Research Theme 3: Panel Data Methods and Applications
Panel data methods have become more common than ever as an econometric tool for modelling individual behaviours (among firms, consumers, households etc.), partly as a result of the development of powerful software for panel data estimation, and partly through the availability of high quality longditudinal data. This third research theme covers arange of research questions that look at panel methods beyond the traditional linear approach to estimation on panels with large N and small T. Our focus is on the following areas of research:
Panel Time Series
Traditional panel data methods are applicable in circumstances where the within-period sample size N is large, but the number of time periods T over which these repeated samples are observed is small. More recently, panel time series methods have been developed to cover the case where both time T and within-period sample size N are large and of the same order of magnitude (for example, cross-country panels). Research on this theme combines theoretical evaluation and development of panel time series methods, and applied work for which such methods are appropriate. Specific questions include:
- How do traditional panel data methods and panel time series methods compare when applied to a common large N, large T datasource?
- How should unobserved individual heterogeneity be dealt with in panel time series?
- What are the properties of tests for unit roots and cointegration in panel time series, compared with traditional time series?
Panel Discrete Choice Models
The use of discrete choice models (particularly conditional, multinomial and nested probit and logit) for public policy evaluation is now commonplace. Such models are often used to model discrete economic decisions among firms and households, and to explore how these decisions are affected by changes in the policy environment. Traditionally, empirical evaluation models for this purpose have ignored intertemporal issues that could have a substantial impact on modelled responses to tax and public policy reform. This second strand of research seeks to explore how panel data methods applied to qualitative choice models can improve our capacity to understand the effects of public policies on economic choices. Some specific questions are:
- How can panel discrete choice models be adapted to take account of serial correlation in unobservables?
- To what extent does genuine inter-temporal state dependence affect modelled outcomes in a properly dynamics pecification of discrete choice?
Semiparametric Estimation of Linear and Non-Linear Panel Data Models
This topic addresses the evaluation and application of recent advances in the use of semiparametric methods applied to non-linear panel data models. Topics under this theme include the following:
- Semiparametric estimation of quantile, proportional hazards, multilevel and LDV regression models.
- Efficiency in the estimation of semiparametric panel data models with dynamic structures, binary dependent variables and latent factors.
- Testing for serial correlations in semiparametric partially linear models (with possibly lagged dependent variables).
- Modelling joint and marginal distributions in the analysis of categorical panel data.
- Treatment of unobserved heterogeneity in non-linear count data models.