서열종속변수모형의 통상최소제곱(OLS) 추정은 적절한가?

초록

The theoretical debate on the OLS estimation of the ordinal dependent variable model(ODVM) is still controversial, and in practice, researchers have used one of the OLS or the ordinal Logit/Probit to estimate the ODVM. In light of this theoretical and practical situation, we review the theoretical debate on the issue, investigate the practice of domestic empirical studies, and examine the appropriateness of the OLS estimator by comparing the results of the OLS with those of the ordinal Logit using the 2017 Social Integration Survey(SIS). We construct the ODVM which regresses the satisfaction with the political situation(SPS) on the female dummy(FD) and the trust in government(TIG) measured by the Likert summated scale. We draw 36 samples from the 2017 SIS, based on the sample size(6), the number of points of SPS(3), and the normality of SPS(2) which are key issues in the theoretical debate. The bias of OLS estimator is examined by estimating the regression model which relates the p-values of FD and TIG and their absolute differences between OLS and ordinal Logit to the sample size, the number of points and the normality of the SPS. The results show that the p-value is affected by the sample size and the normality, and the absolute p-value difference decreases as the sample size increases. These results imply that the bias of the OLS estimator of the ODVM is mainly caused by the sample size.

키워드