(Forthcoming at the Journal of Public Administration Research and Theory, 2013)
Panel data analysis has become a popular tool for researchers in public policy and public administration. Combining information from both spatial and temporal dimensions, panel data allow researchers to use repeated observations of the same units (e.g. government agencies, public organizations, public managers, etc.), and could increase both quantity and quality of the empirical information. Nonetheless, practices of choosing different panel model specifications are not always guided by substantive considerations. Using a state-level panel data set related to public health administration as an example, I compare four categories of panel model specifications: (1) the fixed effects model, (2) the random effects model, (3) the random coefficients (heterogeneous-parameter model), and (4) linear dynamic models. I provide an overview of the substantive consideration relevant to different statistical specifications. I compare, furthermore, estimation results and discuss how these different model choices may lead to different substantive interpretations. Based on model comparisons, I demonstrate several potential problems of different panel models. I conclude with a discussion on how to choose among different models based on substantive and theoretical considerations.