The Socioeconomic Effects of Economic Liberalization Revisited: A Case for Democracy Andrew Dawson AUGUST 31 2005 Series #2005-3
ABSTRACT This study empirically investigates the theoretical doctrines of neo-liberal theory and examines the socioeconomic effects of the level of democratization. A crosssectional time series analysis is performed on 102 countries from Latin America, Africa, Asia and Oceania using data from 1975 to 2003. The results indicate that neo-liberalism has a positive effect on economic growth, but no effect on improvements in health or education. Moreover, democracies are found to have no relation to economic development, but are associated with a more equitable distribution of non-economic resources. A case is made that modern democracies compensate for the shortcomings of neo-liberalism, and alternative forms of political organization are discussed. 2
INTRODUCTION For the last quarter century, economic liberalization, or neo-liberalism, has been an influential model of development. By 1979, both the World Bank and the International Monetary Fund (IMF) had adopted the neo-liberal development philosophy (Corbo and Fischer 1992). With this endorsement by the international financial institutions, economic liberalization emerged as a central component within the discourse of development. More importantly, however, it now had significant financial backing from the international community, which enabled the implementation of formal neoliberal economic adjustment programs in developing countries. This shift to economic liberalization occurred after the state-centered development strategies of the post-war period produced relatively few successes. Some argue that the contemporaneous changes in the international economic environment during this period, most notably the second oil shock, declining terms of trade for developing countries, low export demand for primary products, high world interest rates and the debt crisis, contributed to the change in development strategy (Ghai 1991; Serieux 1999). Given its ascendancy during the last twenty five years, economic liberalization has been a heavily disputed subject among academics, civil society, anti-globalization activists and the general public. Many studies have been commissioned which examine the socioeconomic consequences of this approach to development. However, most cross-national research uses official government policy as their measure of neo-liberalism and does not directly measure the extent of economic liberalization; nor has past research investigated the effects of democracy on economic liberalization. In addition, most 3
studies do not make use of data after the early 1990s, thus it is important to revisit the issue in order to get a more complete picture using data that spans a longer period of time. This study attempts to address these deficiencies by drawing upon more recent data to investigate the following questions: 1) What are the socioeconomic effects of concrete changes in economic liberalization? 2) How does democracy mitigate the effects of economic liberalization? ECONOMIC LIBERALIZATION Economic liberalization can be defined as government policies that promote free trade, deregulation, the downsizing or privatization of public services, and the reduction or elimination of subsidies, price controls and rationing systems (Woodward 1992). Stated differently, these policies follow a non-interventionist, or laissez-faire, approach to economic activity that relies on market forces for the distribution of resources. Government intervention in the market is considered inefficient and should be limited. It is argued that the state may act from a self-interested perspective, using their power and the authority of government for their own selfish ends (Todaro and Smith 2003: 129). However, even if the state acts with good intentions, it is assumed to not have the competence to direct the economy, thus moving scarce resources into less productive economic activities, thereby reducing overall economic growth (Todaro and Smith 2003). Aside from advocating against intervention in the domestic economy, neoliberalism also encourages unrestricted international trade. According to the law of comparative advantage, it is claimed that all countries will mutually benefit from free 4
trade. This law states that a country that is less productive than its counterparts will still benefit from the removal of tariffs and other protective devices, as market forces will induce it into moving resources [within the country] from industries with a comparative disadvantage into those with a comparative advantage (Lal 2002: 37). As such, it is reasoned that the country importing competitive manufactured goods will be able to raise its overall productivity and real income (Lal 2002: 37). Looking beyond economic growth, it is important to highlight the neo-liberal approach to social welfare. Regarding distributive issues and social justice, neoliberalism claims that these matters are managed by the trickle-down effect. Originating in supply-side economic theory, the trickle-down effect maintains that economic growth will eventually benefit all members of a society. Even if wealth creation is not initially equally distributed, in a free market it will nonetheless bring about economic expansion through activities such as employment creation and increased demand for skilled labour which will ultimately improve both the income and the general well-being of the entire society. 1 Literature Review Many scholars have challenged the neo-liberal approach to both economic growth and social development in the third world (Bradshaw, Noonan, Gash, and Serchen 1993; Buchmann 1996; Caves 1996; Chang 1998; Cornia, Jolly, and Steward 1987; Crotty, Epstein, and Kelly 1998; Noorbakhsh 1998; Weisbrot, Baker, Kraev, and Chen 2002). From a strictly economic perspective, some studies have shown that a number of 1 See Canto, Joines and Laffer for a detailed treatment of supply-side economics. 5
