The revolution is dead, long live the demolition: Education and labor market consequences of student riots

The army played a dominant role in Turkish politics by ousting elected governments nearly once in every decade from 1960 to 1980. The 1960 coup marked the beginning of a new era in Turkey. After the military coup, a new constitution was prepared before the free elections in 1961. The new constitution was more liberal and people had more civil rights than ever before. Universities had greater autonomy, students had the freedom to organize their own associations, and workers had the right to strike. As a result, left-wing politics started to gain strength, especially on university campuses. Over time, with the push from the global political events of 1968, Turkey's left became more extremist in the hopes of igniting a revolution. But the left's extremism was soon met and surpassed by the right, which generated conflict and violence (Zurcher, 2004). In March 1971, the army forced the elected government to step down and changed the constitution again (i) to strengthen the state against civil society; (ii) to gain control of the universities and curb radicalism; and (iii) to phase out trade unions (Ahmad, 1993, 2003).

The left soon rallied around the Republican People's Party. In 1973, the Republican People's Party won the parliamentary elections and formed a coalition government with the National Salvation Party. The extremist right-wing parties criticized the government program, which sought to heal the wounds caused by the military regime. Radical leftists responded with acts of violence and political violence became a regular feature of daily life in Turkey in the 1970s. Figure 1 presents the total number of terrorist attacks used as a proxy for conflict intensity in Turkey from 1970 to 1985. The data source is the Global Terrorism Database (GTD), which defines “a terrorist attack as the threatened or actual use of illegal force and violence by a non-state actor to attain a political, economic, religious, or social goal through fear, coercion, or intimidation.” Figure 1 shows that the attacks declined following the 1971 military intervention, but they increased again after 1974 and were quite intense during the turmoil leading up to the 1980 coup.

Figure 1

In April 1977, political parties agreed on an early election, which led to increased violence intensity. Street terror peaked on the May Day (May 1st) of 1977, four weeks before the early elections. The unions organized a huge rally in Istanbul. Shots fired into the crowd killed 36 people and injured hundreds. Additionally, the 1977 election did not produce a strong and stable government because no party won a clear majority. As a result, Turkey experienced one of its darkest periods in terms of political instability and societal chaos. In July 1978, the government started to use the army due to internal security concerns. Despite the increasing use of force, the violence continued until the slaughter reached 20 victims a day in the late 1970s (Ahmad, 1993). From 1978 to 1980, 5,241 people were killed and 14,152 people wounded due to violent political conflict (Kaya, 1981).

The army took control in September 1980 and ruled until the general elections of November 1983. The public welcomed the military intervention, and the army crushed almost all movements from the left and right to de-politicize urban youth (Ahmad, 1993). In the first three months after the coup, 30,000 people were arrested, which increased to 122,600 after a year. By September 1982, 80,000 people were still under arrest, with 30,000 awaiting trial (Zurcher, 2004). Meanwhile, the number of terrorist attacks declined by 90 percent after the intervention (Figure 1).

Violent conflict adversely affected post-secondary educational attainment in Turkey through several channels. First, new enrollments in post-secondary education declined in the 1978–1979 school year (Figure 2). Based on the Turkstat data, the decline in the number of students enrolled is 37,715. This decline was mostly related to the closure of state-controlled educational institutions, which were highly involved in the conflict and violence. Those institutions included academies, vocational schools, and teacher-training institutes that were directly affiliated with certain ministries. Yet, most violence among students was seen in those institutions (Binbasioglu, 2005; Tekeli, 2010), which led the government to close 41 institutions out of 64 in 1978. Students already enrolled in the closed institutions were allowed to complete their programs, but new students were not admitted.

Figure 2

After the 1980 coup, the Council of Higher Education was established as a governing board to plan, coordinate, and review the activities of higher education institutions in Turkey (Dogramaci, 1989). This central body would also determine the enrollment capacity of post-secondary education institutions. Figure 2 indicates that enrollments started to increase right after 1982, when all ministry-affiliated higher education institutions were reorganized under the university system.

