Institutions, freedom, and technical efficiency.

1. Introduction

The impact of institutions on economic performance is currently the subject of much debate among economists and policy makers. In this paper we examine empirically a mechanism through which institutions and economic variables can affect a country's economic performance. It is generally accepted (see, e.g., World Bank 1993) that different countries operate at different distances from the production frontier. We postulate that such technical inefficiency, as measured by deviations from the production possibility frontier, is a function of certain measurable economic and institutional variables. An important contribution of this paper is that it provides statistically significant estimates of several determinants of these deviations.

Although economists have demonstrated that institutions may have large effects on certain aspects of economic performance, these economists have concentrated more on growth and unemployment than on technical inefficiency. For instance, several recent studies have examined the role of institutions in promoting economic growth and productivity. Dawson (1998) estimated cross-country growth and investment regressions and found that economic growth is associated with economic freedom because of the latter's positive effect on investment and the level of total factor productivity (TEP). Aspects of political freedom are associated with higher investment, but there is no indication that they are associated with higher TFP.

Using some of the same data used in our study, Edwards (1998) first estimated a production function for a panel of 93 developed and developing countries and calculated TFP growth. He then estimated a relationship between the degree of trade openness and TFP growth and found that the initial per capita GDP, the initial level of human capital, and the degree of openness are important determinants of TFP growth.

Rodrik (1997) discusses political factors that can affect economic performance. He provides evidence that democracies are associated with (i) more stable long-run growth rates, (ii) greater short-run stability, (iii) better ability to deal with adverse shocks, and (iv) higher wages (Rodrik 1999). He proposes three explanations for the empirical regularities. First, democracies may have greater stability because the preferences of the median voter inhibit radical policy actions that would yield extreme results. Second, a voice in the political process for citizens reduces the amount of internal conflict. Finally, losers in political battles are more likely to avoid economic losses in a democracy than they are under other types of government.

Rodrik (2000) extends his discussion of institutions, democracy, and economic performance by defining five types of institutions that permit markets to work adequately: the institution of property rights, regulatory institutions, institutions for macroeconomic stabilization, institutions for social insurance, and institutions of conflict management. He argues that the building of institutions can be thought of as a form of technology transfer that allows increased productivity. Participatory democracy is a meta-institution that helps build better institutions. Rodrik (2000) provides evidence that participatory democracy improves economic performance in terms of both higher long-run growth rates and increased short-term stability. We study the effect of participatory democracy on economic performance by testing whether political rights or civil liberties affect technical efficiency.

Other research more closely related to ours indicates that planned economies are less efficient than unplanned ones. Bergson (1987, 1989, 1991), Marer (1981), Moroney and Lovell (1997), and others have compared the performances of centrally planned economies with those of western market economies. Bergson (1987, 1989) estimates a constant-returns-to-scale production function via ordinary least squares and a dummy variable identifying planned economies. He finds that planned economies tend to use capital and land less efficiently than market economies do. Moroney (1992) follows a similar approach and shows that planned economies used capital and energy less efficiently than Western European economies did during 1978-1980.

Moroney and Lovell (1997) were the first to use stochastic production frontier panel data techniques to compare the productive performances of market and planned economies. Their goal was to quantify the extent to which market economies are more efficient than planned ones. They found that Western European market economies were much more productive than a group of seven Eastern European planned economies during 1978-1980. They attribute most of the difference to the use of better technology in market economies. The Eastern European economies were no more than 76% as efficient as the Western European economies were during this period.

None of the aforementioned studies accounts for sources of technical inefficiency except with the use of dummy variables indicating planned or market economies. In addition, they focus on OECD countries versus the former USSR or Eastern European economies. In this study, we use panel data to estimate a production frontier and examine the sources of inefficiency among a much broader set of countries than has previously been considered. Our study also differs from previous research in that our concern is the extent to which economic and political institutions contribute to technical inefficiency or deviations from a stochastic frontier rather than how they influence output growth, TFP growth, or unemployment. Importantly, our study also uses a more flexible functional form and includes human capital as an input in the production function. Our results suggest that among a broad set of countries at various stages of economic development, variations in human capital, economic freedom, and development status are li nked to efficiency.

2. The Stochastic Frontier Model

A number of studies have estimated a production frontier and used the difference from the frontier (a measure of the predicted efficiencies) in a second-stage regression to determine reasons for differing efficiencies. In the first stage, the predicted inefficiencies are estimated under the assumption that they are independently and identically distributed. The regression of other variables on the inefficiencies in a second stage is a clear violation of the independence assumption. According to Kumbhakar, Ghosh, and McGuckin (1991), there are at least two problems with such a procedure. First, inefficiency may be correlated with the inputs; if so, the inefficiencies and the parameters of the second-stage regression are inconsistently estimated. Second, the use of ordinary least squares (OLS) in the second stage ignores the fact that the dependent variable (inefficiency) takes on values over the positive domain. Therefore, OLS may yield predictions that are inconsistent with this fact and are therefore not app ropriate.

Kumbhakar, Ghosh, and McGuckin (1991) and Reifschneider and Stevenson (1991) have proposed models of technical inefficiency in the context of stochastic frontier models. In these cross-sectional models, the parameters of the stochastic frontier and the determinants of inefficiency are estimated simultaneously given appropriate distributional assumptions about the model's errors. Battese and Coelli (1995) proposed a stochastic frontier model in which inefficiencies are expressed as specific functions of explanatory variables. The panel specification of this model can be expressed as follows:

[y.sub.it] = [x.sub.it][beta] + ([V.sub.it] - [U.sub.it]); i = 1,...,N t = 1,...,T, (1)

where [y.sub.it] is the logarithm of the output of country i in period t, [x.sub.it] is a k X 1 vector of inputs, [beta] is a vector of unknown parameters, and [V.sub.it] are random variables that are assumed to be independently and identically N(0, [[sigma].sup.2.sub.V]) distributed and independent of [U.sub.it]. [U.sub.it] are nonnegative random variables that account for technical inefficiency in production; they are assumed to be independently distributed as truncations of the N([m.sub.it], [[sigma].sup.2.sub.U]) distribution at zero. The mean inefficiency is a deterministic function of p explanatory variables:

[m.sub.it] = [z.sub.it][delta], (2)

where [delta] is a p X 1 vector of parameters to be estimated. Following Battese and Corra (1977), we let [[sigma].sup.2] = [[sigma].sup.2.sub.V] + [[sigma].sup.2.sub.U] and [gamma] = [[sigma].sup.2.sub.U]/([[sigma].sup.2.sub.V] + [[sigma].sup.2.sub.U]).

The inefficiencies ([U.sub.it]) in Equation 1 can be specified as

[U.sub.it] = [z.sub.it][delta] + [W.sub.it]. (3)

where [W.sub.it] is defined by the truncation of the normal distribution with mean zero and variance [[sigma].sup.2].