developing countries experience difficulties attracting foreign investment, and its ensuing economic growth, long after their ratification of liberal trade policies (Chang 1998; Crotty, Epstein, and Kelly 1998). In addition, it has been argued that favourable tax, wage and regulation concessions are not necessarily the primary factors in international investment decisions. A host of other considerations including quality infrastructure, local demand, education levels, language, culture and proximity to the investment capital's home country have much relevance in the decision making process (Caves 1996). Notwithstanding research on foreign investment and economic growth, most cross-national studies of the socioeconomic effects of neo-liberalism have focused on the lending activities of the international financial institutions. Although there are a few cases of voluntary liberalization in the developing world, for instance in Chile after Pinochet s ascension to power, in many cases commitment to liberalization was primarily a result of pressure from the international community. Aside from political pressure from the industrialized nations, much of the desperately needed financial assistance from both the IMF and the World Bank became conditional upon commitment to economic liberalization. They required that developing countries adopt structural adjustment programs, which require the enactment of specific neo-liberal policies, as a condition of receiving these loans. Further financial pressure was added by private international lenders, who used the IMF as an informal credit rating service by only lending to developing countries that had implemented a structural adjustment program (Bradshaw and Wahl 1991: 254). 6
As a result of the conditional lending programs of the IMF and World Bank, todate the majority of cross-national studies of the effects of economic liberalization in developing countries have examined the impacts of structural adjustment programs. Analyzing data between 1951 and 1990, Pzreworski and Vreeland (2000) find that countries under these programs experienced slower economic growth while controlling for the factors that led to participation in IMF adjustment lending. Using more recent data (from 1975 to 1999) and also controlling for factors that lead to IMF lending, Barro and Lee (2002) discover that loan participation has an insignificant initial effect followed by a significant negative effect on economic growth after a five year lag. In their seminal work, Cornia, Jolly and Stewart (1987) brought to the fore the issue of the social impacts of structural adjustment, including its deleterious effects on health and education, and advocated adjustment with a human face. That research spawned a number of cross-national studies that focused on the social effects of structural adjustment, with mixed results. Among the studies that go beyond analyzing specific segments of the population, Bradshaw and Wahl (1991) examine how IMF structural adjustment lending affects both GNP per capita and the physical quality of life in developing countries. The physical quality of life is an index derived from life expectancy at age one, infant survival rates and secondary school enrolment rates. Comparing the variables both before IMF lending (1975) and after IMF lending (1987), it is found that IMF loans have a strong negative 7
impact on GNP per capita, but have no statistically significant impact on the physical quality of life index (Bradshaw and Wahl 1991). In other study, Crisp and Kelly (1999) examine the socioeconomic impacts of both IMF and World Bank structural adjustment lending in sixteen Latin American countries. Using data for the period 1980-90 and controlling for the extent that each loan recipient actually adjusted, the authors compare indicators of economic growth and inequality before and after structural adjustment lending. They find a weak positive relationship between adjustment and economic growth, and declining income inequality in countries that have adjusted (Crisp and Kelly 1999: 549). Two World Bank sponsored investigations show that the institution s structural adjustment lending has a positive effect on economic growth, no effect on the under-five mortality rate and a negative effect on gross primary school enrolment ratios (Corbo and Rojas 1992; Maasland and Van der Gaag 1992). While controlling for variables that led to participation in the adjustment programs, the authors compared countries that had received intensive lending (two or more structural adjustment loans by 1985), other adjustment lending (one structural adjustment loan by 1985) and no structural adjustment lending. It was discovered that intensive lending countries had increased in GDP growth, when both periods of 1981-84 and 1985-88 where compared to 1970-80 (Corbo and Rojas 1992: 32). Using the same data categories, Maasland and Van der Gaag (1992) found that there were no differences in changes in under-five mortality rates between intensive lending countries and countries with no lending when comparing the two time 8
periods of 1975-80 and 1980-85. The gross primary school enrolment ratios of intensive lending countries were found to have declined between 1980 to 1985, while all other groups had increased (Maasland and Van der Gaag 1992: 58). Another study that used the aforementioned World Bank data found that the structural adjustment programs had no effect on social well-being, but a positive effect on economic growth. Noorbakhsh (1998) compared changes in measures of economic growth, health and education between the periods 1979-85 and 1986-92. Again using the same three country categories, Noorbakhsh found that over time, aside from an increase in GDP per capita PPP, there were no statistically significant differences between infant mortality rate, life expectancy, adult literacy and gross primary school enrolment ratios for all three country groups (1998: 768). In a study focused solely on health, Van der Gaag revisited his earlier research with Barham (1998), using the same data categories of World Bank structural lending recipients. However, each category now comprised of countries that had received loans before 1990, as opposed to 1985. With this re-conceptualization, the countries that received no adjustment lending were shown to have the most economic growth between 1985 and 1990. Comparing infant mortality rates of all country groups between 1972-82 and 1982-92, it was shown that infant mortality declined faster in all country groups except for the intensive lending countries, which slowed its pace of progress (Van der Gaag and Barham 1998: 1005). While comparing the 1970 under-five mortality rate data with the period 1980-90, they found that countries that did not receive any adjustment 9