The second channel that adversely affected educational attainment was that graduation rates declined because of massive student dropouts related to security concerns. Due to the high death toll, many students withdrew their registration in higher education institutions (Kaya, 1981). Some students were unable to finish their education because they were injured or disabled during the turmoil. In addition, some families chose not to send their children to higher education in this period due to heightened risks. Courses were often suspended or canceled during this time. For instance, classes were canceled for 116 days in Ege University and for 421 days in Istanbul University—two of the largest universities in Turkey. The School of Dentistry in Hacettepe University was completely closed during the 1979–1980 academic year (Kaptan, 1986). Faculty offices and student dormitories were often used as weapon warehouses and arsenals (Kaya, 1981; Kaptan, 1986; Gunter, 1989).

And, third, mass student arrests in the wake of the 1980 coup kept many from completing their education. According to a Turkish government report, by 1981, one year after the coup, 9,760 of the state's “captured terrorists” were students. Moreover, 57 percent of the state's 43,140 “captured terrorists” were of age 16–25 (and most were men). In addition to these channels, new enrollments in open education declined by 12,479 between 1977 and 1978.

In this section, we argue that the turmoil generated a substantial decline in post-secondary education among the exposed population—the ones, especially males, born between 1960 and 1965. The decline in higher education as a consequence of a political turmoil in a developing country is an interesting finding in its own. However, we do not stop at this point. In Section 5.2, we use this decline as an instrument based on exposed versus non-exposed cohorts to estimate the effect of post-secondary education on labor market earnings. Finally, in Section 5.3, we show that the decline in post-secondary education during the turmoil had severe negative implications on adult wage and occupational distributions.

In addition to documenting the decline in post-secondary education, this section also provides a background analysis for determining the birth cohorts exposed and non-exposed to the turmoil, e.g., the treatment versus control groups, that will be used in the IV analysis. Although we carry out our baseline empirical analyses focusing on specific treatment and control cohorts, we relax these restrictions in Section 5.2.4, where we perform several robustness checks, and show that the qualitative nature of our baseline results is not overly sensitive to inclusion or exclusion of certain birth cohorts in defining the treatment and control groups. It should also be noted that the baseline analysis is performed using the 2005 wave of the HLFS data set, but the results are robust to including more HLFS waves—as we show in Section 5.2.4.

Post-secondary education. Post-secondary school enrollment rates increased dramatically during the second half of the 20th century all over the world (Psacharopoulos, 1991). Turkey was no exception. However, in Turkey, the enrollment and graduation rates in post-secondary education substantially declined between 1978 and 1982 due to the political turmoil. To analyze the trends in post-secondary education, we use the following regression model: (5.1) si=α+∑c=3054βcdic+Xi′Π+εi, {s_i} = \alpha + \sum\limits_{c = 30}^{54} {{\beta _c}{d_{ic}}} + X_i^\prime \Pi + {\varepsilon _i}, where s_i is a binary variable indicating whether individual i has a post-secondary degree, d_ic is an age dummy indicating whether individual i is c years old, X_i is a vector of covariates, and ɛ_i is an error term.

The age dummies cover the 30–55 age interval. The vector of covariates include 26 NUTS2-level region-of-residence dummies, an urban/rural dummy, and a gender dummy. We omit age 55 in the regressions; therefore, each coefficient β_c can be interpreted as the probability of completing post-secondary education for the corresponding age relative to age 55. In a developing country, one would normally expect β_c to increase as age declines—in the absence of a negative shock on post-secondary education.

