Sense and sensitivity in pooled analysis of political data

Abstract. In recent years, pooled time–series cross–section data analysis has been advocated as a method for overcoming the ‘small N, many variables’ problem in comparative political economy in order to derive valid inferences from statistical comparisons of nation states. Moreover, the approach seemed promising in handling both comparisons among different countries and developments over time. However, due to the complex structure of pooled data sets, this approach cannot simply be regarded as a convenient way of increasing the number of cases and getting more significant results. This paper exemplifies the fallacies of pooled data sets by reanalyzing a study done by Paul Boreham and Hugh Compston on the effect of labour participation in policy formation on unemployment. Reanalysis of their data set controlling for the panel structure of the data by using ‘panel corrected standard errors’ and a more detailed analysis of the bivariate relationships show that the causal effect is less clear–cut than suggested and becomes considerably weaker during the 1980s. These dynamics are simply averaged out by pooled analysis. This leads to the conclusion that the currently most popular approach to pooled time–series cross–section analysis in comparative political economy — the constant–coefficients model — neither solves the small–N problem nor draws attention to dynamics over time. Thus, the article concludes that a sensitive interpretation of the findings obtained by advanced statistical methodology in comparative political economy is still dependent on small–N comparative analysis.