lending were the only ones to experience accelerated progress (Van der Gaag and Barham 1998: 1006). Using the redefined countries categories from Van der Gaag and Barham, Kakwani compares the period average of 1981-85 with the average of 1986-90, and finds that intensive lending countries enjoy faster economic development as measured by GDP per capita (1995: 482). It is interesting to note that taking the period average creates a contradiction with Van der Gaag and Barham who use the exact same data and found that countries with no lending had the most economic growth. Kakwani also found that intensive lending countries had slower progress in life expectancy at birth than all other countries, while there was no statistically significant difference in infant mortality and literacy rates between all country groups (1995: 495). Literature Summary and Analysis Summarizing the cross-national studies, four studies claim the structural adjustment has a positive effect on economic growth and four claim a negative effect; one claims that it has a negative effect on under-five mortality and another claims no effect; one claims a negative effect on infant mortality and two claim no effect; one claims a negative effect on school enrolment rates and another no effect; one claims a negative effect on life expectancy and another claims no effect; two claim no effect on literacy rates; one claims a positive effect on (thereby reducing) income inequality; while one study claims that the overall physical quality of life is not affected by structural adjustment lending. How can these divergent findings be explained? Apart from slightly 10
different datasets and methodologies, I propose two main reasons: 1) the conflation of economic stabilization with economic liberalization, and 2) the operationalization of economic adjustment. Both the IMF and the World Bank have granted economic stabilization loans. Generally, these are short-term loans awarded to get countries out of crises situations, mainly as a result of balance of payments problems, which cause a shortage of foreign exchange necessary to service foreign debt and purchase imports. In most cases, these loans are contingent upon exchange rate devaluation, deficit reduction or stringent monetary policies in order to correct the current account deficit and/or to control hyperinflation (Woodward 1992: 31). However, as Baer and Maloney argue, investigating the socioeconomic effects of stabilization measures is uninteresting, as crisis situations must be corrected regardless of the particular economic model adopted (1997: 315). Stated differently, if a country was running unsustainably large deficits each year, measures to reduce the deficit may cause adverse effects. However, as the original situation was unsustainable, these effects should not be taken into account when evaluating particular economic systems. Thus, stabilization should be separated out from longer-term structural adjustment loans that support economic liberalization and the restructuring of stable economies. Of the aforementioned research, Bradshaw and Wahl, Crisp and Kelly, Pzreworski and Vreeland, and Barro and Lee include stabilization loans when examining structural adjustment lending. 11
Earlier research hinted that loan recipient countries may not adjust their economies as expected (Campbell 1993; Crisp and Kelly 1999). However, it was not formally tested until Easterly (2005) found that structural adjustment lending is not a good predictor of structural adjustment. While ignoring policy and directly measuring the change in macroeconomic indicators, Easterly compared macroeconomic changes with the occurrence of IMF and World Bank adjustment loans for the period 1980-99. It was revealed that although macroeconomic adjustment did occur, it was not related to adjustment lending (Easterly 2005: 14). Even though he only empirically examines the stabilization component of adjustment lending, Easterly postulates that the relationship between economic liberalization and structural adjustment lending should be just as weak as with stabilization (Easterly 2005: 20). Apart from Crisp and Kelly, no other crossnational study controls for the extent of actual adjustment. In response to these shortcomings, I will directly measure the level of economic liberalization, excluding stabilization measures, and examine its socioeconomic effects. Although it is difficult to operationalize concepts such as social welfare, following Sen (1999), this study uses three socioeconomic indicators to measure the level of well-being within society: income, health and education. Hypothesis #1 Investigating the socioeconomic effects of economic liberalization can be framed as a test of the neo-liberal theoretical claims against what happens in practice. Thus, if the theory is correct, we would expect to find that income, health and education all 12
improve with the level of economic liberalization, according to the doctrines of noninterventionism, the law of comparative advantage and the trickle-down effect. THE ROLE OF DEMOCRACY Gourevitch (1992) summarizes the two theoretical positions on how democracy mitigates the operation of free markets and the resultant effects on economic growth. The first theory is that markets require democracy. In sum, concentrated political power disrupts efficient resource allocation through rent-seeking behaviour by those in power. Democracies are considered to have greater controls against this type of abuse of those who hold public office (Gourevitch 1992: 1272). The second theory is that economic liberalization and democracy are antithetical; in fact, it is claimed that free markets require authoritarian regimes. It is argued that democracies are vulnerable to populist pressures which distort markets through taxes and regulation (Gourevitch 1992: 1272). Regarding the social consequences of democracy, Vanhanen (1997) proposes a resource distribution theory. He argues that the distribution of resources (in the form economic wealth, education and the ability to use physical force) is related to the distribution of power within society. Resources are considered a source of political power, although it is acknowledged that one can also acquire resources by political means (Vanhanen 1997: 23). However, since not all resource distribution occurs within the political realm, the distribution of resources is considered the primary determining factor in the distribution of political power. Therefore, the more resources are concentrated in a population, the more political power will be concentrated. As such he contends that 13
