Publication Type: Paper/Poster Proposal
Abstract: Response rates for random digit dial surveys have been falling over recent years. Recent Pew studies (Pew Research Center 2004) have found that national surveys with response rates as low as 27 percent can be as representative as surveys with 51 percent response rates on opinion, civic engagement and attitude items. These studies have pointed to a non-response mechanism that meets the criteria of “missing at random” as opposed to “missing completely at random” (Little and Rubin 1987). We examine whether this holds for health insurance and health care access variables from statewide surveys. Low response rates may lead to biased estimates of state health insurance coverage and access. We examine two recent surveys conducted by the University of Minnesota for the states of Oklahoma (n=5,847, AAPOR response rate #4=45%) and Minnesota (n=13,512, AAPOR response rate #4=56%). Using these data we estimate the probability of being uninsured, having different types of insurance coverage, and lacking access to care by whether the household refused to participate during a previous call, and whether the household took 5 or more days to be completed. Although certain demographic characteristics varied significantly between the two groups such as age (showing the data were not “missing completely at random”), there are no statistically significant differences in multivariate models predicting key health access and health insurance coverage estimates controlling for the demographic differences (i.e., our data meet the criteria for “missing at random.”). Not including the initial refusals and surveys completed after 5 days would result in response rates that are half of the actual rates but would not affect the quality of our estimates after imposing weighting controls for demographic variables. Thus we should consider developing additional summary measures of survey quality that are related to the estimates generated from the survey.