Balancing Cost and Mean Squared Error in Rdd Telephone Surveys: the National Immunization Survey

1. Introduction Nonresponse is a major problem in random-digit-dialing (RDD) surveys. Nonresponse can occur at various phases of data collection, especially if the eligible sample is identified through the selection of a larger screening sample. For example, if we are interested in obtaining data from households containing children in a specific age group, these households will have to be first identified by screening a larger sample of households. The larger sample of households is contacted through calling an even larger sample of telephone numbers. It is well known that a large proportion of telephone numbers is nonresidential or nonworking numbers. Therefore, when there is nonresponse to the initial call attempts to identify whether a selected number is residential, nonworking or nonresidential (e.g., business), there is a possibility that the number is residential but we have failed to identify the number as such (e.g., no answer to all call attempts). The ratio of the number of telephone numbers resolved as either residential, nonresidential or nonworking to the total in the sample is called the resolution rate. Similarly, when a sampled number is identified as residential, we may fail to establish whether the household is eligible for the survey. The ratio of the number of households deemed eligible or ineligible to the number of known households in the sample is called the screener response rate. Finally, there is the usual problem of nonresponse, which occurs in all surveys, when we fail to get data from a household that has been identified as eligible for the survey. The ratio of the number of eligible households that respond to the total number of eligible households is the interview response rate. The overall response rate for the survey is the product of these three rates (Ezzati-Rice et al., 2001). Therefore, it is essential that in RDD surveys we try to achieve a high response rate at each phase so that the overall response rate is reasonable. The response rates, in addition to determining the number of completed interviews on which the survey estimates are based, also determine the sampling weights that are attached to the respondents for obtaining population-based estimates.