Using this parameterization, a test can be constructed to determine whether the estimated frontier is actually stochastic; [gamma] = 0 implies that the variance associated with the one-sided (efficiency) errors, [[sigma].sup.2.sub.U], is zero, meaning that these deviations from the frontier are better represented as fixed effects in the production function. Therefore, a test of the null hypothesis that [gamma] = 0 against the alternative that [gamma] is positive is used to test whether deviations from the frontier are stochastic and whether one should proceed with the estimation of parameters related to the sources of inefficiency within the context of a stochastic production frontier. Failure to reject the null hypothesis suggests that the determinants of inefficiency, [z.sub.ip] should be included in the production function. (1)

The parameters of the model ([beta], [delta], [[sigma].sup.2], and [gamma]) are estimated using the maximum-likelihood estimator; the likelihood function can be found in the Appendix. Then, the technical inefficiency of the ith country at time t is

[TE.sub.it] = exp(-[U.sub.it]) = exp(-[z.sub.it][delta] - [W.sub.it]). (4)

The conditional expectation of [TE.sub.it] is given in Equation A8 of the Appendix and is used to produce predictions for each country in each period, our measure of inefficiency. Computations were performed using the algorithm described in Coelli (1996).

Translog Production Function

We model the production function in Equation 1 with the translog functional form because of its flexibility; it (i) can be interpreted as a second-order approximation of the unknown functional form, (ii) permits returns to scale to vary by country, and (iii) does not require input substitution elasticities to be equal to unity. Following Mankiw, Romer, and Weil (1992), we include human capital along with capital and labor in the input vector. Most other contributions to this literature have adopted the constant-returns-to-scale Cobb-Douglas model (e.g., Bergson 1987, 1989; Moroney 1992; Moroney and Lovell 1997; Dawson 1998). In the translog specification of Equation 1, [x.sub.it] = {ln([L.sub.it]), ln([K.sub.it]), ln([H.sub.it]), 0.5[[ln([L.sub.it])].sup.2], 0.5[[ln([K.sub.it])].sup.2], 0.5[[ln([H.sub.it])].sup.2], ln([L.sub.it])ln([K.sub.it]), ln([L.sub.it])ln([H.sub.it]), ln([H.sub.it])ln([K.sub.it])} where [H.sub.it], [K.sub.it], and [L.sub.it] represent human capital, physical capital, and labor, respecti vely. In the following section, we test the null hypothesis of the Cobb-Douglas specification versus the translog specification and reject the Cobb-Douglas model at the 5% level in every instance. A time trend variable is also included in the model to permit secular changes in output due to technological changes or other events that have affected output over the sample years. (2)

Modeling Inefficiency

In Equation 3, inefficiencies are modeled as functions of several exogenous determinants that are thought to explain differences in efficiency across countries. Several types of factors are considered. The first is the degree of economic freedom experienced by the citizens of a country. As Gwartney, Lawson, and Block (1996, p. 12) note, "[t]he central elements of economic freedom are personal choice, protection of private property, and freedom of exchange." The absence of restrictions on the freedom to choose goods, to supply resources, and to compete in business and in trade has been considered, at least since Adam Smith, central to economic prosperity. As Gwartney, Lawson, and Block (1996, p. 12) suggest, "[a]n index of economic freedom should measure the extent to which rightly acquired property is protected and individuals are free to engage in voluntary transactions." Consequently, we anticipate that the economic freedom index is an important determinant of economic efficiency. Certainly a believer in Ad am Smith's invisible hand would expect nothing else.

Political freedom--participatory democracy--is a different, if not less valuable, characteristic of freedom; it has more to do with how the rules of the economic game are determined than with the actual rules themselves. It exists where adult citizens are free to vote, lobby, choose candidates, and make political contributions of time and resources. In addition to these political rights, in politically free countries, individuals enjoy civil liberties such as freedom of the press, freedom of expression and association, the rule of law, and human rights. As Rodrik (2000) suggests, through its institutions, democracy can promote stability, which may in turn lead to greater economic efficiency.

While economic theory is fairly clear on the impact of economic freedom, the effect of political freedom on technical efficiency is less certain. Political rights and civil liberties may actually promote restrictions on economic activities. Such may be the case in countries like Israel and India, where distributional politics can affect tax rates and administrative regulations. Hence, political freedom and economic freedom can have opposite effects on efficiency.

Developed countries are expected to have other unmeasured externalities that make them more efficient. The externalities might be related to infrastructure, greater marketization of the economy, or the availability of a wider variety of goods. Although these factors are potentially important, continuous measures for them are unavailable; they are accounted for in our model by the dummy variable denoting developed country status. In defining countries as more or less developed, we adopted the World Bank (1992) definition that relies on per capita gross national product in dollars for 1990. The World Bank distinguishes between high-income economies (developed country status [DCS] = 1) and low- and middle-income economies (DCS = 0). Although Singapore and Israel are included in the high-income category, the World Development Report notes that these two countries are classified by the United Nations or otherwise regarded by their authorities as developing. We follow this suggestion in our classification scheme.

Human capital, our final variable, can also influence efficiency. Huffman (1977), building on the insights of Shultz (1975) regarding the ability to deal with disequilibrium, refers to the ability to perceive and respond to changes in economic conditions as allocative ability. Allocative ability is acquired by investments in human capital; Huffman reports econometric evidence that human capital enhances the allocative efficiency of U.S. corn belt farmers. Similarly, countries with more highly educated workers may be quicker in adapting to changing economic conditions. Benhabib and Spiegel (1994) also provide evidence that human capital enhances a country's ability to innovate domestically and to close the "knowledge gap" between itself and the leading technological country. Both forces will contribute toward closing the deviations from the production frontier in our framework. To incorporate these ideas, a measure of human capital is included as a determinant of inefficiency in the model.

In specifying the production frontier and the inefficiency function, a researcher needs to determine whether the frontier is a world frontier or a conditional frontier that depends on a country's institutional characteristics. (3) The world frontier, or, more accurately, a best-practice frontier, is obtained by finding the maximum output obtained from a given input combination for any economy in the world. The variables that we include in the world frontier function are human capital, physical capital, and labor. Institutions as measured by economic freedom, for instance, are excluded from the frontier. With this approach, institutions may influence the distance from the frontier but not its location. Specifically, a variable such as economic freedom would not enter Equation 1 but would be included in Equation 2.

Alternatively, an institutional variable such as economic freedom could enter the production frontier as a determinant of its location. In this case, the frontier is best-practice conditional on the level of economic freedom. On a priori grounds, we believe that the world frontier approach is preferred because we are concerned with how countries could perform if they adopted best-practice institutions. Nevertheless, we nest the world frontier within the conditional one and test for the appropriate specification.

In fact, institutional variables such as economic or political freedom could determine both the location of the frontier and the inefficiency associated with it. If so, an improvement in institutions could have two effects on output. First, it could increase output by moving a country closer to its preexisting frontier, that is, by reducing inefficiency. Two, it could increase output by shifting the frontier out, increasing the productivity of the factors of production.

In our empirical work, we focus on several questions. First, are there inefficiencies associated with aggregate production across countries? This question is assessed using a likelihood ratio test of the null hypothesis [gamma] = [[delta].sub.0] = [[delta].sub.1] ... [[delta].sub.p] = 0; rejection indicates that there is inefficiency and that one or more of the proposed variables contribute to it. Second, given the existence of inefficiency, are the inefficiencies stochastically determined or are they better represented by fixed effects? Rejection of the hypothesis [gamma] = 0 is evidence that they are stochastically determined. Third, which variables in the inefficiency function are individually significant? Fourth, do the political and freedom variables belong in the production function?

