مطالعه تکراری “باز بودن، اندازه کشور و اندازه دولت” A Replication Study of “Openness, Country Size, and Government Size”

There has been wide interest among public economists in exploring and explaining the relationship between a country’s exposure to international trade (openness) and the scope of its government (government size). In an influential paper, Rodrik (1998) advanced the hypothesis that polities respond optimally to the greater macro risks associated with trade openness by putting in place policies of public employment and public transfer spending that provide insurance in the face of external shocks. The evidence buttressed in favor of this hypothesis was a positive partial correlation between government size and openness. However, Alesina and Wacziarg (1998) in response argued that this partial correlation was spurious and reflected the negative relationship between government size and country size, on the one hand, and between openness and country size, on the other hand: when controlling for country size (log population), the effect of openness on government size disappeared.1 Subsequently, Ram (2009) showed that with country- and time-fixed effects, the various effects of country size on either openness or government size were weakened, and that country size could no longer be considered the mediating variable explaining the relationship between openness and government size. Ram’s analysis relied mainly on data from the Penn World Table (PWT) version 6.1 (Heston, Summers, and Aten 2002). The PWT is among the most widely used data sources for cross-country comparisons for the level and growth rate of gross domestic product (GDP) and other macroeconomic variables, resulting from the fact that it accounts for differences in relative prices. However, a number of researchers have raised questions regarding the reliability of newer versions of this data set as a result of data revisions over time. For example, Breton (2012) and Johnson et al. (2013) emphasize that estimates vary substantially across more recent versions of the data set. On the flip side, they acknowledge that the issue is not as pertinent if one uses low frequency data, for example, ten-year averages of variables. Such concerns warrant a replication of Ram’s analysis to determine whether his estimation results are robust to the data revisions. On the estimation side, empirical studies in macroeconomics employing fixed effects tend to find insignificant results with shorter samples, where ordinary least squares (OLS), random effects, or between estimators find significant effects. Examples include the association between democracy and income (Acemoglu et al. 2008) and between growth and human capital (Benhabib and Spiegel 1994). A possible explanation underlying this is the fact that the fixed-effects estimator tends to exacerbate measurement error bias when the right-hand variables are time persistent. If results are driven by measurement error or other data problems, the impact becomes less severe with a longer time period because there is more variation within countries across time in the underlying signal. Even when differencing (which is what fixed effects does), the signal-to-noise ratio may remain sufficiently high to allow the coefficients being estimated to remain significant. For instance, in Acemoglu et al.’s (2008) study, a positive relationship between the two variables is restored even under fixed effects when lengthening the time period beyond the 1960 to 2000 period. More recent versions of the PWT provide us with access to such a longer time frame.