Date of this Version
This dissertation introduces new quantitative methods to comparative politics. These include an approximately unbiased missing data treatment, a first order autoregressive multilevel model for the analysis of cross-sectional longitudinal data, approaches for the separation of cross-sectional and longitudinal predictor effects to identify aggregation bias and a model for the empirical analysis of causal direction. Methods are demonstrated on a model replicating past research on the causes of corruption adding democratic performance as a predictor, than a new model of democratic performance is developed to test Warren’s theoretical propositions that corruption is by nature undemocratic, and finally the causal direction between corruption and democratic performance is tested empirically. The analysis includes 186 countries and between 1984 and 2004. Methodologically the findings show that the missing data treatment is not completely unbiased. Anomalies emerged when cross-sectional and longitudinal predictor effects were separated as country level predictors changes significantly across equivalent models. Substantively, corruption and democratic performance are identified as strong correlates of each other in both models. Contrary to past research, findings show that former British colonies are more corrupt then non-British colonies, underpaid government officials are not more likely to be corrupt, trade liberalization will raise corruption and the level of oil production has no impact on corruption levels. The model of democratic performance does not find level of development to be a significant predictor. Aggregation bias was identified for two correlates of corruption, stability and (in one of the models) democratic performance. Causal analysis showed that causal direction goes from democratic performance to corruption. This suggests that IMF and World Bank policies of lowering corruption before structural adjustment is granted are misguided.