3. Data

Two sets of countries and years are considered. The first data set consists of a broad sample of 73 developed and developing economies with data for 1975, 1980, 1985, and 1990. The second data set consists of 76 countries with data for 1975, 1980, and 1985. The second set uses different measures of economic freedom and human capital. (4) When the estimates for the two samples are compared with each other and with those in Adkins, Moomaw, and Savvides (1998), one can see that the results are robust to changes in data and specification. Summary statistics for the two samples are presented in Tables 1 and 2.

Economic freedom is the first variable considered in the discussion. The two measures used have been constructed in an ongoing project initiated by Gwartney, Lawson, and Block (1996). In this project, the authors construct indices of economic freedom based on the premise that the central elements of economic freedom are personal choice, protection of private property, and freedom of exchange. Currently, the index consists of 21 variables grouped into seven major areas (Gwartney, Lawson, and Block 1996, p. 12): "(I) size of government, (II) economic structure and use of markets, (III) monetary policy and price stability, (IV) freedom to use alternative currencies, (V) legal structure and security of private ownership, (VI) freedom to trade with foreigners, and (VII) freedom of exchange in capital markets." Guided by their understanding of economic freedom, Gwartney et al. (2001) chose various variables for inclusion in the seven categories in light of the data available for a large cross-section of countries o ver time. Principal-components analysis was used to develop an index for each category. Principal-components analysis was used again to develop what Gwartney et al. (2001) refer to as a summary index. Each index varies from 0 to 10, with 10 being the most free. Gwartney et al. (2001) report the indices in five-year increments from 1970 to 1995. The average level of economic freedom appears to have increased between 1975 and 1990. In 1975, the average freedom index was 5.496. It fell to 5.453 in 1980 but began to rise thereafter. In 1985, it was 5.555, and it rose to 6.060 in 1990. Data on this index are available from Gwartney et al. (2001).

In an earlier study (Adkins, Moomaw, and Savvides 1998) and in the second sample considered below, economic freedom is measured using survey index IS1 as defined by Gwartney, Lawson, and Block (1996) on the basis of the same premise; in this case, different variables and a different methodology are used to create the summary index. This index consists of 17 components grouped into four major areas: money and inflation, government operations and regulation, takings and discriminatory taxation, and restrictions on international trade and exchange. The component weights were assigned by Gwartney et al. (2001) on the basis of the responses of knowledgeable people as to the relative importance of each component to economic freedom. The simple correlation between the newer measure of economic freedom and IS1 is 0.78, indicating that the measures are imperfectly correlated. (5)

The two other measures of freedom with broad coverage and availability-civil liberties (CL) and political rights (PR)--are published by Freedom House (2001). Freedom House assigns an index number from one to seven for each country for each measure, with larger numbers indicating less freedom. Political rights refer to the choice of government, election conditions, the party system, and so on. Civil liberties refer to individual rights and their enforcement. The ratings are based on survey information combined with judgments made by the Freedom House survey team. PR and CL are in fact highly correlated with one another in our sample (p = 0.90). Surprisingly, the correlation between political rights and IS1 is only 0.22, and that between civil liberties and IS1 is only slightly higher, at 0.25; however, CL and PR are more strongly correlated with the newer freedom index used in the 73-country sample (the correlation between PR and freedom is 0.52, and that between CL and freedom is 0.57).

Human capital is measured as the average number of total years of schooling (primary, secondary, and tertiary) in the working-age population over 15 years old. The data for the three-year sample are from the Nehru, Swanson, and Dubey (1995) database, which provides annual estimates for the 1970-1987 period. The data for the four-year sample are from the Barro and Lee (2000) database, which provides quinquennial estimates of the stock of human capital. The correlation coefficient between the two human capital series is 0.86.

Output is measured in purchasing power parity dollars (PPP$) and is from the Penn World tables (Heston and Summers 1991). (6) Following the method used in Benhabib and Spiegel (1997), the capital stock is also measured in PPP$. It was obtained by adjusting the output variable for the capital-output ratio. The capital-output ratio was obtained from Nehru and Dhareshwar (1993). Labor is measured as the number of workers and is derived from the Penn World tables.

4. Results

In general, the results indicate that developed countries lie closer to the production frontier and that greater amounts of human capital and economic freedom decrease inefficiency. They also indicate that political rights and civil liberties are not related to technical efficiency.

Specifically, Table 3 contains results and panel data from the 73-country sample covering four years. Time is included in the first stage of the model (the production function) to permit systematic worldwide variations in overall economic conditions and/or technical change. The first and second columns of the table contain the coefficient estimates and t-ratios for the unrestricted model. The unrestricted production function includes human capital, physical capital, and labor as inputs and economic freedom, civil liberties, and political rights as determinants of the technology level. The unrestricted inefficiency function includes economic freedom, civil liberties, political rights, human capital, and development status. The values of various test statistics are given at the bottom of the table along with their approximate distributions under the null hypothesis.

For the unrestricted model presented in Table 3, the parameter [gamma] is statistically different from zero, indicating that the stochastic production frontier approach is appropriate. The economic freedom, political rights, and civil liberties coefficients are not statistically significantly different from zero. The hypothesis that [gamma] and the parameters of the inefficiency function are jointly equal to zero is also rejected, which implies that a stochastic frontier exists and that the inefficiency function provides a meaningful explanation of the sources of inefficiency. Economic freedom has a negative and significant coefficient in the inefficiency function, indicating that increases in economic freedom reduce inefficiency.

Other significant determinants of inefficiency are human capital and development status. Their negative signs support the hypothesis that an increase in the variable reduces inefficiency. We interpret development's effect on inefficiency as arising from the external benefits of more complete markets. Human capital's role in aggregate production is also accounted for by its inclusion as an input for production; it is interesting that it also plays a separate role in reducing inefficiency. We speculate that it does so because better-educated people tend to make better decisions and are able to adapt more rapidly to technological breakthroughs that increase efficiency. For example, better-trained workers may be able to reduce waste in the production process or to suggest other such beneficial improvements in the usage of inputs.

The third and fourth columns of Table 3 contain the results for the specification that we interpret as the world frontier. As such, the production function includes only the three inputs (human capital, physical capital, and labor) and time. As before, the hypotheses of a stochastic frontier and an inefficiency function with jointly significant coefficients cannot be rejected. For this model, we test whether the economic freedom, civil liberties, and political rights variables belong in the production function by nesting a model that excludes the three variables in the unrestricted model. The test for the restriction shows that the joint hypothesis that these variables do not shift the frontier cannot be rejected: the hypothesis that the coefficients on these three variables are jointly equal to zero yields a chi-square statistic of 1.59. (The critical value for the 10% level is 6.25.) It is of interest to note that the coefficients of the production function inputs in the restricted model are similar to thos e in the unrestricted model. Similarly, the coefficients of the variables in the inefficiency function (and their precision) are also similar to those in the unrestricted model.

In the fifth and sixth columns of Table 3, additional restrictions are imposed by omitting political rights and civil liberties from the inefficiency function. The chi-square statistic of 3.32 indicates that the five restrictions cannot be rejected. Although this is our preferred model, it is not much different from the model in the third and fourth columns.

