benefit the group at large will also benefit the individual. In effect, feelings of linked fate result in Black Americans substituting “group utility for individual utility when evaluating policies, parties and candidates” (Gay, 2006). This attachment has often been used as an explanation for many sorts of Black political attitudes, such as perceptions of discrimination toward the in-group members as well as members of other minority-status out-groups (Hurwitz et al., 2015), and behavior, including turnout (e.g., Dawson, 2003; see also Tate 1991).

Due to the belief that the interests of the group benefit the interest of the individual, it is expected that feelings of linked fate might lead Black respondents to overreport their turnout in the most recent election. Greater feelings of Black linked fate should make voting and participating in elections in support of candidates deemed beneficial to the group seem more important than it otherwise would. In turn, individuals with a high degree of linked fate should feel compelled to turn out to vote both to support their group and to support their own well-being, which they view as linked to that of their racial group. Given the effects discussed earlier of internal beliefs about voting’s importance on an individuals’ likelihood of overreporting voting, we should expect those Black respondents with a high degree of linked fate to also feel additional pressure to say they voted in the most recent election, even if they didn’t. Thus, if this is the case we should expect feelings of Black linked fate to be related to the rate of overreporting, regardless of the race of the interviewer. Thus:

Conditional Effects of Race of Interviewer and Black Linked Fate

In addition to the direct relationship between feelings of Black linked fate and overreporting stated by the Linked Fate Hypothesis, we may also expect that feelings of Black linked fate may condition the effects of the race of the interviewer on the likelihood of misreporting. Racialized contexts, such as that created by being in the presence of a same race interviewer, heighten racially relevant attitudes and behaviors (e.g., Anderson et al., 1988; White, 2007), such as Black linked fate, and Black linked fate may heighten attention to racialized contexts as well. Thus, if linked fate explains Black respondents’ overreporting, and also may heighten attention to racialized contexts, we may expect a respondent’s feelings of Black linked fate to attenuate or amplify the effect of having a same race interviewer, and thus we might expect the effect of race of the interviewer to be conditioned by feelings of linked fate.

However, if as we argue, social pressure to conform to Black norms of voting and the desire to avoid social sanctioning are what’s driving the majority of overreporting among Black respondents, then we might expect there to be no conditional relationship between feelings of Black linked fate and overreporting by interviewer race. According to our social norm-based explanation, awareness of the norm and concern about possible sanctions should matter regardless of an individual’s feelings of shared fate with their racial group. Indeed, we believe that because the norm of Black political participation is so strongly embedded in the community basically all members of the racial group are aware of it, regardless of the strength of their feelings of linked fate. Thus, because awareness of norms of Black political behavior are so widely held, understood, and independent of feelings of linked fate we do not expect levels of linked fate to condition the effect of interviewer race on Black overreporting if we are correct that it is social norms, not linked fate, driving behavior. Thus:

We test our hypotheses below.

Study Design

To test these expectations, we rely primarily on data from the nationally representative 2012 ANES Times Series Study face-to-face component. These data are well suited for our task. First, they included a large sample of African American respondents. There are 511 Black respondents in the face-to-face pre-election interview and 481 in the face-to-face post-election interview. There were a total of 2,054 respondents in the pre-election face-to-face interviews and 1,929 in the post-election. The 2012 ANES also has a vote validation component in which self-reported turnout in the 2012 general election was validated against official turnout records. Validation efforts were extensive and completed independently by three separate vendors to ensure greater reliability than in validation efforts where just one vendor was used. All vendors used a variety of methods for matching respondents to official voting records, including matching names, addresses, and birthdates, among other proprietary methods (ANES, 2016). In total, 83.6% of respondents were matched to at least one vendor, yielding a relatively high match rate (ANES, 2016). ² We consider an individual’s turnout to have been validated if at least one of the three vendors validated their turnout, and consider an individual to not have voted if none of the three vendors report a validated turnout. ³

Lastly, the number of Black respondents interviewed by Black and non-Black interviewers in the 2012 ANES post-election interview was roughly similar (206 by a Black interviewer, 175 by a non-Black interviewer). This feature of the survey design affords us the ability to test the effects of racialized survey context on self-reported turnout.

