Weighting mixed mode data for the 2015 Residential Energy Consumption Survey (RECS)
Chen, P., Deng, S., Sathe, N., Dai, L., Harter, R., & Kott, P. (2017). Weighting mixed mode data for the 2015 Residential Energy Consumption Survey (RECS). In Proceedings of the 2017 Joint Statistical Meetings (pp. 393-401). American Statistical Association. http://www.asasrms.org/Proceedings/y2017/files/593809.pdf
The 2015 Residential Energy Consumption Survey (RECS) was a stratified multistage cluster survey of housing units (HUs). RECS was designed for computer-assisted personal interviewing (CAPI) as the method of data collection. Because of difficulties experienced in the field, CAPI data collection was terminated and replaced with a Web/Mail data collection protocol. Nonrespondents and unfinished cases from CAPI were transferred to Web/Mail, and HUs in reserve replicates of sample were released to Web/Mail. This change imposed a challenge for weighting the combined CAPI and Web/Mail data. In this paper we discuss the weighting class method to adjust for bad addresses and drop points, a latent-variable technique to predict the probability of an address corresponding to an occupied HU, and logistic regression models to estimate the probability of a HU being a primary residence. We used a calibration method to adjust for unit nonresponse and to poststratify the nonresponse-adjusted weights to the estimated number of occupied HUs from the 2015 American Community Survey for specified HU characteristics.