Democracy and Education: Does the Type of Government Affect Access to Education? Matthew Tolan, Simon Fraser University April 2011 Abstract: In this article, I seek to find whether the type of government in a country affects the amount of education provided. To do this, I propose a model where political power is distributed throughout the population, and their choice to support a government is based on an observable factor; whether the government provides access to education. Using the results from this model, I propose 3 hypotheses: that post-primary education enrollment will remain constant with changes in governance; that primary education enrollment will increase with an increase in democracy; and that the effect democracy has on primary education enrollment will be greater in predominantly rural countries than in predominantly urban countries. I test these hypotheses using data from Africa between 1990 and 2008. Due to the results from the empirical analysis, I fail to reject the first two hypotheses. Furthermore, I find inconclusive results for the third hypothesis. 1. Introduction This paper seeks to find whether the type of government in a country affects the amount of education provided. In today’s political landscape, western nations are required to interact with all types of governments, ranging from democracies to autocracies. Through this interaction, western nations are gaining influence over the political development of these countries, and as such they must understand the ramifications of decisions they make regarding future political changes. To better understand the effects the type of governance has on the provision of public goods in a country, I address a specific subject within development; the link between type of government in a country and the amount of education provided. To do this, I formulate a model that treats the government as a rent-maximizing agent that is held accountable by the subsections of the population with political influence. Depending on the type of government, certain groups have more or less influence, and from this fact, provision of public goods will cater to the influential groups while neglecting the ones who lack influence. From this model I derive three hypotheses: that post-primary education enrollment will remain constant with changes in governance; that primary education enrollment will increase with an increase in democracy; and that the effect democracy has on primary education enrollment will be greater in predominantly rural countries than in predominantly urban countries. I test these hypotheses using panel data from African countries between 1990 and 2008 and I find that the data fails to reject the first and second hypotheses, while I am unable to draw any conclusion for third hypothesis due to insufficient evidence. The paper proceeds in the following format. Section 1 discusses the importance of the topic and the previous works that have been carried out thus far. In section 2, I present a theoretical model and derive my hypotheses. In sections 3 and 4, I discuss the data and methodology I use to test the proposed hypotheses. In section 5, I analyze my results and discuss their robustness. Finally, in section 6, I provide a conclusion. Section 7 contains references and Section 8 contains the appendices. 2 1.1 Related Literature There are a number of papers that analyze democracy and its effect on the provision of public goods. Ansell (2006) and Stasavage (2005) analyze the effect an increase in democracy has on the distribution and level of education expenditure in a country. They both propose simple models where the government is rent seeking, and both derive hypotheses that democratic governments will spend more on primary education, and less on tertiary education then a corresponding autocracy. Ansell (2006) then continues to analyze the effect democracy has on other public goods, but for the purpose of this paper I need not discuss this further. The difference between the two papers lies in their measurement for democracy. Ansell (2006) uses the polity index, which he argues is the best available data for this measurement. Contrary to this, Stasavage (2005) creates his own index based on dummy variables indicating whether there is no political party, a single political party, or multiple political parties as his measure of democracy. Despite the differences in measurement, they find that democracy does increase primary education expenditure, while keeping secondary and tertiary education expenditure constant. Lake and Baum (2001) use a different approach; they study the effect a change in democracy has on education and health indicators such as the literacy rate, gross enrollment rates, and life expectancy. Similar to Ansell (2006), the authors use the polity index to calculate their results. They find that democracy positively affects literacy rates, gross enrollment rates, and life expectancy. However, the authors do not formulate a mathematical model to explain the effects in this article. Due to this lack of a rigorous representation of the government payoffs, the results are based on qualitative theories. However, in a later paper, Lake and Baum (2003) attempt to explain how democracy affects human capital growth, using a rigorous model as part of their argument. They conclude that democracy has the largest effect on growth through an increase in life expectancy in undeveloped countries, and an increase in secondary school enrollment in developed countries. 