|
Abstract:
|
Interviewer effects result from responses collected by the same interviewer being correlated, reducing the efficiency of survey estimates and decreasing effective sample sizes. While having received attention in early survey methods research, this well-known problem has been much less studied recently. This may be due to the fact that standard approaches for assessing interviewer effects requiring random assignment of cases to interviewers (interpenetrated sampling), which is often infeasible in many sample designs. In the absence of interpenetration, standard clustering methods may overestimate interviewer effects by confounding non-random assignments of respondents with measurement error introduced by interviewers. Here we propose two methods to assess interviewer effects in the absence of interpenetration: "anchoring," which uses variables assumed to have little or no interviewer measurement error to remove intra-interviewer variability due to assignment via joint distribution modeling assumptions; and "propensity adjustment," which uses weights developed from interviewer assignment propensity models to provide pseudo-interpenerated-design estimates.
|
