Elder, Am I Right? Age-Group Differences in Social and Political Interactions in Africa

Abstract

Age-group differences play an essential role in social interaction across sub-Saharan Africa. However, the social effects of these differences remain understudied. We hypothesize that age-group differences will affect response patterns and use Afrobarometer data to test this hypothesis. We also explore three mechanisms through which age-group differences may induce response-pattern variation: in-group loyalty, social acquiescence, and social distance. As hypothesized, we find relatively large and statistically significant effects for age-group differences across a variety of questions. Our findings support in-group loyalty and social acquiescence rather than social distance when questions do not address age-related issues directly. However, social distance may play a more important role when questions address age-specific issues. Additionally, we show preliminary evidence that age-group differences induce larger response-pattern variation than coethnicity. Our findings speak to the importance of age in social interaction in Africa and provide important lessons for the survey research community.

In a village in Sierra Leone, a 22-year-old interviewer announces her presence to conduct a face-to-face survey. A woman in her sixties emerges from the house and agrees to participate in the study. Would her answers be the same if the interviewer had also been in her sixties?

More generally, survey respondents may systematically vary their answers to politically and socially important questions based on many aspects of the survey enumeration process. A well-known potential source of survey response variation emerges from the interaction between interviewers and respondents, which has been investigated in the context of racial (Davis, 1997a; Hatchett & Schuman, 1975; Williams, 1964), ethnic (Adida et al., 2016 ; Weeks & Moore, 1981), religious (Benstead, 2014; Blaydes & Gillum, 2013), gender (Benstead, 2013; Flores-Macias & Lawson, 2008; Huddy et al., 1997), and physical (Eisinga et al., 2011) differences between interviewers and interviewees.

We investigate an understudied element of social interaction: age-group differences. Age-group differences are theoretically important to study because they shed light on social hierarchies and positionality in diverse societies, as well as how they shape respondents’ responses to survey questions. These interactions also suggest concrete ways researchers might choose to organize their survey enumeration teams.

We seek to address one primary and one secondary research question. Primarily, to what extent do age-group differences shape response patterns? Secondarily, what support exists for the underlying mechanisms that may generate these age-group-difference effects?

We argue that, in the African context, there are good theoretical reasons to believe that when two interlocutors are talking, and one person poses a question, the respondent may modify his or her responses to socially relevant questions depending on whether there is an age-group difference. To preview our argument, we hypothesize that the marginalization of African youth and rapid social change across the African continent means that age has become an increasingly salient social identity category and may lead respondents to change their responses to questions depending on the age group of the person posing the question.

While many nonmutually exclusive mechanisms may be at play, we discuss three through which this age-group difference manifests itself. First, according to in-group loyalty, both young and old interlocutors might vary their response patterns when asked questions by people whose age differs from their own to reinforce their social position (i.e., to make their own group look more favorable). Second, social acquiescence posits that only socially nondominant interlocutors—in the case of age-group difference, respondents who are young—might change their response patterns when being asked questions by old, socially dominant interlocutors. Third, social distance suggests that those responding to questions might give the answers they think the individual posing the question wants to hear.

To examine age-group differences, we leverage the Afrobarometer surveys, which provide the age, gender, and ethnic affiliation of all its survey enumerators and respondents starting in Round 3 (2005/2006). Since the interview process itself is a social interaction, the Afrobarometer data allow the unparalleled ability to cross-nationally study age-group differences across Africa.

We conduct three sets of analyses. First, since similar studies have investigated response-pattern variation in the context of ethnicity (Adida et al., 2016), we investigate the effect of age-group differences in the interview process on the same variables and for the same years (Rounds 3 and 4) as the study of coethnicity to benchmark the relative importance of age-group differences and coethnicity. Second, we leverage Round 7 (2016–2018) of the Afrobarometer data, which contains one question about youth across the continent and country-specific questions in Mauritius. In these Round 7 analyses, we investigate the effect of age-group differences on questions we believe would most likely be sensitive to age-group differences: youth-specific questions. Third, we conduct a robustness replication of a previous paper that uses cross-national Afrobarometer data (Gottlieb et al., 2018), where we include our age-group difference variables in the specifications.

To preview our findings, in line with our hypothesis that age-group differences matter, we often find statistically significant and substantively relevant effects of age-group differences between the interviewer and respondent on how the respondents answer questions. While they are certainly not dispositive, these findings provide some support for the mechanisms behind the three theories. We find support for in-group loyalty, particularly on questions related to objective economic well-being, whereas there is the most support for social acquiescence on politically related questions. We only find support for social distance when the survey questions directly address the topic of age. Additionally, we find that for many variables, the effect of at least one of the estimates measuring interviewer–respondent age-group difference is substantively larger than the estimate of having a non-coethnic interviewer. In replicating other previously published work, we continue to find the importance of age-group differences but do not find that these covariates change the authors’ substantive findings.