3 Deacon (2007), Kosack (2009), and Chen (2008) further validate the aforementioned results in their articles. Deacon (2007) looks at a broader spectrum of public goods, and finds that democracy affects many of them positively. Kosack (2009) creates a new model where it is not democracy that directly affects education enrollment, but rather the benefit that such policies have on the leader in terms of greater political power and economic growth. Although the presentation of the argument is different, his empirical analysis yields similar results. Finally, Chen (2008) runs a case study on East Asia, and finds that democracy increases education expenditure and enrollment. In this paper, I seek to build off of these papers but modify their methodology so as to answer a different question. Similar to Stasavage (2005), the model in this paper considers a rural region and an urban region; however, I further allow the regions to contain both elite and commoners. I differentiate the groups based on their demand for education as well as their political influence. Furthermore, I allow for variation in the proportion of the population that lives in rural and urban regions. One of Stasavage’s (2005) assumptions is that the public can view government expenditure. This assumption is likely not valid. In the case of poor countries, the rural population likely does not have the necessary skills to gather or interpret this information; furthermore, in non-democratic countries this information is likely not available. In this regard, using access to education as their deciding factor for political support is more realistic. This would manifest itself in enrollment rates, which allows for a regression specification that is more similar to Lake and Baum’s (2001). Similar to Ansell (2006) I will be using the polity index as my measure of democracy. This is due to its accessibility as well as it being the generally accepted measure in this field. The following model takes strengths from each of these papers, and combines them to make a new model, which better explains how the type of government affects the amount of education provided. Like the above papers, the model only addresses the supply side of education, taking the demand side as given. 4 2. Model 2.1 Structure of Society Consider two societies that differ only in their political systems; one is a democracy and the other is not. Each society contains two regions, a rural region (j=r) and an urban region (j=u). The proportion of the population that lives in one region versus the other varies, and each region has a different distribution of individuals’ demand for education. To account for this, the proportion of the population that lives in the urban region is represented by 𝛽 while the proportion that live in the rural region is represented by 1 − 𝛽. Each individual (i) in each region has a demand for education 𝜃!,! . I assume that individual’s demand for education is uniformly distributed between 0 and 𝜃!! for each region. I further assume that the upper limit is different between the two regions such that 𝜃!! < 𝜃!! . The intuition behind this assumption is that individuals in the rural population are predominantly agriculture based, and their maximum demand for education is lower than those who live in the urban region, who are predominantly manufacturing and service based, which requires more education. This demand for education can be satisfied with a combination of two levels of education (e); primary (p) and post-primary (s) schooling. There is a threshold value of 𝜃 = 𝜃 !→! such that people who have a 𝜃!,! < 𝜃 !→! only demand primary education while those who have 𝜃!,! ≥ 𝜃 !→! demand both primary and post-primary education. I also assume that 𝐸[𝜃!,! ] < 𝜃 !→! < 𝐸[𝜃!,! ]. This assumption states that the average rural person demands primary education while the average urban person demands both primary and postprimary education. Furthermore, there are elite in both the urban and rural regions. These elite are the political, military, and business leaders who have political influence regardless of the political system. The proportion of the elite in the urban region is 𝜀 while the proportion of the elite living in the rural region is 1 − 𝜀. Since these elite are major leaders, I assume that they live predominantly in the urban region, and therefore 𝜀 > 0.5. Furthermore, these elite demand both primary and post-primary education. A graphical representation of the above information can be found in Appendix A, Figure 1. 5 The government (g) can choose whether to provide access to education in each region for both primary and post-primary education separately. I represent this access with an indicator variable, 𝑎!! , that equals 1 if access is provided, and 0 if not. Access can be interpreted as the government providing a specific level of schooling in a specific region at a price that is affordable to all individuals in that region. However, providing this access is costly to the government, which I represent with a cost function, 𝑐(𝑎!! , 𝑎!! , 𝑎!! , 𝑎!! ) where 𝜕𝑐(𝑎!! ) 𝜕𝑎!! > 0. Subtracting this cost from the benefit the government receives (E), the government receives utility, 𝑈! = 𝐸 − 𝑐(𝑎!! , 𝑎!! , 𝑎!! , 𝑎!! ), if the government is in power, and utility, 𝑈! = 0, if it is not in power. Since governments prefer to be in power, I assume that 𝐸 − 𝑐(𝑎!! = 1, 𝑎!! = 1, 𝑎!! = 1, 𝑎!! = 1) > 0 . Depending on which political system is present, the government can only get in power if enough individuals support it. All individuals who’s 𝜃!,! the government satisfies support the government while all those who’s 𝜃!,! they don’t satisfy, do not. 2.2 Democracy In a democracy, both the commoners and the elite have power in the form of a vote. This leads to the following events: 1. The government chooses an education access policy (𝑎!! , 𝑎!! , 𝑎!! , 𝑎!! ). 