The Education Reform of 1980

Prior to independence, access to education for the black community was deliberately restricted. While schooling was compulsory and free for whites (until age 15), black Zimbabweans—who were not required to attend school—had to pay high school fees. In addition, black Zimbabweans were required to pass a series of increasingly difficult exams in order to continue past primary school, while continuation to the first cycle of secondary school was automatic for whites. The education budget for black Zimbabweans was tightly controlled at 2% of GDP, while out-of-pocket secondary school tuition and boarding fees cost almost two months of the average annual wage.Footnote ⁶ The government spent about 12 times more per capita on primary schooling for whites than for blacks (Dorsey Reference Dorsey1989). King (Reference King2013) documents that such discriminatory policies were implemented in many African colonies, which had come to associate education with greater unrest.

Starting in April 1980, the ZANU government implemented a wide-ranging set of educational reforms. Primary education was made free and compulsory for all Zimbabweans, regardless of color. While some fees were applied for secondary school, automatic progression from primary to secondary school was decreed. Furthermore, age barriers were removed for older children, allowing those who did not start school on time to attend. The government also undertook a large-scale school building campaign and reopened schools that had been closed during the independence war. Between 1980 and 1986, the number of primary schools almost doubled, from 2,401 to 4,291, while the number of secondary schools increased dramatically, from 177 to 1,276 (Bourne Reference Bourne2011).

The reform had an immediate effect: overall student enrollment doubled in one year (Narman Reference Narman, Darnolf and Laakso2003). As Figure 2 illustrates, this increase was most dramatic for secondary enrollment, which rose from 66, 215 students in 1979 to 537,427 in 1986. The change is also apparent in the primary-secondary progression statistics: while in 1979 only 25% of primary school leavers continued to secondary schools, by 1986 78% did (Bourne Reference Bourne2011).

FIGURE 2.

Number of Enrolled Students in Form 1 (First Year of Secondary Education) and in All Secondary Education Grades by Year, 1979–1986

Source: Government of Zimbabwe, Annual Report of the Secretary of Education, 1986.

Based on the nature of Zimbabwe’s education reform, we focus on the expansion of secondary education.Footnote ⁷ Although primary school fees were formally banned, the reform did not substantially affect primary educational attainment. This is likely because 80% of black Zimbabweans were already enrolled in primary school even under white rule, and because some primary schools continued charging informal fees (Nhundu Reference Nhundu1992). Had the 1980 reforms significantly affected primary enrollment, we should observe a sharp increase in education among the cohorts starting primary school in 1980, i.e., those born from 1972 to 1974. However, we find no evidence of a jump around those birth years (see Figure 5 below).Footnote ⁸ The reform’s small effect on primary school completion principally reflects the remedial education of individuals whose education was interrupted by the war (Narman Reference Narman, Darnolf and Laakso2003).Footnote ⁹

Given the rapid expansion, qualified teachers could not be hired quickly enough, instructional quality often suffered, and school construction lagged behind enrollment, leading to overcrowded classrooms.Footnote ¹⁰ While some slippage in quality was perhaps inevitable given the speed of the reform, quality remained sufficiently high to deliver substantial material benefits to the reform’s beneficiaries, as we demonstrate below. Notwithstanding these challenges, Zimbabwe was widely recognized as a leader in expanding access to education in Africa during the 1980s (Dorsey Reference Dorsey1989), and provides an excellent setting in which to examine the role of education in electoral authoritarian regimes that allow some restricted political contestation.

Data

We combine all available rounds of the Afrobarometer surveys conducted in 1999, 2004, 2005, 2009, 2010, and 2012.Footnote ¹¹ Since the Afrobarometer questions change across survey rounds, different rounds may be used to test different outcome measures (see Online Appendix). We focus exclusively on black respondents, who comprise the overwhelming majority of Zimbabwe’s population and were the targets of the 1980 education reform.Footnote ¹²

Education is our key (endogenous) explanatory variable, which is measured using the following seven-point ordinal scale: no schooling, incomplete primary, complete primary, incomplete secondary, complete secondary, incomplete college, complete college. A one-category increase in the education measure is equivalent to about 2–4 years of education, given the discrete nature of the variable. Figure 3 shows the distribution of this measure in our sample, indicating that the modal level of schooling is incomplete secondary school.

FIGURE 3.