democratization takes place under conditions in which power resources have become so widely distributed that no group is any longer able to suppress its competitors of maintain its hegemony (Vanhanen 1997: 24). Vanhanen s resource distribution theory endeavors to explain the level of democratization based on the extent of the distribution of resources in society. According to the theory, it logically follows that, ceteris paribus, democracies will have a more equitable distribution of resources than dictatorships. That is, we would expect to find resources such as wealth, education and health more equitably distributed within democratic societies. Hypothesis #2 Despite the theoretical debate on democracy and economic growth, empirical research has found that there is no significant relationship between the two concepts. 2 As such, I hypothesize that the same relationship will hold true in this study. The relationship between democracy, health and education can be modeled using Vanhanen s resource distribution theory. Of the three dependent variables, health and education measures can be considered as rough indicators of the distribution of resources within society. Average income is not a valid proxy for distribution, as it is possible for the average income to increase while the majority of the population experience stagnant, or even decreasing, income levels. This is not the case with measures of health and education, thus they are more accurate measures of resource distribution. Accordingly, it 2 See Przeworski, Alvarez, Cheibub and Limongi. 14
is expected that democracy will contribute to improvements in the levels of health and education. DATA AND METHODS The units of observation in this study are developing countries, defined as countries classified as either low or middle income by the World Bank (World Bank Group 2005b). The regions under investigation include Africa, Latin America, Oceania and Asia (including the Middle East, but excluding the former Soviet republics and Cyprus). In addition, all countries in the study have an estimated population greater than five hundred thousand inhabitants in 2003. In total, this study examines 102 different countries. Methodology The fixed effects cross-sectional time series model is used to test the hypotheses. In essence, the only concern of this statistical test is to determine how changes in the independent variables relate to changes in the dependent variables by primarily drawing upon the time-series data. As such, the fixed effects model is somewhat unique in crosssectional time series analysis in that it excludes all time-invariant variables within each country in the calculations of the coefficients. One drawback of the model is that time-invariant controls, such as dummy variables for region or whether a country exports oil, cannot be used. It is also difficult to make inferences beyond the data values of the independent variables in the sample; 15
however, such problems are minimized in cross-national studies where the sample being analyzed contains almost all of the countries for which the results are to be generalized. An advantage of using fixed effects is that the model can handle unbalanced datasets (missing data). Another benefit of this methodology is that it is not necessary to control for selection bias, as countries are in effect being compared to themselves using the timeseries data. Therefore, no complex formula is required to match countries that have and have not undergone economic liberalization in order to estimate the counterfactual. The study uses three time series cross-sections. For the independent variables, these years are 1975, 1985 and 1995. The ten year gap between each cross-section is a result of the absence of annual data for some of the variables. Regarding the time lag between the independent and dependent variables, some authors argue that a five year period is sufficient in order to determine the effects of economic changes (Stokes and Anderson 1990); others maintain that 10 years is more appropriate (Bradshaw, Noonan, Gash, and Serchen 1993; Buchmann 1996). This study takes the latter approach, emphasizing the investigation of long-term socioeconomic effects. Therefore, the corresponding dependent variables for the above independent variable cross-sections are 1985, 1995 and 2005 respectively. However, at the time of writing, 2005 data for the dependent variables were not available to measure the effects of the independent variables in 1995. Thus, all three dependent variables use the 2003 data in its place. 16
Data Definitions Dependent Variables As mentioned, income, education and health are used as proxies for social wellbeing. Although the United Nations Development Programme s Human Development Index captures these data categories neatly in one index (United Nations Development Programme 2003), it is important to independently examine the effects of each variable in order to properly test the hypotheses and separate out the specific consequences. Income: Income is measured by gross national income (GNI) per capita, PPP (formerly gross national product or GNP). GNI is the sum of value added by all resident producers plus any product taxes (less subsidies) not included in the valuation of output plus net receipts of primary income (employment and property) abroad on a per capita basis. GNI PPP is gross national income converted to international dollars using purchasing power parity rates. An international dollar has the same purchasing power over GNI as a U.S. dollar has in the United States. The data are expressed in current international dollars (World Bank Group 2005b). Health: The under-five mortality rate is used as the measure of health. It measures the probability of dying between birth and the age five expressed per 1000 live births (UNICEF 2005). Although the data required for this study are for the years 1985, 1995 and 2003 (representing 2005), the under-five mortality rate data are only available for 1980, 1990, 1995 and 2003. Thus, I derived the 1985 data by taking the average of the 1980 and 1990 figures for each country. 17