The model presented in the seventh and eighth columns of Table 3 differs from the unrestricted model in the first and second columns in specifying a Cobb-Douglas rather than a translog production function. The six restrictions necessary for acceptance of the Cobb-Douglas specification are rejected at any reasonable level of significance. The chi-square statistic for the restrictions is 116.54, and the 5% critical value is 12.59. The importance of testing for the appropriateness of the Cobb-Douglas production function relative to that of the translog production function is revealed by a comparison of the inefficiency function presented in the first and second columns with that presented in the seventh and eighth columns. In contrast to the unrestricted model, in the restricted model the coefficient of human capital is insignificant and the coefficient of civil liberties is significant.

Table 4 presents the results for the original data set (76 countries for 1975, 1980, and 1985). (7) In addition, this sample uses different measures of human capital and economic freedom. The results presented here are remarkably similar to those presented in Table 3. For the unrestricted model, one possible difference is that the coefficients of economic freedom, political rights, and civil liberties are larger in absolute value and have t-ratios greater than 1. The coefficient of economic freedom has a t-ratio of -- 1.70 but an implausible negative sign. Even so, the inefficiency functions are remarkably similar for the two tables. Moreover, the joint hypothesis that the three variables do not belong in the production function cannot be rejected; the chi-square statistic is 5.64, and the critical value for the 10% level is 6.25. Keeping in mind the differences in the two data sets, the coefficients and t-statistics for the world frontier model are also similar in the two tables. We conclude that the importa nt reductions in efficiency associated with increases in economic freedom and human capital stand up to different samples and different variable definitions. (8)

Table 5 shows the absolute efficiency levels for each country in our sample for 1975, 1980, 1985, and 1990. The reported efficiencies for Iraq, Ivory Coast, Madagascar, Morocco, and Nigeria are based on the estimates for the 76-country sample, while the remainder are from the 73-country sample. (9) (Efficiency estimates are unavailable for any country having a missing observation for any of the variables, which accounts for the various blanks in the table.) The results appear quite "sensible" and in accord with intuition. The southeastern Asian economies (Indonesia, Korea, Malaysia, the Philippines, Thailand, and Singapore) have some of the highest efficiency levels among the low- and middle-income economies. These countries were the subject of the extensive World Bank (1993) study and were praised for their stable macroeconomics and solid institutions yielding (until recently) high growth rates. We confirm that in addition to high growth, these economies operate closer to the world frontier. It is interestin g that among the southeastern Asian economies, the economy of the Philippines is the least efficient (in terms of the average level over the four years of our study), and the Philippines is also notorious for extensive rent seeking and a weak institutional framework. (Indonesia and the Philippines are ranked as the two most corrupt economies of those in our sample for the 1980-1985 period according to the Corruption Perceptions Index of' Transparency International [2001].) We also note that economies that have adopted more outward regimes, such as Chile and Mauritius, were able to increase their efficiencies. We speculate that access to the world economy through trade has enabled these economies to adopt innovative practices that have improved the utilization of resources.

Although we have not estimated an OECD-only model, we can compare our results for the 17 selected OECD countries for which Fare et al. (1994), as a by-product of their programming study of productivity growth, compute efficiency levels assuming constant returns to scale and a labor-capital production function. Four of the top six countries in their study (Canada, Sweden, the United Kingdom, and the United States) are also four of the top six (based on average rank) in this study. (We omitted Iraq from this comparison because data existed for only one year, 1975.) Australia and Greece complete Fare et al.'s top six, and Belgium and Ireland complete ours. The remaining 11 countries have very different ranks in the two studies. The most inefficient OECD country in Fare et al.'s sample, Ireland, is 67% less efficient than the leading country, the United States, while in our study the most inefficient of the 17 countries in question is Norway, which in 1990 was 10% less efficient than Canada, the leading country i n this study. The overlap among the top six countries in the two studies is heartening; moreover, we believe that the variation in inefficiency among OECD countries found in our study is more in line with intuition.

5. Conclusions

Although our results should be regarded as tentative, we believe that they provide new insight into the way that institutions, in particular institutions that promote economic freedom, affect economic performance. We find that increases in economic freedom are associated with improved economic performance in that increases in economic freedom move countries closer to the production frontier. Our approach also allowed economic freedom to shift the frontier and to move countries closer to it, but the probability of its doing so was rejected on statistical grounds. This finding is also consistent with our belief that studies such as ours that attempt to measure economic inefficiency should measure it against the best-practice, or world, frontier.

Our specific findings are similar in spirit to those of Moroney and Lovell (1997), who found that planned economies operate farther from the production frontier. Our results, however, show that gradations of planning or the lack of planning--economic freedom--promotes inefficiency. Moreover, our results are consistent with those of researchers who, like Dawson (1998), find that economic freedom promotes total factor productivity; in addition, we show that one of the mechanisms through which economic freedom operates is increased efficiency rather than advances in knowledge.

Another important finding in this study is that of the dual role of human capital. We model it as a productive input and as a variable that affects efficiency. In particular, we provide evidence that one benefit of investment in human capital is improved allocative ability in the economy.

In conclusion, although problems associated with the estimation of the frontier production function suggest that these efficiency results must be treated with care, our results indicate that countries with higher degrees of economic freedom and higher levels of human capital tend to lie closer to the world production frontier.

Appendix

This appendix reproduces key results from Battese and Coelli (1993) and is intended to aid the reader in interpreting and using the results contained in this paper. See Battese and Coelli (1993) for additional details.bThe pdf of iw [v.sub.it] is

[FORMULA NOT REPRODUCIBLE IN ASCII] (A1)

The pdf of the truncated normal density is

[FORMULA NOT REPRODUCIBLE IN ASCII] (A2)

where the subscripts i and t have been omitted for convenience and the function [PHI]( ) is the distribution function for the standard normal random variable.

Let E denote the overall equation error of the linear model and note that V = E + U. Given the statistical independence of V and U, the joint density of E and U is obtained by multiplication, which yields

[FORMULA NOT REPRODUCIBLE IN ASCII] (A3)

Using the reparameterization [[mu].sub.*] = ([[sigma].sup.2.sub.V]z[delta] - [[sigma].sup.2.sub.U]e)/([[sigma].sup.2.sub.V] + [[sigma].sup.2.sub.U]) and [[sigma].sup.2*] = [[sigma].sup.2.sub.U][[sigma].sup.2.sub.V]([[sigma].sup.2.sub.V] + [[sigma].sup.2.sub.U]) yields

[FORMULA NOT REPRODUCIBLE IN ASCII] (A4)

The marginal density of E is then obtained by integrating U out of the joint density, which yields

[FORMULA NOT REPRODUCIBLE IN ASCII] (A5)

The density function for production [Y.sub.it] is then

[FORMULA NOT REPRODUCIBLE IN ASCII] (A6)

where [d.sub.it] = [z.sub.it][delta]/[[sigma].sub.U], [d.sup.*.sub.it] = [u.sup.*.sub.it]/[[sigma].sup.*], and [u.sup.*.sub.it] = [[[sigma].sup.2.sub.V][z.sub.it][delta] - [[sigma].sup.2.sub.U]([y.sub.it] - [x.sub.it][beta])]/([[sigma].sup.2.sub.V] + [[sigma].sup.2.sub.U]). With [[sigma].sup.2.sub.S] [equivalent to] [[sigma].sup.2.sub.V] + [[sigma].sup.2.sub.U] and [gamma] [equivalent to] [[sigma].sup.2.sub.U]/[[sigma].sup.2.sub.S], it follows that the log likelihood is

[FORMULA NOT REPRODUCIBLE IN ASCII] (A7)

where [d.sub.it] = [z.sub.it][delta]/([[[gamma][[sigma].sup.2.sub.S]).sup.1/2], [d.sup.*.sub.it] = [[mu].sup.*.sub.it]/[[[gamma](1 - [gamma])[[sigma].sup.2.sub.S]].sup.1/2], [[mu].sup.*.sub.it] = (1 - [gamma])[z.sub.it][delta] - [gamma]([y.sub.it] - [x.sub.it][beta]), and [[sigma].sup.*] = [[[gamma](1 - [gamma])[[sigma].sup.2.sub.S].sup.1/2]. To predict efficiencies, the following conditional expectation is used:

[FORMULA NOT REPRODUCIBLE IN ASCII] (A8)

For additional details, see Battese and Coelli (1993).

