All Times ET
Program is Subject to Change
The Effects of Interview Length and Rest Period on Response In Establishment Surveys (308430)*Joseph Rodhouse, National Agricultural Statistics Service
Heather Ridolfo, National Agricultural Statistics Service
Pamela McGovern, National Agricultural Statistics Service
Zachary Terner, National Institute of Statistical Sciences
Keywords: Interview Length, Rest Period, Response Propensity, Burden
Research has shown that a sample unit’s likelihood to respond can be predicted using the estimated burden of the previous survey they were sampled in, and the length of the rest period between the previous survey and the current one. Using the Office of Management and Budget’s (OMB) Burden Hours Estimates (BHE) to distinguish two surveys as being higher burden and lower burden, sample units from the higher burden survey (BHE = 70 minutes) were more likely to respond to a subsequent survey (BHE = 20 minutes) than the sample units from the lower burden survey (BHE = 20 minutes). Some questions remain, however, about this phenomenon. This paper builds on previous research by examining the role of actual interview length experienced in a previous survey on the likelihood to respond to a subsequent survey. We also examine how the likelihood to respond to the subsequent survey is impacted by the rest period in between the two surveys. Rest periods are strategic tools many organizations use to try to reduce burden for those in the survey population. However, the evidence regarding the significance of rest periods on response likelihoods are mixed, and to date, have not accounted for actual interview lengths experienced by respondents when evaluating the impact of rest periods. To carry out this research, we use data from the National Agricultural Statistics Service, linking respondent data together from two surveys: the Agricultural Resource Management Survey – Phase 3 (ARMS 3) and the June Acreage and Production Survey (June APS). Interview length of the prior survey (ARMS 3) is calculated using paradata from web interviews (CAWI) and field enumerated in-person interviews. We use variables for interview length, rest period, mode, and farm characteristics (i.e., size and type) in a logistic regression model to predict response likelihood to the June APS. The findings may inform nonresponse weighting adjustments and delineate subgroups for adaptive data collection designs.