We contribute to the literature by empirically measuring how young and old people change their views when interacting. Showing these differences sheds light on how diverse social mechanisms can influence respondents’ replies to questions. For practitioners in the sub-Saharan African context, we suggest that deploying only teams of young interviewers with appropriate sensitization may be the most optimal strategy.

The rest of the paper proceeds as follows. We first provide a conceptual background for understanding age-group differences in the survey enumeration process as a social interaction and lay out the empirical expectations associated with the three theories we discuss. Second, we explain the Afrobarometer data we use for the study and provide summary statistics. Third, we provide our results benchmarking age-group differences with coethnicity difference from Afrobarometer Rounds 3 and 4. Fourth, we discuss replication results. The article concludes with a discussion of our results and their implications.

Conceptual Background

Youth in Africa often see themselves as marginalized and socially distinct from their elders, both in rural and urban settings. Yet, how youth and their elders perceive each other and how these perceptions shape African public opinion remains understudied. Indeed, as Jon Abbink writes, “The cognitive dimension of age and generational difference in African societies is underestimated” by social science (2005, p. 24). These cognitive dimensions are important because they help to understand the changing views in African societies on key issues such as views on gender equality, respect towards elders, and citizens’ relationship to governments and democracy.

This high level of social and political distinctiveness between age groups has emerged in Africa from a variety of sources, many related to the difficulty of achieving traditionally idealized concepts of adulthood (Dawson, 2014). Adulthood is often related to marriage and employment (Juárez & Gayet, 2014). Youth unemployment and underemployment are particularly common across the African continent (Honwana, 2014), and affording marriage has become more difficult (Masquelier, 2005). The inability to obtain these signs of adulthood has made youth see themselves as increasingly socially and politically distinct from their elders.

Despite their inability to acquire traditional signposts of adulthood, youth maintain high aspirations to achieve upward social and economic mobility through education and internal migration (Kabiru et al., 2013). In addition, youth population bulges and growing inequality have exacerbated intergenerational tensions, as many youths feel rejected by their elders, whom they often see as corrupt and as not providing space for their voices within the political system (Honwana, 2012). These youth often seek out alternative forms of action and participation outside of mainstream politics (Diouf, 2003; Van Gyampo & Anyidoho, 2019).

Youth marginalization in the African context may lead to certain types of response-pattern variation, both in general social interaction and in the survey enumeration process. Because of their marginalized position in society, youth may behave in a systematically different manner when they are asked their opinions by elder society members compared to when they are asked their position by someone of the same age group. Old individuals can also reinforce their position of dominance when discussing with young people and, therefore, might speak young people differently than when interacting with people from their own age group.

To study response-pattern variation, we conceive of age-group difference as a dyadic effect in that it does not depend only on the age of one interlocutor but on the ages of both parties in a conversation. We highlight that this differs from most scholarship in African politics that considers only the age of the respondent (e.g., Resnick & Casale, 2011) or other social science research that looks only at interviewer characteristics (the direct interviewer effect), including age (see Hox et al., 1991; West & Blom, 2017).

To interpret these dyadic effects, we draw on extensive public opinion literature that suggests that while respondents will always potentially vary their answers depending on whom they are talking to, respondents are generally more comfortable answering questions when the person posing the question has similar demographic features (Johnson et al., 2000).

From a causal framework, we can think of a survey (or other discussion) assignment process whereby a respondent is assigned an interviewer in a group that is either socially close to or far from them. We consider an interviewer–respondent dyad as socially close when the two interlocutors come from the same age group. For young respondents, this means they will be socially close with young interviewers and socially far from old interviewers. For old respondents, this means they will be socially close to old interviewers and socially far from young interviewers. Hence, the comparison group for old respondents (old interviewers) differs from the comparison group for young respondents (young interviewers). We show this in Table 1. We hypothesize that these age-group differences will systematically change response patterns.

The psychosocial mechanisms through which respondents situationally vary their answers are complex and often yield observationally equivalent effects. Moreover, these mechanisms may also be nonmutually exclusive such that two mechanisms can simultaneously drive the results. We, therefore, do not conduct hypothesis testing directly on these mechanisms but seek to qualitatively investigate the extent to which evidence aligns with these different mechanisms. While there are different ways to classify these dyadic mechanisms, Benstead (2013, 2014) distinguishes between mechanisms of power relations and those of social desirability.

Power relations mechanisms involve respondents first understanding their own social positionality regarding the interviewers. Because respondents are concerned with social hierarchy, they will be motivated to respond in ways that reinforce that hierarchy. In the case of age-group differences in surveys in Africa, we assume that old interviewers are socially dominant and young interviewers are socially nondominant.

On the other hand, social desirability mechanisms involve respondents seeking to conform to the opinion they expect the interviewers to hold, regardless of their social status. In this paper, we propose two power relations mechanisms and one social desirability mechanism (social distance) that may be at play with respect to age-group difference.