Education: Education is expressed as the gross primary school enrolment ratio. It is the total enrolment in primary education, regardless of age, expressed as a percentage of the population in the official primary school age group (UNESCO 2005). Independent Variables Data for some of the independent variables are available on an annual basis. For those that are better represented as period averages, the 1985 value is the average of the ten year period 1976-85 and the 1995 value is the average of the ten year period 1986-95; however, the 1975 value, which represents the initial state of affairs, is simply the value corresponding to that year and not a period average. The superscript symbol denotes measures that use period averages. All other variables use the values corresponding to the particular year. Population: Previous studies have found that population growth has a negative impact on per capita income (Kakwani 1995; Przeworski, Alvarez, Cheibub, and Limongi 2000). Logically, this makes sense as per capita income is a function of both total economic growth and the size of the population. Thus, population is included as a control. It is defined as all residents of a territory, regardless of legal status, except nonpermanently settled refugees, who are considered part of the population of their country of origin (World Bank Group 2005b). Debt Service Ratio : Following Bradshaw and Wahl (1991), a measure of debt dependence is used as a control. The debt service ratio is the total debt service divided by 18
exports of goods and services (including workers' remittances), expressed as a percentage (World Bank Group 2005a). Democracy: As the definition of democracy is a contentious issue among academics, the operationalization of the concept is equally, if not more, controversial. Dahl (1971) labels contemporary representative democratic systems as polyarchies, which he defines as polities with high participation rates (in both elections and public office) and high levels of political competition. Contrarily, Przeworski et al. argue that participation should not be included as a definitional feature of democracy (2000: 34). Consequently, they advocate a purely institutional approach to measuring the level of democratization. Given the disagreement, this study employs two different measures of democracy, tested separately, that correspond to both definitions. This allows a test of robustness across the two different measures. For the time periods used in this study, the two measures of democracy are highly correlated (0.836), but there is enough disparity to warrant testing both measures. 1. Index of Democracy : Following Dahl, Vanhanen (1997) created an index of the level of democracy (or polyarchy) based on the extent of political participation and competition. With the index, a higher score corresponds to a greater level of democratization. The index is calculated by multiplying the competition variable by the participation variable (both expressed as a percentage) and dividing the product by 100. The competition variable is equal to the percentage of votes won by the all smaller political parties, calculated by subtracting the percentage of votes received by the largest 19
political party from 100. In the event that voting percentages are not available, the index uses the distribution of seats in parliament. There is an upper limit of 70% for the proportion of votes won by smaller parties, in order to reduce bias of higher scores achieved by those countries that use proportional electoral systems, which can lead to a large number of small parties. The participation variable is calculated by the number of people who voted divided by the total population (International Peace Research Institute 2005). 2. Polity IV Polity Score : Ted Gurr s Polity Project is an institutional measure of the level of democratization. This study uses the Polity 2 time series indicator, which measures both the level of democracy and autocracy, with scores ranging from ten (full democracy) to negative ten (full autocracy). 3 The indicator is calculated by subtracting the autocracy score (ranging from 0-10) from the democracy score (also ranging from 0-10). Periods of interregnum are given a neutral score of zero and transition periods are prorated over the transition years (Polity IV Project 2003). Inflation : As a control for hyperinflation, the consumer price index is used as a measure of inflation. It reflects changes in the cost of acquiring a fixed basket of goods and services by the average consumer expressed in percentage change per annum (International Monetary Fund 2005). International Interest Rate : World interest rates are a potential source of external economic shocks. Therefore, the Special Drawing Rights (SDR) rate of interest is 3 See Appendix A for a breakdown of the Polity IV democracy and autocracy indices. 20
included as a control. SDR value is determined on a daily basis from a basket of currencies, with each country (mainly U.S., Germany, France, Japanese and the United Kingdom) being assigned a weight in the determination of value. The SDR interest rate is based on the interest rates of the combined short-term money market instruments of these countries (International Monetary Fund 2005). Economic Liberalization: Two variables are used to measure the extent of economic liberalization within countries. Government size is used as a proxy for intervention within the domestic economy. Freedom of international trade is used to measure the level restriction-free international trade in each country. 1. Government Size: This measure is derived from the size of government area rating from the Economic Freedom of the World index. It is an index that examines the extent that countries rely on individual choice and markets, as opposed to the government, for the allocation of resources. Components of the index are: government consumption as a share of total consumption; transfers and subsidies as a share of GDP; the extent that countries use private rather than public enterprises to produce goods and services; and the fourth component is an average of the top marginal income-tax rate including the income threshold at which it applies. As some index components were missing in earlier periods, I recalculated the index based on data available across all three time periods for each country. Thus, the changes are comparable over time. The scale ranges from 0 to 10; with a higher rating signifying a lower level of government intervention in the market (Gwartney and Lawson 2004). 21
