OBJECTIVE: To determine differences in statistical inferences when evaluating drug therapy and other interventions with typical left censored (nonnegative) health care cost data and different econometric models. METHODS: One of the common problems confronting health economists is the censored (nonnegative) nature of health care cost data. Using simulated health care utilization survey data we examined three alternative econometric methods to model the effect of a therapeutic intervention on patient ambulatory costs. The models included a simple Tobit model, a two-part model, and a Heckman two-stage model. RESULTS: Results from the simple Tobit suggested that the usage of institutional care had a significant negative effect on outpatient costs (-672, p=0.011). The results from a two-part model further found that the negative effect was because patients who used institutional care had lower likelihood of using outpatient care (-1.14, p=0.0025). Neither model found a significant coefficient for the therapeutic intervention. A Heckman two-stage model detected even more significant cost determinants than the other two models. For example, it showed that Medicaid eligibility had a significant negative effect (-1.40, p=0.0014) on patient ambulatory cost, and that the therapeutic intervention had a significant negative effect (-1.03, p=0.021) also. CONCLUSION: In evaluation of patient ambulatory costs, the Heckman two-stage model detected more explanatory variables with significant effects than the simple Tobit model and two-stage model. In contrast to the simpler econometric models, the Heckman two-stage model found that Medicaid eligibility and the therapeutic intervention was associated with reduced outpatient costs. This comparative evaluation underscores the value of using appropriate econometric methods when estimating the effects of drug therapy or other medical interventions on treatment costs.