Item Nonresponse: Modeling the Impact of Probing Don't Knows

• Activity Number: 467
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Survey Research Methods Section
• Author(s): Chris Skinner*+ and Jouni Kuha and Myrsini Katsikatsou and Sarah Butt
• Companies: London School of Economics and London School of Economics and London School of Economics and City University
• Keywords: item nonresponse ; missing data ; response propensity

Abstract

It is common in surveys to probe don't know (DK) responses in order to reduce item nonresponse. There is a risk, however, that too much probing will lead to measurement error. This negative impact needs to be balanced against the potential benefit that probing will elicit meaningful responses from initial DK respondents and thus improve overall data quality. This paper will consider how modeling may be used to disentangle the different possible impacts of probing, positive or negative, in a multi-item survey setting where missing data occurs on all items. In such a setting, we suppose that item nonresponse may be affected by a latent response propensity as well as a latent variable of interest. We use data from an experiment on the use of probing in the innovation sample of the European Social Survey in three European countries in 2014. Multi-item scales include attitudes to recipients of welfare benefits, where DKs may arise from the sensitivity of the questions, and subjective judgements about national levels of welfare need, where DKs may arise from the complexity of the task.