We also use data from the 1992, 1996, and 2008 years of the ANES Time Series Cumulative File to conduct additional analyses and demonstrate that the race of interviewer effects we observe are not unique to 2012. We do not include the 2016 ANES data in our analyses because it did not include a sufficient number of Black respondents in the face-to-face mode (N = 119), with even fewer (N = 93) who had either a white or Black interviewer, and thus as a result we would have substantially less certainty about the nature of any results from that study. Thus, we choose to exclude the 2016 data from this analysis. The ANES Times Series Cumulative File is a compilation of every cross-sectional election study conducted from 1948 through 2016. The nature of these data are nationally representative, randomly sampled, repeated cross-sections of the American public for every presidential election, and most congressional elections. ⁴ These years were selected in addition to 2012 because data on the race of the interviewers is available to researchers and they each contain respondents in both the same- and different-race interviewer categories.

We first conduct a series of difference of proportions tests comparing the rates of overreporting among members of different racial and ethnic groups based upon the race of the interviewer. We then investigate the role played by feelings of linked fate in causing overreporting by using difference of proportion tests to compare overreporting rates among respondents at different levels of linked fate.

Next, due to the difficulty posed by controlling for confounders in univariate analyses such as difference of proportions tests, we employ logistic regression to model the propensity to overreport as a function of the race of the respondent’s interviewer using the 2012 ANES data. In addition, we control for a number of other factors that prior work has suggested may be related to overreporting including respondent gender (e.g., Hill & Hurley, 1984; Silver et al., 1986; but see Traugott & Katosh, 1979), age (e.g., Hill & Hurley, 1984; Sigelman, 1982; Traugott & Katosh, 1979), education (e.g., Bernstein et al., 2001; Silver et al., 1986), strength of partisanship (e.g., Silver et al., 1986), feelings of civic duty (Hill & Hurley, 1984; Sigelman, 1982), and feelings of linked fate with their racial in-group (Anderson et al., 1988). We control for these using the standard ANES measures for each, more information of which is available in the 2012 Times Series Documentation (ANES, 2014), and also in Supplemental Appendix D. Following previous research into overreporting interested in differences across races, we estimate separate models for Black, Hispanic, and white respondents (e.g., Bernstein et al., 2001; Leighley & Vedlitz, 1999).

One potential limitation of designs such as this is that interviewers are not randomly assigned to respondents, however we believe the potential difficulties posed by this are minimal. First, other work on race of interviewer effects regarding social norms (e.g., Laird et al., 2016) report that although interviewers may be assigned to respondents in areas proximate to where they live, neither the interviewer nor the interviewee may select one another. Similarly, White and Laird (2020, p. 149), when discussing their use of race of interviewers suggest there is, “no reason to believe that Blacks interviewed by Black interviewers are fundamentally different from those interviewed by non-black interviewers.” Second, others (e.g., Anderson et al., 1988) have reported that the SRC and ANES do not have policies encouraging the matching of interviewers and respondents based on race. Third, even if the relationships observed in 2012 were due to the idiosyncrasies of somewhat proximate geographies of interviewer and interviewee, we would not expect similar patterns to exist across multiple years of data because the interviewers and interviewees change from survey to survey. However, this is precisely what we demonstrate—that the relationship demonstrated in 2012 is robust over multiple years of ANES data. Fourth, the controls included in our multivariate model account for most of the confounders one might expect to be related to overreporting due to geographic proximity, thus this should help control for any difficulties posed by the assignment process. Finally, in Supplemental Appendix E we include additional robustness checks demonstrating that other potential pre-treatment ways in which Black respondents who receive a same-race and different race interviewer may be different do not seem to affect the results. Therefore, we feel confident that the effects we present are true race of interviewer effects based on social pressure and not artifacts of the interviewer assignment process.