We first discuss two different submechanisms of power relations theory about social interaction: in-group loyalty theory (Benstead, 2014; Blaydes & Gillum, 2013)¹ and social acquiescence theory (Carr, 1971; Davis, 1997b; Lenski & Leggett, 1960).²

The first power relations mechanism comes from in-group loyalty theory, which posits that respondents will agree with the stereotype of their group when posed a question since changing their response patterns can help them signal their social position within the social hierarchy (Benstead, 2014).³ As in-group loyalty theory pertains to the survey enumeration process, respondents from a socially dominant (e.g., old) group can signal their position of dominance by changing their response pattern to reinforce their dominant position in the face of a nondominant interviewer. Socially nondominant respondents can also reinforce their stereotyped socially inferior position when talking to a socially dominant interviewer.

For example, let us imagine two Kenyan schoolteachers. The 25-year-old interviewer asks the 45-year-old respondent whether she has faced financial hardship in the past year. On average, in situations like this, in-group loyalty would predict that old respondents would be less likely to say they suffered financial hardship when questioned by young interviewers, compared to if they interacted with old interviewers. In other words, old (i.e., more socially dominant) individuals would vary their responses so as not to reveal they had faced financial hardship to young interviewers because their stereotyped position is that they are wealthier. If the respondents were young (25 years old) and the interviewers old (45 years old), in-group loyalty would predict that the young teachers would reinforce that they were less well off by saying they had faced financial hardship (when they would not have said so to a teacher of the same age).

Second, the social acquiescence mechanism within power relations theory suggests that respondents in socially nondominant positions will be more likely to vary their response patterns when being interviewed by a socially dominant interviewer. As this mechanism pertains to the survey enumeration process, if the interviewer comes from a socially dominant group, respondents from socially nondominant groups might edit their answers to align with what they think socially dominant interviewers want to hear (Williams, 1964). However, according to the social acquiescence mechanism, members of the socially dominant group are unlikely to edit their answers when interacting with an interviewer from a nondominant group, as they do not need to show deference to the nondominant interlocutor.

For example, let us now imagine an interview carried out by a 45-year-old Kenyan schoolteacher employed as a survey enumerator. The 45-year-old teacher contacts a 25-year-old schoolteacher to carry out an interview. The interviewer asks whether the respondent supports recent street protests. In situations like these, we might suspect that young Kenyan respondents (such as our schoolteacher), who counterfactually would have said they supported the street protests if they were talking to interviewers of the same age, might modify their responses to say they do not support the street protests when interviewed by old interviewers because the young respondents expect the old interviewers posing the question not to support the street protests.⁴ However, if the ages of the interviewer and the respondent were reversed and a 45-year-old teacher (who is now the respondent) did not support protest, under the social acquiescence mechanism, we would not expect her to change her response regardless of the interviewer’s age. That is, it would not matter whether she was talking to a 45-year-old teacher (an age peer) or a 25-year-old teacher because her position of social dominance would allow her to state their belief freely.

Within social desirability, we focus on one often-discussed mechanism: social distance. The social distance mechanism suggests that regardless of the social hierarchy of the two interlocutors, the respondent who is being questioned will want to conform to the imputed behaviors and attitudes of the interviewer posing the questions when the respondent and interviewer are socially distant. We assume that when there is an age mismatch between the interviewer and respondent, they are more socially far apart from each other than when the respondent and interviewer are of similar ages. Thus, we would expect respondents with an interviewer of a different age to vary their responses compared to when they are being interviewed by someone of the same age to be more in line with the views the respondent imputes the interviewer to possess.

These three mechanisms are laid out in Table 2. As seen in the first row, we first highlight that in-group loyalty suggests the effects of age-group differences will be differentially signed for your young and old respondents, which is encapsulated in Expectation 1.

E1: Differentially signed effects of having an interviewer from a different age group for young respondents (negative) and old respondents (positive) will provide evidence for in-group loyalty.

Indeed, in-group loyalty is the only theory that would predict the different signs, which, therefore, distinguishes it from the other two theories.

Social acquiescence and social distance should yield similar predictions for young respondents. Unlike in-grouployalty, under both mechanisms, estimates for young respondents should be positively signed. However, as shown in Expectations 2 and 3, they would yield different predictions for old respondents (no effect for social acquiescence and a negative effect for social distance).

E2: Only positively signed effects of having an interviewer from a different age group for young respondents but not for old respondents will provide evidence of social acquiescence.

E3: Differentially signed effects of having an interviewer from a different age group for young respondents (positive) and old respondents (negative) will provide evidence of social distance.