2. An election is held. Individuals who support the government vote for it, and those who do not support the government, do not vote for it. The outcome of the election is determined by whether the median voter supports the government. 3. Regardless of the outcome of the vote, the elite decide whether to revolt and overturn the government depending on whether their needs are satisfied. Whether this revolt occurs depends on whether the median elite supports the government. 4. Government provides (𝑎!! , 𝑎!! , 𝑎!! , 𝑎!! ) if it succeeds past step 2 and 3. Solving this model for the government’s optimal access policy, two possible solutions occur; one for the case where 𝛽 > 0.5, and one for the case where 𝛽 < 0.5. 6 From event 3, we see that the optimal policy must satisfy the urban elite demand for education (since the median elite is in the urban region). This leads to an access policy which includes 𝑎!! = 1 and 𝑎!! = 1. By providing access to both primary and secondary education in the urban region, the government gets support from all of the urban voters. This means that the government is supported by 𝛽 of the population. If there are more people in the urban region than there are in the rural region (𝛽 > 0.5), the government will receive the median voters vote; therefore, the government does not need to provide any access to education in the rural region. This means that the government’s optimal access policy is (𝑎!! = 0, 𝑎!! = 0, 𝑎!! = 1, 𝑎!! = 1). However, if there are more people in the rural region (𝛽 < 0.5), the government will need to gain support from some people in the rural region to receive the median voters vote. Let 𝜃!∗ be the median voters demand for education. The optimal access policy will satisfy the following equation: !!∗ 𝛽+ ! 1−𝛽 1 𝑑𝜃 = , 0 ≤ 𝛽 < 0.5 𝜃!! 2 or 𝜃!∗ = 1 − 1 𝜃 ! , 0 ≤ 𝛽 < 0.5 2(1 − 𝛽) ! Since, over the domain 0 ≤ 𝛽 < 0.5 , 𝜃!∗ (𝛽, 𝜃!! ) is continuous in 𝛽 and 𝜕𝜃!∗ 𝜕𝛽 < 0 , the range of 𝜃!∗ is 0 < 𝜃!∗ ≤ 0.5𝜃!! . By the stated assumption that 0.5𝜃!! = 𝐸[𝜃!,! ] < 𝜃 !→! < 𝐸[𝜃!,! ], and the fact that the median and the mean are equal in uniform distributions, the optimal access policy is (𝑎!! = 1, 𝑎!! = 0, 𝑎!! = 1, 𝑎!! = 1). Restating the above two cases as one solution, the optimal education access policy under a democratic system, dependent upon the distribution of the population, is: ! 𝑂𝑝𝑡𝑖𝑚𝑎𝑙 𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛 𝑎𝑐𝑐𝑒𝑠𝑠 = ! (𝑎! = 1, 𝑎!! = 0, 𝑎! = 1, 𝑎!! = 1), ! ! (𝑎! = 0, 𝑎!! = 0, 𝑎! = 1, 𝑎!! = 1), 7 0 ≤ 𝛽 < 0.5 0.5 ≤ 𝛽 ≤ 1 2.3 Non-Democracy In a non-democracy, the commoners (non-elite) have no political power and therefore event 2 in the democratic situation is not necessary. This leads to the following events occurring in a non-democracy: 1. The government chooses an education access policy (𝑎!! , 𝑎!! , 𝑎!! , 𝑎!! ). 2. The elite decide whether to revolt and overturn the government depending on whether their needs are satisfied. Whether this revolt occurs depends on whether the median elite supports the government. 3. Government provides (𝑎!! , 𝑎!! , 𝑎!! , 𝑎!! ) if it succeeds past step 2. From the above events, the optimal government access policy must only satisfy the median elite’s requirements. Therefore, (𝑎!! = 0, 𝑎!! = 0, 𝑎!! = 1, 𝑎!! = 1) is the government’s optimal education access policy. Since the commoners have no power, and the median elite is in the urban region, the government has no incentive to provide education access to anyone in the rural region, regardless of the distribution of population between the rural and urban regions. 2.4 Hypotheses I pose 3 hypotheses based on the findings from the above model: 1. A change from a non-democratic system to a democratic system will have no effect on post-primary access and therefore enrollment rates. This hypothesis is drawn from the changes in the 𝑎!! variables when a government moves from a non-democracy to a democracy. With regards to urban access to postprimary education, 𝑎!! , the government must provide access to stay in power under both systems of government. Furthermore, the government will not offer access to postprimary education in the rural region regardless of the form of government. From this, I conclude that, with an increase in democracy, access to post-primary education does not change and therefore enrollment rates do not change. 8 2. A change from a non-democratic system to a democratic system will lead to more primary school access and therefore higher primary enrollment rates. This hypothesis is the result of the difference in the rural access to primary education variable, 𝑎!! , between non-democratic and democratic governments with countries that have a predominantly rural population. Since most of my sample is composed of predominantly rural countries, an increase in democracy should lead to an increase in primary education enrollment rates. 3. The effect an increase in democracy has on primary education enrollment will be greater in predominantly rural countries than in predominantly urban countries. I propose the third hypothesis based on the two different outcomes that occur in democracies, depending on the percentage of the population that is rural. Since the optimal education access policy in a democracy is dependent upon the percentage of the population that is rural, a positive effect on the primary enrollment rate should only be observed for countries with predominantly rural populations. There should be no significant difference in primary enrollment rates between in countries with predominantly urban populations. 