Results

We begin by investigating the prevalence of overreporting in modern survey work using the 2012 American National Election Study. Table 1 demonstrates the rate of overreporting in the 2012 ANES for both the whole sample and broken down by race. Here we measure overreporting by the percentage of validated non-voters, those whose vote wasn’t validated by any of the three vendors, who reported voting in the 2012 ANES. As Silver et al. (1986) note, validated non-voters are the only group at risk of overreporting. Thus, we restrict our analyses of the 2012 data to only these individuals, resulting in 465 respondents across the three racial and ethnic categories of interest at risk of overreporting. ⁵ Overall, the percentage of validated non-voters who reported that they voted in the 2012 ANES was 36.9%. The rates of overreporting vary dramatically by race. While 55.7% of Black respondents who did not turn out to vote reported voting, only 32.8% of white validated non-voters and 27.9% of Hispanic voters who did not vote reported turning out to vote. Notably, Black non-voters reported having voted at substantially higher rates than either white or Hispanic respondents, suggesting, as prior work has, that there is something unique about the phenomena of Black overreporting. Indeed, as we can see in Table 2, Black turnout in the 2012 election as measured by self-reports in the 2012 ANES goes from significantly greater than that of whites (6.8-point difference, p < .05) to essentially identical to that of whites when we examine validated turnout. Interestingly, in the 2012 CCES (Ansolabehere, 2013), in which all respondents complete the survey online, the rates of overreporting between Black, White, and Hispanic respondents are substantially smaller, with 70% of African American validated non-voters reporting having voted, 64% of White validated non-voters reporting having voted, and 62% of Hispanic validated non-voters reporting having voted. ⁶ That the differences in overreporting rates by race and ethnicity are more prominent in the face-to-face ANES survey may indicate, as we suggest, that social pressure is one of the mechanisms accounting for these differences.

OPEN IN VIEWER

Table 1.

Percent of Validated Non-Voters Who Reported Having Voted by Race, 2012 American National Election Study.

	Overreporting rate (%)	95% CI	N
Black	55.7	46.4, 64.9	115
White	32.8	26.4, 39.0	214
Hispanic	27.9	20.3, 35.6	136
Total	36.9	32.5, 41.3	465

OPEN IN VIEWER

Table 2.

Self-Reported and Validated Turnout Estimates by Race.

Race	Self-report (%)	Validated (%)	Difference (%)	p-Value
Black	86.6	71.5	15.1	.000
White	79.8	72.0	7.8	.000
Hispanic	68.8	57.9	10.9	.045
Total	76.12	68.8	7.3	.000

As we suggest above, part of this phenomena of Black overreporting can be accounted for by social pressure to conform to norms of either generalized civic behavior or norms of Black political behavior. However, simply looking at differences in the rates of overreporting across the races tells us very little about how the race of the interviewer might help account for these differences. We expect that if racialized norms of voting unique to the Black community affect overreporting (Hypotheses 1a and 1b), then we should observe greater levels of overreporting behavior among Blacks in the presence of a Black interviewer compared to white interviewers. Looking at the results presented in Table 3 we find that this is the case. African American respondents in the presence of a same-race interviewer overreport at significantly higher rates than those Black respondents without a Black interviewer. Specifically, non-voting Black respondents who were interviewed by a Black interviewer reported voting at a rate of 71.4%, whereas only 42.8% of Black non-voters who were interviewed by a white interviewer reported voting in the previous election. Thus, there is a difference of 28.6% points in the rate of overreporting between non-voting Black respondents interviewed by a same race interviewer and those interviewed by a white interviewer, a substantively and statistically significant difference with a p-value of .007. The substantive difference is particularly notable when compared with the differences in overreporting by race of interviewer for the other two race or ethnic groups.

OPEN IN VIEWER

Table 3.

Percent of Validated Non-Voters Who Reported Having Voted by Race and Race of Interviewer, 2012 American National Election Study.