2. Freedom of International Trade: A variable also derived from the freedom to trade internationally area rating from the Economic Freedom of the World index. Free trade is measured by the extent that a country has low tariffs, a trade sector larger than expected, a freely convertible currency and few controls on capital. The index is calculated by averaging the following scores: the actual versus the expected size of the trade sector; the difference between official and black market exchange rates; restrictions in foreign capital market exchange; and the average of both the international trade tax revenue score and the mean tariff rate score. The area rating and its components range in value from 0 to 10; with a higher rating corresponding to an increased level of free trade (Gwartney and Lawson 2004). As with the government size index, some of the freedom to trade internationally index components were missing in earlier periods. I recalculated the index based on data available across all three time periods for each country, making the data comparable over time. Descriptive Statistics The means, standard deviations, maximum and minimum values, and the number of observations for each variable are presented in Table 1. As there are three time series cross-sections and 102 countries (or groups), a variable without missing data would have 306 observations. Unfortunately, for a variety of reasons, this is difficult to achieve in practice. However, of the eleven variables, only four have fewer than 260 observations. The two control variables, debt service ratio and inflation, have 206 and 236 observations, respectively. The indicators of economic liberalization, government size and freedom of international trade, have the fewest observations. This is a direct result of 22
the calculations used to ensure backward compatibility with the time series data. Nonetheless, both measures provide roughly two thirds of the expected number of observations across two thirds of the expected number of countries. Table 1. Cross-Sectional Time Series Descriptive Statistics Variable Mean Std. Dev. Min Max Observations Income Per Capita overall 2,991 2580.05 330.0 13230.0 N = 267 between 2380.38 486.7 11446.7 n = 92 within 1035.98-982.5 7017.5 Under-Five Mortality Rate overall 110.6 74.59 7.0 320.0 N = 305 between 72.12 11.8 298.7 n = 102 within 19.53 57.4 168.9 Gross Primary School Enrolment Rate overall 93.7 27.23 13.6 191.7 N = 262 between 25.62 13.6 155.3 n = 99 within 11.85 42.7 138.6 Population overall 35,400,000 131,000,000 210,000 1,200,000,000 N = 304 between 129,000,000 381,333 1,060,000,000 n = 102 within 18,500,000-121,000,000 197,000,000 Debt Service Ratio overall 18.8 12.30 0.0 64.5 N = 206 between 10.90 2.6 46.8 n = 91 within 6.11 1.1 36.5 Vahanen's Index of Democracy overall 4.5 6.77 0.0 40.1 N = 295 between 5.77 0.0 23.3 n = 101 within 3.49-11.4 28.7 Polity IV Polity Score overall -2.9 6.15-10.0 10.0 N = 295 between 5.53-10.0 10.0 n = 101 within 2.72-10.9 8.1 Inflation overall 70.5 316.39 0.8 3296.5 N = 236 between 204.05 1.1 1127.4 n = 92 within 249.15-1028.3 2239.6 International Interest Rates overall 6.1 1.33 4.4 7.6 N = 306 between 0.00 6.1 6.1 n = 102 within 1.33 4.4 7.6 Government Size overall 5.5 1.65 1.5 9.2 N = 204 between 1.43 2.0 8.8 n = 68 within 0.83 2.8 8.0 Freedom of International Trade overall 4.7 2.02 0.0 8.5 N = 198 between 1.56 0.0 8.0 n = 70 within 1.29 1.0 9.0 Note: "N" is the number of observations; "n" is the number of groups 23
Testing for Violations of Model Assumptions A test of multicollinearity between the covariates using OLS regressions for each cross-section of every time series model found that the independent variables are not highly collinear with one another (all variance inflation factor scores were under 3). Regarding linearity, each independent variable was tested separately to determine whether it was better represented by a power term or polynomial function. The freedom of international trade variable was found to be better represented by a polynomial function by adding trade to the second power as a covariate in under-five mortality rate model using the Polity IV measure of democracy. In addition, it was found that taking the natural log of GNI per capita significantly improved the fit of both its models, as well as using the natural log of the under-five mortality rate for the model that uses Vahanen s index of democracy. As heteroskedasticity cannot be directly corrected using the fixed effects cross-sectional time series statistic in STATA, the data was transformed and modeled using OLS. The gross primary school enrolment rate model using Vahanen s index of democracy was the only model that proved to be heteroskedastic. When correcting the OLS model using Huber-White standards errors, only the population coefficient changed in significance (became significant). However, this is of little consequence for this model as both the OLS and fixed effects statistics have F-test p- values above 0.05, hence the models on aggregate not add to the explanation of the phenomena under investigation. 24
