Can interviewer observations of the interview predict future response?

"Interviewers made four observations related to future participation, respondent cooperation, enjoyment and whether the respondent found the questions difficult, for a large sample of face-to-face interviews at wave four of the UK Millennium Cohort Study (MCS). The focus of the paper is on predicting response behavior in the subsequent wave of MCS, four years later. The two most predictive observations are whether the respondent is likely to participate in the next wave and whether they enjoyed the interview. Not only do these predict non-response at the next wave, they do so after controlling for other explanatory variables from earlier waves in a response propensity model. Consequently, these two interviewer observations improve discrimination between respondents and non-respondents at wave five as estimated by Gini coefficients generated by a Receiver Operating Characteristic curve analysis. The predicted probabilities of responding at wave five are also used to estimate R-indicators, particularly to address the question of whether, hypothetically, conversion of 'frail' respondents would lead to improved representativity and reduced bias in longitudinal estimates of interest. The evidence from the R-indicators and partial -Rindicators suggests that successful conversions could achieve those aims although the cost of so doing might outweigh the benefits." (author's abstract)