Evaluation of a Procedure Based on Interruptions in Telephone Service for Reducing Coverage Bias in RDD Surveys

It is well known that telephone surveys are subject to coverage bias because of noncoverage of nontelephone households. Although only around 5% of the households in the U.S. do not have a telephone, rates of telephone coverage show substantial variation by geography, and demographic and socioeconomic factors. In particular, lack of telephone service is more common among households that contain ethnic and racial minorities or that have lower socioeconomic status. There is some evidence to suggest that the percent of households without telephone service is higher in households with children below 3 years than other households. The National Immunization Survey (NIS) is a random-digitdialing (RDD) survey of telephone households with children between 19 and 35 months that is used to estimate vaccination coverage rates for 78 Immunization Action Plan (IAP) areas, consisting of the 50 states, the District of Columbia and 27 metropolitan areas. These rates have to be adjusted for noncoverage of nontelephone households. One method of adjustment is a widely-used estimation technique known as simple poststratification which involves adjusting the sampling weights of children with completed interviews such that the sum of the weights agrees with known population control totals by child’s age, mother’s race/ethnicity and mother’s education. This adjustment partially reduces the bias. A drawback of this method is that it assumes that the percentage of children vaccinated within each poststratification cell is the same for telephone and nontelephone households. But, there is evidence to suggest that telephone status is related to vaccination status. Therefore, in the NIS this adjustment is currently done through a method known as modified poststratification which aims to reduce noncoverage bias to a greater degree than is possible with simple poststratification. For a detailed description of the NIS methodology please see Smith et al. (2001) and Zell et al. (2000). The basic idea in modified poststratification is to split each poststratification cell into two subcells; number of children who are up-to-date and the number who are not up-to-date for 4:3:1:3 vaccination series which is a combination of 4 or more doses of DTP, 3 or more doses of polio virus, 1 dose of measles-containing vaccine (MCV) and 3 doses of Hib. An assumption underlying this method is that the ratio of vaccination coverage rate for 4:3:1:3 series for children in nontelephone households to telephone households in the NIS is the same as the ratio of rates for nontelephone to telephone households in the National Immunization Provider Record Check Study (NIPRCS) in the National Health Interview Survey (NHIS). For a description of NIPRCS, please see Peak and Cadell (1996). Since NHIS/ NIPRCS covers both telephone and nontelephone households, this ratio is known. The unknown coverage rate for nontelephone households in the NIS is estimated by multiplying ratio of 4:3:1:3 vaccination coverage rates in telephone households by the ratio of 4:3:1:3 vaccination coverage rates for children in nontelephone to telephone households in the NHIS/NIPRCS. In this method, 4:3:1:3 up-to-date rate serves as a proxy for up-to-date rates for other vaccines. Further details of the modified poststratification method can be found in Battaglia et al. (1995). Over the last few years, we have been investigating another method of adjusting for noncoverage of nontelephone households in the NIS. This method is based on the data obtained from telephone households in the NIS that had interruptions in their telephone service during the year preceding the survey period. This method is based on the observation by Keeter (1995), that at any given time, the population of telephone households includes households that have had interruptions in telephone service. This means that if the survey had been conducted at some previous point in time, the households that reported interruptions would be considered as part of the population of nontelephone households if they had interruptions at that time. Essentially, we think of the population of households in scope for the survey as consisting of four subpopulations or groups as shown in Table 1. Group T/NI contains persons from households with telephone service (T) at the time of the survey and no interruption in telephone service (NI) during the entire year. Group T/I contains persons from households with telephone service at the time of the survey but interruptions in telephone service (I) during the year. Group NT/I consists of persons from households that did not have telephone service (NT) at the time of the survey but had telephone service at some time during the previous year and finally, Group NT/NI contains persons from households with no telephone service during the entire year. The last group is considered as permanent nontelephone households with no chance of being covered by the NIS. Some of the households in the group NT/I would have had a chance of selection, if the NIS had been conducted at some previous point in time. Previous studies (Keeter, 1995) have shown that the households with interruptions in telephone service are very similar to nontelephone households at the time of the survey. Therefore, these households may be used to represent the nontelephone households.