Assessing Interviewer Observations in the NHIS

Face-to-face surveys provide interviewers the opportunity to gather information beyond the scope of the survey. These observational paradata may prove useful for streamlining the data collection process and enhancing post collection weighting adjustments, both improving the quality of survey estimates. However, interviewer observations can be error-prone especially when the observation requires a judgment based on limited information. This research uses survey paradata to evaluate new interviewer observations made during the collection of the 2013 National Health Interview Survey (NHIS). Multilevel models discern the variation in compliance with the new task attributable to interviewers, finding that the characteristics of the case, not the interviewer, reduced variation in compliance. We then examine variation in the observations by whether contact was made when recording the observations, finding inferential but not factual observations vary based on making contact during the observation. When interacted with contact, 11 of the 15 observations are predictive of either the number of contact attempts made on the case or the odds of refusal—two measures of level of effort.