Figure 4 plots the estimated β_c's for the whole sample. The solid line shows the probability of completing post-secondary education, while the dashed lines indicate the 95-percent confidence interval. The substantial drop in post-secondary education between ages 40 and 45 is not typically observed in developing countries and may be hard to justify in the absence of a big negative shock affecting post-secondary educational outcomes for those cohorts. Figures 5 and 6 plot the estimated coefficients for men and women, respectively. Although the trends for completing post-secondary education look similar in both figures, the effect of the turmoil is much more pronounced for men than that for women. Based on these results, individuals of age 40–45 are marked as “exposed individuals” throughout the analysis. Those individuals were born between years 1960–1965, and were about 13 to 18 years old as of 1978.

Figure 2 shows that first-year enrollments in higher education declined significantly for the first time in 1978 and remained low until 1982. This decline probably affected young adults of age 17 and 18 from 1978 to 1982. Therefore, the affected group was approximately from 13 to 18 years old in 1978, which is line with the findings reported in Figures 4–6. In addition, student dropouts related to security concerns and mass student arrests after the coup also affected the educational attainment in this age group. Based on these findings, we assert that men of age 40–45 (in the 2005 survey) were the ones most severely affected by the turmoil.

Figure 4

Figure 5

Figure 6

The 40–45 age group has 6 age categories. We construct a comparison group which also has 6 age categories: the 46–51 age group. We choose this older group as the comparison group for two main reasons. First, after the coup, the educational institutions had changed and many new universities were constructed; so, the younger birth cohorts cannot be used as a relevant comparison group. Second, older cohorts were exposed to similar political conditions except violence. Given a natural upward trend in a developing-country context, one would expect a higher post-secondary educational attainment for the 40–45 age group relative to the 46–51 age group. Figure 4 shows that the probability of completing post-secondary education for the exposed group is clearly less than that for the comparison group. Alternative age groups are used in Section 5.2.4 for robustness purposes.

Different levels of education. Next we compare the educational attainment of age groups 40–45 and 46–51 using the following regression model: (5.2) si=α+βzi+Xi′Π+εi, {s_i} = \alpha + \beta {z_i} + X_i^\prime \Pi + {\varepsilon _i}, where s_i is a binary variable indicating whether individual i has graduated from a school (post-secondary, high school, elementary/primary school) or not, z_i is a dummy variable taking 1 if the individual i is in the age group 40–45 and 0 if 46–51, X_i is a vector of covariates, and ɛ_i is an error term.

We use the same vector of covariates as in Equation 5.1.

Table 3 presents three sets of estimates from Equation 5.2. Column 1 displays the result for postsecondary degree, column 2 for only high school degree, and column 3 for only elementary/primary school degree. Column 1 shows that the probability of completing post-secondary education is 1.5 percentage points lower for the 40–45 age group. In contrast, the probabilities of graduation from elementary/primary school and high school increase significantly as expected—see columns 2 and 3. However, the increase in the probability of graduation from high school is 4.5 percentage points, or about twice the increase in the probability of graduation from elementary/primary school. This suggests that those individuals affected from the protests would have normally gone to or completed a post-secondary education, but could not do so due to the 1978–1982 turmoil. So, the number of high school graduates increased more than it normally would, which implies that potential college graduates have remained as high school graduates as a consequence of the turmoil.

We also compare the age groups 46–51 and 52–57. The high school graduation probability is 2.9 percentage points higher for the 46–51 age group. The difference between 40–45/46–51 and 46–51/52–57 differences is 1.6 percentage points—approximately equal to the percentage point decline in completing post-secondary education for the 40–45 age group. This tells us that the main group affected from the student protests were those who would have normally continued post-secondary education in the absence of the turmoil. Figures 7 and 8 confirm this point.

Figure 7

Figure 8

Probability of wage employment. We check whether the turmoil had any effect on the probability of wage employment. In the regression, the dependent variable is a dummy variable taking 1 if an individual is wage employed (regular employee or casual employee) and 0 otherwise (employer, self employed, or unpaid family worker). The sample includes all employed individuals. Figure 9 plots the estimated coefficients of age dummies. The trend is plausibly smooth over the age horizon; so, the turmoil had no effect on wage employment.

