The fade-away of an initial nonresponse bias in panel surveys
Jul 26, 09:00
The term ''Fade Away Effect'' was coined to name a phenomenon that can be observed in panel surveys which were sampled from registers. If one has access to the register information of all units, including the non-respondents of the initial wave of the panel, one will observe that a non-response bias in the first wave declines in the sub-sequent waves by itself. Alho et al. (2017) have used a Markov chain approach to model this fade-away of the initial nonresponse. The key idea is the existence of a steady state distribution of the Markov chain. Initital nonresponse is then interpreted as a deviation from the steady state distribution. The distribution on the state space swings into the steady state distribution with a velocity that depends on the transition matrix of the Markov chain. In this work we investigate a sample that is far away from a steady state distribution. Also the transition matrix is not stable over time. So in a strict sense the steady state distribution is not present in this case. Yet it can be demonstrated that a Fade Away Effect exists in this case.
The empirical data base is the German Panel on Labour Market and Social Security (PASS). Here a sample of social benefit receivers was selected from the German Social Security Register. It was possible to get access to the employment information from also for the non-responders of the first wave. The analysis covers the first 5 waves of the PASS. The non-response rate at the start was about 80 percent. We analyse transitions into and out-of social aid payments. We also present extensions to regression analysis and discuss implications for the design of longitudinal surveys.