Data and Method

We use Afrobarometer data from sub-Saharan Africa for all our analyses. To facilitate interpretation and benchmark our results, we use linear models and present the outcomes in standard deviation units. Supplementary Tables G3, G6, G9, G12, and G15 present the regression results for logistic and ordered logistic models, which yield nearly identical results.⁵ We also reverse the scales on some questions, such that the socially dominant position is always the higher value on the x-axis.

In our first set of analyses, we use the same Afrobarometer data set (from Rounds 3 and 4) used by Adida et al. (2016), which contains 38,381 responses from 14 countries (see Supplementary Tables A1 and A2 for the complete list). Adida et al. (2016) study coethnic interviewer–respondent dyads, but they do not examine or control for age-group differences. To conduct a comparable analysis and avoid fishing for results, we continue to use the same 28 outcome variables across four question categories used by Adida et al. (2016). These include three sets of questions posed to respondents on economic, political, and ethnic topics.⁶ The fourth set of questions focuses on interviewer self-assessments of the interviews. All these variables are either binary outcomes or variables on Likert scales with between three and five options.

In our second set of analyses, we focus on outcome variables that specifically address youth issues and, therefore, might be more likely to have age-group effects. Round 7 (2016–2018), which Afrobarometer carried out across the continent in 31 sub-Saharan African countries and contains 42,190 responses (see Supplementary Table A3 for the complete list), asks: “How well or badly would you say the current government is handling the following matters, or haven’t you heard enough to say?” One of the matters was: “Addressing the needs of youth.” This question is on a 4-point Likert scale from “very badly” to “very well.”

Additionally, in Mauritius, one of the countries least studied in political science (Wilson & Knutsen, 2022), the Afrobarometer posed further questions on the same Likert scale about the government’s handling of youth unemployment.⁷ Moreover, the issue of youth unemployment is particularly pressing in Mauritius where approximately a quarter of youth are unemployed (Tandrayen-Ragoobur & Kasseeah, 2015).

In Mauritius, the Afrobarometer added six questions about the government’s handling of youth-related issues: drug abuse, teenage pregnancy, underage consumption of alcohol, smoking, and youth delinquency. These issues mirror concerns of the Mauritian government and international bodies, which stem from societal changes in Mauritius as it became a major drug trafficking hub. As a result, socially undesirable behaviors, which particularly affected youth, increased (Rambaree, Mousavi, & Ahmadi, 2018). The descriptive statistics for all outcome variables we use in both our first and second analyses pooled and by respondent age group are provided in Supplementary Tables B1, B2, C1, and C2.

The Afrobarometer data contain the ages of the interviewers and respondents from Round 3 onward.⁸ From this information, we construct our main independent variable for all models, which is a four-category nominal variable. This variable is created in two steps. First, we independently divide both interviewers and respondents into two groups: those who are 35 or under as “young,” and those who are older than 35 or “old.” We choose the cutoff of 35 years old, a typical designation of youth in the African context (Abbink, 2005, p. 6).⁹ We then cross-classify all respondent–interviewer combinations, which align with Table 1.¹⁰

Table 3 shows the descriptive statistics for respondents’ and interviewers’ ages, and Figure 1 shows the proportions of the sample that fall into the four categories in Table 3. Not surprisingly, given that the distribution of Afrobarometer interviewers skews much younger than the rest of the population distribution, the two most common categories are the one in which both dyad members are in the young (35 and under) age group (53%) and the one in which the interviewer is in the young group (35 and under), and the respondent is in the old (over 35) age group (34%).

An ideal design might involve randomly assigning respondents to interviewers from the same or different age groups. However, for many good reasons, Afrobarometer does not randomly assign interviewers to respondents. Adida et al. (2016) document reasons for nonrandom assignment of coethnic interviewer–respondent dyads, including the need for interviewers who can speak the language the respondents most often speak at home and the desire to minimize contentious interethnic relations.

Unlike coethnic dyads, Afrobarometer’s protocols for assigning interviewers to certain localities, selecting households, and selecting respondents within households should not result in similar levels of systematic bias in terms of which respondents are assigned someone from the same age group and which are assigned someone from a different one. Yet, there might be some deviations from random assignment. For example, if interviewers are deployed in teams, young interviewers, who might be more willing to travel, might be sent to more remote settlements with predominantly higher numbers of old people (because the youth left for jobs in the city). It is also theoretically possible that individuals’ potential willingness to open the door and consent to a survey varies based on their age match with the interviewer.

To check for such possible biases, we compare means on several key indicators that are unlikely to be biased by interviewer–respondent age matches. Tables 4 and 5 show balance on a range of observables comparing the means for respondents when they were interviewed by someone in the same age group and someone in a different age group.

As seen in Tables 4 and 5 (see Supplementary Table B3 for Mauritius only), there is a substantively small but statistically significant imbalance across seven and eight of the nine demographic variables in at least one of our two paired comparisons in Rounds 3 and 4 (Table 4) and in Round 7 (Table 5), respectively.¹¹ The variable in-home language (a dummy variable indicating whether the interview was conducted in the language the respondent most often speaks at home) stands out as being particularly large, most notably in Rounds 3 and 4.

