Regional Development of Employment and Deconcentration Processes in Eastern Germany. An analysis with an econometric analogue to shift-share techniques

The process of unification of Germany 1989/90 was accompanied by a transformation crisis in the East which severely affected the labour market. Although in East Germany the average loss of employment between October 1989 and June 1993 was 37.7% there were marked re-gional differences. In rural areas this loss was higher (up to a maximum of 49.7%) and lower in city centres (the minimum was 18.1%). In 1998, the average unemployment rate was 21.9%, which varied between 10.0% and 29.2%. Since 1993 there has been a more stable development of regional employment which is analysed in the paper. Theoretical basis is an integration of the product cycle thesis (influ-enced by ideas of Schumpeter, Appelbaum & Schettkat) into a general regional-economic model which is used to explain the destiny of regions. The dynamics of the regions are ex-plained by an interaction of changes of productivity and of product demand. The effects of innovations with respect to the production process are ambivalent, since they are associated with gains of employment if the product demand is elastic, and with employment losses if the demand is inelastic. The data base for the empirical analyses is the employment statistics of East Germany. The regional units are relatively small districts, 114 are defined for Eastern Germany. Data on 28 industries are included. The individual observation is the growth of employment of an indus-try in a region for a specific year (1993-1999) (N = 19152). The method used for the analyses is based on a regression approach which is analogous to shift-share techniques and extends these traditional methods. According to a common criti-cism shift-share analysis is mechanical and deterministic (non-stochastic). Significance tests are not possible since shift-share includes a mechanic decomposition of solely a few effects. An interpretation with respect to causality is problematic. As alternatives several authors (Patterson 1991, Knudsen, Barff 1991, Möller, Tassinopoulos 2000) proposed linear models which allow significance tests and the inclusion of additional explanatory variables. In the paper a special variant of the method based on restricted regression is used.