Performance evaluation of some propensity score matching methods by using binary logistic regression model

OLMUŞ H., Bespinar E., Nazman E.

COMMUNICATIONS IN STATISTICS-SIMULATION AND COMPUTATION, 2019 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Basım Tarihi: 2019
  • Doi Numarası: 10.1080/03610918.2019.1679181
  • Dergi Adı: COMMUNICATIONS IN STATISTICS-SIMULATION AND COMPUTATION
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus
  • Anahtar Kelimeler: Binary logistic regression model, Matching, Propensity score, Standardized bias, Percent reduction bias, BIAS, SUBCLASSIFICATION
  • Gazi Üniversitesi Adresli: Evet

Özet

The unit selection bias in treatment and control group affects negatively in evaluation of treatment effects. In some studies, the random units selected for the treatment and control group are out of control of the researcher and there may be differences between the units under consideration. This will cause the estimates obtained to be biased. Propensity score matching (PSM) has been used to reduce bias in estimation of treatment effect in observational data. Therefore, nearest neighbor (1:1), caliper, stratification, mahalanobis metric, full and combining propensity score and mahalanobis metric matching, which have been widely used as PSM methods, were compared in terms of the correct classification rates conducting a detailed Monte Carlo simulation study. In addition, standardized and percent reduction bias of covariates were evaluated for each of the PSM methods. It is suggested that stratification and full matching methods should be considered to study with high correct classification rate whereas caliper matching method should be prefered due to the low bias to make statistical inferences.