RESULTS Two models were run against each dependent variable. The first model (Model 1) uses Vahanen s index of democracy; the second model (Model 2) uses the Polity IV polity score. Table 2. Fixed Effects Cross-Sectional Time Series of Per Capita Income Per Capita Income Independent Variables Model 1 Model 2 Government Size 0.070** 0.081*** (2.33) (2.78) Freedom of International Trade 0.073*** 0.072*** (3.98) (3.82) Level of Democracy 0.010* 0.011 (1.92) (1.48) Population 0.000** 0.000** (2.63) (2.63) International Interest Rates 0.098*** 0.102*** (5.13) (5.26) Debt Service Ratio 0.003 0.003 (1.1) (0.93) Inflation -0.000* -0.000 (-1.82) (-1.65) Constant 6.302*** 6.303*** (33.88) (33.02) R-Squared (Within) 0.670 0.662 N 132 132 Number of Groups 61 61 *p<.10 **p<.05 ***p<.01 Model 1 uses the Index of Democracy; model 2 uses the Polity Score Note: T-values in parentheses. Note: Due to rounding, the population and inflation rate coefficients are represented as zero. However, the actual values are greater than zero. Analysis of Income The results in Table 2 show the effects of the independent variables on the natural log of per capita income using data that spans 1975 to 2003. These results support the neo-liberal claim that economic liberalization is associated with increased economic growth. Across both models, both decreases in government intervention in the economy 25
and increases in the freedom of international trade are strongly related to growth of per capita income. With respect to democracy, the data also support the second hypothesis. Model 1 indicates that democracy has a weak positive effect economic growth at a 10% significance level, while Model 2 indicates that it has no effect. However, the removal of India from Model 1 changes democracy s coefficient from borderline significant to completely insignificant (from p=0.059 to p=0.180). Accordingly, across both models changes in the level of democratization have no statistically significant impact, excluding outliers, on changes in per capita income. Somewhat counterintuitively, population growth is shown to have a significant positive effect on economic development. Analysis of Health The results reported in Table 3 show the effects of the independent variables on under-five mortality rates, again using data the spans 1975 to 2003. Large disparities appear when matching the coefficients of both models, as Model 1 was better represented using the natural log of under-five mortality rates, while Model 2 was not. Before interpreting the results, it is important to note that Model 1 was quite susceptible to influential cases. The absence of India from the model changes the statistical significance of the population coefficient to having a strong significant negative effect (from p=0.370 to p=0.005). Freedom of international trade is the most volatile variable. Without India, the trade coefficient is transformed from having a significant negative effect to an insignificant effect (from p= 0.020 to p=0.055). The absence of both India and Malaysia change it back to having a significant negative effect again (p=0.023). However, trade is ultimately statistically insignificant (p=0.104) if all three influential 26
cases (India, Malaysia and Sri Lanka) are removed from the model. Moreover, as the freedom of international trade variable in Model 2 was better represented as a quadratic equation, it is important to note the F-test of trade and the trade square term combined in the second model resulted in a p-value that is only significant at a 10% significance level (p=0.09). Table 3. Fixed Effects Cross-Sectional Time Series of Under-Five Mortality Rates Under-Five Mortality Rates Independent Variables Model 1 Model 2 Government Size -0.046 0.034 (-1.25) (0.02) Freedom of International Trade -0.053** -7.668* (-2.38) (-1.98) Freedom of International Trade Squared Term n/a 0.602 n/a (1.55) Level of Democracy -0.019*** -1.877*** (-3.04) (-3.97) Population -0.000-0.000** (-0.9) (-2.45) International Interest Rates -0.079*** -4.359*** (-3.38) (-3.32) Debt Service Ratio -0.002-0.288 (-0.59) (-1.31) Inflation 0.000 0.002 (0.75) (0.4) Constant 5.323*** 144.642*** (23.25) (9.79) R-Squared (Within) 0.485 0.507 N 134 134 Number of Groups 62 62 *p<.10 **p<.05 ***p<.01 Model 1 uses the Index of Democracy; model 2 uses the Polity Score Note: T-values in parentheses. Note: Due to rounding, the population and inflation rate coefficients are represented as zero. However, the actual values are greater than zero. For the most part, these results contradict the first hypothesis and the claims of neo-liberalism. According to the trickle-down effect, in the long-term economic liberalization will result in economic growth followed by an amelioration of health within 27
society. Both models report that changes in the amount of government intervention in the economy has no long-term statistically significant effect on changes in under-five mortality rates. The other measure of economic liberalization, freedom of international trade, shows a mildly significant negative relationship with under-five mortality in the first model. That is, an increase in the freedom of international trade corresponds to a decrease in under-five mortality rates (or an improvement in health). The polynomial function of trade in the second model suggests a negative relationship between trade and under-five mortality rates until the freedom of international trade score surpasses 6.4, then the mortality rate begins to increase as trade becomes more liberalized. However, as mentioned, the absence of outliers in the first model turns the trade coefficient insignificant, while in the second model the F-test shows that trade is only significant at a 10% significance level. Thus, there is a very weak relationship between the two variables that, excluding outliers, is not statistically significant at a 5% significance level. Democracy has a strong statistically significant effect on under-five mortality rates, which lends support to the second hypothesis. Both measures of democracy show that it is negatively associated with the dependent variable. Namely, an increase in the level of democratization results in improved health levels by decreasing the under-five mortality rate. Analysis of Education The results of running the fixed effects cross-sectional time series on gross primary school enrolment rates using data from 1975 to 2003 are presented in Table 4. 28
