Abstract
Panel data often serve as the basis for innovation analysis on corporate R&D activities and their economic impacts. Unlike simple time series or cross section data, panel data permit analysis of dynamic aspects of innovation activities involving cross-sectional (e.g. cross-enterprise) developments. This article reviews the methodology of panel data analysis as exemplified by an estimation of production function at the corporate level. The explanatory variables in the production function at the corporate level may include factors of production, such as labor and capital, quality of the management and employees, brand image, organizational management, and many more. Such qualitative factors are not usually treated as explicit explanatory variables in estimation of the production function because of difficulty in statistical treatment, but they do bias estimation. An advantage of estimation based on panel data is freedom from such biases. In practice, "fixed effect" models are often used in which unobserved variables are assumed to be constant over the period of observation. It is known, however, that the fixed effect model enlarges the influence of errors in explanatory variables on the estimation results. The article describes a model for evaluating the relative influences of unobserved variables and of data errors. It also presents an overview of the instrumental variable technique, which is indispensable for dealing with the problem of endogenicity of explanatory variables, taking into account the characteristics of panel data. Since the estimated production function is highly dependent of the estimation technique used, choosing an appropriate method is of prime importance in statistical estimation using panel data.