Table 1

Summary Statistics for a Sample Consisting of Observations for 73
Countries for 1975, 1980, 1985, and 1990

                                   Standard
Variable           N      Mean     Deviation   Minimum     Maximum

PR                275   3.2327273  2.0919602   1.0000000   7.0000000
CL                275   3.3054545  1.8247506   1.0000000   7.0000000
Economic freedom  275   5.6003636  1.6103112   2.1000000   9.1000000
DCS               275   0.3309091  0.4713983   0.0000000   1.0000000
ln(Y)             275  24.5632609  1.7375475  21.3634400  29.1395809
ln(K)             275  25.5036830  1.7649147  21.3835925  30.1157380
ln(L)             275  15.4709661  1.5168731  11.5462732  20.3459501
ln(H)             275   1.6033645  0.5166253   0.2476410   2.4735929
Table 2

Summary Statistics for a sample Consisting of Observations for 76
Countries for 1975, 1980, and 1985

                           Standard
Variable   N      Mean     Deviation   Minimum     Maximum

PR        225   3.4800000  2.1256932   1.0000000   7.0000000
CL        225   3.5244444  1.8445329   1.0000000   7.0000000
IS1       225   4.3226667  1.3030089   1.2000000   8.0000000
DCS       225   0.3066667  0.4621379   0.0000000   1.0000000
ln(Y)     225  24.3702961  1.6867418  21.3634400  29.0084899
ln(K)     225  25.2661510  1.7514359  21.3835925  29.9791404
ln(L)     225  15.3574841  1.3993062  11.5462732  19.5041045
ln(H)     225   1.5341557  0.6039560  -0.7721904   2.6646555
Table 3

Maximum-Likelihood Estimates of the Production Frontier and Determinants
of Technical Inefficiency for 73 Countries for 1975, 1980, 1985, and
1990 (a)


                                       Unrestricted Conditional
                                            Frontier Model
Variable                            Coefficient     t-Ratio

Production function
  Constant                            14.5940       12.3509
  ln(L)                                1.3784        6.3715
  ln(K)                               -1.1132       -5.8691
  ln(H)                                2.6665        4.3187
  0.5[[ln(L)].sup.2]                  -0.0193       -0.4877
  0.5[[ln(K)].sup.2]                   0.1110        4.0512
  0.5[[ln(H)].sup.2]                   1.0849        5.8794
  ln(L)ln(K)                          -0.0387       -1.1568
  ln(L)ln(H)                           0.1145        1.5866
  ln(H)ln(K)                          -0.2438       -3.9238
  Economic freedom                     0.0115        1.0517
  Political rights                    -0.0016       -0.0827
  Civil liberties                      0.0007        0.0338
  Time                                -0.0211       -2.9424
Inefficiency function
  Constant                             1.2388        3.7168
  DCS                                 -1.6091       -2.0534
  ln(H)                               -0.5514       -3.7460
  Economic freedom                    -0.1211       -3.1656
  Political rights                    -0.0002       -0.0053
  Civil liberties                      0.0418        0.7795
  [[sigma].sup.2]                      0.1194        5.1740
  [gamma]                              0.9383       60.8449
Loglikelihood                        100.3848
[H.sub.0]: [gamma] =                 134.97 *
 [[delta].sub.0] = ... =
 [[delta].sub.p] = 0
[H.sub.0]: [gamma] = 0               137.90 *
[H.sub.0]: parametric restrictions


                                     Unconditional World      Restricted
                                                                 World
                                       Frontier Model          Frontier
                                                                 Model
Variable                            Coefficient  t-Ratio     Coefficient

Production function
  Constant                            14.5848    10.8812       14.2727
  ln(L)                                1.4006     6.7317        1.3551
  ln(K)                               -1.1121    -5.4677       -1.0680
  ln(H)                                2.4716     4.1139        2.5651
  0.5[[ln(L)].sup.2]                  -0.0122    -0.3136       -0.1045
  0.5[[ln(K)].sup.2]                   0.1148     4.3833        0.1123
  0.5[[ln(H)].sup.2]                   1.0885     6.5969        1.1198
  ln(L)ln(K)                          -0.0447    -1.3926       -0.0420
  ln(L)ln(H)                           0.1212     1.7390        0.1266
  ln(H)ln(K)                          -0.2404    -4.1204       -0.2497
  Economic freedom                      --         --            --
  Political rights                      --         --            --
  Civil liberties                       --         --            --
  Time                                -0.0198    -2.2436       -0.0190
Inefficiency function
  Constant                             1.3175     4.8481        1.5511
  DCS                                 -1.5227    -2.2280       -1.6449
  ln(H)                               -0.5233    -3.9186       -0.5760
  Economic freedom                    -0.1398    -4.2551       -0.1361
  Political rights                     0.0024     0.0678         --
  Civil liberties                      0.0401     0.8846         --
  [[sigma].sup.2]                      0.1138     6.4641        0.1160
  [gamma]                              0.9354    64.4667        0.9368
Loglikelihood                         98.5877                  98.7245
[H.sub.0]: [gamma] =                 153.42 *                 151.69 *
 [[delta].sub.0] = ... =
 [[delta].sub.p] = 0
[H.sub.0]: [gamma] = 0               133.82 *                 132.06 *
[H.sub.0]: parametric restrictions     1.59                     3.32

                                                   Unconditional
                                    Restricted       Cobb-Douglas
                                      World
                                     Frontier       Frontier Model
                                      Model
Variable                            t-Ratio   Coefficient  t-Ratio

