Theoretical and empirical literature

Given the pervasiveness of gender inequality and gender-unequal norms across the African continent, it is likely that an interviewer’s gender will play an important role in shaping survey responses. According to the 2024 Human Development Report’s Gender Inequality Index, the region of ‘sub-Saharan Africa’ (as classified by the United Nations Development Programme (UNDP)) exhibits the highest gender inequalities with an average score of 0.565, which is greater than all other regions (Conceicao et al. Reference Conceicao2024: 296).Footnote ² While there is substantial heterogeneity across African countries – with the index ranging from a low of 0.360 in Cabo Verde to a high of 0.679 in Nigeria (based on 2016 data corresponding to the Afrobarometer data used in this article) – Cabo Verde remains highly unequal, exhibiting a 52.8 percentage point gap in the share of seats in parliament and a 11.5 percentage point gap in the labour force participation rate.Footnote ³

In this context, gender inequalities and unequal gender norms across political, economic and social realms likely heighten the salience of interviewer gender and associated norms, shaping respondents’ self-reported attitudes on key issues related to women’s rights and democracy. This is not inconsequential; policymakers and development actors, including domestic political decision-makers as well as multilateral and bilateral aid agencies such as the United States Agency for International Development (USAID), often use public opinion survey results as key indicators and inputs for policies and programmes.Footnote ⁴

In the literature on gender-of-interviewer effects, the existing studies theorize that response bias can arise due to social desirability pressures. Timothy Johnson and Fons van de Vijver (Reference Johnson, Van de Vijver, In Harkness, Van de Vijer and Mohler2003: 194) define social desirability as the ‘tendency of individuals to “manage” social interactions by projecting favourable images of themselves, thereby maximizing conformity to others and minimizing the danger of receiving negative evaluations from them’. In the case of interviewer characteristics, the literature suggests that respondents project these images to appease the interviewer.

Gundula Zoch (Reference Zoch2021) and Lindsay Benstead (Reference Benstead2014) emphasize two models of social desirability that are likely to operate in the context of gender-of-interviewer effects: the social attribution model and the conditional attribution model. Zoch (Reference Zoch2021: 628) writes: ‘The social attribution model argues that respondents attribute values and views to the interviewer that are based on social stereotypes, and solely linked to observable characteristics of the interviewer.’

In contexts where gender inequality is pervasive in social, economic and political realms, it is likely that women lean towards more egalitarian gender views than men. As a result, respondents are more likely to assert egalitarian views when interviewed by a woman and more traditional, conservative gender views when interviewed by a man, conforming to the perceived social norm among men (Zoch Reference Zoch2021: 628). Because this effect hinges solely on the gender of the interviewer, social attribution theory anticipates that male and female respondents are more likely to express more egalitarian views to females and inegalitarian views to males, constituting a direct effect.

The conditional attribution model (also known as the social distance model) extends this first model and posits that respondents adjust their responses based on both the characteristics of the interviewer and themselves, aiming to minimize the perceived relative social distance (Benstead Reference Benstead2014; Zoch Reference Zoch2021). The perceived relative social distance is smaller when respondents and interviewers share the same gender, thus respondents will feel more comfortable and adjust their responses to a smaller extent. However, when there is a gender mismatch, this model predicts that males will offer more egalitarian responses to female interviewers while female respondents will provide more traditional, conservative responses to male interviewers to conform to the perceived social norm (Benstead Reference Benstead2014; Liu and Stainback Reference Liu and Stainback2013; Zoch Reference Zoch2021). To test this model, it is necessary to examine the interaction effect of the interviewer–respondent (mis)match.

While both of these models fall under the social desirability framework, it is crucial to examine the distinctive social psychological mechanisms that activate social desirability pressures in face-to-face interviews. In a review of the literature published between 1997 and 2002 on compliance and conformity, Robert Cialdini and Noah Goldstein (Reference Cialdini and Goldstein2004: 593) highlight research suggesting that ‘individuals avoid or alleviate feelings of shame and fear via public compliance’. Other mechanisms or normative conformity motivations include gaining social approval, building relationships with others or elevating self-esteem (Cialdini and Goldstein Reference Cialdini and Goldstein2004). Both injunctive social norms (what is usually approved or disapproved of) and descriptive norms (what is typically done) are likely to influence behaviour and survey responses, especially when uncertainty is high (Cialdini and Goldstein Reference Cialdini and Goldstein2004). In the context of gender-of-interviewer effects, patriarchal (or otherwise gender-unequal) norms are likely to influence many respondents to adjust their answers due to a combination of social conformity pressures, a desire for social approval or a goal of reducing negative evaluations from a male interviewer.