With respect to home language, among old respondents, both Tables 4 and 5 show a higher proportion of interviews conducted in the home language of the respondent when the interviewer and respondent were in the same age group, compared to when the interviewer was in the young group. The opposite pattern emerges with young respondents in Rounds 3 and 4 (but not Round 7)—a lower proportion of interviews was conducted in the respondent’s home language when the interviewer and respondent were in the same age group, compared to when the interviewer was in the old group.

Beyond random variation, there are several possible, nonmutually exclusive explanations for the fact that in-home language is not balanced. It could be that multilingual old interviewers speak in-home languages more fluently than multilingual young interviewers, and therefore, multilingual respondents choose to speak with old interviewers more often in in-home languages, either to mimic the interviewers’ use of in-home languages or because respondents prefer to speak in-home languages. It could also be that old interviewers, who may have more facility with in-home languages, are deployed systematically at a local level to target respondents who may be more comfortable with in-home languages, leaving young interviewers, who may have less facility with in-home languages, to conduct relatively more interviews in national languages.

For all four sets of outcomes, in addition to a four-category age-group difference variable and a dummy variable for coethnicity, following Adida et al. (2016), we run linear regression specifications with robust standard errors and control for age of respondent, sex, education, urban/rural location, ethnic regional minority,¹² and in-home language. Additionally, we include fixed effects for the Afrobarometer round and country.¹³ This regression specification is represented by the model shown in the equation below, where the α coefficients are those representing our age-group differences; Both Old is the excluded category, α₁ is the coefficient for Interviewer Young (D₁), α₂ is the coefficient for Both Young (D₂), and α₃ is the coefficient for Interviewer Old (D₃). Additionally, ρ is the coefficient for coethnicity (C); $Xi⊤β$ represents the vector of control variables, and their associated coefficients; γ_r represents the fixed effects for Afrobarometer rounds, δ_c represents the fixed effects for countries, and ε_i is the error term.

Our primary quantities of interest in all model specifications are two linear contrasts based on our age-group coefficients and the coefficient on the binary coethnicity variable. More specifically, for the age-group differences, the first contrast compares old respondents interviewed by old interviewers (control) to old respondents interviewed by young interviewers (treatment), which is captured by the coefficient α₁. The second contrast compares young respondents interviewed by interviewers from the same age group (control) to young respondents interviewed by old interviewers (treatment), captured by the difference between coefficients two coefficients (α₃ – α₂). See Supplementary Tables E1–E10 for the tabular version of these estimates and Supplementary Tables G2, G5, G8, G11, and G14 for the underlying regression tables.

To preview our findings, as hypothesized, age-group differences between interlocutors change response patterns: across all 28 questions, we find that 42 out of the 56 coefficients (75%) are statistically significant, and many of these are greater than 0.1 SD units, which is larger than many other standardized estimates of the effects of interviewer–respondent differences (e.g., gender, ethnicity) on response patterns. Additionally, at least one of the coefficients on age-group differences is substantively larger than the coethnicity effect in 23 out of 28 cases, and coefficients on age-group differences are 18 percentage points more likely to be statistically significant (57% of coethnic estimates vs. 75% of age-group difference estimates).¹⁴

Results

Age-Group Differences: Replication Results From Round 3 and 4

Economic and well-being questions

The Afrobarometer poses a series of questions related to respondents’ views of the overall economy as well as economic and physical well-being. Do responses vary depending on the interaction between respondent and interviewer age group? Much research in American politics has shown that even when questioned about objective realities, subjects may vary their answers based on a wide variety of factors (Bullock & Lenz, 2019; Prior et al., 2015). Therefore, respondents can also vary their answers to such questions.

Figure 2 plots our two sets of contrasts of interviewer–respondent dyads along with estimates of the effects of coethnicity, showing that age-group differences shape response patterns as hypothesized. The figure, scaled such that higher values are those more associated with the socially dominant group, provides evidence in favor of E1 derived from the in-group loyalty mechanism. In six of eight questions, old respondents report a position that reinforces their position in the social hierarchy, and in all questions, young respondents, on average, give lower responses to an old interviewer compared to a young one. Of these 14 estimates in line with the theory, 10 are statistically significant. The evidence for this pattern is most substantively large (and statistically significant) for respondents’ assessments of not fearing crime in their own home (EST = 0.119, SE = 0.027 old/EST = −0.131, SE = 0.027 young) and saying their own living conditions are good (EST = 0.072, SE = 0.023 old/EST = −0.119, SE = 0.023 young). These findings align with the idea that young respondents’ answers, on average, reinforce their socially inferior position to old interviewers.