3. Data 3.1 Sample My sample includes all the African countries between 1990-2008. The full list of countries is available in Appendix C. Africa is the natural choice to test my hypotheses due to the high variability in democracy both between countries and over time, and the variability of educational enrollment between countries and over time. Furthermore, by using such an important sample with regards to development, the need for external validity is not as strong as that if I used a different, less prominent sample. 9 3.1 Dependent Variables In this paper, my dependent variables are net enrollment in primary school, net enrollment in secondary school, and gross enrollment in tertiary schools (including universities and technical colleges) from the World Development Indicators database by The World Bank (2011). Net enrollment is defined as the quotient of the number of students enrolled in school who are school aged and the total number of school aged people. Gross enrollment is defined as the quotient of the number of students enrolled in school and the number of school aged people. Gross enrollment is used for tertiary school because there is not a defined age limit to tertiary education, and therefore identifying whether the students are of an appropriate age is difficult. Since the denominator of the fraction is the school aged population, and the limits to this population are hard to identify, I must assume that there is more measurement error in the measure of gross tertiary school enrollment than in the primary and secondary school enrollment measures. 3.2 Explanatory Variable My explanatory variable is the Polity IV index created by the Polity IV Project (2011). This index has a range from -10 to 10; however, I standardized the index so the regression coefficients represent the change if the polity index increases 1 standard deviation. This allows for better interpretation of my results. I further create a second explanatory variable that is a democracy dummy variable. This democracy dummy has a value of 1 when a country has received a Polity IV index value of 6 or greater, otherwise it is 0. The value of 6 was chosen because this is the threshold value for democracy recommended by the Polity IV project (2011) authors. 3.3 Control Variables I use the following control variables in my empirical analysis: GDP per capita (PPP, 2008 USD), net official foreign aid and official foreign assistance received, rural population as a percentage of total population, and a variable called “New Regime” that designates whether a regime is 3 or fewer years old. The GDP per capita, official foreign aid and foreign assistance, and rural population percentage are from the World 10 Development Indicators database (The World Bank, 2011). The new regime variable is created from the measure of durability in the Polity IV (2011) dataset. The durability measure in the Polity IV dataset is the number of years since a regime change. The “new regime” dummy is equal to 1 when the durability measure is less than or equal to 3, and equal to 0 otherwise. 3.4 Omitted Data Points There are two main causes for omitted data points; missing data, and conflict indicators in the polity index. Unfortunately, due to data limitations, data is not available for all countries and over all years in the sample. In instances where there is a missing data field, I omit the data point; however, since all the regressions contain country fixed effects, differences in the sample due to a country systematically not recording data would show up in the country fixed effect. The second instance when I omit data points is when the polity index indicates that the country is in a domestic conflict such as a civil war or other internal conflicts. During these periods, education enrollment rates should be systematically lower and measurement error should be more prevalent. By omitting these data points, I reduce the potential for bias due to this. 4. Methodology To test my 3 hypotheses I run 16 regressions; 12 regressions to test hypotheses 1 and 2, and 4 regressions to test hypothesis 3. Section 4.1 contains the methodology for Hypothesis 1 and 2, while Section 4.2 contains the methodology for Hypothesis 3. 4.1 Hypothesis 1 and 2 To test hypotheses 1 and 2, I run 12 regressions, 4 regressions for each level of education on the full dataset. In regressions (1) and (2) I use the polity index for the country as the explanatory variable. This index allows for a higher level of variation in the level of democracy than the democracy dummy. To test whether these results hold true with greater measurement error, I replace the polity index with the democracy dummy for regressions (3) and (4). In all four regressions, I lag both of these two 11 variables two years. I chose a two-year lag because it leads to the best fit for the data (highest R2). The R2 for regressions (1) and (2) using lags from 0 to 5 years can be seen can be seen in Appendix B, Table 2. For all 12 regressions, I control for the natural logarithm of GDP, natural logarithm of foreign aid, and the percentage of the population who are rural. These control variables vary over time and affect education enrollment and, as such, need to be included. GDP is included because countries with higher GDPs may have higher skilled industries that require more education to support; thus leading to a higher natural amount of education in the country. Foreign aid is included because governments may choose