Construction of Revealed Relative Policy Strengths – Basic Idea
We begin this section with an introduction of our methodology for constructing revealed relative policy strengths. Our basic idea can be illustrated with the help of Fig. 1, which shows a hundred cells into which a country may theoretically fall. The y-axis might represent one of our three policy focus variables (health, social safety, or human capital) and is divided into deciles. The higher the value on the y-axis, the better a country performs in a certain policy area relative to other countries. The x-axis represents income per capita, but in terms of reverse deciles.
Imagine now one country in cell one and another one in cell ten. Then, both countries perform equally “badly” in terms of the policy strength variable (health, social safety, or human capital). Both countries find themselves in the lowest 10 percentile of policy performance. However, the country falling into cell one is a high-income country (percentile > 90%) while the country falling into cell ten is a low-income country (percentile < 10%). Then, because the low-income country, with fewer resources, accomplishes the same policy outcome as the high-income country, the low-income country, we argue, has a revealed relative policy strength in comparison to the high-income country (10 > 1).
Assume next that the low-income country from cell ten moves beyond the 90th percentile in terms of policy performance to cell one hundred. Cell one hundred then represents the greatest possible revealed relative policy strength, because, loosely speaking, the country accomplishes the maximum policy performance with the least resources. On the other hand, if the country from cell one moves beyond the 90th percentile in terms of policy performance, then again it has the highest possible policy score, but accomplishes such with the maximum resources, which is reflected by a lower policy score (91 < 100).
A numerical example may clarify this further. Assume that a country x has a percentile score of 0.413 in a certain policy variable, meaning that 41.3 percent of all the countries in the sample perform as well as country x, or worse. Assume further that country x’s GDP percentile is 0.271 so that 27.1 percent of all the countries in the sample have the same income as country x, or less. Country x would then receive a revealed relative policy strength score of 48. How?
In a first step, we convert country x’s policy percentile into a scale between zero and nine, simply by multiplying the percentiles, which are between zero and one, by nine, and round the result to zero digits. Country x then receives a policy score on a scale between zero and nine of four. We do the same for the income variable, which gives country x initially a value of two. Yet, because countries with higher incomes receive lower income scores, we reverse the scale and subtract the initial value of two from nine, to receive for country x a final income score of seven. The final revealed relative policy strength score, which is between one and one hundred, is then simply calculated as
$$Revealed Relative Policy Strength Score=Policy Score\times 10+Income Score+1$$
1
Equation (1) assures that revealed relative policy strength scores lie between one and a hundred, as in Fig. 1. For country x, the score is accordingly 48 (4 ×10 + 7 + 1).
This methodology provides a ranking of countries where higher ranking countries have achieved a level of policy effectiveness with fewer resources. Our approach, we argue, informs about governments’ commitments towards assuring social mobility that allows for a more nuanced comparability of states with similar prevailing developmental levels.
Data
Dependent Variable “Free & Equal”
This paper tries to identify what separates countries with high economic freedom and equitable social development from countries with high economic freedom and inequitable social development. Our dependent variable is called Free & Equal. In constructing Free & Equal, we calculate the percent rank of the Heritage Foundation’s Index of Economic Freedom (Overall Score) between 1 and 100, where higher percent ranks indicate more freedom. We label the resulting transformation “Free.” In constructing “Equal,” we use from the World Inequality Database (World Inequality Lab, 2019) the variable Gini (pre-tax national income, total population, adults, equal split), and calculate the percent rank between 1 and 100 in reverse order, so that higher percent ranks indicate more equality. We then calculate Free & Equal as the geometric mean of our “Free” and “Equal” variables.
Focus Independent Variables Revealed Relative Policy Strengths
Our focus explanatory variables are Revealed Relative Public Health, Social Safety and Human Capital Strengths. The variable GDP per capita, PPP (constant 2017 international $) is used to position countries on the x-axis of Fig. 1.
On the y-axis we transform the variable Infant mortality per 1,000 livebirths to calculate the Revealed Relative Public Health Strength score. As for Revealed Relative Social Safety Strength, we acknowledge that different levels of economic development correlate with different social safety challenges. Vulnerable segments of society in countries with low incomes are typically more vulnerable to nutritional deficiencies. Yet, as countries’ incomes increase, other forms of health risks replace nutritional deficiencies. These health risks often result from substance abuse and self-harm. Because differently developed countries have different correlates of social insecurity, we used the maximum value associated with either nutritional deficiency, substance abuse, or self-harm to calculate the Revealed Relative Social Safety Strength score. Lastly, we use the variable Citable Documents per Million Population to construct the Revealed Relative Human Capital Strength score. All variables were transformed on a scale between zero and nine, with higher values indicating lower rates of infant mortality, lower prevalence of either nutritional deficiency, substance abuse or self-harm, and higher citable documents per million, and then used to calculate the three revealed relative policy strength scores as outlined above.
