Patterns of survey attrition and reluctant response in the 1996 medical expenditure panel survey

In panel designs with multiple waves of data collection, the overall survey response rate is a multiplicative function of the wave specific response rates. The 1996 Medical Expenditure Panel Survey (MEPS) follows this model, requiring five rounds of data collection with the same panel of sampled households, to acquire data on health care use, expenditures, insurance coverage and sources of payment that cover two consecutive calendar years. Gaining an understanding of the factors that distinguish the cooperative respondents, the survey participants that require use of nonresponse conversion techniques to maintain their cooperation (reluctant respondents), and the initial participants that eventually drop out of the survey (part-year respondents) is essential from both an estimation and data collection perspective. To inform the specification of nonresponse adjustment strategies in MEPS to correct for survey attrition, this study attempts to identify the characteristics that distinguish survey participants across rounds from the part-year respondents. In addition, the study identifies factors that distinguish the cooperative respondents, the reluctant respondents and the part-year respondents, to better inform the MEPS data collection effort. The investigation also examines the implications of a data collection strategy that would not convert initial survey refusals by studying the effect, on survey estimates and their precision, of excluding reluctant survey respondents. Our findings revealed that reluctant respondents in the first round of the survey were significantly more likely to become non-respondents in the second round. In addition, the round two non-respondents were also more likely to be located in large metropolitan areas, to reside in the larger households with 5 or more members, to be elderly, and more likely to be either married or separated. Many of these characteristics were similar to those found to be at higher risk for nonresponse to the round one interview. Reluctant respondents as a whole appear to be a distinctly separate group, sharing one set of characteristics with the cooperative respondent group, another set with those who refused during the second round of the survey, and yet a third set of characteristics that are uniquely their own. If no effort had been made to convert reluctant participants, the precision of our survey estimates would have declined, but not substantially.