The findings clearly fail to support the first hypothesis that the trickle-down effect will in due course increase the level of education. Changes in the amount of government intervention in the economy have no statistically significant effect on changes in gross primary school enrolment rates across both models. Likewise, both models show that changes in the liberalization of trade are not significantly correlated with changes in school enrolments rates. Table 4. Fixed Effects Cross-Sectional Time Series of Gross Primary School Enrolment Rates Enrolment Rates Independent Variables Model 1 Model 2 Government Size 0.546-0.148 (0.34) (-0.1) Freedom of International Trade -0.387-0.655 (-0.4) (-0.7) Level of Democracy 0.090 0.864** (0.34) (2.38) Population 0.000 0.000 (0.78) (0.78) International Interest Rates -0.733-0.639 (-0.72) (-0.66) Debt Service Ratio 0.141 0.161 (0.86) (1.04) Inflation 0.003 0.003 (0.7) (0.56) Constant 97.891*** 103.195*** (10.22) (11.12) R-Squared (Within) 0.054 0.140 N 121 121 Number of Groups 59 59 *p<.10 **p<.05 ***p<.01 Model 1 uses the Index of Democracy; model 2 uses the Polity Score Note: T-values in parentheses. Note: Due to rounding, the population coefficients are represented as zero. However, the actual values are greater than zero. The results for democracy are mixed. Changes in Vanhanen s index of democracy do not have a statistically significant effect on changes in primary school enrolment rates. Conversely, changes in the Polity IV measure of democracy have a 29
significant positive relationship (at a 5% significance level) with enrolment rates. In other words, higher levels of democracy correspond to increased enrolment rates. Therefore, on the whole the data do not confirm, nor deny, the second hypothesis. In sum, economic liberalization lives up to its expectations with respect to economic growth, but its argument that supply-side economics handles issues of social justice regarding the equitable distribution of increased wealth is not borne out by the data. As with other empirical studies, democracy seems to have no effect on economic growth. The results also appear to be consistent with resource distribution theory in that an increase in the level of democracy seems to result in improvements in both the level of health and education within society, with the caveat that its positive effect on education only holds for one of the two measures. However, having a statistically significant positive effect on three of the four measures, while the other shows no effect (i.e. it does not have a negative effect), is nonetheless a significant finding. Therefore, aggregating both the health and education results as a proxy for resource distribution, it seems reasonable to conclude that democracy has an effect on the distribution of resources well beyond mere chance. DISCUSSION The results reinforce the importance of acknowledging that different variables affect income and social welfare. Although an argument can be made that combining economic growth and measures of social justice provides a useful overall picture of the state of development, causal analysis calls for the separation of these concepts for a 30
clearer picture of the influential factors of each. Thus, assuming that health or education is strongly associated with income not only has the potential to bias academic research, but it may also result in misguided policy prescriptions. The finding that different factors effect income and socioeconomic distribution seem inconsistent with recent research on income distribution. Dollar and Kraay (2002) found that the relationship between economic growth and the income of the poorest quintile of society is nearly perfectly correlated and that democratic institutions do not have an effect on either economic growth or income distribution. However, these findings do not necessarily contradict the results of the present study. Although income growth may be enjoyed evenly by the entire population, it is not inevitable that an increase in income will result in an improvement in access to other resources such as health and education. Numerous intervening factors, such as a decrease in government provisioning of social services or an increase in the cost of health and education relative to income, may result in such a discrepancy. Further research is needed that examines the relationship between income distribution and the inequalities of other socioeconomic resources. As the data shows that income and the levels of health and education are not dependent on the same factors, it logically follows that both health and education can be improved in the absence of economic growth. Coburn (2004) pursues this line of inquiry and contends that there are other factors beyond income inequality that determine the level of health inequality. He cites Kerala, Costa Rica and Cuba as instances in the 31
developing world where government or collective action, not economic growth, was the main driver towards improvements in healthcare (2004: 53). In this sense, democracy may act to facilitate collective action through government to implement programs that lead to a reduction in social inequalities, regardless of the level of economic growth. More research is required to determine the exact mechanisms by which democracy may lead to improvements in health and education. In addition to a more evenhanded distribution of resources, Rodrik finds that democracies have other benefits, mainly that they yield more predictable long-term growth rates, they produce more stable economic performance, they are better at handling adverse shocks and they pay higher wages (1997: 2-3). However, before advocating the promotion of democracy, it is important to consider its potential drawbacks. Robinson (1996) persuasively argues that polyarchy is an ideological front that serves the purpose of silencing subordinate classes to the benefit of the transnational capitalist elite. Although this may be the case, the data demonstrate that, controlling for the extent of neo-liberalism, modern representative democracies are much more socially just than autocratic forms of government. However, regime type is not necessarily a simple dichotomous selection. To avoid the problems of both authoritarianism and polyarchy, Robinson advocates another alternative, popular democracy, defined as a fusion of representative government and participatory democracy that holds states accountable beyond the indirect mechanism of periodic elections (1996: 57). He argues that popular democracy will end the domination of the transnational elite by laying the foundation for a more equitable distribution of socioeconomic resources. 32