Clustering analysis of panel data

The panel data method has become quite common in recent years because of the two dimensional nature of the panel data set; i.e. cross section and time section. Over the dimension of time, panel data are related to individuals, firms, countries and so forth. Thus, heterogeneity may occur in these units. However, it is assumed that the slope parameter of interest in the panel data model is homogeneous with respect to the units. Ignoring the heterogeneity of the slope parameters will cause incorrect results. In this study, panel data cluster analysis was applied considering that the panel data set is heterogeneous according to the units. A new distance measurement has been defined to cluster the mixed variable panel data. The cluster analysis was performed by integrating Gower, k-prototypes distances and the proposed new distance measurement into the agglomerative hierarchical clustering algorithms. In addition, clustering has been done with the k-prototype algorithm and the results have been examined. Similar units are grouped in the same clusters by clustering analysis. Panel data analysis was applied to the obtained groups and parameter estimations were made.