Production function
  Constant                           9.8552      1.9642     9.3494
  ln(L)                              8.0483      0.3005    13.6685
  ln(K)                             -5.9530      0.7209    35.0036
  ln(H)                              4.4436     -0.0387    -0.8341
  0.5[[ln(L)].sup.2]                -0.3741       --         --
  0.5[[ln(K)].sup.2]                 4.5061       --         --
  0.5[[ln(H)].sup.2]                 6.7054       --         --
  ln(L)ln(K)                        -1.3380       --         --
  ln(L)ln(H)                         1.8775       --         --
  ln(H)ln(K)                        -4.4352       --         --
  Economic freedom                    --          --         --
  Political rights                    --          --         --
  Civil liberties                     --          --         --
  Time                              -2.4008     -0.0355    -3.3298
Inefficiency function
  Constant                           7.6423      0.8149     3.1803
  DCS                               -1.9814     -1.3471    -2.5956
  ln(H)                             -4.4628     -0.0886    -0.7896
  Economic freedom                  -4.4457     -0.1906    -3.9307
  Political rights                    --        -0.0406    -1.0141
  Civil liberties                     --        -0.1242     2.1579
  [[sigma].sup.2]                    5.1495      0.1610     4.0270
  [gamma]                           61.5367      0.9276    45.4843
Loglikelihood                                   42.1104
[H.sub.0]: [gamma] =                           127.97 *
 [[delta].sub.0] = ... =
 [[delta].sub.p] = 0
[H.sub.0]: [gamma] = 0                          98.30 *
[H.sub.0]: parametric restrictions             116.54 *

(a) Test statistics marked with an asterisk indicate rejection of the
null hypothesis at the 5% level.
Table 4

Maximum-Likelihood Estimates of the Production Frontier and Determinants
of Technical Inefficiency for 76 Countries for 1975, 1980, and 1985 (a)

                                        Unrestricted Conditional
                                             Frontier Model
Variable                            Coefficient        t-Ratio

Production function
 Constant                               15.4382        12.4028
 ln(L)                                   1.7900         7.8487
 ln(K)                                  -1.3469        -7.2208
 ln(H)                                   1.3107         2.0393
 0.5[[ln(L)].sup.2]                      0.0718        1.30994
 0.5[[ln(K)].sup.2]                      0.1672         4.8423
 0.5[[ln(H)].sup.2]                      0.6994         4.5057
 ln(L)ln(K)                             -0.1158        -2.6612
 ln(L)ln(H)                              0.1564         1.7332
 ln(H)ln(K)                             -0.1987        -2.6815
 Economic freedom                       -0.0204       - 1.7028
 Political rights                        0.0310         1.3076
 Civil liberties                        -0.0272        -1.0522
 Time                                   -0.0352        -2.4850
Inefficiency function
 Constant                                1.8168         4.2672
 DCS                                    -1.5605        -2.3882
 ln(H)                                  -0.8544        -3.9088
 Economic freedom                       -0.2057        -4.5169
 Political rights                        0.0550         1.0080
 Civil liberties                         0.0016         0.0255
 [[sigma].sup.2]                         0.1524         8.3341
 [gamma]                                 0.9352        47.5906
Loglikelihood                           49.66
[H.sub.0]: [gamma] = [[delta].sub.     105.88 *
 0] = ... = [[delta].sub.p] = 0
[H.sub.0]: [gamma] = 0                 103.28 *
[H.sub.0]: parametric restrictions       --

                                     Unconditional World      Restricted
                                                                 World
                                       Frontier Model          Frontier
                                                                 Model
Variable                            Coefficient  t-Ratio     Coefficient

Production function
 Constant                               14.4586  14.5422         14.1816
 ln(L)                                   1.7658   7.8996          1.7099
 ln(K)                                  -1.2965  -7.3614         -1.2244
 ln(H)                                   1.7802   2.8010          1.5444
 0.5[[ln(L)].sup.2]                      0.0810   1.3599          0.0716
 0.5[[ln(K)].sup.2]                      0.1700   4.7690          0.1610
 0.5[[ln(H)].sup.2]                      0.7115   4.5050          0.7359
 ln(L)ln(K)                             -0.1207  -2.6385         -0.1123
 ln(L)ln(H)                              0.1715   1.9746          0.1536
 ln(H)ln(K)                             -0.2262  -3.2428         -0.2088
 Economic freedom                          --       --              --
 Political rights                          --       --              --
 Civil liberties                           --       --              --
 Time                                   -0.0398  -2.4322         -0.0389
Inefficiency function
 Constant                                1.7735   4.4718          2.0330
 DCS                                    -1.5077  -2.7940         -1.6526
 ln(H)                                  -0.8301  -3.6709         -0.9294
 Economic freedom                       -0.1926  -4.3838         -0.1848
 Political rights                       -0.0233  -0.5467            --
 Civil liberties                         0.0579   1.1072            --
 [[sigma].sup.2]                         0.1427   8.3285          0.1519
 [gamma]                                 0.9157  46.6451          0.9256
Loglikelihood                           46.84                    46.56
[H.sub.0]: [gamma] = [[delta].sub.     110.72 *                 110.15 *
 0] = ... = [[delta].sub.p] = 0         97.64 *                  98.32 *
[H.sub.0]: [gamma] = 0
[H.sub.0]: parametric restrictions       5.64                     6.20

                                    Restricted    Unconditional
                                      World       Cobb-Douglas
                                     Frontier    Frontier Model
                                      Model
Variable                            t-Ratio   Coefficient

Production function
 Constant                           13.9630        2.1861
 ln(L)                               7.4992        0.2955
 ln(K)                              -6.7960        0.7299
 ln(H)                               2.4702       -0.2049
 0.5[[ln(L)].sup.2]                  1.3269          --
 0.5[[ln(K)].sup.2]                  4.8602          --
 0.5[[ln(H)].sup.2]                  5.5178          --
 ln(L)ln(K)                         -2.6684          --
 ln(L)ln(H)                          1.8186          --
 ln(H)ln(K)                         -3.0560          --
 Economic freedom                      --            --
 Political rights                      --            --
 Civil liberties                       --            --
 Time                               -2.3651       -0.0596
Inefficiency function
 Constant                            7.4289        1.4486
 DCS                                -2.7950       -1.2970
 ln(H)                              -5.2368       -0.4643
 Economic freedom                   -4.5959       -0.1484
 Political rights                      --         -0.0273
 Civil liberties                       --          0.0470
 [[sigma].sup.2]                     7.8238        0.1441
 [gamma]                            47.8243        0.8868
Loglikelihood                                      5.05
[H.sub.0]: [gamma] = [[delta].sub.                84.29 *
 0] = ... = [[delta].sub.p] = 0                   68.14 *
[H.sub.0]: [gamma] = 0
[H.sub.0]: parametric restrictions                89.22 *

                                     Unconditional
                                     Cobb-Douglas
                                    Frontier Model
Variable                            t-Ratio

Production function
 Constant                            7.7030
 ln(L)                              10.2781
 ln(K)                              28.5851
 ln(H)                              -3.0733
 0.5[[ln(L)].sup.2]                    --
 0.5[[ln(K)].sup.2]                    --
 0.5[[ln(H)].sup.2]                    --
 ln(L)ln(K)                            --
 ln(L)ln(H)                            --
 ln(H)ln(K)                            --
 Economic freedom                      --
 Political rights                      --
 Civil liberties                       --
 Time                               -2.8534
Inefficiency function
 Constant                            5.1253
 DCS                                -2.6323
 ln(H)                              -4.2753
 Economic freedom                   -3.1158
 Political rights                   -0.5630
 Civil liberties                     0.8033
 [[sigma].sup.2]                     4.8515
 [gamma]                            28.1703
Loglikelihood
[H.sub.0]: [gamma] = [[delta].sub.
 0] = ... = [[delta].sub.p] = 0
[H.sub.0]: [gamma] = 0
[H.sub.0]: parametric restrictions
Table 5