education policies that lead to receiving higher levels of foreign aid; specifically, providing more access to education and therefore become eligible to receive more foreign aid. Such a policy is not chosen on the basis of being democratic but rather on receiving as much foreign aid as possible. This effect is not an effect I am interested in and therefore I control for it. I control for the percentage of the population that is living in rural areas because rural regions tend to focus on agriculture, which has a lower education requirement than urban occupations. This variable is included to make countries with different levels of urbanization comparable. In addition, I control for whether the regime is 3 years old or less, as well as an interaction term between being having a new regime and the level of democracy. This interaction term is included because the effect a new democratic regime has on the decision to attend school may be different from the effect a new autocratic regime would have on such a decision. In regression (1) and (3) I use only a country fixed effect. This specification is chosen because there is no reason to observe a trend value for either education enrollment or democracy. Absorbing the changes in enrollment over time with a time fixed-effect variable forces democracy to explain the variation in excess of this trend. I am arguing that this trend is due to changes in democracy, and therefore should not be included. However, to further test for robustness, I include both a time and country fixed-effect variables in regressions (2) and (4). 12 In the following regressions I have used the following labels: the polity index is P, democracy dummy is D, whether the regime is new is N, GDP is Y, foreign aid is A, and the percentage of the population that is rural is R. With this, my regressions specifications are: 1. 𝐸𝑛!" = 𝛼! + 𝛽! 𝑃!,!!! + 𝛽! 𝑃!,!!! ∗ 𝑁!,!!! + 𝛽! 𝑁!,!!! + 𝛽! 𝑙𝑛(𝑌!,! ) + 𝛽! 𝑙𝑛(𝐴!,! ) + 𝛽! 𝑅!,! + 𝜀!,! 2. 𝐸𝑛!" = 𝛼! + 𝛾! + 𝛽! 𝑃!,!!! + 𝛽! 𝑃!,!!! ∗ 𝑁!,!!! + 𝛽! 𝑁!,!!! + 𝛽! 𝑙𝑛(𝑌!,! ) + 𝛽! 𝑙𝑛(𝐴!,! ) + 𝛽! 𝑅!,! + 𝜀!,! 3. 𝐸𝑛!" = 𝛼! + 𝛽! 𝐷!,!!! + 𝛽! 𝐷!,!!! ∗ 𝑁!,!!! + 𝛽! 𝑁!,!!! + 𝛽! 𝑙𝑛(𝑌!,! ) + 𝛽! 𝑙𝑛(𝐴!,! ) + 𝛽! 𝑅!,! + 𝜀!,! 4. 𝐸𝑛!" = 𝛼! + 𝛾! + 𝛽! 𝐷!,!!! + 𝛽! 𝐷!,!!! ∗ 𝑁!,!!! + 𝛽! 𝑁!,!!! + 𝛽! 𝑙𝑛(𝑌!,! ) + 𝛽! 𝑙𝑛(𝐴!,! ) + 𝛽! 𝑅!,! + 𝜀!,! These regressions are repeated for net primary education enrollment, net secondary education enrollment, and gross tertiary education enrollment for a total of 12 regressions. 4.2 Hypothesis 3 To test Hypothesis 3, I create two different data sets from my original dataset; one that contains countries with predominantly rural populations and one that contains countries with predominantly urban populations. The rural dataset is composed of data points where the percentage of the population that is rural is greater than 50%. The urban dataset contains all the remaining data points. I then run regressions (1) and (2) on each data set with primary enrollment as the dependent variable. The specifications remain identical to those used in section 4.1. I omit regressions (3) and (4) because the datasets are much smaller. Since with this reduced size, increasing the measurement error will further understate the coefficients, leading to a large amount of bias. 13 5. Results 5.1 Descriptive Statistics The descriptive statistics for my full dataset are presented in Appendix B, Table 1. The mean and median for primary enrollment are 69.17% and 71.51%, which indicates that primary enrollment is negatively skewed. This is apparent due to a minimum value of 21.24 and a maximum value of 99.60. Furthermore, a standard deviation of 20.66 signifies that there is a high level of variability in this measure. The means for secondary and tertiary enrollment are 28.8% and 5.67% respectively. Despite the low means, standard deviations of 18.77% and 7.94% indicate that there is sufficient variation to test my hypotheses. The skews for secondary enrollment and tertiary enrollment are opposite that of primary enrollment, with means greater than their medians. The Polity IV index is standardized with a maximum value of 1.815 and a minimum value of -1.665. The democracy dummy indicates that 35 percent of my data points are democratic, which is a sufficient amount of variation to make the results of those regressions meaningful. Finally, the mean percentage of rural population is 62.67% with a median of 63.66%. This high mean value indicates that the majority of my data points have predominantly rural populations. From the results of the model, this means that changes in democracy should have an effect on primary education enrollment for the majority of the data points. A maximum value of 94.60% and minimum value of 12.70% show that there is a high level of variation in this sample, thus leading to more valid results. The descriptive statistics for the reduced data sets are presented in Appendix B, Tables 3 and 4. Variations in the democracy measures by country for the predominantly urban data set are presented in Appendix B, Table 5. Table 3 indicates that the countries in the rural dataset are more democratic, have lower primary enrollment rates, and significantly lower secondary and tertiary enrollment rates than countries in the urban dataset. They also have a lower average GDP. Table 4 contains descriptive statistics for the urban dataset. One notable observation is that the distributions of education enrollment are very negatively skewed. This is likely due to the small sample size. This leads to a major issue with the urban dataset; there is not much variation in the level of 14 democracy. Table 5 contains all the countries in the urban dataset, the standard deviations of their democracy variables, and the number of data points present. Of the 12 countries in the dataset, only 4 have a standard deviation greater than 0.2 for their polity index. This low level of variation in the data will cause the regression results to be very inaccurate and therefore conclusions cannot be drawn when using this dataset. 