We show in Section 5.3 that there is also no impact on labor force participation, formal employment, and informal employment.

Figure 9

Although the probability of being a wage-earner did not change, the school attainment among wage-earners of different age groups might have changed as a consequence of the turmoil. To test this conjecture, we estimate the probability of post-secondary educational attainment for wage earners (column 1) and non-wage earners (column 2), separately. Table 4 reports the results. The estimates clearly suggest that wage-earner men of age 40–45 are the ones whose post-secondary educational attainment have been affected the worst. We conclude that the group most affected from the turmoil is wage-earner men of age 40–45 (in the 2005 survey, with birth years from 1960 to 1965).

The following Mincerian setting is often used to estimate the effect of education on wages: (5.3) wi=α+βsi+Xi′Π+εi, {w_i} = \alpha + \beta {s_i} + X_i^\prime \Pi + {\varepsilon _i}, where w_i is a measure of labor income, s_i is a measure of schooling, X_i is a vector of observables, and ɛ_i is an error term assumed to be independent of the explanatory variables (Griliches, 1977). In this setting, the causal effect of schooling on labor income may not be consistently estimated because of the classical omitted variables problem—also known as the ability bias. A possible solution to this problem is to use an instrument, which requires an exclusion restriction, i.e., at least one observable covariate that affects labor income only through schooling.

In Section 5.1, we show in detail that post-secondary educational attainment declined significantly for individuals born between 1960 and 1965 due to the student protests in the late 1970s and the subsequent military coup. We also argue that we set the non-exposed cohort—the ones born between 1959 and 1954, i.e., the 46–51 age group in the 2005 HLFS wave—as the comparison group.

The 34–39 age group may not be an appropriate comparison group as they were subject to post-coup educational institutions, which enhanced post-secondary education opportunities through new university openings and reorganization of tertiary education system. However, our findings are robust to using preceding or succeeding cohorts as IV. Table 5 reports summary statistics for three age groups: 34–39, 40–45, and 46–51. The 40–45 age group clearly has lower average log hourly wage, fewer average years of schooling, and lower average post-secondary educational attainment than the younger and older groups. Moreover, this group has a higher average high school graduation rates compared to other age groups.

In a heterogeneous-outcome framework, the IV method potentially estimates the average treatment effect (ATE) of schooling on earnings for the sub-group whose schooling attainment is changed by the instrument—i.e., the local average treatment effect (LATE) (Imbens and Angrist, 1994; Angrist et al., 1996; Card, 2001). There are two key conditions (Imbens and Angrist, 1994). The first one is the existence of a valid instrument. Because an individual's year of birth—within a reasonably narrow year of birth interval—is randomly assigned and probably unrelated to individuals’ innate ability, personal characteristics, or family characteristics, it seems reasonable to assert in our case that the wage decline for the 40–45 age group relative to the 46–51 age group is due to the exposure to turmoil and the associated educational disruptions—after standardizing labor market experience and controlling for other observables. Thus, potential outcomes should be independent of the instrument and the exclusion restriction assumption should be satisfied. We show in Section 5.1 that the probability of completing post-secondary education is related to z_i in a non-trivial way. The second condition is monotonicity. This condition ensures that the instrument affects the post-secondary education in a monotonic way (Imbens and Angrist, 1994; Angrist et al., 1996). We also document in detail that the 1978–1982 turmoil negatively affected all sub-samples of the relevant population, which suggests that the monotonicity condition should also be satisfied. Based on these arguments, our IV estimates can be interpreted as the local average treatment effect for those who did not continue post-secondary education due to the 1978–1982 turmoil, but who would have normally obtained a post-secondary degree.

One potential concern about the relevance of our instrument is the possibility that the events leading to post-secondary education disruptions might be a response to declining returns to higher education. We believe that this is not the case as the student protests were not specific to Turkey and happened parallel to global political developments. In fact, it is well known that the college premium significantly increased across the world during that period (Juhn et al., 1993; Oreopoulos and Petronijevic, 2013).