Although small, the empirical literature has found mixed support for both models. Brady West and Annelies Blom (Reference West and Blom2017: 187) conducted a research synthesis of previous studies on gender-of-interviewer effects on the outcome of quality of survey responses rather than bias. They found that 10 out of 23 studies obtained null effects; nine studies suggested that female interviewers collect higher quality responses, four found that male interviewers do so, while there were three studies where the gender effects were moderated by the respondent’s gender (in line with the conditional attribution model). In Morocco, Benstead (Reference Benstead2014: 369) examined survey items related to gender equality in politics and found support for the conditional attribution model for survey response bias: ‘males reported more egalitarian views to female interviewers’ and sought ‘to reduce social distance with females’. In parallel, Flores-Macias and Lawson (Reference Flores-Macias and Lawson2008) discovered that respondents in cosmopolitan Mexico City offered more egalitarian responses when interviewed by women, and that these effects were stronger for male respondents than female respondents for gender-sensitive items. However, they did not find any differences in the rest of Mexico. Similarly, drawing on panel data from Germany, Zoch (Reference Zoch2021: 633) concluded that survey respondents reported ‘less traditional gender ideologies to female interviewers’, but these results were primarily driven by male respondents and related to questions about gender equality in work and technology.

Findings using the Afrobarometer survey are also mixed. Drawing on data from 20 countries, Lau (Reference Lau2018) found that respondents surveyed by women were less supportive of democratic procedures and more supportive of authoritarianism. Sundström and Stockemer (Reference Sundström and Stockemer2022) found strong support for the direct (social attribution model) effect and limited support for the conditional attribution model based on a single question of whether women should have an equal chance of being elected to political office.

Based on the theoretical literature and the mixed results in the empirical literature, I hypothesized that the presence of a male interviewer (relative to a female interviewer) would lead to a significant underestimation of public support for greater gender equality, as respondents of all genders aim to conform to the predominant gender-unequal norm and appease the male interviewer. Because of the high levels of gender inequality and unequal gender norms in the sample of African countries, I speculate that male interviewers often activate a psychological awareness of the situation and induce respondents towards a socially desirable response (i.e. what they think the average male would want to hear based on societal norms). At the same time, it is also possible that respondents are providing more egalitarian descriptive and normative views when they are interviewed by a woman (which may be in line with what they believe a woman interviewer would find socially desirable). In other words, both male and female interviewers likely generate pressures to conform socially and avoid negative evaluations (in line with Cialdini and Goldstein Reference Cialdini and Goldstein2004).

In line with Sundström and Stockemer’s (Reference Sundström and Stockemer2022) finding for support for equal rights to political office, my conjecture was that these effects would apply to questions about equal rights to owning and inheriting land as well as equal rights to a job. In other words, I anticipated that the models would provide support for both the social attribution model (a direct effect) and the social distance model (an interaction effect), with the effect appearing stronger for male respondents.

My preliminary hypothesis for the stronger effect among male respondents is that they may experience greater social desirability pressure linked to their perception of how their response might influence their social approval (or negative evaluation) depending on whether they align (or clash) with conservative gender norms when speaking to a male interviewer. This pressure is particularly relevant when answering gender-specific questions about men’s rights to own and inherit land or receive preferential treatment in the labour market, as men may feel more compelled to provide responses that they believe align with their male counterpart’s expectations about the dominant unequal social norm. In contrast, while female respondents may also experience social desirability pressures to conform to the unequal gender norm when interviewed by men, the effect is likely weaker due to the absence of a shared gender identity. This hypothesis is a slight revision of the conditional attribution model, because I contend that men are more likely than women to adjust their responses based on both the characteristics of the interviewer and themselves.