Respondent race	Same race interviewer (%)	Different race interviewer a (%)	Difference	p-Value	N
Black	71.4	42.8	28.6	.007	91
White	32.0	26.9	5.1	.609	176
Hispanic	28.2	28.9	−0.7	.940	123
Total	35.9	35.0	0.9	.860	390

Note. Approximately 2.5% of interviewers refused to provide their racial background.

a

For Blacks this was a Black respondent with a white interviewer; for whites this was a white respondent with a Black interviewer; for Hispanics this was a Hispanic re with a white interviewer.

Thus, the pattern of overreporting based upon the race of the interviewer appears to hold only among Black respondents, suggesting as we hypothesized in the Black Respondent Hypothesis that there is a unique social norm within the Black community regarding political participation that is activated when in the presence of another member of the community. This results in social pressure to conform that leads to overreporting by Black respondents faced with Black interviewers.

These differences in overreporting resulting from same-race interviewers for Black respondents are not isolated to the 2012 election. Table 4 presents Black Americans’ self-reported turnout as measured by the American National Election Study by race of interviewer for presidential elections ranging from 1992 to 2012. Unfortunately, for many of these surveys there were few Black interviewers and sporadic efforts at vote validation. As a result, we are only able to examine self-reported rates of turnout among Blacks across race of interviewer for four elections. Across each of these elections Black respondents interviewed by a same race interviewer were more likely to say they voted than respondents interviewed by white interviewers. Across all four elections, the average difference in self-reported turnout between Black respondents interviewed by a Black interviewer and a white interviewer was just about 13-points (p < .05), with respondents interviewed by a Black interviewer self-reporting turnout at higher rates.

OPEN IN VIEWER

Table 4.

Black Self-Reported Turnout by Interviewer Race, 1992 to 2012 Presidential Elections.

Year	Black interviewer	White interviewer	Difference (BI − WI)
1992	79.5%	65.8%	13.7-points a
N	39	243
1996	90.9%	65.9%	25-points a
N	11	144
2000
N
2004
N
2008	85.1%	81.3%	3.8-points
N	121	360
2012	89.7%	74.4%	15.3-points b
N	214	176
Total	87.3%	73.9%	13.4-points b
N	386	1,074

Note. Data from ANES Time Series Cumulative Data File, includes only face-to-face interviews. This table is limited only to respondents who had a Black or white interviewer. Race of interviewer data is not available in the ANES cumulative data in 1988 and prior years. 2004 omitted because race of interviewer only coded as white/non-white. 2000 omitted because there is only one Black respondent who was interviewed by a Black interviewer, however, we do include it in the pooled analysis at the bottom of the table.

a

Indicates a difference of proportions two-tailed p-value of .10 or better.

b

Indicates .05 or better.

In Table 5 we present the percentage of validated non-voters who reported they voted by the strength of their feelings of linked fate with their racial or ethnic group. As discussed above, it is possible that because the struggle for African American rights focused so centrally on securing Blacks’ right to vote Blacks feel the need to say they voted in order to show a commitment to upholding group interests. That is, Blacks who are highly racially identified may just be principally committed to the idea that voting is an essential civic duty to their group and thus feel a need to report voting even if they didn’t. Similarly, we should expect that respondents who are not particularly identified with their racial group would feel less of a need to overreport their voting. Using a measure of racial in-group linked fate we see in Table 5 suggestive evidence that Blacks with stronger feelings of linked fate are more likely to report voting when they did not. Here we see that among Black respondents the percentage of validated non-voters who have high levels of linked fate are about 11% points more likely to overreport having voted than those with lower levels of linked fate with their racial group. However, these differences are not statically significant (p > .23). This may be due to the small number cases that we have once we isolate validated non-voters or because the difference is not meaningful. The differences for whites and Hispanics are even smaller at 6-points and 1-points, respectively, both with p-values for the difference above .34, indicating no statistically meaningful difference between the groups.

OPEN IN VIEWER

Table 5.

Percent of Validated Non-Voters Who Reported Having Voted by Race and In-Group Linked Fate.

