Temporary and permanent unit non-response in follow-up interviews of the Health and Retirement Study


  • Pierre-Carl Michaud Université du Québec à Montréal
  • Arie Kapteyn RAND Corporation
  • James P Smith RAND Corporation
  • Arthur van Soest Tilburg University




Selection bias, attrition, panel data, propensity scores


We study the effects of attrition and other unit non-response in the HRS on inferences about the distribution of socio-economic variables. A feature of the HRS is that efforts are made to bring non-respondents in a particular wave back in the next wave. For cross-section distributions of socio-economic variables of interest in 2004, we find much larger selection effects when discarding this group than when temporary non-respondents are included. A similar conclusion is obtained from our analysis of examples of panel data models, explaining changes in wealth, health, or labor force participation. This has implications for users and designers of the HRS as well as other surveys.