In addition, the literature on interviewer effects has typically found that people with less education are more vulnerable to interviewer effects (Blaydes and Gillum Reference Blaydes and Gillum2013; Sundström and Stockemer Reference Sundström and Stockemer2022). Given the significance of traditional social norms surrounding gender inequality and the potential for these norms to diminish with higher education levels, I examine the interaction between the respondent’s education level and the interviewer’s gender (following Sundström and Stockemer Reference Sundström and Stockemer2022). I hypothesize that more-educated individuals will be less inclined to appease the interviewer by conforming to perceived social norms associated with the interviewer’s gender – expressing pro-male positions with male interviewers and pro-egalitarian views with female interviewers.

Previous research on the gender-of-interviewer effect has mostly been limited to gender-specific questions. One exception is Lau (Reference Lau2018), who found that respondents were less supportive of democratic procedures and more supportive of authoritarianism in the presence of a female interviewer (a direct effect). Building on this study, I draw on a larger sample of 34 countries and also include previously unstudied political questions such as feeling close to opposition (versus ruling) parties. In order to construct the variable, I follow Fjelde and Olafsdottir’s (Reference Fjelde and Olafsdottir2024) approach, with a few slight revisions. I code ‘ruling’ party based on which party held executive power at the time of the survey dates in each specific country. If the head of state was an independent (e.g. Patrice Talon in Benin), I coded the parties that supported that candidate as the ruling party.Footnote ⁵

Although theoretical explanations for gender-of-interviewer effects on responses to questions about democracy are limited, I anticipated that, in African countries – where significant gender disparities in political authority and political engagement exist, with men disproportionately occupying positions of power (Rwanda’s parliament being a notable exception) – the gender of the interviewer would significantly influence responses to political questions.Footnote ⁶ Specifically, I contend that male interviewers, rather than female interviewers, are more likely to activate social desirability pressures to align with pro-democratic norms, which are widely endorsed across the continent irrespective of the actual level of democracy in the country.Footnote ⁷ Survey data across African countries have shown that men tend to be more supportive of democracy than women (García-Peñalosa and Konte Reference García-Peñalosa and Konte2014) and have higher rates of political participation (Coffe and Bolzendahl Reference Coffe and Bolzendahl2011).

There are several reasons put forward to explain these gender gaps, including women’s ‘preference for social expenditures such as basic infrastructure (e.g. water supplies), health and education that impact the production of household goods, including children, on which women tend to specialize’ (García-Peñalosa and Konte Reference García-Peñalosa and Konte2014: 105). If women perceive a trade-off between social expenditures on essential public goods and funding for democratic institutions or multiple political parties, they may express lower support for democracy. Alternatively, if they believe that promoting democracy and pluralistic competition hinders their access to these services, they may also be less inclined to support democracy.

Relatedly, another theoretical reason offered is ‘women’s greater risk aversion’ and that women may be more likely to ‘accept traditional roles, prefer the order and security of authoritarian rule and are less willing to accept plurality’ (García-Peñalosa and Konte Reference García-Peñalosa and Konte2014: 105–106). Indeed, based on their regression analyses of the Afrobarometer data, Cecilia García-Peñalosa and Maty Konte (Reference García-Peñalosa and Konte2014: 115–116) conclude that women may perceive the ‘presence of many parties’ as a potential source of conflict, which could explain why women are less supportive of pluralistic democratic competition. In related research, Amanda Clayton and Pär Zetterberg (Reference Clayton and Zetterberg2021: 874) assert: ‘Women may be more risk adverse than men as they choose political careers and thus more likely to affiliate with more established and stable ruling parties.’ Taken together, this suggests that women interviewers may be perceived as less politically engaged, more likely to support stable ruling parties and more focused on household welfare and social stability, such as the absence of political conflict. As a result, both male and female respondents may adjust their answers based on what they assume female interviewers would find favourable, such as weaker support for pluralistic democratic competition or opposition parties.

Because these are not gender-specific questions about women’s rights, I hypothesize that respondents will be made less aware of their own gender and relative position in society vis-à-vis the interviewer’s gender. In other words, any social desirability pressure is less likely to be activated differently for male and female respondents.

In addition, based on the importance of traditional social norms and their likely erosion with greater education, I interact the respondent’s level of education with the gender of the interviewer (following Sundström and Stockemer Reference Sundström and Stockemer2022). I hypothesize that individuals with more education should be less likely to appease the interviewer by conforming to the socially desirable norm of supporting democracy in the presence of a male interviewer or by expressing lower support for democracy and pluralistic competition in the presence of a female interviewer.