Figure 2 shows these results are weakest on the question about knowing someone who has died of AIDS. Under in-group loyalty theory, there is no reason to expect age-group effects on this question because it is not clear that knowing or not knowing someone who has died of AIDS is a mark of one’s status in a social hierarchy, in the way that economic security is. Finally, we also note that in each of the seven questions focused on economic status (we exclude the AIDS question), at least one age-group difference effect is larger than the coethnicity effect.

Politics questions

In this section, we present the results of political questions. With economics questions, the dominant position is clear. To be able to evaluate whether respondents are adjusting their responses to highlight their dominance for political questions, we need to identify the socially dominant positions. Regarding politics, we suggest that socially dominant positions—those held by old society members—generally reinforce the institutions of the current political system. In all the sampled countries, this is multiparty electoral democracy, albeit with electoral authoritarian traits in some countries, with many elements of a free market economy. Therefore, regardless of who is interviewing them, we would expect old respondents to trust all incumbent political parties and government officials more than young respondents because they represent institutions of the current political system. They should also, therefore, distrust opposition parties more and show relatively higher levels of support for the executive (president) and for democracy. Finally, we would expect old respondents to report higher levels of civic involvement. Supplementary Table C1 demonstrates that the data generally support these suppositions.

Again, for these political questions, we plot our two sets of dyadic age contrasts and the estimates of interviewer coethnicity for these political questions, shown in Figure 3, on the same scale of standard deviation units as reported by Adida et al. (2016). As in the economics questions, as hypothesized, age-group differences in the interview process is associated with variation in response patterns by respondents.

While certainly not dispositive, these findings appear most in line with social acquiescence theory (E2), since, more often than not (in seven out of nine questions), young respondents increase their evaluations to be more in line with views held more commonly by respondents over 35, and six of these differences are statistically significant. Old respondents, on the other hand, mostly slightly decrease or do not vary their responses at statistically significant levels (with the exception of distrust in opposition parties and exposure to vote buying). Notably, at statistically significant levels, young respondents say they have a greater preference for democracy (EST = 0.147, SE = 0.024), have a higher interest in community affairs (EST = 0.126, SE = 0.025), attend more community meetings (EST = 0.081, SE = 0.023), trust the ruling party more (EST = 0.06, SE = 0.02), and approve of the president’s performance (EST = 0.053, SE = 0.025).

Our estimations yield one finding that we do not believe is explained by social acquiescence, in-group loyalty, or social distance mechanisms. This finding addresses a core issue in political science: exposure to vote buying. Old respondents are more likely to report exposure to vote buying to a young interviewer (relative to an interviewer of the same age). While we show standard deviation units in Figure 3 (EST = 0.125, SE = 0.028), this represents a 0.12 increase in reported vote buying on a 0–3 scale.

This finding has important implications for survey research in Africa and bears further scrutiny. Questions related to vote buying are generally suspected to be subject to underreporting (Kramon, 2016). Those who engage in vote buying generally know it is socially undesirable and are often aware it is illegal and, therefore, may be hesitant to answer a survey question honestly (Erlich, 2020). Our results suggest that reporting of vote buying may vary depending on who is asking the question. One explanation is that old respondents may feel more comfortable divulging socially unacceptable behavior, such as vote buying or not voting, to young interviewers compared to interviewers of the same age.

The question related to vote buying brings attention to the potential for response bias in estimates highlighted by other scholars. If old respondents are more honest with young interviewers about this behavior, then this provides evidence to generally recruit young interviewers for the survey enumeration process. We return to this point in our discussion below.

While we focus on the importance of age effects, we again benchmark them against ethnicity. Overall, for these political questions, at least one age effect is larger in magnitude than the coethnicity effect on all but one question (knowledge of MP’s name). Additionally, 63% (14/22) of the estimates are significant on our age variables relative to 6/11 (54%) of the coefficients on coethnicity.¹⁵ This finding suggests that age-group differences may affect social discussions more than coethnicity. At a minimum, it suggests that the age of interviewers may be important to consider when Afrobarometer chooses interviewers.

Ethnicity questions

What about questions related to ethnicity itself, where ethnic differences will most likely have an effect? On this set of questions, it is unclear what the socially dominant outcomes for old versus young individuals are, and the extent of the socially dominant position may vary by context, so our theoretical framework does not yield ex-ante predictions. In their original specification with more fine-grained ethnic control variables, Adida et al. (2016) only find that three of seven of these questions are statistically significant. Moreover, many of the effects are estimated to be substantively null. In our specification, even relaxing these controls, we find similar results for coethnicity (see Figure 4 and Supplementary Figures D5 and D6).

