When do authoritarian rulers educate: Trade competition and human capital investment in Non-Democracies

See, for example, Baum (2003), Krueger and Lindahl (2001), Glaeser et al. (2004), Sala-i Martin et al. (2004), Sianesi and Van Reenen (2003), Mankiw et al. (1992), Romer (1990), Lucas (1988), Helliwell (1994), Feng (1997), Levine and Renelt (1992), and Webber (2002).

However, recent work has started to question the importance of the democracy vs. non-democracy dimension in understanding distributive policymaking in general. For instance, Mulligan et al. (2004) find no difference in public policies between economically similar democracies and non-democracies. Ross (2006) shows that in policy outcomes such as infant and child mortality rates, democracies are not better than non-democracies despite more money spent on education and health.

Percentages are calculated by authors, using Wright (2008)’s extension of Geddes’ data of authoritarian regime types (Geddes 1999). These numbers are likely to underestimate the actual number, because there are non-democratic regimes that are not yet classified by the data set.

See, for example, Wantchekon et al. (2012), Galston (2001), Campante and Chor (2012), Finkel (2002), Evans and Rose (2012), and Aldomd and Verba (1963).

See Glaeser et al. (2007), Castelló-Climent (2008), Gibson et al. (1992), Glaeser et al. (2004), Papaioannou and Siourounis (2008), and Sanborn and Thyne (2014).

Note that the two aforementioned examples – Malaysia and Indonesia – were both stable authoritarian regimes, the former ruled by a dominant party and the latter a military dictator.

See Ansell (2008), Brown (1999), Brown (2004), Nooruddin and Simmons (2009), Kaufman and Segura-Ubiergo (2001), Avelino et al. (2005), Stasavage (2005), Hecock (2006), Ames (1987), Rudra and Haggard (2005), and Huber et al. (2008).

However, evidence on the connection between democracy and tertiary education is often mixed.

As pointed out by Elis (2011), the origins of public education among many advanced countries predate democratization by a significant amount of time. These examples include Japan, Austria, Germany, and Denmark (Galor and Moav 2006; Lindert 2004).

The exceptions could be countries that monopolize a certain sector of the global market such as some large oil-exporting countries.

Education spending, given its direct effect on human capital, is different from other components of social spending. Take primary education as an example, on the one hand, there is unlikely to be pressure for more spending on primary education to compensate citizens facing increased insecurity due to globalization (compensation hypothesis). On the other hand, the efficiency hypothesis might not work, either, because governments are likely to be pressured by business leaders to improve human capital via education spending. The effects of globalization on education spending might also depend on government development strategies and the comparative advantage of the economy: those specializing in labor intensive production might try to cut social spending while those requiring increased human capital or upgrading might need further investment in education (Hecock 2006).

Many studies show a positive effect of trade openness (Kaufman and Segura-Ubiergo 2001; Avelino et al. 2005). Moving to a lower, sub-national level of analysis, Hecock (2006) finds that maquila exports positively affect education spending.

Our empirical analysis finds that trade openness has no effect on education spending in authoritarian states; nor does it mediate the effect of the trade competition variable defined as the weighted average level of education spending in trade competitor countries.

Recent studies show that diffusion mechanisms are present in a variety of areas such as social welfare policies (Brooks 2005; Gilardi 2007; Cao 2010, 2012b), economic liberalization (Way 2005; Simmons 2004; Elkins et al. 2006), financial regulations (Brooks and Marcus 2012), and environmental policies (Busch et al. 2005; Ward and Cao 2012; Cao and Prakash 2010, 2012a).

We focus on export competition in this study. On how import competition creates policy diffusion between a country and its import-competitor countries, see López-Cariboni and Cao (2015).

As De Mesquita and Smith (2010) summarized: “educated people with access to transport and knowledge of the market are more productive than ignorant and isolated people.”

See, for example, Glaeser et al. (2004), Glaeser et al. (2007), Przeworski et al. (2000), Castelló-Climent (2008), Sanborn and Thyne (2014), and Feng and Zak (1999).

“Economic backwardness” is more likely when the ruling elite is somewhat entrenched but still fears replacement (Acemoglu 2006, ).