Data and method

In this study, I draw on the Round 7 Afrobarometer dataset, which includes 45,823 households across 34 African countries. The survey took place between September 2016 and August 2018 (Afrobarometer Network Reference Afrobarometer Network2017). The countries included in the dataset are: Benin, Botswana, Burkina Faso, Cabo Verde, Cameroon, Côte d’Ivoire, Eswatini, Gabon, The Gambia, Ghana, Guinea, Kenya, Lesotho, Liberia, Madagascar, Malawi, Mali, Mauritius, Morocco, Mozambique, Namibia, Niger, Nigeria, São Tome e Principé, Senegal, Sierra Leone, South Africa, Sudan, Tanzania, Togo, Tunisia, Uganda, Zambia and Zimbabwe.

In each wave, between 1,200 and 2,400 households are interviewed in each country. Within each of these households, only one adult household member is randomly selected to be interviewed. Afrobarometer employs a clustered, stratified, multistage, probability sample design that resulted in a sample composed of 50.06% women and 49.94% men. Afrobarometer’s sampling method adheres to a gender quota by explicitly alternating respondents by gender, ensuring a balanced sample. Round 7 is the first round in which the Afrobarometer Network decided to switch to computer-assisted personal interviewing (CAPI). The chosen male or female respondent was read questions from the screen of a handheld tablet (Afrobarometer Network Reference Afrobarometer Network2017).

One threat to the validity of gender-of-interviewer effects is that male and female interviewers are not randomly assigned and there may be differences between the characteristics of male and female interviewers which could also affect responses and the estimate of the gender-of interviewer effects. The second related concern is linked to the non-random assignment of interviewers to respondents, which could introduce bias if there are systematic differences in the respondents’ characteristics which are correlated with the gender of the interviewer (for example, if female interviewers are assigned to high-educated respondents and male interviewers are assigned to low-educated respondents).

To address these concerns, Sundström and Stockemer (Sundström and Stockemer Reference Sundström and Stockemer2022: 676) conducted interviews with Afrobarometer and its national partner organizations, concluding that ‘gender of interviewer effects in the data do not stem from assignment bias or a systematic skewness in the assignment of male or female interviewers to particular areas or respondents.’ Additionally, they note that because the survey employs a gender quota in its sampling, ‘the assignment of interviewers to respondents should not create gendered distortions’ (Sundström and Stockemer Reference Sundström and Stockemer2022: 676).

I assess these differences in the Round 7 data and, like Sundström and Stockemer (Reference Sundström and Stockemer2022), find very small albeit statistically significant differences in interviewer characteristics (age, education and rural/urban residence) by interviewer gender (see Table A1 in the Appendix in the Supplementary Material). To address this particular threat, all models control for interviewer age, interviewer education and whether the interviewer comes from a rural or urban area. Similarly, Table A2 in the Appendix shows very small statistically significant differences for most of the respondent characteristics that are controlled for in the model.Footnote ⁸ For example, the last rows of the table show that female interviewers are slightly more likely than men to be assigned to urban areas. Given that there are small differences, the model also accounts for these potential confounders including the respondent’s gender, age, education and urban/rural residence.

Moreover, in line with other studies in the literature, there are strong theoretical and empirical reasons linking these factors to the outcomes of interest in this article such as support for women’s rights, support for democracy and partisan identity (Flores-Macias and Lawson Reference Flores-Macias and Lawson2008; Lau Reference Lau2018; Lau et al. Reference Lau, Baker, Fiore, Greene, Lieskovsky, Matu and Peytcheva2017). In addition, the models include covariates for whether one’s spouse is present during the interview (following Flores-Macias and Lawson Reference Flores-Macias and Lawson2008; Lau et al. Reference Lau, Baker, Fiore, Greene, Lieskovsky, Matu and Peytcheva2017; Zimbalist Reference Zimbalist2022) and for whether the respondent believes the interviewer was sent by the government (as opposed to being sent by a non-governmental institution) (following Lau Reference Lau2018; Tannenberg Reference Tannenberg2022; Zimbalist Reference Zimbalist2018).Footnote ⁹ These covariates have been shown to be associated with biased responses for a range of survey questions that overlap with the ones used in this study. Of most importance to this article, these studies have found that the perception of a government interviewer is associated with less support for democracy and greater support for the ruling party, possibly due to fears of repercussions from the perceived government interviewer. (For further description of these independent variables as well as the outcome variables, see Table A8 in the Appendix.) All models include country-fixed effects to control for unobserved confounding factors that vary across countries (e.g. relatively constant or slowly changing national cultural attitudes that may affect gender norms and the outcome variables).Footnote ¹⁰