Sample Distribution of Educational Attainment (8,014 observations)

Our key dependent variable, political participation, is operationalized using four binary variables.Footnote ¹³ Voted indicates whether the respondent reported voting in the most recent legislative or presidential election. We also examine indicators for directly contacting one’s local government councilor (Contacted local councilor), attending a community meeting (Attended community meeting), or joining other community members in raising an issue (Raised issue at meeting) within the past 12 months.Footnote ¹⁴

In electoral authoritarian settings like Zimbabwe, political action such as voting, attending and participating in community meetings, and contacting local leaders, comprise a coherent set of behaviors encapsulating noncontentious political participation. First, these are common types of political behavior, rather than unusual forms of participation restricted to elites. Respectively, 64%, 38%, 66%, and 58% of respondents engaged in such activities.Footnote ¹⁵ The Online Appendix shows that these levels of participation broadly represent the median country in Afrobarometer surveys.

Second, mobilization of observable mass participation in public contexts such as voting and public meeting attendance is a key component of regime strategy in electoral authoritarian contexts, since they represent a signal of the regime’s strength (Magaloni and Kricheli Reference Magaloni and Kricheli2010). Marongwe (Reference Marongwe2013), for example, notes that active participation in political meetings was a key marker of regime support over the period covered in this analysis. Mass participation is a powerful signal that active resistance for any individual is pointless because opposition to the regime is only viable if many others join the struggle (Kuran Reference Kuran1991). Conversely, deliberately abstaining from such forms of participation is a meaningful political act and a genuine sign of implicit resistance in this context.

Finally, we focus on noncontentious participation because these actions are low cost and low risk forms of participation that increased education might reasonably affect. By contrast, consistent with their comparatively higher costs, only 17% and 11% of respondents report contacting their MP or attending a demonstration. Unreported results also indicate that education reduces contact with MPs and increases demonstration attendance, although neither reaches statistical significance.

We further combine the four binary participation variables, which are positively correlated with a Cronbach’s alpha of 0.65, into a simple summative rating scale (Participation scale).Footnote ¹⁶ Although we also present the results for each component separately, we place greatest weight on our scale measure because it averages over any noise contained in our individual indicators.

Variation in Access to Secondary Schooling

In order to identify the causal effect of educational attainment, we exploit cross-cohort variation in access to secondary schooling arising from Zimbabwe’s 1980 reform. Specifically, we compare black citizens from cohorts that were just young enough to be affected by the reform to black citizens from cohorts that were just too old to benefit from the educational expansion.

We define those born in 1967 or later, who were 13 or younger when the reform was implemented, as fully “treated” ( $\mathit {Secondary \ access}=1$ ). Those born in 1963 or earlier, and thus aged 17 or older in 1980, are defined as our control group that was not affected by the reform ( $\mathit {Secondary \ access}=0$ ). Finally, those aged 14–16 at the time of the reform’s onset are considered “partially treated.” Such individuals are coded according to the number of additional years of schooling available to them; for example, a black Zimbabwean aged 15 in 1980 is coded as receiving a “dosage” of half treatment, while individuals aged 14 and 16 are coded as receiving one-quarter and three-quarter dosages, respectively.Footnote ¹⁷ This coding scheme, shown graphically in Figure 4, defines Secondary access, our source of exogenous variation.

FIGURE 4.

Operationalization of Secondary Access, the Study’s Key Treatment Variable

Figure 5 provides preliminary evidence that the reform increased average educational attainment across cohorts. The education scale (top left) shows that cohorts fully treated by the reform exhibit substantially higher levels of education relative to cohorts born in 1963 or earlier. The increase is large and almost equivalent to moving from complete primary to incomplete secondary, or from incomplete to complete secondary education. The figure also demonstrates that the reform increased the education levels of partially treated cohorts, but by less than fully treated cohorts.

FIGURE 5.

Trends in Educational Attainment by Cohort

Notes: Each gray dot represents average education for a given cohort (birth year). Large dots reflect larger sample sizes. Black lines are local polynomials fitted on either side of the reform (indicated by the vertical dashed line). The vertical gray dashed lines indicate the bandwidth used for our main analysis.

