Quantitative Cross-national Research Methods

Quantitative cross-national comparisons usually are based on smaller N's. This implies that theory needs to be stronger and that counterfactuals need to be made explicit. Bayesian estimation is, in this situation, an attractive possibility. Because dependent variables are often categorical or limited, it is often preferable to use nonlinear models, like logit and probit. Selection bias is potentially acute due to nonrandom sampling. This is a problem of identification that usually cannot be overcome by quasi-experimental designs. It calls, again, for strong counterfactual thinking. Cross-national comparisons often cannot assume independence between observations on a variable (nations form 'families'), and this provokes biased coefficients and likely heteroskadisticity. Finally, the article examines the endogeneity problem (. X is influenced by Y, or both jointly by an unobserved Z) which is potentially serious in cross-national comparisons because the meaning of a variable is embedded in the nation's entire history. The bias can derive from variable omission in which case correction calls for added controls. The article also examines the tradeoffs between large-. N and smaller-. N comparisons in terms of generalization and context, and stresses the need for stronger theory the smaller the number of cases being sampled.