Table 1 provides descriptive statistics for respondent and interviewer characteristics included in the model and for the outcome variables.Footnote ¹¹ Respondents are 37 years old on average, evenly split between males and females and possess between primary school completed (= 3) and some secondary education (= 4) on average. Respondents are more likely to be rural residents (55%). Furthermore, respondents believe that interviewers are state representatives nearly 40% of the time. A respondent’s spouse is present in 6% of interviews. With regard to interviewer traits, their average characteristics based on the 45,823 household respondents’ data are as follows: interviewers are slightly younger than respondents on average (the average age of the interviewer is 30 years old across all interviews), more likely to be urban residents (56% of interviews are conducted by an urban interviewer), possess between secondary school completed (= 5) and post-secondary qualifications (= 6) and are slightly more likely to be male (52% of interviews were conducted by a male interviewer).Footnote ¹²

Table 1. Descriptive Statistics

Notes: The data are weighted.

Following Sundström and Stockemer (Reference Sundström and Stockemer2022), the analysis has multiple steps. I first report cross-tabular statistics for gender-specific questions, and use a chi-squared test of independence to examine whether those differences are statistically significant. Since the outcome variables are ordered – except for support for democracy, which is dichotomized due to the lack of a natural category ordering, and closeness to opposition parties, which is dichotomized as closeness to ruling versus opposition parties – I use ordered logistic regression models to assess gender-of-interviewer effects.Footnote ¹³ In addition to the tables, I present marginal effects plots to graphically display the gender-of-interviewer effects for the interaction terms.

Findings

Tables 2 and 3 show the cross-tabular results for two previously unstudied questions in the Afrobarometer survey that ask about normative support for women’s equal rights to land and job opportunities. Women respondents are, unsurprisingly, more supportive of equal rights across the board. However, the chi-squared tests of independence show that there are clear gender-of-interviewer effects. In line with Hypothesis 1, respondents are more likely to support gender equality when interviewed by women, and less likely to do so when interviewed by men. Table 2 shows that there are substantively smaller statistically significant row-level differences for female respondents (compared to male respondents) in support of equal rights to land, which aligns with the multivariate interaction effects observed in the subsequent regression models. By contrast, the magnitude of the differences for male and female respondents in support for equal rights to a job are comparable, which aligns with the statistically insignificant multivariate interaction effects observed in the subsequent regression models.

Table 2. Agreement with Equal Rights to Own and Inherit Land

Table 3. Agreement with Equal Rights to Job

The multivariate regression models presented in Table 4, including an appropriate set of controls, provide stronger support for Hypothesis 1 (an unconditional direct effect) for the outcome of equal rights to a job. There is a large negative statistically significant association between male interviewer and support for equal rights (ranging from −0.33 to −0.44 depending on the specification). Based on the smallest coefficient of −0.33, the odds of weaker support for equal rights to a job increase by 1.39 times on average if one is interviewed by a man relative to a woman.Footnote ¹⁴

Table 4. Logit Regressions for Support for Equal Rights to a Job

Notes: ** p < 0.05, *** p < 0.01.

The models are ordered logit models with individuals as the unit of analysis. All models include country-fixed effects. The data are unweighted. Standard errors in parentheses.

For the binary predictors, only being a male respondent has a larger association (ranging from −0.53 to −0.56) with the outcome variable. By contrast, column II shows that there is no support for Hypothesis 2 (a conditional interaction effect). There are no statistically significant differences for the male interviewer effect conditional on the gender of the respondent. This can be seen graphically in Figure 1, where the gaps for female and male respondents roughly mirror each other (comparing lines of the same colour across the two panels). The figure shows that, with all other factors held constant in the model, women have a roughly seven-percentage-point higher chance of strongly agreeing with the statement that women should have equal rights to a job as men when the interviewer is female compared to when the interviewer is male (a change from 41% to 34%). Meanwhile, men exhibit a statistically similar six-percentage-point lower probability of strongly agreeing with the same statement that women should have equal rights to a job as men when interviewed by a woman than when interviewed by a man (the predicted probability to give this answer decreases from 29% to 23%, as shown in Figure 1).

