The interaction between unit and item nonresponse in view of the reverse cooperation continuum

The paper addresses two questions. First, is item nonresponse (INR) a precursor of panel attrition (UNR), as predicted by the theory of a latent cooperation continuum, or is the interrelation of another type? Second, are the results in models of item nonresponse behavior affected by a selectivity bias due to panel attrition? We test the hypothesis of a latent cooperation continuum with data taken from the German Socio-Economic Panel (GSOEP) - and can not find evidence for it. In contrast, we hypothesize that the relationship of both nonresponse types may be inverse in principle (the reverse cooperation continuum) and both types of cooperation may coexist. Besides unit nonresponse we analyze questionnaire nonresponse, i.e. participating but refusing a whole questionnaire with specific items in a multi-questionnaire survey. We find evidence for negative correlation of INR with questionnaire nonresponse. The correlation between item and unit nonresponse is inverse U-shaped which supports the hypothesis of coexistence of both types of cooperation. Addressing the second question we test whether panel attrition causes endogenous sample selection in regressions of INR by means of a bivariate probit model for selection correction. Additionally we use Monte Carlo simulations to test the influence of alternative assumptions for INR-behavior of attriters. The results show that attrition bias is item-specific. Existence and magnitude of the bias differs with the analyzed subsample.