To identify the causal effects of the reform we must assume that black Zimbabweans on both sides of the reform cutoff are identical, with the exception that only some cohorts were eligible to enjoy access to secondary education. However, independence undoubtedly brought about many important changes, and socialization processes could operate differently at different stages of life (Alwin and Krosnick Reference Alwin and Krosnick1991; Sears and Valentino Reference Sears and Valentino1997). To address such concerns, we limit our comparison to cohorts that were born just too early to be directly affected by the reform and the first cohorts to be affected by the reform. Specifically, our main analysis focuses on a bandwidth of five cohorts on either side of the reform cutoff years of birth (1963 and 1967). This is a powerful design even though it dramatically reduces our sample size: the cohorts around the reform are subject to similar economic, social and political environments, but differ in their ability to take advantage of a schooling reform that could not have been anticipated by parents more than a decade earlier.

Using a subsample of five cohorts around the eligibility cutoff, there are good reasons to believe that control cohorts only differ from treated cohorts with respect to secondary schooling eligibility. First, Figure 5 indicates that trends in education are relatively flat once we focus on the five cohorts on either side of the reform’s cutoff point (inside the gray dashed lines). Below, we also document flat trends in our political outcomes. Second, by varying the size of the bandwidth, using placebo reforms, and controlling flexibly for cohort trends on either side of the reform cutoff, our robustness checks demonstrate that trends across cohorts cannot be driving our results. Third, Figure 6 indicates that cohorts on either side of the reform are balanced across treatment groups with respect to gender, age at the date of the survey, district-level political aggregates, and distance to the borders from which the rebels entered during the war of independence. Although there is a slight imbalance with respect to tribe, we show below that our findings are robust to the inclusion of pre-treatment covariates and that the effects of access to education do not vary across tribes. Moreover, previous studies indicate that cohorts affected by the reform are uncorrelated with good proxies for family income in early childhood, such as adult height (Agüero and Ramachandran Reference Agüero and Ramachandran2014).Footnote ¹⁸ Finally, the frequency of surveyed individuals by age is not affected by the treatment, suggesting that there is no differential fertility or migration across cohorts around the reform. Tellingly, the proportion of educated respondents in our surveys does not change even after hyperinflation began in the mid-2000s.

FIGURE 6.

Trends in Pretreatment Variables by Cohort

Note: See Figure 5.

Estimation Strategies

Building on our key identifying assumption that access to secondary schooling is exogenous across cohorts on either side of the reform cutoff, we utilize two main approaches to identify the long-run effects of secondary education on political participation. Our first approach includes partially treated respondents, which enables us to examine outcomes as a function of treatment intensity (i.e., differential access to secondary schooling) across cohorts that were subject to essentially identical common shocks. We first estimate the reduced form effects of increasing the availability of secondary education—which is equivalent to an “intent-to-treat” (ITT) analysis—by estimating the following regression equation using ordinary least squares (OLS):

(1) $$\begin{eqnarray} Y_{icdt} = \gamma \mathit {Secondary \ access}_{c} + \eta _t + \epsilon _{icdt}, \end{eqnarray}$$

where Y_icdt is an outcome measure, and $\mathit {Secondary \ access}_{c}$ is our key treatment variable. We include survey fixed effects, η _t , to account for time-varying shocks that impact respondents differently across survey rounds, and cluster standard errors by district to account for spatial correlations between respondents.Footnote ¹⁹

Access to public education, however, does not necessarily entail enrollment. Rather, the link between educational access—an opportunity that equally affects all cohort members—and educational outcomes is probabilistic. This is because not all primary students continue to secondary school, and because some older individuals returned to school after the war. To identify the effects of actual education among Zimbabweans who only received additional education because of the 1980 reform, we use access to secondary schooling to instrument for education. In our first stage, we estimate the effect of access to secondary education on a respondent’s educational attainment:

(2) $$\begin{eqnarray} \mathit {Education}_{icdt} = \delta \mathit {Secondary \ access}_{c} + \eta _t + \xi _{icdt}, \end{eqnarray}$$

before estimating the following structural equation using two-stage least squares (2SLS):

(3) $$\begin{eqnarray} Y_{icdt} = \beta Education_{icdt} + \eta _t + \varsigma _{icdt}. \end{eqnarray}$$

The instrumental variable (IV) estimates thus rescale the reduced form to estimate the effect for black students who only remained in school because of the reform.

Our linear coding of education follows Marshall (Reference Marshall2014), who shows that coding an endogenous education variable as binary can significantly upwardly bias estimates if greater education at lower levels—which do not register in the first stage—also affects the outcome.Footnote ²⁰ Since any additional education may affect political behavior, we use the seven-point education scale (described above) as our endogenous independent variable. This allows us to consistently estimate the average effect of an additional unit of education.

