<論文>スノーボール・サンプルによる比率の推定

Abstract

Non probability sampling is often employed in many disciplines including sociology. Despite this, their sampling errors are understudied. We calculated errores from a type of non probability sampling: snowball sampling. After generalizing the standard error for non probability sampling, we did monte carlo simulations that five to fifty respondents are snowball-sampled from artificial populations where error correlations exist between friends. Their results show (1) ESS decreases as its sample size increases, (2) the starting respondent of snowball sampling does not affect ESS, and (3) ESS from simple estimates of rate with binomial distribution is smaller than estimates assuming error correlations between referring and referred respondents, especially when sample size is small.