Figure 1. Predicted Average Marginal Effect of Interviewer Gender, by Interviewer Gender

Finally, to evaluate Hypothesis 3, I interact interviewer gender with the respondent’s level of education, and find a positive statistically significant association in column III of Table 4 (see the results displayed graphically in Figure 2). In other words, at higher levels of education, Figure 2 shows that respondents are more likely to agree with support for equal rights to both male and female interviewers, especially for the ‘strongly agree’ response, shown in maroon and yellow (the predicted probabilities rise substantially as education increases). Similarly, the predicted probabilities for ‘strongly disagreeing’ with support for equal rights fall dramatically for both male and female interviewers as education rises (depicted in purple and dark blue). The bars around the point estimates display the confidence intervals, showing that the predicted probabilities are statistically different from one another across most education levels for the ‘strongly agree’ and ‘strongly disagree’ responses.

Figure 2. Predicted Average Marginal Effect of Interviewer Gender, by Respondent’s Education Level

The findings for the second gender-specific outcome of interest, agreement with support for equal rights to own and inherit land, are also interesting. The results in Table 5 provide support for Hypothesis 1 (an unconditional direct effect) and, in contrast to the first outcome, Hypothesis 2 (a conditional attribution effect). There is a large negative statistically significant association between male interviewer and support for equal rights (ranging from −0.10 to −0.30 depending on the specification). For the binary predictors, only being a male typically has a larger association (ranging from −0.35 to −0.52) with the outcome variable. Similarly, columns II and III show that there is support for Hypothesis 2 (a conditional interaction effect). This can be seen graphically in Figure 3, where the slopes of the lines are steeper for male respondents for all response options except for ‘agree’. The figure shows that, with all other factors held constant in the model, men exhibit a roughly five-percentage-point lower probability of strongly agreeing with the statement (shown in yellow) that women should have equal rights to own and inherit land as men when interviewed by a woman than when interviewed by a man (the predicted probability to give this answer decreases from 34% to 29%, as shown in the right-hand panel of Figure 3). By contrast, there is a statistically smaller, roughly three-percentage-point higher chance of women strongly agreeing with the same statement when the interviewer is female compared to when the interviewer is male (44% to 41%, as shown in the left-hand panel of Figure 3).

Figure 3. Predicted Average Marginal Effect of Interviewer Gender, by Interviewer Gender

Table 5. Logit Regressions for Support for Equal Rights to Own and Inherit Land

Notes: ** p < 0.05, *** p < 0.01

The models are ordered logit models with individuals as the unit of analysis. All models include country-fixed effects. The data are unweighted. Standard errors in parentheses.

Because the interaction between interviewer gender and respondent gender was statistically significant in column II, I also ran a triple interaction between these terms and respondent education (presented in column III). While the triple interaction coefficient is not statistically significant, the two-way interaction between interviewer gender and respondent gender remains statistically significant and doubles in magnitude (in line with Hypothesis 2), and the interaction between interviewer gender and education is statistically significant and positive (in line with Hypothesis 3). This latter association is greater in magnitude in column IV, when the triple interaction term is omitted. I present the results graphically of this two-way interaction effect in Figure 4. The figure demonstrates that, at higher levels of education, all respondents are more likely to ‘strongly agree’ with support for equal rights in the presence of both male and female interviewers. The predicted probabilities for the ‘strongly agree’ response (represented in maroon and yellow) increase significantly as education levels rise. Meanwhile, as respondent education increases, one can also observe that the predicted probabilities for disagreeing with support for equal rights decline for both male and female interviewers.

Figure 4. Predicted Average Marginal Effect of Interviewer Gender, by Respondent Education Level

To summarize, the analyses supported a direct unconditional effect (Hypothesis 1) and a declining gender-of-interviewer effect with increasing respondent education (Hypothesis 3). There was support for the conditional attribution effect (Hypothesis 2) for the equal rights to owning and inheriting land outcome, but not for the equal rights to a job outcome. I interpret the divergent results on the conditional effects hypothesis as follows: due to the significant gender inequality regarding women’s rights to own and inherit land,Footnote ¹⁵ male respondents may perceive that they have more at stake when answering this question in the presence of a male interviewer. This dynamic could lead them to respond in ways that align more closely with traditional gender norms, especially if they believe their responses might weaken or reinforce their standing within those norms.