Race	Some/A lot LF (%)	DK/none/not very much LF (%)	Difference	p-Value	N
Blacks	62.0	50.8	11.2	.230	115
Whites	36.3	30.1	6.2	.340	214
Hispanics	26.2	27.4	−1.2	.880	134
Total	39.6	34.5	5.1	.258	463

As noted earlier, there are a number of other factors that might be related to the race of the interviewer that may also be related to whether or not a non-voter reports having voted when they did not. Using logistic regression and controlling for respondents’ gender, age, education level, partisan strength, and belief in the civic duty to vote, we model whether or not a validated non-voter reports having voted in the 2012 election (“overreports”) for Black, white, and Hispanic respondents as a function of whether or not the interviewer is of the same race and feelings of linked fate with their racial or ethnic group. Table 6 presents the results from the logistic regressions.

OPEN IN VIEWER

Table 6.

Predictors of Vote Overreporting by Racial Group.

Overreporting	Coef	Coef	Coef
	White	Black	Hispanic
Female	−.53	−.48	.23
	.36	.57	.49
Age	.11	.06	−.12
	.15	.09	.10
Education	.30	.57*	.64*
	.16	.27	.25
Strong party identification	.71	1.70*	.42
	.46	.60	.56
Voting a duty a	.59	.81	1.35*
	.48	.58	.57
Same race interviewer	.16	1.81*	−.29
	.54	.59	.51
Linked fate with racial in-group	.12	.70	.01
	.38	.57	.49
Constant	−2.55	−3.35	−2.26
	.82	1.15	−.96
Pseudo R2	.10	.25	.21
N	167	86	118

a

Effect relative to believing voting is a choice.

*

Statistically significant at p < .05, two-tailed test.

The results of the logistic regression confirm the earlier support for our hypotheses (Hypotheses H1a and H1b) about racialized social norms underlying Blacks’ overreporting behavior. Even after controlling for a number of other factors related to overreporting, African-American non-voters are more likely to report having voted when they are interviewed by a Black interviewer. Looking at Table 7, which presents the average marginal effect of each of our central independent variables, we see that the effect of having a same-race interviewer for Black respondents is statistically significant and meaningful, with a marginal effect suggesting a 32-point increase (p-value .000) in the likelihood of overreporting voting with a same-race interviewer relative to a white interviewer. The effect of linked fate in explaining Black overreporting, however, is much smaller in magnitude and fails to achieve statistical significance, suggesting that an internalized belief in Black unity is likely not responsible for Black overreporting. For Hispanic and white non-voters, having a same-race interviewer appears to have no effect, as the coefficient on the same race interviewer variable is small and not statistically significant at traditional levels for either group. Thus, we find additional support for the racialized norms expectations of the Same Race Interviewer and Black Respondent Hypotheses beyond the initial tabulations presented in Table 3, suggesting that Black rates of overreporting systematically differs according to the race of the interviewer. ⁷

OPEN IN VIEWER

Table 7.

Marginal Effects of Predictors of Overreporting by Race.

	ME	p-Value
Blacks
Same race interviewer	0.32	.000
Linked fate with racial in-group	0.12	.200
Civic duty to vote a	0.14	.158
Whites
Same race interviewer	0.03	.767
Linked fate with racial in-group	0.02	.750
Civic duty to vote a	0.12	.230
Hispanics
Same race interviewer	−0.04	.572
Linked fate with racial in-group	0.00	.978
Civic duty to vote a	−0.27	.016

Note. Marginal effects calculated using observed values approach (Hanmer & Kalkan, 2013).

a

Effect of responding “voting is mainly a duty” relative to responding “voting is mainly a choice.”

Despite the lack of an effect of linked fate on Black overreporting, it is plausible that it may be the case that rather than directly affecting the likelihood of non-voters to report voting, the effects of a belief in Black linked fate conditions the effect of the race of the interviewer. This would be the case if, for example, only Black respondents with high levels of linked fate are influenced by the presence of another Black individual, such as when being interviewed by a Black interviewer. Thus, if this is the case, the combination of a Black interviewer and a high degree of linked fate should lead to a greater likelihood of overreporting because being in the presence of another Black individual may be more influential on the decision on whether or not to overreport for those individuals with high feelings of linked fate than those with lower levels. On the other hand, if as we argue this effect is truly the result of social pressure resulting from being in the presence of a member of one’s in-group, we should observe no difference in the effect of having a same versus different race interviewer on overreporting between respondents with both high and low levels of linked fate, as awareness of the group norms and the expected consequences should be equal across each group (No Conditional Effects Hypothesis).

