Online Program

Voted “least likely to respond”: Using classification trees to identify likely non-respondents and proactively manage data collection in NASS’s Quarterly Agricultural Survey

Keywords: predictive models, nonresponse, propensity scores, classification trees

NASS has recently developed predictive models to identify survey non-respondents prior to data collection. Classification tree models based on several classes of auxiliary data were used to rank sampled establishments based on their likelihood to be either refusals or non-contacts. These scores were provided to local NASS field offices for use in managing their data collection activities for several quarters of NASS’s Crops/Stocks Survey. An initial evaluation to determine the usefulness of these scores will be discussed, but this proved difficult as each field office used unique strategies to employ the scores. How model scores worked in the field will be discussed and the paper will conclude with a summary of our efforts to standardize the use of the non-response scores. Finally, considerations unique to using them in establishment surveys will be discussed.