The causal chain is a long one: education spending improves productivity, which in turn increases competitiveness in global markets and ultimately increase government revenue. We acknowledge the presence of intermediate variables along the causal chain, but the fact that we find significant results shows that our theory works despite potential intermediate conditions.

For instance, in a sub-national context, Hong (Forthcoming) shows that Chinese local governments with a large natural resource sector have few incentives to invest in labor productivity enhancing social services because abundant resources decrease the need to attract outside investments which often favor higher labor quality.

We make a simplifying distinction between sectors of the economy that are resources-based and those that depend more on a productive population. However, oil, natural gas, and mineral extraction industries are different from agriculture and other primary activities based on relatively large land-endowment. We focus on the former type of natural resources because oil and minerals usually are geographically concentrated and easier for the state to control. They are often important components of non-tax government revenue.

However, Jerry J. Rawlings of Ghana and Joaquim A. Chissano of Mozambique actually won two presidential elections after democratization of the country.

See http://​www.​personal.​psu.​edu/​jgw12/​blogs/​josephwright/​WrightEscribaBJP​SAppendix.​pdf, accessed on May 30, 2013.

The other type of threats for the ruler are those that emerge from within the ruling elite. They are often dealt by establishing narrow institutions such as consultative councils, juntas, and political bureaus (Gandhi and Przeworski 2007).

How authoritarian legislature works still is an ongoing research topic, for an online discussion from experts, see http://​themonkeycage.​org/​2012/​12/​what-do-legislatures-in-authoritarian-regimes-do/​.

Legitimizing the authoritarian regime through popular mobilization in legislative elections could be an additional potential mechanism (Levitsky and Way 2002; Schedler 2006).

Data on education spending are from the UNESCO Institute for Statistics, the Edstats dataset from the World Bank. Results for spending as a percentage of total government spending are similar and available upon request.

For a more detailed explanation of the spatial lag variable, see Section A of the Online Appendix.

Note that the ECM is arithmetically equivalent to a general ADL specification (De Boef and Luke 2008; Keele et al. 2016), and that the spatio-temporal autoregressive model (STAR) is in fact an ADL model with spatial lags. Model selection for TSCS depends on both theoretical and empirical considerations. Potential concerns are the order of integration and whether there exists equation balance in the model (Grant and Lebo 2016). We performed different panel unit root tests for both the education spending data and the spatial lag of education spending. The tests are panel unit root test either based on a pooled statistic (Levin et al. 2002), or a group-mean test averaging augmented Dickey-Fuller regressions for each time series (Im et al. 2003). The evidence suggests we are dealing with stationary data on both sides of the equation. We also performed these tests with unbounded time series data (Lebo and Grant 2016; Grant and Lebo 2016), by taking education spending measured in constant US dollars. Again, unit root tests confirm that our data is most likely to be I(0). This strengthens our confidence in reporting reliable hypothesis testing and long-run multipliers for the substantive interpretation size effects (Grant and Lebo 2016; Keele et al. 2016).

In some of our specifications where we still observe serially correlated disturbances due to the the persistence of education spending data and different sample sizes, we include lagged differences of the dependent variable as a means of purging the remaining autocorrelation.

Results remain unchanged after controlling for aggregate signs of external competitiveness like the external trade balance and consumption prices.

Including the Polity score in specifications with other institutional variables such as legislatures or authoritarian regime types might be redundant, because for some countries, it might pick up certain level of the same information contained in other institutions variables. Excluding the Polity score, however, does not change the main results: regression tables are available upon request from authors.

The per capita income data comes from expanded times series by Gleditsch (2002), updated in 2013. The output gap is estimated as the difference between real GDP per capita and the underlying growth trends, as a percentage of the trend. A Hodrick-Prescott filter (H-P) is used to estimate the underlying growth trend. The H-P filter implements long-run moving average to de-trend the output series. See: (Kaufman and Segura-Ubiergo 2001, 584).

Data are from the World Bank, World Development Indicators (The World Bank 2012).

Using total government spending does not changes the results, but it significantly shrinks the sample size. Data are from the IMF-GFS and the World Development Indicators (The World Bank 2012).

We follow the standard Bewley transformation of error correction model to calculate the long-run multipliers and their corresponding standard errors. See De Boef and Luke (2008).

