This paper provides a tutorial on using generalized boosted models (GBM) to estimate propensity scores for multiple treatments in non-randomized or observational studies. The authors aim to address the lack of practical guidance for multiple treatments, which is a common scenario in many research contexts. They define the causal quantities of interest, derive weighted estimators, and present a detailed plan for using GBM to estimate propensity scores and weights. The paper also includes tools for assessing balance and overlap of pretreatment variables among treatment groups. A case study on the effects of three treatment programs for adolescent substance abuse is used to demonstrate the methods. The authors highlight the advantages of using GBM over parametric models, such as automated variable selection and more stable weights, which can improve the accuracy and reliability of causal effect estimates.This paper provides a tutorial on using generalized boosted models (GBM) to estimate propensity scores for multiple treatments in non-randomized or observational studies. The authors aim to address the lack of practical guidance for multiple treatments, which is a common scenario in many research contexts. They define the causal quantities of interest, derive weighted estimators, and present a detailed plan for using GBM to estimate propensity scores and weights. The paper also includes tools for assessing balance and overlap of pretreatment variables among treatment groups. A case study on the effects of three treatment programs for adolescent substance abuse is used to demonstrate the methods. The authors highlight the advantages of using GBM over parametric models, such as automated variable selection and more stable weights, which can improve the accuracy and reliability of causal effect estimates.