5.2 Panel Regression Results The regression results for the specifications proposed to test my first two hypotheses (in section 4.1) are located in Appendix B, Tables 6 and 7. From regressions (1), (3), and (5) in Table 6, a 1 standard deviation increase in the polity IV index is associated with a 4.98% increase in primary education enrollment, no significant increase in secondary education enrollment, and a 0.70% decrease in tertiary education enrollment. These regressions explain between 33% and 43% of the variation in the data. However, in regressions (2), (4), and (6), which include a time fixed effect variable, a similar change yields an increase of 2.42% in primary enrollment, no significant change in secondary enrollment, and a 1.29% decrease in tertiary enrollment. However, the amount of variation explained by the model decreases to values between 11% and 14%. In both cases, the results support my first and second hypotheses as primary education enrollment does increase with democracy, and secondary education enrollment remains the same. The decrease in tertiary enrollment is not explained in my theoretical model, however this is likely due to the budget constraint of the government and the greater emphasis on primary education spending. If the government is required to spend more on primary education, and also spend on non-education related programs, tertiary education expenditure will likely decrease, and therefore the number of seats in tertiary institutions would decrease with it. Since my model does not take the government budget constraint and expenditure patterns into account, this result is not observed. There are two possible reasons for the change in the R2 between the two specifications. The first possibility is that the time fixed-effect variable is absorbing most of the variation due to an increase in education enrollment over time, despite the increase in democracy over time. The second possibility is that there is an omitted variable that is increasing 15 enrollment over time and if we exclude a time-fixed effect, we are wrongly attributing this change to changes in democracy. The first possibility occurs because both democracy and education enrollment is increasing over time. By including a time fixed-effect, we are forcing changes in democracy to explain the variation in excess of the trend; however changes in democracy may be the cause of the trend itself. In this case, if we exclude the time fixed-effect we will get a less biased measure of the effect. The second possibility is that an omitted variable that is relatively constant across all countries is increasing enrollment over time, by including a time fixed-effect, we are allowing it to absorb an appropriate amount of variation, leading to the coefficient on the democracy variable to be less biased than without it. In this case, the numerical figures in the regressions with the time fixed-effect are less biased. Due to these conflicting possibilities, the numerical values of the regression coefficients cannot be assumed to be the correct values; however, since in both cases the results maintain their signs and significance, I can conclude that democracy does have a significant effect on education enrollment rates. The results in Appendix B Table 7 show that, in the regressions using the democracy dummy, a change from a non-democratic system to a democratic system leads to a 9.38% (or 5.32% with a time-fixed effect) in primary education enrollment, no significant change in secondary education enrollment, and a 1.56% (2.05%) decrease in tertiary education enrollment. Despite the increase in measurement error, the sign and significance of the coefficients remain, further supporting the results in Table 6. The same possible explanation behind the effect the time fixed-effect has had on the results remains from the previous regressions; however, the fact that the coefficients have all retained their sign and significance indicates that changes in democracy have an effect on education enrollment. From these results, I fail to reject both hypotheses 1 and 2. This is true regardless of whether I am using a country fixed-effect or both country and time fixed-effects. The results from the regressions proposed in section 4.2 are in Table 8. From regressions (1), we can see that an increase in the level of democracy leads to a positive and significant change in primary enrollment. However, when a time fixed-effect variable is included, this relationship disappears. The reasons for this are the same as those for the 16 results in Tables 6 and 7. The results in regression (1) are more valid than those in regression (2) for two reasons; theoretically there should be no trend in education enrollment and therefore a time fixed-effect variable is not necessary, and the R2 of the regressions drops significantly when the time fixed-effect is included. Furthermore, despite the coefficients in regression (2) being insignificant, the sign remains the same. In regressions (3) and (4) there are no significant results. This seemingly validates my third hypothesis; however, this is not true. As stated in section 5.1, the urban dataset is too small and does not contain enough variation in the democracy measures to draw conclusions. This is apparent in the signs of the coefficient, which alternate for the interaction term, the new variable, and the rural variable. In the case of the rural variable, both coefficients are significant at the 10% and 5% level, however the signs are opposite. From this, I conclude that the regressions (3) and (4) do not provide adequate evidence that there is no significant effect between democracy and primary education enrollment in predominantly urban countries. Therefore, I am unable to test hypothesis 3 adequately with this dataset, and as such, the results in Table 8 are inconclusive. 