OBJECTIVES: Endogeneity is a common problem in retrospective claims data studies because patients in claims data were not assigned to treatment by randomization. Propensity score matching (PSM), instrumental variables (IV), and difference-in-difference (DID) have been used to control for selection bias in evaluating the impact of treatment on outcome measures. This study compares the estimated incremental costs between typical and atypical antipsychotic medication users in patients with schizophrenia using these three methods. METHODS: Patients ages 18-64 years old with at least one prescription of typical or atypical antipsychotic medication and at least one diagnosis claim of schizophrenia (ICD-9-CM diagnosis 295.xx) within 90 days of the antipsychotic medication were identified in MarketScan® Multi-State Medicaid Database 2002–2009. The index date was the first prescription date of antipsychotic medication. All patients had ≥12-month continuous enrollment prior to and post the index date. Outcomes were total all-cause expenditures and psychiatric-related expenditures during the 12-month follow up.  The incremental costs associated with the use of typical antipsychotic medication versus atypical medication was estimated using six regression models–three estimation methods (GLM, IV, and DID) on two samples (matched and non-matched sample). IV was prescribing physicians’ characteristics. RESULTS: A total of 447 atypical and 4544 typical antipsychotic users met all study criteria, with a mean age of 38.9.  Five out of the six models (with the exception of the DID model on the unmatched sample) did not find a significant relationship between types of antipsychotic medications and total costs, and five out of the six models (with the exception of the IV model on the full unmatched sample) found atypical medication users had significantly higher psychiatric-related costs than typical medication users. CONCLUSIONS:   The PS-based approach combined with the DID or IV methods may be better than each approach alone.