Adding changes and their interactions with domestic conditional variables makes no difference to the conclusions we arrive from the empirical analysis.

Another way that trade openness could affect education spending is through its mediating effect with trade competitors’ education expenditures. Not all countries are equally open to trade. The same level of spending increase in competitor countries might have a larger impact on a more open economy. We need to test an interactive effect between trade openness and trade competitors’ education expenditure changes. We have tried various openness variables (imports, exports, and total trade) and trade liberalization index (e.g., the KOF index Dreher 2006; Gygli et al. 2018); they do not mediate the effect of interdependence. However, we do find that our results regarding education spending races among trade competitor countries are more important for the period of 1990-2009 (high global economic integration) than for the period of 1970s and 1980s (low economic integration) for developing nations. See Section I of the Online Appendix for more details.

Oil and natural resource rents data are from the World Bank, World Development Indicators. Since both variables are highly skewed, we log transform the data before estimating the models. Taking into account of production costs is very important given the significant cross-country variation in the cost of producing a barrel of oil: in the United Kingdom, it costs $52.50 to produce a barrel of oil; in Brazil, it costs nearly $49; on the other hand, Saudia Arabia and Kuwait can pump a barrel of oil for less than $10 – see http://​money.​cnn.​com/​2015/​11/​24/​news/​oil-prices-production-costs/​index.​html, accessed on November 30, 2017. Ross (2012) offers an alternative measure of oil and gas wealth by dividing the total value of oil and gas production by a country’s population. We think this is a very important measure for a country’s overall oil and gas wealth. But it does not take into account the aforementioned cross-country variation in the cost of production which significantly affects the amount of rents that can be captured by an autocratic ruler. Nevertheless, we conducted robustness checks using this oil income per capita variable. The detailed results are in Section D of the online appendix: we observe a very similar finding – rulers with no natural resources engage in education races; as oil and gas income per capita increases, the effect of interdependence becomes insignificant.

The results here need to be interpreted with caution regarding the substantive effect, especially for areas around extreme values such as those close to the maximum value of the oil variable, because on the maximum side, there are not many observations in the data. As a function of this, the confidence intervals of coefficient estimates when the resource variables approach their maximum values are often large and overlap with the confidence intervals when the values of the resource variables are close to 0 (Fig. 3).

See also Cheibub (1998) who uses predicted hazard rates for leadership failure.

Also, shortened time-horizons may affect education spending negatively if education spending among trade competitor countries is high.

Data are from the World Bank, World Development Indicators.

Section H of the online appendix has more detailed discussion on model estimates.

We thank one reviewer for raising this interesting alternative causal mechanism.

It is possible that these two proxies might not be able to pick up enough variation in the political strength of of teachers’ unions, especially in cross-country context.

We have also replicated estimates from the main paper clustering standard errors to correct for potential time-wise heteroscedasticity and cross-sectional heteroscedasticity and correlation. As shown by Section L and Table 14 of the online appendix, there is no significant difference compared to spatial OLS models in the main paper.

This type of quality improvements can happen in relatively short periods, which allow for increased market shares in export destinations by selling better quality products with only moderate changes in price.

Even at the extensive margin, geographic diversification is more important than the upgrading process that leads to product diversification.

We have argued that partisan legislatures matter because they better enable the two important instruments for authoritarian regime survival: policy concessions and cooptation. Future analysis should look at via which instrument(s) partisan legislature affects regime survival.

For instance, Boix (1997) finds that left-wing governments spend heavily in physical and human capital formation to raise the productivity of factors and the competitiveness of the economy.

Interestingly, we find no evidence for an education spending race in developed country democracies and a negative and significant long-term effect of trade competition in developing country democracies. Details are reported in the Online Appendix, Section C. Why developing country democracies would spend less on education when their trade competitor countries spend more? Our speculation is that when trade competitors invest in education which causes actual or perceived loss of competitiveness for a developing democracy, the country reacts by increasing compensation to trade losers using “short-term-solution” policy instruments (e.g., subsidies) at the expense of the education spending.

Of course, mass mobilization is not a perfect measure of collective action. But in autocracies, mass mobilization is probably among the few ways for the public to express political demands.

This shift in focus to educational inequality is also raised by a recent review essay by Gift and Wibbels (2014).