Similar to coethnicity, for the estimates of the effects of age-group differences, we find four questions with substantively large and statistically significant results. This again supports our hypothesis that age-group differences are related to response patterns. For the first two questions, old respondents are more likely and young respondents are less likely to identify nationally (EST = 0.123, SE = 0.025 old/EST = −0.084, SE = 0.025 young) and positively rate their ethnic group’s economic conditions (EST = 0.081, SE = 0.023 old/EST = −0.072, SE = 0.023 young).¹⁶ In the latter two questions, although the estimates are less robust, the effects are reversed. When respondents are old, they are less likely and when respondents are young, they are more likely to say they trust coethnics (EST = −0.093, SE = 0.034 old/EST = 0.096, SE = 0.36 young), and leaders should help their home community (EST = −0.088, SE = 0.025 old/EST = 0.105, SE = 0.026 young).¹⁷ We suggest that more work needs to be done to understand why this might be the case.

Interviewer questions

In a final set of analyses, we investigate how interviewer assessments of the interview process vary. These questions do not assess response-pattern variation but rather investigate interviewer assessment variation; therefore, relevant comparisons have different interpretations than the ones above. Given that much of our explanation of the response-pattern variation relies on the assumption that respondents see interviewers differently based on their respective ages, our claims should be further strengthened if interviewers also perceive respondents differently if they are in different age groups rather than in the same age group. The fact that interviewers may behave differently would also suggest that some of the response-pattern variation may be induced by interviewer behavior.

Overall, interviewers rate respondents quite positively, rating the experience on average between 0.13 and 0.28 on a 0–2 scale, where higher values are associated with a negative experience (see Supplementary Table B1). Nevertheless, as shown in Figure 5, the interviewer assessment data also provide clear evidence that interviewers’ self-assessment patterns vary depending on whether they interacted with respondents in the same or different age group. Young interviewers tend to report that old respondents treat them less well compared to respondents of the same age. For example, old respondents perceive young interviewers as less suspicious (EST = −0.149, SE = 0.026) and less impatient (EST = −0.155, SE = 0.026), and young interviewers see old respondents to be more suspicious (EST = 0.147, SE = 0.024) and impatient (EST = 0.137, SE = 0.024).

One explanation for this finding, in line with our theoretical framework, is that old respondents are not deferent or respectful toward young interviewers. On the other hand, old interviewers who interview young respondents tend to rate these young respondents more positively compared to respondents of the same age. This may speak to the deference that young respondents give old interviewers. Again, both effects are much larger than coethnic effects.

Age-group differences play an essential role in social interaction across sub-Saharan Africa. However, the social effects of these differences remain understudied. We hypothesize that age-group differences will affect response patterns and use Afrobarometer data to test this hypothesis. We also explore three mechanisms through which age-group differences may induce response-pattern variation: in-group loyalty, social acquiescence, and social distance. As hypothesized, we find relatively large and statistically significant effects for age-group differences across a variety of questions. Our findings support in-group loyalty and social acquiescence rather than social distance when questions do not address age-related issues directly. However, social distance may play a more important role when questions address age-specific issues. Additionally, we show preliminary evidence that age-group differences induce larger response-pattern variation than coethnicity. Our findings speak to the importance of age in social interaction in Africa and provide important lessons for the survey research community.

Age-Group Differences: Results From Round 7

We now turn to our second set of analyses, where we might expect to find larger dyadic effects about questions specifically related to age. The descriptive statistics by respondent age group for the one question asked across all 30 countries, and the six questions addressed only for Mauritius, are presented in Supplementary Table C2. These tables show that, unsurprisingly, old respondents have better views of the government’s addressing of youth issues than young respondents, which is consonant with the socially dominant position that the government is doing a good job. This difference in old versus young respondents’ views, therefore, raises questions about how much these attitudes vary by interviewer–respondent dyad.

As seen in Figure 6, there does appear to be variation in response patterns, supporting our main hypothesis. The pattern is the same for all seven youth-related variables, at least in Mauritius. In this country, old people appear to move towards the socially nondominant position when speaking to young interviewers, and these effects are much larger than when we examine the cross-national data with estimates and range from 0.136 to 0.299 SD units. Young respondents also appear to respond differently to old interviewers, though these results are generally not statistically significant and are much smaller in magnitude. We note that the scale of this graphic differs from the previous figures from Round 3 and 4 because the effects are up to a third of a standard deviation unit, which is much larger than the effects for other variables where the question posed is not directly related to youth. On the cross-national level, however, the results are more ambiguous, and both the sign and the statistical significance vary by model specification (see Supplementary Figures D9 and D10 and Supplementary Tables E9 and E10).

Unlike the results from Afrobarometer Rounds 3 and 4, the results from Mauritius appear most in line with social distance theory rather than social acquiescence or in-group loyalty. Indeed, it is only social distance that would predict that old respondents change their behavior in the face of young interviewers, and the data most clearly show such a trend. The data from Mauritius, therefore, add more nuance to the study of age-group effects and suggest that social distance may be more likely to occur when an issue is of high importance and directly affects the nonsocially dominant group.

