check; the response rate achieved was 84 percent after excluding the exit cases (National Institute on Aging, 2011b).
Panel Study of Income Dynamics
The PSID is a longitudinal household survey directed by the University of Michigan Survey Research Center. The first round of data collection took place in 1968; the sample consisted of 18,000 individuals living in 5,000 families. From 1968 to 1997, data were collected every year. In 1999 PSID started collecting data biennially. The mode of data collection changed from in-person interview to telephone interview in 1973 to reduce costs. Exceptions are made for households with no telephone or in other circumstances that do not permit a respondent to give a telephone interview. The average in-person interview was around one hour long. Despite attempts at streamlining the survey over time, the cumulative response rate dropped from 76 percent in the baseline year, 1968, to 56 percent in 1988 (Hill, 1999). Further changes were made in 1993 when PSID moved from using paper-and-pencil telephone interviews to CATI. CATI has been used for PSID ever since.
Table 1-9 shows response rates for different segments of the PSID sample for the 2003 and 2005 waves.10 A response rate of 97 percent is consistently achieved for the core segment and 88 percent for the immigrant segment. McGonagle and Schoeni (2006) documented the factors they believe are responsible for the high response rates in the PSID. These include incentive payments, payments for updating locating information during non-survey years, the use of experienced interviewers, persuasion letters sent to reluctant respondents, informing respondents about upcoming data collection waves, sending thank-you letters to responding households, and training interviewers to handle refusals. Even though response rates are exemplary in the case of the PSID relative to other panel surveys, it does suffer from high cumulative nonresponse rates due to an attrition rate of 50 percent (Cellini et al., 2008).
The reasons given by nonrespondents for not taking part in surveys have not changed much over time. A comparison of nonresponse reasons from two U.S. surveys—the 1978 National Medical Care Expenditure Survey (NMCES) and the 2008 NHIS—shows that the reasons reported
10The response rates are shown for two different family types—those that have remained stable (non-split-offs) and those in which family members have split off from the basic family unit to form a new family unit between waves.