IV estimation requires several additional assumptions. First, the relationship between the instrument (secondary access) and the endogenous variable (education) must be strong. The first stage estimates in Table 1 show that the reform substantially increased education among affected cohorts, principally at the secondary level. Reinforcing the results in Figure 5, the estimate for our education scale in column (1) indicates that being fully treated by the reform increases education by two-thirds of a level on average. This entails a large first stage F statistic of 71, which far exceeds the standard critical value of 10 required to minimize weak instrument bias (Staiger and Stock Reference Staiger and Stock1997). Second, the exclusion restriction requires that our instrument only affects political outcomes through increased education. We discuss this assumption in greater detail below.Footnote ²¹

TABLE 1.

Estimates of Education Reform on Educational Attainment

Notes: All specifications are estimated using OLS, include survey fixed effects, and cluster standard errors by district. All specifications include five cohorts on either side of the cohorts that were fully affected or fully unaffected by the reform (panel A), or five cohorts on either side of the first cohort to receive any treatment (panel B). *p < 0.1, **p < 0.05, ***p < 0.01.

Our second approach drops all partially treated respondents, and thus compares only untreated respondents who were born in 1963 or earlier (i.e., too early to be affected by the reform) to respondents who were fully treated. This allows us to focus on a sharp change in treatment assignment, and is thus similar to regression discontinuity (RD) designs relying on the weaker assumption that potential outcomes are smooth across the discontinuity.Footnote ²² For this second approach, we again estimate Equations (1) and (3), but exclude all partially treated respondents. Finding consistent results across both approaches should increase confidence in the study’s findings.

Main Estimates

Contrary to the positive effects of education documented in advanced democracies (Sondheimer and Green Reference Sondheimer and Green2010) and democratic developing country contexts (Larreguy and Marshall Reference Larreguy and Marshall2014), we find that in Zimbabwe, education substantially and significantly reduces levels of political participation. Column (1) in Table 2 reports the estimates for our participation scale across all estimation strategies. Relative to its sample mean of 0.65, panel A shows that access to secondary education reduces participation by around 10%. Panel C shows that this estimate is barely affected by excluding partially treated respondents. The IV estimates in panels B and D reveal a similar story: regardless of whether partially treated respondents are included, a one-unit increase in education reduces participation by around 15% relative to its sample mean. Finally, we report the simple correlation between education and political participation in panel E in the sample containing all cohorts, which also shows a significant negative correlation.Footnote ²³

TABLE 2.

The Effects of Education on Political Participation

Notes: All specifications in panels A, C, and E are estimated using OLS, and include survey fixed effects. All specifications in panels B and D are estimated using 2SLS, in which access to schooling is used to instrument for education, and include survey fixed effects. All specifications include five cohorts on either side of the cohorts that were fully affected or fully unaffected by the reform; panels C and D exclude partially treated cohorts born between 1964 and 1966. Standard errors are clustered by district in all specifications. *p < 0.1, **p < 0.05, ***p < 0.01.

Importantly, as Table 2 and Figure 7 indicate, the negative effects of education on the participation scale are not simply due to an exceptionally large impact on one constituent variable. Rather, a one-unit increase in access to secondary schooling and a one-unit increase in education both cause substantial declines in each noncontentious form of political participation. Specifically, our estimates suggest that being fully eligible for secondary education reduces voting by 8 percentage points, contacting one’s local councilor by 6 percentage points, attending a community meeting by 6 percentage points, and raising issues with others at a meeting by 4 percentage points. Similarly, a one-unit increase in educational attainment reduces voting by 12 percentage points, contacting a local councilor by 8 percentage points, attending a community meeting by 9 percentage points, and raising an issue by 6 percentage points. Across all our specifications, only the decrease in raising an issue is not statistically significant.

FIGURE 7.

Trends in Political Participation by Cohort

Note: See Figure 5.

Robustness Checks

Given that these findings challenge the conventional wisdom that education increases participation, it is essential to demonstrate their robustness. In this subsection we describe several checks that more formally test our identifying assumptions and that examine whether our results hold when using alternative specifications. All robustness results are presented in Table 3.

TABLE 3.

