2012 年 27 巻 2 号 p. 104-117
Using two Japanese survey datasets sharing the same variables that were collected at about the same time during the election of the House of Councillors in 2007, we investigated the effectiveness of propensity score adjustment to internet survey data of voting behavior. One dataset was from an internet survey based on purposive sampling, and the other was from an in-person interview survey based on probabilistic random sampling. Setting party identification and actual votes as dependent variables, the covariates for calculating the propensity scores were selected on the basis of the Strongly Ignorable Treatment Assignment (SITA) condition. The results indicated that while the adjustments were effective in some cases (e.g. the votes for proportional representation seats), those for other variables were ineffective and the divergence from the probability sampling survey became even larger. Conditions in which propensity scores can effectively adjust internet survey data are discussed.