5.3 Robustness There are three potential causes for bias in my results; reverse causality, measurement error, and endogeneity. By calculating the optimal lag time in Appendix B, Table 2, the data indicates that reverse causality is likely not an issue. This is because the data is best fit with a 2-year lag between changes in democracy and changes in enrollment. The second best fit is for a lag of 3 years. The regressions with a country fixed-effect as well as those with both time and country fixed-effects support this. These results support the assumption that changes in democracy cause changes in enrollment. The Polity IV index is created by academics on a subjective basis. This means that the indicator is soft and therefore contains measurement error. This measurement error causes my results in Appendix B, Tables 6, 7, and 8 to be understated; however, since the most important aspects of these results are the signs and significance, and they are constant throughout all estimations for the full dataset, measurement error is not an issue. Furthermore, despite injecting more measurement error by creating the democracy 17 dummy, the signs and significance of my variables remains the same. This further validates that these results are not affected by measurement error. Endogeneity is an issue that would cause the results to be biased. There are two sources of endogeneity possible in this analysis; endogeneity due to historical factors that do not vary over time, and endogeneity due to factors occurring during the period analyzed. The first form of endogeneity is not an issue since it would not vary over time. This means that the omitted endogenous variation in the polity index would be accounted for in the country fixed effect variable. The second form of endogeneity is not accounted for in my paper. In other literature, Ansell (2006) uses two instruments to address this issue; the average regional polity index, and a 5 year lagged term of the polity index. Unfortunately, the lack of strength both these instruments have when using this sample makes them unsuitable for this application. However, the results Ansell (2006) finds, when estimating the effect democracy (using the polity index) has on public school expenditure, do not vary much in magnitude or significance between the IV regression and the regular fixed-effect regressions. This indicates that the endogenous portion of the polity index does not cause much bias when analyzing education indicators. This issue is not exclusive to this paper; Baum & Lake (2003) and Chen (2008) indicate that they were unable to find a suitable instrument for their polity index. 6. Conclusion In this paper, I provide a theoretical framework that helps explain the relationship between the type of government and access to education. From this I propose 3 Hypotheses: that post-primary education enrollment will remain constant with changes in governance; that primary education enrollment will increase with an increase in democracy; and that the effect democracy has on primary education enrollment will be greater in predominantly rural countries than in predominantly urban countries. I test these hypotheses using and empirical analysis of African nations between 1990 and 2008. Using a fixed-effects model I find evidence that supports the hypotheses 1 and 2 and this evidence holds up to my robustness checks. Due these results, I fail to reject the first two hypotheses. Finally, due to limitations of my dataset, I am unable to draw a conclusion on 18 the third hypothesis. Further empirical analysis, with an expanded dataset or using a different time period, is required to test hypothesis 3. Despite this, the results from this paper show that the type of government does have an effect on access to education. Therefore, the type of government should be considered when creating policies that seek to affect access to education in developing countries. 19 7. References Stasavage, D. (2005) Democracy and Education Spending: Has Africa’s Move to Multiparty Elections Made a Difference for Policy? American Journal of Political Science, (Vol 49. No. 2) Ansell, B. W. (2006) Traders, Teachers, and Tyrants: Democracy, Globalization, and Public investment in Education, Weatherhead Center for International AffairsWorking Paper Series, (06-01) Lake D. A., Baum M. A. (2001) The Invisible Hand of Democracy: Political Control and the Provision of Public Services, Comparative Political Studies (24:587) Lake D. A., Baum M. A. (2003) The Political Economy of Growth: Democracy and Human Capital, American Journal of Political Science (Vol 47. No 2.) Deacon R. T. (2007) Public Good Provision under Dictatorship and Democracy. Public Choice, (139: 241-262) Kosack S. (2009) Realizing Education for All: Defining and Using the Political Will to Invest in Primary Education, Comparative Education (45:4) Chen J. (2008) Democratization and Government Education Provision