Conferences

Panel data analysis presents great opportunities for testing causal relationships. At the same time, misuse of analytic strategies can lead to misspecification of models and misinterpretation of results. This session presents key issues in the causal analysis of multi-wave panel data, for which the Polish Panel Survey [POLPAN], the British Household Panel Survey [BHPS], and the National Longitudinal Survey of Youth [NLSY]) are typical examples.

We will consider papers based on panel data with constant coefficients models, fixed effects models, and random effects models. However, the priority will be given to types of models that are dynamic, and explicitly deal with autocorrelation from one temporal period to another. It is assumed that an autoregression on lags of the residuals indicates if autocorrelation is present, and thus if dynamic panel analysis is necessary.

Various models can account for dynamic effects including, for example, those of general methods of moments (GMM) (with instrumental variables), Arellano and Bond (with lagged dependent variables), or based on a Prais-Winston transformation or a Cochrane-Orcutt transformation (with a first partial differencing).

Three statistical packages – STATA, LIMDEP, and SAS – appear to have a particularly rich variety of panel analytic procedures. Comparisons of specific procedures in these packages are of a great interest for panel data users. Also, the potentials of panel data analysis with R are awaiting detailed exploration.