Objective: In the situation that randomization is not avaliable, to minimize the biasness in treatment arm assignments, the use of propensity score weighting method and the assessment of performances related to results obtained from generalized boosted and multinomial logistic regression (MLR) of propensity score weighting are aimed. Method: Results obtained from MLR and GBM are to compare with the help of a simulation study. In simulation study, data with n=500, 1000, 2000 sample size will be derived using 1000 repetitions on seven scenarios with three categorized treatment group, continuous outcome variable and continuous/binary covariates. The propensity weights will be found with the help of Propensity scores obtained from MLR and GBM and using these weights, the balance will be assessed using balance metrics with average treatment effect estimation (ATE). In study, Â“twangÂ” package in R program is used. Results: As the number of samples increases, the balance values decreases more, so it seems that the biasness has fallen. As the scenarios become more complex, GBM produces better balance results. There are better results for MLR at main effect model. Trimming or removing excess weights ensures improving of balance.
Propensity score weighting;GBM;Multinomial Logistic Regression