Robustness Checks

Notes: Panels A and B include 3 and 10 cohorts, respectively, on either side of the reform. Panel C multiply imputes the outcome variables. Panel D treats cohorts born between 1957 and 1961 as treated, and compares them to cohorts born between 1952 and 1956. Panel E includes 10 cohorts on either side of the first and last cohorts on either side of the reform and includes linear birth-year polynomials on either side of the reform. Panel F includes only cohorts that turned 18 between 1981 and 1984. Panel G includes Shona, Ndebele, and male dummies as controls, as well as controls for the district incumbent vote share and district turnout at the nearest election. Panel H includes district fixed effects. Standard errors are clustered by district in all specifications. *p < 0.1, **p < 0.05, ***p < 0.01.

We first show that our results are not an artifact of specification choices or cohort trends. Panels A and B show that the reduced form estimates are similar when we include either three or ten cohorts on either side of the reform eligibility threshold. Figure 3 in the Online Appendix demonstrates that our results are similarly robust to any bandwidth choice between 1 and 10. Second, we test whether our findings could be explained by potential biases arising from missing data. Panel C shows that if anything the results are stronger when imputing the missing data across our main participation outcomes.Footnote ²⁴

Third, to address the possibility that our findings are driven by long-term changes across cohort, we employ placebo tests and control flexibly for cohort trends. In panel D, we examine a placebo reform in which we estimate the reduced form effects of a (hypothetical) reform in 1970 and compare cohorts five years on either side of this arbitrary cutoff. Reassuringly, we do not observe any significant reduction in political participation around the placebo reform. Furthermore, we repeat this exercise and find no effects for placebo reforms in any year between 1960 and 1972.Footnote ²⁵ Fourth, panel E implements an RD design where we redefine our treatment to include the “partially affected” citizens; i.e., all respondents born since 1964 are counted as treated. Using a 10-cohort bandwidth, and including linear cohort trends on either side of the discontinuity to account for cohort trends, our RD estimates similarly show large negative effects of access to the reform.Footnote ²⁶

As noted above, plausible confounding explanations must relate to political differences between the cohorts immediately around the reform eligibility cutoff. First, one potential concern is a “first election” effect, such that respondents with different levels of treatment behave differently because they first voted in different elections (Meredith Reference Meredith2009). To show that this cannot explain our results, in panel F we restrict attention to respondents born between 1963 and 1966—who were first eligible to vote (at age 18) in the 1985 election—and find that the intensity of secondary access continues to significantly decrease political participation. Second, although our design minimizes differences in citizen characteristics around the reform cutoff, we also show that our results are robust to the inclusion of other potentially confounding omitted variables. Panel G, which includes the pretreatment variables described in Figure 6, yields similar results. In particular, these results suggest that participation is not being driven by compositional changes in the proportion of Shona and Ndebele respondents. Panel H demonstrates the robustness of our results to the inclusion of district fixed effects, although contacting a local councilor slightly falls outside statistical significance. Finally, while the inclusion of age fixed effects substantially decreases the precision of our estimates by removing considerable cross-cohort variation, we show in the Online Appendix that, if anything, the magnitudes of our negative estimates increase.

While our reduced form estimates do not require that the exclusion restriction holds, the IV estimates do. There are, however, good reasons to believe that the secondary education reform only affects participation through its effect on educational attainment. First, because education is highly proximate to the reform itself, most downstream behavioral responses—such as fertility, marriage, and vocation—are a function of a respondent’s education. Second, the fact that the decrease in participation levels for partially treated respondents is lower than for fully treated respondents, but higher than for untreated respondents, increases our confidence that participation is responding to changes in actual schooling. If political responses to being affected by the reform itself were driving the results via some other channel, then it is hard to see why it would have differentially affected those receiving different instrument dosages. Third, a typical concern with such reforms is the possibility of cross-cohort spillovers. However, if cohorts on either side of the reform cutoff interact with one another, spillovers are likely to reduce the effects of schooling as behavior becomes more homogeneous.

Nevertheless, we examine the sensitivity of our results to violations of the exclusion restriction by calculating the extent of the violation required to nullify our finding. Using Conley, Hansen, and Rossi (Reference Conley, Hansen and Rossi2012)’s most conservative sensitivity test, we find that 50% of the reduced form effect must operate through channels other than education for our negative estimate of education’s effect on the participation scale to become statistically insignificant.Footnote ²⁷

Mechanism

We begin by examining the mechanisms described in the introduction. First, we examine whether education increases citizens’ material resources, which allow for enhanced engagement in politics, as well as their political interest. Second, we examine the relationship between education attainment and value change, i.e., whether education has a positive effect on support for democratic institutions. Third, we assess whether voters’ increased interest and ability to participate in politics, and higher premium on democratic values, lead to a more critical assessment of the incumbent’s performance by testing whether education has a negative effect on perceptions of incumbent performance, and ultimately on the level of support for the incumbent authoritarian regime.

