4
derived from a 1% sample of the insurance accounts that the German Federal Labour Office
(Bundesanstalt für Arbeit) maintained in respect of employees liable to pay social security
contributions between 1975 and 1995.
1
These ‘process-produced data’ are supplemented by
information on periods of unemployment during which a claimant received benefits and on
certain characteristics of the establishments that employed individuals in the subsample dur-
ing the period of observation. This corresponds to about 7.8 million employment or benefit
payment notifications, with each individual record containing 35 variables (cf. Bender, Haas
and Klose 2000).
For our estimations we will use cox proportional hazard rate models (“cox models”). Com-
pared with other parametric methods the semi-parametric cox model has one particular advan-
tage to estimate transition rates: By using cox models we can certainly calculate the influ-
ences of the interesting covariates on the transition rate as a mathematical function, but we
need no further assumptions about the time dependency of the transition rate. Therefore, the
cox model is a very robust and flexible method to analyse transition processes (Blossfeld and
Rohwer 2002).
The main aim of this paper is to answer the question how job stability and its determinants
have changed in the course of time. Therefore, cox proportional hazard rate models will be
estimated for two different periods in time both including the same set of explanatory socio-
economic variables. First, we will estimate an unspecified transition model for the general
event of leaving a job. The end of an employment spell (‘exit’) is defined as the termination
of an existing insurable job. Second, we will estimate a competing risk model in which we
distinguish between different kinds of events:
1
Hence, the analysis is restricted to West-German employees and the self-employed, civil servants and those
in marginal part-time employment are not included in the following analysis.
Convention