Abstract

The LISS panel has taken extraordinary efforts to recruit and retain households which were not regular users of the internet into the study. These households were provided with computers, internet connections, and training. Leenheer and Scherpenzeel (2012) showed that including these cases made the panel more representative of the
overall Dutch population by bringing in responsdents who were more likely to be older, to live in single-person homes and to have migration backgrounds than those who already had home computers and internet access. This study extends this analysis beyond demographics and household characteristics by exploring how the inclusion of these cases affects model estimates derived from LISS panel data. In particular, we replicate two analyses by van Wilsem (2011) on crime victimization, and explore how the conclusions in these papers would have been different had the non-internet households not been brought into the LISS survey. In this way, we estimate the reduction in coverage bias for these analyses due to inclusion of these households.