Control variables
In our regression analysis, we also control for various other variables which might explain Free & Equal. These variables are Manufactures and Services Export Share as a Percentage of GDP, Population Share of Catholics, a measure of Democracy, and the presence of Armed Conflict (for data and sources see Appendix A-1). Manufactures and services exports, we argue, are indicative of productive economic competitiveness and representative of a spirit of economic freedom, which provide social mobility and economic opportunities. We also include the population share of Catholics to control for countries’ colonial legacy. In countries where Catholicism is more widespread, land inequality is often greater and liberal economic thoughts often face a more difficult environment to shape economic policy. To capture the interrelation between political and economic freedom, we use a measure of democracy. Socioeconomic grievances, we posit, can be more effectively remedied in democracies than in autocracies. Lastly, we control for the presence of armed conflict, which, by definition, replaces freedom by coercion.
Final Dataset
Our final dataset is a panel of 142 countries. Each country has five five-year average observations beginning in 1996–2000 (1996–2000, 2001–2005, …, 2016–2020). We opt for a panel with five-year averages to reduce potential bias from countries with many observations relative to countries with fewer observations. Whenever a country does not have at least one observation for Economic Freedom, Gini, GDP per capita, Infant Mortality, Nutritional Deficiencies, Self-Harm, Substance Abuse, or Citable Documents for any of the five time periods, we did not include the country in our sample. In other words, if we could not construct at least one observation for Free & Equal (dependent variable) or our focus explanatory variables (Revealed Public Health Strength, Revealed Social Safety Strength, and Revealed Human Capital Strength), we did not include the country in our dataset. Appendix A-2 provides a list of countries included in our dataset, sorted by region.
It is also important to note that we construct the revealed relative policy strength scores not over all observations in the entire panel dataset in one calculation, but for each of the five time periods separately, which allows for a more accurate comparison of relative standings over time.
Estimation Strategy
We test our hypothesis that countries with higher Free & Equal scores have, on average, a greater Revealed Relative Public Health, Social Safety, and Human Capital Strength using a pooled OLS model. Our three independent focus variables, however, are highly colinear, so we combine them into one Revealed Relative Social Mobility indicator.
We opted for pooled OLS after running our models first as a fixed-effects panel, which generated unexpected signs and non-significant results. This suggests that the country fixed-effects correlate highly with some of the right-hand side variables, many of which are relatively time-invariant as well, such as, for example, Democracy, Manufactures and Services Export Shares, and Population Share of Catholics.
Serial correlation leading to spurious regression results is a concern in any panel data estimation, which is normally ameliorated by the introduction of a lagged dependent variable on the right-hand side. Yet we decided against using a lagged dependent variable on the right-hand-side for three reasons. Firstly, when running our models with the lagged dependent variable, it absorbed so much explanatory power that the remaining variation was insufficient for other variables to generate meaningful results. Secondly, we suspect on theoretical grounds that shocks to “Free & Equal” and the “Free & Equal” variable itself are in fact cointegrated and thus subject to a long-run equilibrium relationship, even though we cannot provide a corroborating test in its support. Thirdly, the value of a lagged dependent variable on the right-hand-side comes at the expense of a loss of degrees of freedom, which we consider to be quite costly given the fact that we work with only five time periods, especially if cointegration cannot be fully ruled out. In evaluating all these aspects, we argue that a pooled OLS estimation strategy with regional fixed effects (rather than country fixed-effects as in a panel fixed-effects model) and no lagged dependent variable is (among all the imperfect options) still the least costly compromise.
Lastly, our dependent variable Free & Equal might be simultaneously determined with the revealed relative policy strengths. We therefore conduct a Hausman test for endogeneity by instrumenting Revealed Relative Social Mobility with the variables Life Expectancy and Natural Resources Rents. Citizens in countries with high life expectancy can be plausibly assumed to benefit from good public health, social safety, and educational systems. On the other hand, in predominantly natural resources-extracting countries, working conditions are often harsh, demand for workers non-competitive, and productive and diversified economic opportunities scarce. Life Expectancy and Natural Resources Rents therefore indicate to explain our Revealed Relative Social Mobility well (relevance condition). Yet there is no immediate reason to assume that decisions to implement economic freedom and institutions for equitable social development simultaneously determine life expectancy, much less a country’s endowment with natural resources rents (exclusion condition). Appendix A-3 provides summary statistics of all our variables (Mean, Median, Minimum, Maximum, Standard Deviation, and Inter-Quartile Range). We also report the kind of data transformations that we introduced to improve the distributional characteristics of some of our variables. Appendix A-4 shows the correlation matrix with the variables as they entered our empirical analysis. We will also gladly provide our dataset upon request. We conduct our empirical analysis with the open-source software program gretl (Gnu Regression, Econometrics and Time-series Library), available at http://gretl.sourceforge.net/.