Robert F. Belli
Date of this Version
In this dissertation, I seek to develop a tool for the enhancement of time-use and wellbeing measures from a total survey error perspective. In particular, I evaluate the quality of the time use data produced in the American Time Use Survey (ATUS), by exploring its indicators and identifying its main predictors, including interview rapport. Results from these analyses are then used to evaluate the extent to which certain variables correlate, as predicted, with expected levels of wellbeing.
The first specific objective was to investigate the data quality of the 2010 ATUS by constructing a data quality index. In my dissertation, data quality was operationalized as the degree of completeness with which the ATUS diary was completed. The second objective was to examine whether interview rapport predicts data quality. Interview rapport is understood as the conversational interaction quality that contributes to motivate respondents to be more thorough in their reports and to help them access the required information. Finally, the third objective was to assess the predictive power of activitybased wellbeing measures and other variables assumed to affect overall wellbeing, controlling for the impact of data quality in the prediction model.
Two factors of data quality were found through a confirmatory factor analysis model, one related to the degree of motivation to report and the second one related to memory processes that impact the level of accuracy and detail of activity reports. Gender, age, and education are significant predictors of both factors. Through a structural model, it was also found that interview rapport predicts motivation and memory though in opposite directions: While rapport appears to benefit the motivation to respond, it can be detrimental to remembering details. Finally, when predicting overall health (taken as a proxy for wellbeing), it was found that only when controlling for the memory-related data quality factor was there a relative increase in the amount of variance explained, although it was not a practically significant increase. Further research would be helpful in validating the measurement of data quality and rapport constructs, as well as to more efficiently incorporate data quality and rapport in the prediction of wellbeing.