In contrast, for the question regarding equal rights to a job, male and female respondents do not appear to be influenced differently by the gender of the interviewer. I speculate that this outcome reflects the relatively balanced labour force participation rates between men and women, particularly in sub-Saharan Africa, which has one of the highest rates of female labour force participation worldwide. According to UNDP data, the female labour force participation rate (ages 15 and older) in 2016 – the year of the Afrobarometer survey – was 60.8% in sub-Saharan Africa, surpassing all other regions identified by the UNDP, including the ‘Very High Human Development’ category, which had a rate of 53.6%.Footnote ¹⁶ Still, there are major gender disparities in types and quality of work, with women more likely to be employed in lower wage, informal, or self-employment categories (Woldemichael Reference Woldemichael2020). Because this outcome exhibits relative parity – making the descriptive norm more gender-equal in stark contrast to land rights – male respondents may feel less pressure than in the land rights question to align their answers with what they believe their male interviewer counterparts find socially desirable. As a result, their responses do not differ significantly from those of female respondents, as both groups state more gender-unequal attitudes with male interviewers and more gender-equal attitudes with female interviewers.

Table 6 presents the results for five questions about support for democracy, one question about whether the respondent feels close to opposition parties (versus ruling parties), and a question about whether people can disobey laws of governments they did not vote for. The gender-of-interviewer effect was statistically significant for all outcomes except for one: rejection of one-man rule. In line with Hypothesis 4, these effects reveal that, in the presence of a male interviewer, respondents consistently overstate their support for democracy, rejection of military rule or one-party rule and support for freedom of movement as opposed to supporting government restrictions on movement in the interest of public security. Similarly, there were statistically significant gender-of-interviewer effects for two previously unstudied questions about feeling close to an opposition party, and whether one should obey government laws even if the respondent did not vote for that government (presented in Table 6).

Table 6. Logit Regressions for Support for Democracy and Opposition Parties

Notes: ** p < 0.05, *** p < 0.01

The models are logit (when binary dependent variable) or ordered logit models (when ordinal dependent variable) with individuals as the unit of analysis. All models include country-fixed effects. The data are unweighted. Standard errors in parentheses.

In the presence of a male interviewer, respondents asserted higher levels of closeness to opposition parties (versus the ruling party) and were more likely to agree with a statement that one does not need to obey the laws of the government if they didn’t vote for it. It is important to highlight that all of these gender-of-interviewer effects operate in the opposite direction of the perceived government interviewer effects. Specifically, male interviewers elicit responses biased towards support for democracy and close affiliation with opposition parties, while perceived state interviewers prompt biases favouring non-democratic governance and close ties to ruling parties.Footnote ¹⁷ These results lend some support to the notion that respondents seek to align with the injunctive norm of pro-democracy and pluralistic competition, which male interviewers – who are more likely to be involved in politics and are perceived to value democratic competition – may make more salient. Meanwhile, respondents interviewed by women may instead adjust their answers to align with what they perceive to be socially desirable to female interviewers – prioritizing social and political stability, even if it means expressing less support for democracy or opposition parties.

The magnitudes of these effects are also non-trivial. The odds of stronger support for democracy increase by 1.38 times on average if one is interviewed by a man relative to a woman.Footnote ¹⁸ With regard to feeling close to an opposition party, an individual is 1.21 times more likely on average to state that they feel close to an opposition party if they are interviewed by a male compared to a female.Footnote ¹⁹

While there are direct, unconditional effects associated with interviewer gender, no statistically significant differences were found between male and female respondents when the interaction between respondent and interviewer gender was included. The inclusion of these interaction terms does not alter the direct, unconditional effects. I present those results in Appendix Table A4.

Finally, there is mixed support for Hypothesis 5. The interaction term between interviewer gender and respondent education is statistically significant for less than half of the outcomes (three out of seven), suggesting that more-educated respondents do not consistently show less susceptibility to social desirability pressures to align with prevailing norms in favour of democracy and support for opposition parties when interviewed by men (see Table 6).