5.2.1

Other effects as confounding factors? A test for instrument validity

One potential threat to instrument validity is the presence of other confounding variables, such as taste differences for risk or discount factors, psychological factors, a fall in the quality of schooling for the exposed cohorts, or different entry effects into the labor market during the turmoil that would affect wages of the exposed individuals independent from schooling quantity outcomes. If confounding factors, other than schooling, influence the wages of the 40–45 age group due to the turmoil, one would observe this by looking at wages “within the same education group.” The individuals affected from the turmoil are high school graduates, who could not complete their postsecondary education. To address this concern, we run separate regressions (similar to Equation 5.1) for both high school and college graduates in which the dependent variable is the log hourly wage standardized to 26 years of labor market experience following Altonji et al. (2012) and the explanatory variables are the age dummies, dummies for region of residence, and an urban/rural dummy. We omit age 51 in the regressions; therefore, the coefficients of age dummies can be interpreted as the gap between the wage of the corresponding age relative to age 51. By these regressions, we also test whether the composition of high school and college graduates changed due to the turmoil in terms of wages. In the regressions, the first sample includes male wage earners of age 40–51 with only a high school degree, the second sample includes only college graduates for male wage earners of age 40–51. The coefficients of age dummies are plotted in Figures 10 and 11, respectively. They lie almost on a straight line and none of the age coefficients are statistically significant. Therefore, the log hourly wages of high school and college graduates born between 1960–1965 (e.g., the exposed ones) are not statistically different from those of the non-exposed cohort. In other words, the wage pool of high school and college graduates did not change due to the turmoil. This suggests that the decline in earnings for the exposed cohorts is solely due to the decline in their post-secondary educational attainment, not due to other confounding factors.

Figure 10

Figure 11

To assess whether the decline in post-secondary education for men of age 40–45 is reflected on earnings, we run a third regression for the high school and college graduate men because we already show that the instrument affects only the post-secondary education. The estimated coefficients of ages are plotted in Figure 12 and, clearly, log hourly wages begin to decline for younger cohorts after age 47, similar to the trend in post-secondary educational attainment in Figure 4. The substantial drop in wages between ages 40 and 45 are statistically significant. We also show the average wage for males for only high school graduates, only college graduates, and both high school and college graduates in the same plot in Figure 13. Clearly, the average log hourly wages of the exposed and non-exposed groups follow almost a straight line for only high school and only college graduates. However, the mean wage of the exposed group is smaller than that for the comparison group. This difference is also statistically significant based on Figure 12. These findings suggest that there are likely no permanent psychological or other confounding effects of the political turmoil on wages.

Figure 12

Figure 13

As a side note, since we analyze wage earnings measured about 25 years after the 1978–1982 turmoil, any initial/temporary effects might have vanished over time.

We implement the semi-parametric procedure developed by DiNardo et al. (1996) to analyze the impact of the political turmoil on the wage distribution of men born between 1960–1965, i.e., the exposed individuals (the 40–45 sample of males). To be consistent with our baseline analysis, the 2005 wave of the HLFS is used in this exercise. We re-weight the 40–45 sample to have the same distribution of post-secondary education as the 46–51 sample. We then compare how labor income is distributed in the re-weighted (counterfactual) 40–45 sample versus the actual 40–45 sample. This comparison roughly demonstrates how the decline in post-secondary educational attainment due to the turmoil affected the density of wages in the treatment group. A formal description of how we formulate and implement the density estimation are presented in Appendix A.

We run a probit model, as in DiNardo et al. (1996), to estimate the re-weighting function and plot the weighted kernel density estimates of the counterfactual (dotted line in Figure 14) and the actual (solid line in Figure 14) densities. We use log hourly wages standardized to 26 years of labor market experience for men—as in the previous sections. Both lines are superimposed in Figure 14.