Estimated Efficiencies by Country

                                                                Rank
Country              1975    1980    1985    1990   Average  Average (a)

Algeria                     0.5020  0.5930  0.5683  0.5544       64
Argentina           0.9283  0.8998  0.8618  0.8890  0.8947       32
Australia (b)       0.9475  0.9459  0.9519  0.9569  0.9505       16
Austria (b)         0.9608  0.9635  0.9595  0.9671  0.9627       11
Bangladesh          0.3098  0.5209  0.6023  0.7635  0.5491       65
Belgium (b)         0.9749  0.9770  0.9769  0.9795  0.9771        2
Bolivia                     0.6072  0.5979  0.6794  0.6281       60
Brazil              0.6642  0.6348  0.6403  0.6686  0.6520       58
Cameroon                    0.7935  0.8062  0.6860  0.7619       49
Canada (b)          0.9804  0.9790  0.9776  0.9762  0.9783        1
Chile               0.8146  0.9458  0.8986  0.9574  0.9041       31
China                       0.7513  0.8414  0.8776  0.8234       42
Colombia            0.9236  0.9399  0.9241  0.9505  0.9345       23
Costa Rica          0.9598  0.9465  0.9086  0.9315  0.9366       22
Cyprus              0.9119  0.8709  0.9091  0.9411  0.9082       29
Denmark (b)         0.9561  0.9584  0.9686  0.9684  0.9629       10
Ecuador             0.7890  0.8292  0.8327  0.8307  0.8204       43
Egypt               0.6160  0.8669  0.9351  0.9131  0.8328       38
El Salvador                 0.8960  0.8447  0.8809  0.8739       34
Finland (b)         0.9390  0.9481  0.9553  0.9587  0.9503       17
France (b)          0.9629  0.9632  0.9597  0.9636  0.9624       12
Germany (b)         0.9504  0.9602  0.9552          0.9553       15
Ghana               0.6448  0.6412  0.6620  0.6252  0.6433       59
Greece              0.9261  0.9305  0.9195  0.9239  0.9250       26
Guatemala           0.7481  0.9134  0.8487  0.9084  0.8546       36
Haiti                       0.7164  0.6711          0.6937       54
Honduras            0.7519  0.8028  0.8026  0.8502  0.8019       45
Iceland (b)         0.9539  0.9724  0.9726  0.9762  0.9688        6
India               0.5634  0.6073  0.6815  0.7943  0.6616       56
Indonesia           0.9169  0.9113  0.8554  0.7802  0.8659       35
Iran                0.9168  0.6445  0.8557  0.8917  0.8272       41
Iraq (c)            0.9703                          0.9703       (2)
Ireland (b)         0.9702  0.9711  0.9686  0.9785  0.9721        3
Israel              0.9490  0.9550  0.9623  0.9697  0.9590       14
Italy (b)           0.9237  0.9478  0.9493  0.9604  0.9453       19
Ivory Coast (c)     0.3327  0.4338  0.5082          0.4249      (68)
Jamaica             0.5251  0.4672  0.4715  0.5543  0.5045       67
Japan (b)           0.9447  0.9311  0.9273  0.9224  0.9314       25
Jordan              0.8915  0.9793  0.9636  0.9407  0.9438       20
Kenya               0.3178  0.4578  0.4934  0.5556  0.4562       69
Korea               0.9619  0.9177  0.9270  0.9279  0.9336       24
Madagascar (c)      0.6298  0.6547  0.6385          0.6410      (59)
Malawi              0.3835  0.3785  0.4195  0.4679  0.4124       70
Malaysia            0.9388  0.9191  0.8906  0.9045  0.9133       28
Mali                0.4853  0.5390  0.5020  0.4852  0.5029       68
Mauritius           0.5869  0.6599  0.7894  0.9358  0.7430       50
Mexico              0.9046  0.9202  0.8938  0.9044  0.9057       30
Morocco (c)         0.4077  0.5186  0.5665          0.4976      (67)
Myanmar                     0.4892  0.6292          0.5592       63
Netherlands (b)     0.9650  0.9663  0.9636  0.9673  0.9655        8
New Zealand (b)     0.9470  0.9296  0.9366  0.9402  0.9383       21
Nigeria (c)         0.2667  0.2583  0.3335          0.2862      (75)
Norway (b)          0.8978  0.9308  0.9419  0.8826  0.9133       27
Pakistan (b)        0.4419  0.5947  0.6443  0.9446  0.6564       57
Panama              0.8572  0.9051  0.8990  0.7196  0.8452       37
Paraguay                    0.7985  0.8258  0.7139  0.7794       47
Peru                0.8567  0.8144  0.7691  0.6421  0.7706       48
Philippines         0.8847  0.8682  0.6756  0.7371  0.7914       46
Portugal            0.5869  0.7487  0.7079  0.7770  0.7051       53
Rwanda              0.6602  0.8624          0.6212  0.7146       51
Senegal                     0.5723  0.6357  0.6726  0.6268       61
Sierra Leone        0.5406  0.5657  0.5497  0.6319  0.5720       62
Singapore           0.9407  0.9580  0.9291  0.9639  0.9480       18
Spain (b)           0.9619  0.9619  0.9560  0.9656  0.9614       13
Sri Lanka                   0.7635  0.8461  0.8200  0.8099       44
Sweden (b)          0.9708  0.9647  0.9703  0.9700  0.9690        5
Switzerland (b)     0.9707  0.9625  0.9607  0.9613  0.9638        9
Tanzania            0.4136  0.4229  0.3665          0.4010       71
Thailand            0.7813  0.8169  0.8283  0.8931  0.8299       40
Tunisia             0.5588  0.6626  0.6922  0.7737  0.6718       55
Turkey              0.6630  0.6543  0.7178  0.7982  0.7083       52
Uganda                      0.2043  0.4206  0.5722  0.3990       72
United Kingdom (b)  0.9686  0.9690  0.9722  0.9745  0.9711        4
United States (b)   0.9622  0.9643  0.9685  0.9703  0.9663        7
Uruguay             0.8972  0.9034  0.7995  0.9114  0.8779       33
Venezuela           0.8780  0.8798  0.7777  0.7950  0.8326       39
Zambia              0.2779  0.3348  0.3580  0.4088  0.3449       73
Zimbabwe                    0.4377  0.5413  0.5572  0.5121       66

(a) Ranks in parentheses are from the 76-country, three-year regressions
and are not directly comparable to the others.

(b) The predicted efficiencies for these countries are based on the
results presented in the fifth column of Table 4. The predicted
efficiencies for the other countries are based on the results presented
in the fifth column of Table 3.

(c) Developed countries.

Received June 2000; accepted August 2001.

(1.) If [gamma] = 0 and [z.sub.it] includes variables that also appear in the production function, then the duplicate variables cannot be added to the mean function without inducing perfect collinearity. From a practical standpoint, this means that the test statistic of the null hypothesis [gamma] = 0 has a [chi square] distribution where the degrees of freedom are determined by the number of duplicate variables in [x.sub.it] and [z.sub.it] plus one (for the parameter [gamma]).

(2.) Some experimentation was done with different time specifications (e.g., time and time squared, quinquennial dummies, and nonneutral technical change), but none of the choices had much impact on the overall results we obtained.