in East Asia, Journal of East Asian Studies (8) The World Bank (2011) World Development Indicators, accessed February 2011 at http://data.worldbank.org/ Polity IV Project (2011) Global Report 2009, accessed February 2011 at http://www.systemicpeace.org/polity/polity4.htm 20 8. Appendices 8.1 Appendix A Figure 1: Demand for Education 21 8.2 Appendix B Table 1: Descriptive Statistics for Full Data Set Variable Mean Median Std. Dev. Min. Max. Primary Enrollment 69.17 71.51 20.66 21.24 99.60 Secondary Enrollment 28.80 24.97 18.77 2.55 80.10 Tertiary Enrollment 5.67 2.903 7.94 0.00 55.743 Polity IV 0.00 -0.27 1.00 -1.665 1.815 Democracy Dummy 0.35 ~ ~ ~ ~ ln(GDP) 7.41 7.10 0.96 5.72 10.35 ln(Foreign Aid) 19.63 19.84 1.24 15.28 23.28 Rural Population 62.67 63.66 17.65 12.70 94.60 New Government 0.31 ~ ~ ~ ~ Table 2: R2 for Regressions (1) and (2) for lags of 0 to 5 time years Lag Primary 0 0.388 1 0.416 2 0.424 3 0.408 4 0.356 5 0.394 Country Fixed Effect Secondary Tertiary 0.447 0.313 0.432 0.303 0.426 0.339 0.417 0.336 0.391 0.324 0.426 0.311 Sum 1.148 1.151 1.189 1.161 1.071 1.131 22 Time and Country Fixed Effect Primary Secondary Tertiary Sum 0.062 0.112 0.162 0.336 0.094 0.116 0.136 0.346 0.112 0.120 0.147 0.379 0.114 0.101 0.136 0.351 0.067 0.096 0.118 0.281 0.083 0.041 0.098 0.222 Table 3: Descriptive Statistics for Predominantly Rural Countries Variable Mean Median Std. Dev. Min. Max. Primary Enrollment 66.88 69.29 19.29 21.24 99.60 Secondary Enrollment 24.69 22.50 15.85 2.55 80.10 Tertiary Enrollment 3.77 2.41 5.00 0.29 35.18 Polity IV 0.06 -0.10 1.01 -1.66 1.81 Democracy Dummy 0.39 ~ ~ ~ ~ ln(GDP) 7.14 6.95 0.78 5.83 10.35 ln(Foreign Aid) 19.78 19.95 1.14 16.74 23.28 Rural Population 69.95 69.38 10.70 50.10 94.60 New Government 0.34 ~ ~ ~ ~ 23 Table 4: Descriptive Statistics for Predominantly Urban Countries Variable Mean Median Std. Dev. Min. Max. Primary Enrollment 76.89 87.46 23.20 26.54 98.66 Secondary Enrollment 49.75 60.73 18.71 13.98 71.85 Tertiary Enrollment 12.58 10.75 11.91 0.00 55.74 Polity IV -0.21 -0.62 0.93 -1.32 1.64 Democracy Dummy 0.19 ~ ~ ~ ~ ln(GDP) 8.39 8.64 0.94 5.72 9.65 ln(Foreign Aid) 19.07 19.31 1.43 15.28 21.44 Rural Population 36.05 40.54 11.37 12.70 50.00 New Government 0.19 ~ ~ ~ ~ Table 5: Democracy Variation by Country for Predominantly Urban Countries Country Algeria Angola Botswana Cameroon Djibouti Gabon The Gambia Liberia Libya Morocco South Africa Tunisia Polity Std. Dev. 0.48 0.07 0 0 0.80 0 0 0.39 0 0.08 0.23 0.13 Democracy Dummy Std. Dev 0 0 0 0 0 0 0 0.38 0 0 0 0 24 n 19 8 12 8 19 18 8 9 19 16 17 19 Table 6: Polity Index Panel Regression Results Education Enrollment by Type Variables Primary Secondary (3) (4) (1) (2) Polity (t-2) 4.98*** (0.09) 2.42** (0.97) 0.26 (0.95) Polity*New (t-2) -2.36** (1.12) -1.34 (1.02) New (t-2) -4.52*** (1.13) ln(GDP) Tertiary (5) (6) -0.66 (1.01) -0.70** (0.35) -1.29*** (0.35) 1.46* (0.88) 1.89** (0.87) 1.00*** (0.38) 1.51*** (0.39) -2.76*** (1.04) -4.75*** (0.89) -3.65*** (0.87) -0.33 (0.38) -0.39 (0.38) 5.65* (3.04) -5.31* (2.98) 13.66*** (4.3) 6.46 (4.26) 10.9*** (1.34) 7.25*** (1.50) ln(Aid) 5.15*** (1.11) 4.44*** (1.03) -1.29 (0.79) -1.88** (0.77) 0.75** (0.37) 0.62 (0.39) Rural -1.53*** (0.20) 0.19 (0.26) -0.55** (0.24) 0.21 (0.29) -0.20** (0.09) 0.11 (0.11) Fixed Effect Country Country Time Country Country Time Country Country Time R2 0.42 0.11 0.43 0.12 0.33 0.14 N 397 397 223 223 365 365 Standard errors are in parenthesis (*=p<0.1; **=p<0.05; ***=p<0.01) 25 Table 7: Democracy Dummy Panel Regression Results Education Enrollment by Type Variables Primary Secondary (3) (4) (1) (2) Democracy (t-2) 9.38*** (2.10) 5.32*** (1.93) 2.05 (1.63) Democ.*New (t-2) -2.85 (2.00) -0.38 (1.83) New Regime (t-2) -3.31** (1.32) ln(GDP) Tertiary (5) (6) 0.72 (1.60) -1.56** (0.72) -2.05*** (0.73) 1.69 (1.43) 2.21 (1.41) 1.24* (0.70) 1.91*** (0.70) -2.77** (1.20) -4.84*** (1.01) -3.90*** (1.01) -0.71 (0.45) -1.05** (0.45) 5.78* (3.04) -5.58* (2.96) 12.53*** (4.18) 5.58 (4.22) 10.79*** (1.34) 7.48*** (1.52) ln(Aid) 4.79*** (1.12) 4.18*** (1.04) -1.40* (0.79) -1.99** (0.77) 0.79** (0.37) 0.66* (0.39) Rural -1.69*** (0.19) 0.17 (0.26) -0.52** (0.23) 0.17 (0.29) -0.18** (0.08) 0.13 (0.11) Fixed Effect Country Country Time Country Country Time Country Country Time R2 0.43 0.12 0.44 0.13 0.33 0.12 N 397 397 223 223 365 365 Standard errors are in parenthesis (*=p<0.1; **=p<0.05; ***=p<0.01) 26 Table 8: Reduced Sample Panel Regression Results Primary Enrollment Variables Rural Urban (1) (2) (3) (4) Polity (t-2) 3.72*** (1.36) 1.63 (1.22) 3.28 (2.21) 0.58 (2.76) Polity*New (t-2) -0.80 (1.44) -0.43 (1.31) -0.01 (1.68) 2.62 (2.58) New (t-2) -5.49*** (1.31) -3.10** (1.23) 0.26 (1.84) -0.61 (2.08) ln(GDP) 5.04 (3.54) -5.73* (3.41) 22.58*** (6.70) 3.95 (10.73) ln(Aid) 5.28*** (1.30) 4.89*** (1.18) -0.88 (1.95) -1.06 (2.24) Rural -2.47*** (0.28) -0.82** (0.33) -0.15** (0.37) 2.17* (1.15) Fixed Effect Country Country Time Country Country Time R2 0.50 0.15 0.33 0.12 N 301 301 78 78 Standard errors are in parenthesis (*=p<0.1; **=p<0.05; ***=p<0.01) 27 8.3 Appendix C Sample Countries Algeria Angola Benin Botswana Burkina Faso Burundi Cameroon Cape Verde Central African Rep. Chad Comoros Cote d’Ivoire Congo, Dem. Rep. Congo, Rep. Djibouti Egypt Equatorial Guinea Eritrea Ethiopia Gabon The Gambia Ghana Guinea Guinea-Bissau Kenya Lesotho Liberia Libya Madagascar Malawi Mali Mauritania Mauritius Morocco Mozambique Namibia 28 Niger Nigeria Rwanda Sao Tome and Principe Senegal Seychelles Sierra Leone Somalia South Africa Sudan Swaziland Tanzania Togo Tunisia Uganda Zambia Zimbabwe