Analysis The relationship between democracy and growth may be tested in many ways. We take as our point of departure the cross-country regression format. Yet, this scarcely limits the methodological field since any of the multitudinous approaches employed in current growth regressions might also be applied to this particular question.3 The researcher faces choices about which time intervals to consider, how to correct for serial and spatial autocorrelation, and how to resolve issues of specification, simultaneity, and endogeneity, among other matters. Fortunately, there is agreement about how to measure the dependent variable, economic growth, which is generally understood as the percentage change in gdp per capita.4
In Table 3.1, growth is regressed against two measures of democracy – a contemporary measure (“democracy level”) and a stock measure, along with various controls. The results shown in models 1 and 2 confirm the standard finding. If democracy is understood as having a contemporaneous or short-term relationship with growth (lagged by only one time-period), it has no statistically significant effect on economic performance. This nonrelationship is robust across a wide range of democracy indicators and model specifications (not shown). It matters not how one measures the level of democracy in a given year; it still has no effect on subsequent economic performance.
In the remaining models we investigate democracy as a stock variable. Model 2 shows level and stock measures of democracy side by side. [Comment on the results.] Model 3 presents what we regard as our benchmark model, with only one generally recognized control variable, GDP per capita (to account for convergence effects). Model 4 drops that control variable, demonstrating that democracy stock enhances growth performance even when convergence effects are ignored. Model 5 includes a small set of additional control variables chosen by virtue of their theoretical significance (in the growth literature) and/or their robust performance in this specification (see chapter appendix for variable definitions). Note that because some of these factors may be endogenous to a country’s regime-type, their inclusion may pose identification problems. Model 6, a “kitchen sink” specification, adds another set of control variables that are available for a large sample of countries over the postwar period and are suggested by the literature.5 Here, problems of potential endogeneity (vis-à-vis democracy stock) are even greater.
Although these various specifications affect the coefficient and standard error of democracy stock, the variable remains statistically significant. When measured as a stock variable, democracy appears to have a strong positive relationship to growth performance regardless of the specification of the growth equation.6 Further specification tests are included in the chapter appendix and in chapter fourteen.