William Gnam

William Gnam

Dissertation Title:  "The Impact of Managed Behavioral Carve-Outs on Quality of Depression Care"

This thesis consists of three empirical papers pertaining to the economics of mental health. The first paper examines the impact of managed behavioral health care carve-outs on the quality of outpatient depression care. The dataset analyzed features claims files from four plans of a national health care insurance company that independently carved out mental health benefits during 1994-1995, and 3 additional comparison plans that did not carve out. Regression models, structural break searches, and difference-in-difference estimates suggest that implementation of the carve-outs was associated with a higher probability of receiving guideline-level outpatient depression care in 3 of 4 plans. These results suggest that managed behavioral carve-outs can lead to greater efficiency in the delivery of mental health services.

The second paper examines the effects of alcohol disorders on labor market supply and income, using data from a Canadian population-based household survey of mental disorders. Using single-equation and instrumental variables methods, and estimators robust to weakly correlated instruments, men with a lifetime history of a clinical alcohol disorder did not experience reductions in work hours or income, but did have reduced employment. Women with a lifetime history of alcohol disorders had increased income, although the instrumental variable estimates were inconclusive. These results do not support the view that alcohol disorders lead to decrements in income or work hours for men or women.

The third paper is an empirical analysis of the effect of a managed behavioral health carve-out on depression treatment choice, based upon data from a health plan that carved out mental health benefits in 1995. Depression treatment choice is specified as a multinomial discrete choice problem modeled by conditional logit, multinomial probit, and mixed logit. The analysis suggests that the higher probability of receiving an antidepressant drug following the carve-out is largely explained by temporal trends and a higher likelihood of receiving treatment from a network provider. The model results from the conditional logit specification are surprisingly similar to those of the mixed logit and multinomial probit, suggesting that the choice problem modeled here does not violate the Independence of Irrelevant Alternatives assumption.





Graduation Year


Dissertation Committee Member