BACKGROUND: The California Medicaid program (Medi-Cal) added fluoxetine and paroxetine to its formulary in May 1996. This policy change created two types of selection bias that must be accounted for in comparisons across drugs. First, new patients may have sought treatment in response to the change (woodwork effect). Second, prescribing patterns may have also changed (drug selection bias). OBJECTIVES: This study used a Dual-? method to adjust jointly for these two selectivity biases in evaluating the effects of alternative antidepressant therapies on health care costs. METHODS: All patients with a recorded diagnosis of major depression (MDD) between 04/01/1994 and 01/31/1999 were selected from the Medi-Cal paid claims system. Results from the Dual-? method were compared with the standard Heckman-Lee method. RESULTS: The formulary policy change was a statistically significant factor in the woodwork effect and altered antidepressant selection patterns. In the second-stage cost models with double ?s, the biases associated with the woodwork effect were statistically significant (fluoxetine p<0.0055, paroxetine p<0.0041, and sertraline p=0.3261). None of antidepressant selection biases were significant. No statistical significance differences were found in total health care costs for patients initiating therapies with these three SSRIs as compared with TCAs. However, in models with a single ? to adjust only for treatment selection bias, the fluoxetine group had significant drug selection bias (p<0.0418) and significant high total health care cost as compared with TCAs (p<0.0076). CONCLUSION: Double-? method better resolved the selection biases in antidepressant therapy in an evaluation of the impact of alternative therapies on post-treatment costs following a formulary change than the Heckman-Lee method.