Resources: Socioeconomic Status and Political Interest

We first examine whether access to secondary education increases the economic welfare of the cohorts that benefited from it. Specifically we are interested in exploring the relationship between education and socioeconomic status (SES), since SES can be expected to lead to greater political participation either because some forms of participation are costly, or because higher economic status leads to greater involvement in social networks, which are entry points into such participation (Verba, Schlozman, and Brady Reference Verba, Schlozman and Brady1995). If we find that education did not increase SES, this would suggest that the economic returns to education in Zimbabwe are not as consequential as they are in mature democracies, reducing the likelihood that better educated Zimbabweans have lower participation rates due to deliberate disengagement.

We examine the economic returns of education for black Zimbabweans in terms of employment (Employed), self-reported living conditions (Good living conditions), and a more objective poverty scale (Poverty).Footnote ²⁸ We also combine these three measures to produce an economic outcomes scale (Economic scale). The Cronbach’s alpha for this scale is 0.41. As Table 4 shows, education improved Zimbabweans’ economic well-being. Consistent with human capital models, treated adolescents are more likely to be employed two to three decades after the reform’s onset (column 2), and they have higher income, at least as reflected in objective measures of living standards (column 4). Although not quite statistically significant, treated respondents are also likely to rate their living conditions more highly (column 3). Together, this evidence sharpens our theoretical puzzle: despite greater economic resources, which should facilitate greater participation according to the current literature, we observe lower levels of participation.

TABLE 4.

Estimates of Secondary Education on Economic Outcomes and Political Interest

Note: See Table 2.

We have also argued that education likely leads to greater interest in politics, which is important for citizen behavior. Specifically, informed citizens with greater interest in politics will have more accurate assessments of the incumbent regime, and better understanding of the limited responsiveness of the regime to the needs, preferences and interests of its constituents. Political interest is operationalized using two distinct measures. First, News scale combines indicators for whether respondents consume news from radio, television, or newspapers at least once a week. The scale has a Cronbach’s alpha of 0.74. Second, Interest in public affairs is an indicator variable for the 64% of respondents who express interest in politics and public affairs. In both cases, higher values suggest greater political interest.

Consistent with a large number of studies from Western democracies, Table 4 shows that education in Zimbabwe has a positive effect on political interest. For political interest, a one-unit increase in education raises the likelihood that an individual regularly obtains political news by around 9 percentage points, or 25% relative to the sample mean (column 5). Similarly, we find a positive, if weaker, effect of education on interest in public affairs (column 6). Our estimates thus demonstrate that reduced participation is not simply accounted for by reduced interest in politics. In fact, educated Zimbabweans are more likely to follow politics, at the same time that they are less likely to participate.

Value Change: Support for Democratic Institutions

Education is thought to also increase support for democratic institutions. Dating back to Aristotle, through Thomas Jefferson and de Tocqueville, it has been argued that education supports democratic institutions by breeding tolerance and acceptance of others’ opinions (Dewey Reference Dewey1916). By contrast, Lerner (Reference Lerner1958) highlights the impact of education on self-assessment and self-confidence. In Lerner’s model, educated people in modernizing societies start developing opinions about public issues, which leads them to believe that they have the ability, and thus should have the right, to provide input on matters of importance that affect their welfare. This psychological change, Lerner (Reference Lerner1958) argues, translates into growing support for inclusive political institutions. We therefore test whether education increases support for democratic institutions.

We argue that respondents who support democratic institutions would be less interested in legitimizing the autocratic regime with their participation. A null finding on education’s effect on support for democratic institutions would therefore be at odds with our “deliberate disengagement” argument. We measure support for democracy in two ways. First, we examine the relationship between education and Support democracy, an indicator for the 72% of respondents claiming to support or strongly support democracy. Importantly, this question is not asking respondents whether they approve of democracy as practiced in Zimbabwe, but rather democracy in the abstract. Second, to better capture support for specific liberal institutions associated with democracy, we group the following nine indicators into a scale: do you agree that parties are needed, do you reject one-party government, do you reject one-man government, are you against government banning civil society organizations, are you against government closing news stations, are you against presidential discretion, are you in favor of parliament making the laws, do you agree that the president should obey the laws, and do you support term limits. The variables that make up this Support liberal institutions index are positively correlated with a Cronbach’s alpha of 0.74. Again, larger values suggest greater support for democratic institutions.