To test for the conditioning effects of Black linked fate on the race of the interviewer, we conduct two separate analyses. First, using the Black respondent model from above, we estimate the marginal effect of having high levels of linked fate on the likelihood of a respondent overreporting voting when their interviewer is Black and when they are white. The results are presented in Table 8. The effect of a respondent holding high levels of Black linked fate relative to low levels on overreporting is statistically insignificant regardless of the race of interviewer. Further, the difference between the two is substantively and statistically insignificant with a difference of 0.02 and p-value of .384.

OPEN IN VIEWER

Table 8.

Marginal Effect of Linked Fate for African American Respondents by Interviewer Race.

Linked fate on overreporting	ME	p-Value
Same race interviewer	0.11	.204
Different race interviewer	0.13	.206
Difference	−0.02	.384

Note. Marginal effects calculated using observed values approach (Hanmer & Kalkan, 2013).

Second, to explicitly test the No Conditional Effects Hypothesis, we modify the Black respondent model and interact linked fate with the variable measuring if a respondent was interviewed by a member of the same race or not. The results from this analysis are in Table 9. Since the output from binary response models is not directly interpretable when using interaction terms (Ai & Norton, 2003), we focus on comparing the marginal effect of having a same race interviewer (relative to a different race) on the probability of overreporting between Black respondents with high levels of linked fate and low levels of linked fate. This comparison yields a statistically insignificant difference (difference of 0.12 in marginal effects, p-value of .499), indicating that for Black respondents, the effect of having a same-race interviewer on overreporting is not conditioned by feelings of linked fate. This suggests that the effects we find of having a same-race interviewer on overreporting for Black respondents is not due to feelings of linked fate, and as the No Conditional Effects Hypothesis suggests, it is also not the result of feelings of linked fate conditioning the effect of a same-race interviewer on overreporting. This is clear evidence of the fact that for Black respondents an interviewer’s race matters regardless of the individuals’ level of linked fate. This suggests that linked fate doesn’t explain Black overreporting in the face of Black interviewers, nor does it condition the effects of interviewer race, and instead it is likely that this phenomenon is the result of social pressures to comply with norms of Black political behavior occurring when in the presence of another member of the racial group.

OPEN IN VIEWER

Table 9.

Predictors of Black Overreporting with Linked Fate Black Interviewer Interaction, 2012 ANES.

Overreport	Coef
Female	−.40
	.58
Age	−.06
	.09
Education	.54*
	.33
Strong party identifier	1.63*
	.60
Black interviewer	1.43*
	.75
Voting dutya	.82
	.58
Black linked fate (BLF)	.38
	.69
Black interviewer × BLF	.95
	1.21
Constant	−3.21
	1.16
Pseudo R2	.26
N	86

Note. ^aEffect relative to believing voting is a choice.

*

Statistically significant at p < .05, two-tailed test.

We provide evidence in support of the Same-Race Interviewer, Black Respondent, and No Conditional Effects Hypotheses, which together demonstrate evidence that one cause of Black overreporting is social pressure resulting from the unique social dynamics of the interview process when a Black respondent is presented with a same-race interviewer. Black non-voters are consistently more likely to overreport voting when interviewed by another member of their race than when interviewed by a member of another race. In addition, we find that the race of interviewer effect is not conditional on the individual’s levels of linked fate. We don’t find race of interviewer effects for white or Hispanic non-voters, suggesting that as expected there is something unique about the norms of Black political participation driving this effect. In contrast to the robust findings on the effect of having a same-race interviewer for Black respondents’ overreporting, we fail to find evidence that racial group ties as measured by feelings of linked fate are responsible for such overreporting, presenting further evidence in favor of our survey-context and social norm-based explanation.

