Search

Dual-frame telephone surveys that sample from both landline and cell-phone frames have become common because the US population is increasingly abandoning landline phones. Designing a dual-frame telephone survey requires a decision on the sample allocation, and ultimately the allocation of respondents, from each of the two frames. The allocation of the sample to the two frames influences the unequal weighting effect (UWE), and consequently, the precision of estimates calculated from the survey data. The sample allocation also influences cost, since obtaining cell-phone respondents is often more expensive than obtaining landline respondents. In this paper, we present results that designers of dual-frame studies can use to determine an allocation of responders that optimizes the precision of the survey estimates for a given cost. [ABSTRACT FROM AUTHOR]

Given the theoretical promise of these auxiliary data for overcoming the challenge of nonresponse, survey researchers have shown an increased interest in collecting paradata, data from screening interviews, and other contextual information. However, very few studies have systematically assessed the use of these data for post-survey nonresponse adjustments. Those studies that do exist generally do not identify auxiliary variables that correlate with response propensity and key survey variables, a necessary condition for auxiliary data to be effective tools for reducing nonresponse bias. Using the National Survey of Family Growth (NSFG), this paper leverages a large set of auxiliary variables available for the full NSFG sample to assess their potential as independent tools for post-survey nonresponse adjustments. We begin by using this auxiliary information to predict response propensity (RP) for each person in the full sample. We then display descriptive estimates for a variety of attitudes and behaviors measured in the NSFG, using post-stratification weighting adjustments as well as RP adjustments followed by post-stratification. The results show that accounting for RP in the weighting adjustment often produces noteworthy differences in the estimates, thus supporting the collection of these types of auxiliary variables in practice. These results also suggest that standard post-stratification adjustments may not be entirely effective at removing nonresponse bias from all survey estimates, and that some subgroup analyses may be especially subject to bias when adjusting survey estimates using post-stratification alone. [ABSTRACT FROM AUTHOR]