Belying an explanation rooted in limited demand for democracy, we find that education increases support for democracy. An additional unit of education significantly increases the likelihood that an individual professes support for democracy by 8 percentage points (Table 5, column 1). The positive, albeit not always quite statistically significant, effect on support for liberal institutions similarly suggests that voters possess a genuine understanding of the building blocks required to support liberal democracy (column 2). In sum, these results suggest that, consistent with our deliberate disengagement argument, support for democracy increases with education.

TABLE 5.

Estimates of Secondary Education on Support for Democracy

Note: See Table 2.

Lower Support of the Incumbent Regime

We have argued above that to the extent that education increases interest in politics and support for democratic institutions, we should expect that education reduces support for the incumbent autocratic regime. If true, this should in turn reduce participation in noncontentious political action among citizens that do not wish to legitimize the regime by actively participating.

To explicitly test this mechanism, we assess support for the government using four measures. Our first and second measures, Close to ZANU-PF and Close to MDC, indicate whether respondents report feeling close or very close to the ruling party and the main opposition party; 27% and 24% of respondents in our sample reported being close to ZANU-PF or the MDC, respectively. Third, we create a variable Incumbent trust and performance, which is a summative rating scale combining three indicators for trusting the president, the ruling party, and its MPs, and three indicators for whether the respondent believes that the president, MPs, and the local government are performing well in office (alpha of 0.84). Fourth, Perceived government corruption is a summative rating scale (alpha of 0.80) that combines four indicator variables asking whether the respondent believes the president, MPs, local councilors, and government officials are corrupt. Finally, we combine these four variables to produce our View of government scale (alpha of 0.62).

The results in Table 6 support our theoretical argument: across all specifications in column (1), access to secondary education has a negative effect on the view of government scale. Furthermore, columns (2) and (3) show a significant decrease in support for ZANU-PF and a significant increase in support for the MDC. Trust in government also broadly decreases with education (column 4). Consistent with the idea that political interest might decrease support for the regime, perceptions of corruption significantly increase with education. These findings suggest that better-educated citizens are more critical of Mugabe’s regime and cognizant of its problems, and consequently less likely to support it.

TABLE 6.

Estimates of Secondary Education Reform on Support for the Government

Note: See Table 2.

Testable Implication: Participation and Election Competitiveness

We now move to examine a central implication of our theoretical argument. Specifically, we test whether the relationship between education and participation varies as the nature of elections in Zimbabwe has changed over time.

Zimbabwean elections during the study period (1996, 2002, 2005, and 2008) varied significantly in their competitiveness. Notably, the first round of the 2008 election allowed a higher level of contestation than previous elections.Footnote ²⁹ Despite the intense violence preceding the second round of voting, the ultimate result of the 2008 election was that the MDC (together with its splinter group, MDC-M) gained a majority in the House of Assembly, a majority of municipal councils, and some level of executive power through the internationally brokered unity government. This contrasts sharply with the 2002 and 2005 elections, which were marked by severe repression and in 2005 an initial opposition boycott (Frankel Reference Frankel2010), which left ZANU-PF in complete control of government.Footnote ³⁰ Thus, if better educated voters are more likely to disengage when they feel that participation is futile, merely serving to legitimize the government, or when boycotts have been called specifically to highlight the illegitimacy of the regime, we should also expect such voters to re-engage when elections are able to more meaningfully reflect political preferences. We thus compare the effects of education on political participation for respondents who were surveyed before and after 2009 (the first survey since the 2008 election).Footnote ³¹

The results, reported in Table 7, show that education had different effects before and after 2008. Consistent with our theoretical argument, the effect of education is negative and very large during the uncompetitive period before 2008. As demonstrated by the positive interaction term for post-2009 survey responses, the effect of education is essentially zero when elections meaningfully affected the distribution of executive power. In no case is access to secondary education statistically significant for respondents surveyed since 2009, while the difference in coefficients is statistically significant for voting. These results are also important because they cannot be easily reconciled with alternative explanations grounded in relatively stable characteristics such as income or early-life socialization.

TABLE 7.

The Effects of Education on Political Participation, Before and After 2008

Note: See Table 2.

