抄録
This paper presents the results of experimental telephone surveys conducted via two kinds of sampling methods; namely, telephone directory sampling (TD) and random digit dialing sampling (RDD). The TD sample is restricted to those who list their phone number in the directory. This restriction increases the proportion of the aged and the rural sample and causes results which are definitely favorably biased toward a conservative direction when compared with the results of the RDD which includes both the listed and the unlisted samples. Even if we compare TD with RDD limited to the listed respondents, there still remain a number of differences. In order to explain the discrepancies, we propose to divide each sample into two latent classes; one common class which would respond in both survey methods and one unique class which would not be obtained through the alternative method. The results given by a normal mixture model indicate the listed RDD respondents seem contain a moderate size of the unique class.
