Predicting Wave Nonresponse from Prior Wave Data Quality

• Activity Number: 41
• Type: Topic Contributed
• Date/Time: Sunday, August 8, 2004 : 4:00 PM to 5:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Brian J. Meekins*+ and Roberta L. Sangster
• Companies: Bureau of Labor Statistics and Bureau of Labor Statistics
• Keywords: panel attrition ; data quality ; nonresponse

Abstract

Prior research on nonresponse and data quality in panel surveys is largely focused on the effect of reducing unit nonresponse on survey estimates. This paper models panel attrition or subsequent wave nonresponse, from both noncontact and refusal, using aspects of data quality from previous waves of a panel survey. Data from the Telephone Point-of-Purchase Survey (TPOPS) for the years 2001 to 2003 are used in this analysis. Respondents for the TPOPS are initially recruited using an RDD design and are subject to three more waves of interviewing. A total of 35,477 respondents are used. Panel attrition is a problem with about 76% of the initially completed respondents participating in wave 2, decreasing to 68 percent in waves 3 and 4. Measures of data quality include: expenditure rounding, outlet address completeness, number of outlets/unique outlets, duration of interview, and item nonresponse, especially to income and race. Factor analysis is used to create a single factor of data quality. Covariates also include the difficulty of contacting the respondent and the respondent's reluctance to complete the interview on the prior waves.