The Stata optimal bandwidth and Gaussian kernel function are chosen; but, note that the results are not sensitive to the choice of bandwidth and alternative kernel functions.

Figure 14

The difference between actual and counterfactual densities represents the effect of the decline in post-secondary educational attainment—due to the political turmoil—on the distribution of wages for the 40–45 age group. Strikingly, the decline in post-secondary education pushed these individuals from the higher-income group toward the minimum-wage group. Those individuals who would have otherwise completed a post-secondary degree would have earned much more than their actual earnings if the turmoil had never occurred.

Next we explore the impact of the turmoil on the occupational structure. Before this analysis, we address the following question: does the turmoil affect other labor market outcomes such as labor force participation, employment, and labor informality? To answer this question, we run three regressions based on the following simple model: (5.5) si=α+∑c=3450βcdic+Xi′Π+εi, {s_i} = \alpha + \sum\limits_{c = 34}^{50} {{\beta _c}{d_{ic}}} + X_i^\prime \Pi + {\varepsilon _i}, where s_i is a binary variable for labor market status, d_ic is a dummy variable indicating whether individual i is c years old, X_i is a vector of covariates, and ɛ_i is an error term. The HLFS data used in this analysis is restricted to men of age 34–51. In all regressions, we use region-of-residence dummies and an urban/rural dummy as the vector of covariates. Age 51 is the omitted age category.

The first regression is for labor force participation. The estimated coefficients of ages are plotted in Figure 15, which shows that the trend for labor force participation is smooth over the age horizon. Thus, the decline in post-secondary educational attainment did not affect labor force participation for men. The second and third regressions are for employment and formal employment (defined as being registered with the social security institution in the current job), respectively. The results are reported in Figures 16 and 17. Both figures point out that the decline in post-secondary educational attainment did not affect employment and formal employment. We conclude that the turmoil did not have any statistically significant effect on the main employment outcomes.

Figure 15

Figure 16

Figure 17

We show, however, that the turmoil largely affected the occupational structure in the labor market—as hinted by the counterfactual shift in wages. Our data set contains 27 sub-major divisions of occupations and they are classified according to International Standard Classification of Occupations (ISCO-88). We calculate the mean of log wage for each occupation for men and, then, we separately find the share of individuals in each occupation for three age groups (34–39, 40–45, and 46–51). We classify the occupations based on average log wage values and, accordingly, we construct five broad occupation groups. The first two groups can easily be defined, because they contain similar sub-major divisions. The occupations in the last three groups are in different majors. Thus, we classify them based on mean log wage values. The results are presented in Table 12.

The top group is corporate managers and professionals and their mean log wage value is over 1.5—approximately corresponding to the second peak of the distribution of wages in Figure 14. 12 percent of those in the 40–45 age group are within this category. This ratio is much less than the other two age groups—being nearly 7 percentage points lower than the 46–51 age group. This difference is consistent with the difference between actual and counterfactual wage density estimations. We observe that individuals in the 40–45 age group—i.e., the exposed ones—have less attractive jobs on average.

The second occupation group consists of technicians, associate professionals, and clerks. These occupations have less education requirements than the top group. The percentage of this group in the 40–45 age group is higher than the other two age groups and this result is also consistent with the counterfactual density estimation. The third and fourth groups have also confirmed the same result—as the percentages for the 40–45 sample are higher. If we combine the sub-major divisions whose mean log wage values between the minimum wage and 1.5 [the second peak in Figure 14], the fractions of individuals in age groups 34–39, 40–45, and 46–51 become 68.7, 72.5, and 66.3, respectively. This suggests that the 40–45 age group have a higher fraction of low-pay occupations. Therefore, we conclude that the decline in post-secondary attainment led to a shift in occupations from high-pay to low-pay ones. These findings suggest that violent political turmoils or other large-scale turbulent events affecting large fractions of the productive population may have permanent negative consequences on countries’ growth potentials.