Propensity and Prediction in Adaptive Design

Time: 
30.09.2014 - 17:15
Location : 
A 5,6 Raum A 231
Type of Event : 
AB A-Kolloquium
Lecturer: 
Prof. James M. Lepkowski
Lecturer affiliation: 
University of Michigan
Description: 

Household surveys based on carefully selected samples from a populations face uncertainty during household recruitment about a myriad of design criteria. The uncertainty arises in such processes as contacting households, identifying eligible subjects in a sample household, gaining household and subject cooperation, and finding mutually agreeable times to interview eligible subjects. Responsive or adaptive survey designs are being investigated to learn whether paradata observed during data collection can be used to monitor costs and response characteristics and develop interventions to change the outcome of data collection processes. This paper reviews a production model for a data collection process that suggests a set of paradata that might be useful in managing a survey responsively. The paradata are used to estimate response propensity and to predict values of key characteristics for households and subjects who have not yet been successfully recruited or interviewed. The paper presents examples of management interventions based on paradata predictions in the National Survey of Family Growth in the United States. The effectiveness of the interventions is assessed in terms of changes in interviewer performance, changes in response rates, and predicted and observed differences between respondent and nonrespondent characteristics.