Implementation of weighting system for mixed-mode social surveys

点击翻译按钮获取中文摘要

The purpose of the present study was to propose a method for making a weight variable to adjust for the covariate distribution when mixed-mode surveys were administered simultaneously on different subsets of population. We also apply our method to real survey data to discuss its utility. Specifically we pay attention to comparison of survey by mail or face-to-face interview administered to probability samples and web survey administered on non-probability samples. We use the latter to approximate the former. The core of our method is estimating the propensity score, based on the framework of covariate adjustment.As the result of our study, we found that in approximating the results of surveys on probability sample by results of survey done on non-probability samples by different mode, choosing good sets of covariate is very important. We proposed a four-step procedure of variable selection for better covariate adjustment. In addition, we also emphasized that obtaining results of high quality in web survey itself is another key for success in that approximation. We discussed various aspects of quality control in web survey administration. Based on these results, we will further investigate into the task of appropriately integrate the results of mixed-mode surveys into single estimates of population parameter.