Finally, one potential question regarding our findings is that in addition to the race of the interviewer, the gender of the interviewer may matter for whether or not an individual overreports. This follows from a growing literature suggesting that African American women are unique in their political participation, by both participating at higher levels than men and exhibiting even greater cohesion around the norm of support for Democratic candidates (e.g., Gillespie & Brown, 2019). In addition, other work suggests that gender, race, and linked fate are related (e.g., Stout et al., 2017), with Black men, for example, more likely to identify with their race than are Black women (Simien, 2005). Thus, there may be reason to think that the gender of the interviewer plays a role in explaining overreporting as well. Unfortunately, ANES data are not well-suited for testing these expectations. This is because the vast majority of ANES interviewers are female (86.7% in the 2019 release of the ANES cumulative file), including the 2012 ANES in which almost 81% of post-election interviewers were women.

However, we can offer one initial test from the model with additional covariates presented in Supplemental Appendix E before moving on. One of the covariates added as part of the robustness check is a control for the gender of the interviewer. This variable fails to achieve statistical significance, and an estimate of the marginal effect of having a female versus male interviewer is small and fails to near statistical significance (effect of 0.09, p = .441). We recognize that this test fails to fully analyze the intersection of Black respondents’ gender with the race and gender of the interviewer, however, attempting to impose models designed to appropriately tease out these relationships is impractical with these data. ⁸ Therefore, instead, we call for future research to investigate this question further using different and additional data better suited for this purpose.

Conclusion

For years, the issue of overreporting by Black respondents has puzzled researchers. The lack of sufficient data and theoretical development around the issue of Black political decision-making has greatly impeded efforts to understand the nature of vote overreporting. Using data with relatively large numbers of Black respondents and interviewers we have been able to identify one source of Black vote overreporting: social pressure to conform with norms of voting stemming from the survey interview context. The results presented here provide clear evidence that there exists considerable social pressure among Blacks to overreport voting in the presence of other members of the Black community. These norms, born out of Black Americans’ unique political history that was often centered around the struggle for rights to participate equally in the political process, come with clear expectations for how to behave politically.

These findings should encourage survey researchers to think more carefully about how they design surveys of the Black community. Implied here is that one way to reduce overreporting among Black respondents in surveys would be to limit the use of same-race interviewers. Indeed, it does seem that white interviewers elicit more accurate turnout responses from African-Americans than Black interviewers. However, before we abandon the use of Black interviewers for such tasks it is important to consider why we began to match the race of interviewers and respondents in the first place. In the 1970s and 80s it became clear among survey researchers that Black respondents were providing more racially conservative positions on racial issues when interviewed by a white interviewer. Thus, we instead suggest that in designing and analyzing surveys it is important to think carefully about how the racial context might affect Blacks’ responses to a broad array of questions, not just those on race.

Finally, we believe these findings also speak more broadly to understanding the behavior of Black individuals in day-to-day interactions. The interactions between a survey interviewer and respondent, although an unusual setting, may replicate the real-world interactions Black Americans have with one another on a day-to-day basis. Thus, we might expect that in everyday interactions Black Americans are more likely to overreport their turnout—or possibly, other forms of participation—to another African American than they are to those individuals from another race. As we reference earlier, in-person survey modes are where racial differences in overreporting are most pronounced relative to other modes, such as web-surveys. This is precisely the context in which we would expect social pressure to overreport to exist—particularly that social pressure resulting from same-race interactions that we argue encourages overreporting by African American respondents. Thus, similar effects resulting from social pressure are likely to be most pronounced in every-day in-person interactions between Black Americans, where the threat of norm enforcement is most likely to occur. Therefore, we believe these findings not only provide insight into the phenomenon of African American overreporting, but also serve as a point for understanding the way in which Black individuals may discuss their political participation with others outside of the survey context. Thus, in addition to our call for future work to continue investigating the intersection of interviewer race and gender, we suggest that future work, such as that of White and Laird (2020), should continue to investigate these dynamics outside of the context of the survey interview.

Supplemental Material