It is possible that the differences between the pre- and post-2008 period could reflect changes in the characteristics required for deliberate disengagement, such as support for democracy and disapproval of the regime. We find, however, that none of the potential mechanisms of deliberate disengagement—which are relatively long-term processes that should not substantially fluctuate across elections—changed across the pre- and post-2009 periods (see Online Appendix). While other factors may have differentiated the 2008 election from earlier elections in Zimbabwe, it is not obvious how these interact with education. Though not conclusive, these findings are suggestive of the possibility that meaningful political contestation can reverse the negative effects of education on participation.

Alternative Explanations

Finally, we eliminate three alternative interpretations of our deliberate disengagement argument. First, we test whether less educated voters are disproportionately the targets of turnout mobilization drives, because vote-buying efforts either target the poor or the regime’s core supporters (Stokes et al. Reference Stokes, Dunning, Nazareno and Brusco2013). Better educated voters seem a priori to be less likely to be included in clientelistic networks. However, as shown in columns (1) and (2) of Table 8, greater education does not decrease the likelihood that voters receive a gift from a political party before the most recent elections,Footnote ³² or perceive their vote not to be free. In fact, better-educated voters are more likely to report receiving a gift. Furthermore, if mobilization were driving our results, we might expect the negative effect of education to be largest in locations dominated by ZANU-PF or where turnout is high. However, we find no significant negative coefficient on the interaction of access to secondary school for either the district-level ZANU-PF vote share or the turnout rate at the most recent election (columns (3) and (4)). We conclude that even if better-educated voters are less likely to be mobilized—and we find no such evidence in our data—this cannot be driving the key results reported herein.

TABLE 8.

Mobilization Explanations

Notes: All specifications are estimated using OLS, include survey fixed effects, and cluster standard errors by district. Specifications include five cohorts either side of the cohorts that were fully affected or fully unaffected by the reform. *p < 0.1, **p < 0.05, ***p < 0.01.

A second alternative explanation is that better educated citizens are simply more likely to be repressed by the regime. This channel seems plausible given that in both the 2002 and 2008 elections the regime targeted significant violent repression at suspected MDC supporters. Similarly, educated citizens may preemptively disengage to avoid facing violence by signaling that they are not troublemakers. However, columns (1) and (3) in Table 9 show that education does not affect a respondent’s fear of repression nor their belief that voting is monitored. More so, given that President Mugabe has historically regarded the Ndebele as the opposition, if education induces greater fear then we should expect this to be greatest among the Ndebele. Interacting access to secondary education with an indicator for Ndebele respondents, columns (2) and (4) do not support this possibility. Finally, we show that in districts with more instances of violence against civilians by ZANU-PF—as measured by the Armed Conflict Location and Event Data Project—educated voters are no less likely to participate in politics (column 5). Together, these findings suggest that targeted repression is not the main driver of our key results.

TABLE 9.

Repression Explanations

Notes: All specifications are estimated using OLS, include survey fixed effects, and cluster standard errors by district. Specifications include five cohorts either side of the cohorts that were fully affected or fully unaffected by the reform. Event is district number of instances of violence against civilians by ZANU-PF as measured by the Armed Conflict Location and Event Data Project. *p < 0.1, **p < 0.05, ***p < 0.01.

Last, a subtle “coming of age” alternative explanation is that older students may have been more cognizant of the independence movement, and their more intense support for Mugabe’s regime could be manifested in greater participation that has persisted until today. However, we find little support for the key predictions of this argument. First, columns (1) and (2) of Table 10 show that, contrary to the idea that engagement in the war cultivated nationalist sentiments among relatively older cohorts of young teenagers, expression of national identity is instead positively (and insignificantly) associated with secondary access. Second, columns (3)–(5) similarly show that respondents living closer to the borders where the ZANLA and ZIPRA independence groups entered the country—and are thus likely to have been most exposed to the armed independence movements or directly exposed to conflict—are no less likely to disengage upon receiving access to education.Footnote ³³ Third, members of the Shona tribe—the group most strongly associated with the victorious ZANLA group that ultimately took power—who had access to secondary education are similarly no less likely to disengage. Finally, persistent differences in support for Mugabe’s regime across cohorts cannot convincingly explain the differential change in the participation of younger (better-educated) cohorts following the more competitive 2008 election.

TABLE 10.

Coming of Age Explanations

Notes: See Table 8.