Replication

A typical question is whether not controlling for age-group differences may affect the results of other studies where the primary variable of interest is not age itself. From an omitted variable bias perspective in observational studies, therefore, a researcher would want to know whether age-group differences are correlated with both the explanatory and the outcome variable. We suggest that checking these differences is worthwhile but should be motivated by theory and that, in many cases, dyadic age-group differences will not have an effect on other variables of theoretical interest.

For example, we reproduce the seemingly unrelated regression (SUR) analysis of Gottlieb et al. (2018), who assess gender differences in policy prioritization using Rounds 4 and 5 of the Afrobarometer. We keep their model identical except for incorporating age and gender differences between interviewers and respondents in the analysis. We do not have any strong reason to believe that age-group differences induce bias in the estimated coefficients. Indeed, as shown in Supplementary Figure H1, Gottlieb et al.’s coefficients remain stable after controlling for age-group differences.

Nevertheless, as seen in Figure 7, age-group differences appear to independently affect some of Gottlieb et al.’s outcomes. Old respondents are more likely to prioritize violence and less likely to prioritize education when talking to young interviewers—two positions that likely also reinforce the current social hierarchy. On the other hand, young respondents are less likely to prioritize violence and more likely to prioritize the economy when talking to old respondents. These results appear in line with in-group loyalty.

We suggest this replication provides further evidence on the utility of checking (and potentially including) age-group differences between interviewers and respondents in observational regression specifications while highlighting that there may also be unrelated effects of age-group differences in the interview process.

Conclusion

Scholars have long posited that age dynamics are important to studying African societies. Our study reaffirms this finding by examining personal interaction in face-to-face surveys. Moreover, we theorize about three mechanisms that could generate response-pattern variation in these face-to-face interactions: in-group loyalty, social acquiescence, and social distance. While certainly not dispositive, the data provide more evidence for theories of power relations rather than social desirability, with a mixture of findings that appear most in line with social acquiescence theory and, particularly in the context of economic questions, in-group loyalty theory.

The social acquiescence finding of young respondents deferring to old interviewers on political questions is in line with scholarship discussing youth’s marginalization. This finding suggests that survey estimates of youth discontent on issues such as trust in political parties and preference for democracy may be systematically underestimated. Since response patterns are situational, we caution analysts to think carefully about how age-group differences may influence response patterns, particularly in situations where age-group differences may be important.

Our research provides evidence that the effects of age-group differences between interviewers and respondents are substantively larger than the effects of coethnicity. These findings reiterate the importance of age-group differences benchmarked against another group category (ethnicity) often discussed in African studies.

Future research could explore the mechanisms we discuss more deeply and in different contexts. First, meta-analysis and other research with even more outcome variables could attempt to more rigorously test the comparative evidence for the different mechanisms. Second, additional research could examine whether the same mechanism driving age-group differences in response patterns in sub-Saharan Africa exist in other types of societies. In many Asian societies, contrary to the sub-Saharan African ones we study, the populations are rapidly aging. Therefore, mechanisms underlying the social effects of age-group differences will likely differ from those in sub-Saharan Africa. However, it will also be essential to see if the theorized deference to authority (social acquiescence) in Asian societies affects survey response patterns as it has views towards elders (North & Fiske, 2015). While their populations are also aging, Eastern European post-communist societies, on the other hand, may share a similarity in that age-group differences are associated with growing up in different types of regimes, and, for many young respondents, their childhood was shaped by a rebelling against the authoritarian past (Nikolayenko, 2007; Pop-Eleches, & Tucker, 2017). Therefore, it may be interesting to compare whether power relations theories as they relate to age-group differences function the same or differently in these societies.

Given our findings, from a practical perspective, at least in sub-Saharan Africa, employing young interviewers rather than combining a mixture of old and young interviewers may induce less response-pattern variation. However, given young interviewers’ reports of worse treatment by old respondents, more sensitization and training for young interviewers in dealing with elders could help improve their work experience.

Acknowledgments

We thank Aengus Bridgman, Costin Ciobanu, Jeff Conroy-Krutz, Nicholas Kerr, Amanda Robinson, and attendees of the 2019 Annual Meeting of the Southern Political Science Association (SPSA) and the Centre for Population Dynamics (CPD) at McGill University speaker series for sharing data and feedback. We would also like to thank Lawrence Plastina for his research assistance.

Funding

None declared.

Conflict of Interest

None declared.

Data Availability

The replication code and data underlying this article are available in the Harvard Dataverse, at https://doi.org/10.7910/DVN/M3LPXD.

Biographical Notes

Aaron Erlich is an Associate Professor of Political Science at McGill University, where he is a member of the Centre for the Study of Democratic Citizenship and the Centre on Population Dynamics. He is interested in the role information plays in developing democracies and survey research methods.

Andrew McCormack completed his M.A. degree in Political Science at McGill University. He currently works in the public opinion polling industry.

References

Footnotes