(3.) Fare et al. (1994), for instance, use linear programming techniques to decompose productivity growth in 17 OECD countries into the part due to technical change and the part due to catch-up. In doing so, they specify a world production function that depends only on capital and labor. No other country data are included in the production function.

(4.) In an earlier version of the paper (Adkins. Moomaw, and Savvides 1998). we did not include human capital as an input in the production function. We thank an anonymous referee for suggesting that we include human capital as an input. We were also encouraged by a referee to extend the period under study into the 1990s if possible. The original human capital series and freedom measure did not permit this, and we were required to use alternative measures.

(5.) The correlation is measured on the basis of 213 overlapping observations we have for the two measures. A correlation of 0.78 is, in our opinion, surprisingly low given the similarity of components in the indices. Fortunately, the results of this paper are robust with respect to this choice.

(6.) The current data (version 5.6) are available from http://pwt.econ.upenn.edu/.

(7.) In this sample, the following countries are added: Iraq (1975 only), Ivory Coast, Madagascar, Morocco, and Nigeria. China and Myanmar are not in this sample.

(8.) Adkins, Moomaw, and Savvides (1998) estimate similar models without including human capital as an input in the production function and obtain results that are very similar to the ones obtained here, which indicate that economic freedom (measured using IS1 from Gwartney, Lawson, and Block 1996) reduced inefficiency among 23 OECD countries during 1975, 1980, and 1985 sample periods.

(9.) Ranks are also given in Table 5. Ranks in parentheses are from the 76-country, three-year regressions and are not directly comparable to the others. A correlation of 0.969 was computed between the ranks of the two sets of predicted efficiencies.

References

Adkins, Lee C., Ron Moomaw, and Andreas Savvides. 1998. The determinants of intemational variation in technical efficiency: Estimates from a stochastic frontier production. Presented at the 1998 Southern Economic Association Meetings, Baltimore.

Barro, Robert, and Jong-Wha Lee. 2000. International data on educational attainment: Updates and implications. Harvard University, Center for International Development Working Paper No. 42.

Battese, G. E., and T. J. Coelli. 1993. A stochastic frontier production incorporating a model for technical inefficiency effects. University of New England Working Papers in Applied Statistics No. 69.

Battese, G. E., and T. J. Coelli. 1995. A model for technical inefficiency effects in a stochastic frontier production function for panel data. Empirical Economics 20:325-32.

Battese, G. E., and G. S. Corra. 1977. Estimation of a production frontier with application to the pastoral zone of eastern Australia. Journal of Agricultural Economics 21:169-79.

Benhabib, J., and M. Spiegel. 1994. The role of human capital in economic development. Journal of Monetary Economics 34:143-74.

Benhabib, J., and M. Spiegel. 1997. Growth and investment across countries: Are primitives all that matter? New York University Working Paper.

Bergson, Abram. 1987. Comparative productivity: The USSR, Eastern Europe, and the West. American Economic Review 77:342-57.

Bergson, Abram. 1989. Planning and performance in socialist economies. Boston: Unwin Hyman.

Bergson, Abram. 1991. The USSR before the fall: How poor and why. Journal of Economic Perspectives 5:29-44.

Coelli, T. J. 1996. A guide to FRONTIER 4.1: A computer program for stochastic frontier production and cost estimation. University of New England Centre for Efficiency and Productivity Analysis Working Paper No. 96/07.

Dawson, John W. 1998. Institutions, investment, and growth: New cross-country and panel data evidence. Economic inquiry 35:603-19.

Edwards, Sebastian. 1998. Openness, productivity, and growth: What do we really know? Economic Journal 108(447):383-98.

Fare, R., S. Grosskopf, M. Norris, and Z. Zhang. 1994. Productivity growth, technical progress, and efficiency change in industrialized countries. American Economic Review 84:66-83.

Freedom House. 2001. Freedom in the world: Freedom House country rankings. http://www.freedomhouse.org/ratings/index.htm.

Gwartney, James, Robert Lawson. and Walter Block. 1996. Economic freedom of the world: 1975-1995. Vancouver: Frazier Institute.

Gwartney, James, Robert Lawson, Walter Park, and Charles Skipton. 2001. Economic freedom in the world 2001. Vancouver: Fraser Institute (available at http://www.freetheworld.com/ download.html).

Heston, Alan, and Robert Summers. 1991. The Penn world table (mark 5): An expanded set of international comparisons, 1950-1988. Quarterly Journal of Economics 106:327-68.

Huffman, Wallace E. 1977. Allocative efficiency: The role of human capital. The Quarterly Journal of Economics 91(1):59-79.

Kumbhaker, S. C., S. Ghosh, and T. McGuckin. 1991. A generalized production frontier approach for estimating determinants of inefficiency in U.S. dairy farms. Journal of Economics and Business Statistics 9:279-86.

Mankiw, Gregory N., David Romer, and David N. Weil. 1992. A contribution to the empirics of economics growth. Quarterly Journal of Economics 107:407-37.

Marer, Paul. 1981. Economic performance and prospects in Eastern Europe: Analytical summary and interpretation of results. Papers submitted to the Joint Economic Committee, Congress of the United States. Washington, D.C.: Government Printing Office.

Moroney, John R. 1992. Energy consumption, capital, and real output: A comparison of market and planned economies. Resources and Energy 14:363-380.

Moroney, John R., and C. A. K. Lovell. 1997. The relative efficiencies of market and planned economies. Southern Economic Journal 63:1084-93.

Nehru, Vikram, and Ashok Dhareshwar. 1993. A new database on human physical capital stock: Sources, methodology, and results. Revista Analisis Economico 56(2):37-59.

Nehru, Vikram, Eric Swanson, and Ashutosh Dubey. 1995. A new database on human capital stock in developing and industrial countries: Sources, methodology, and results. Journal of Development Economics 46:379.

Transparency International. 2001. Corruption perceptions index: Internet center for corruption research. http://www.gwdg.de/index.html (in German).

Reifschneider, D., and R. Stevenson. 1991. Systematic departures from the frontier: A framework for the analysis of firm productivity. International Economic Review 32:715-23.

Rodrik, Dani. 1997. Democracy and economic performance. Prepared for a conference on democratization and economic reform in South Africa, Cape Town, January 16-19, 1998. December 14, 1997.

Rodrik, Dani. 1999. Democracies pay higher wages. The Quarterly Journal of Economics 114:707-38.

Rodrik, Dani. 2000. Institutions for high-quality growth: What they are and how to acquire them. NBER Working Paper No. 7540.

Schultz, Theodore W. 1975. The value of the ability to deal with disequilibria. Journal of Economic Literature 13:827-46.

World Bank. 1992. World development report 1992. New York: Oxford University Press.

World Bank. 1993. The east Asian miracle: Economic growth and public policy. New York: Oxford University Press for the World Bank.

Lee C. Adkins, * Ronald L. Moomaw, + and Andreas Savvides +

* Department of Economics and Legal Studies in Business, College of Business Administration, Oklahoma State University, Stillwater, OK 74078, USA; E-mail [e-mail]; corresponding author.

+ Department of Economics and Legal Studies in Business, College of Business Administration, Oklahoma State University, Stillwater, OK 74078, USA.

We thank two anonymous referees and Kevin Grier for helpful comments and suggestions.