125 The effects of supported housing for individuals with mental disorders an overidentified first-stage. The many weak instruments problem could be masked by the use of the leniency, which is a continuous instrument composed of the assessorlevel averages of the endogenous variable. Concerns are that, as the dimensionality of the underlying variation is not one but the number of assessor fixed effects, first-stage F-statistics might overstate the instrument strength [37]. To investigate this concern we estimate our results using UJIVE14, a 2SLS estimator that is “consistent for a convex combination of LATE under many instrument asymptotics (and) that also allow for many covariates and heterocedasticity” [36]. Results obtained using UJIVE confirm that the main results are unlikely strongly biased due to weak instruments (Table A11 and A12 of the Appendix). DISCUSSION This paper studies the impact of supported housing eligibility for people with mental disorders on care use, mortality, employment and income, and on spillovers to their parents. We report four main findings. First, compliers who are being assigned to a more lenient assessor are more likely granted eligibility to supported housing. This, in turn, increases the probability of an admission to supported housing and reduces the likelihood of getting home care. Second, public care spending increases mainly via higher expenditures on supported housing. Findings do also suggest increased total mental health care expenditure and specifically higher outpatient mental health care expenditure. One potential mechanism behind this finding could be that in more supervised environments individuals increase their adherence to medical appointments, psychotherapy or medication. The effect on inpatient mental care is not significant, suggesting that supported housing does not considerably prevent inpatient admissions. Finally, expenditures on other types of care do not change. Third, eligibility for supported housing decreases income from work and personal income, and likely the probability of having paid work. Fourth, both parents are more likely to work in the long-run, and these effects are stronger among parents living with their child before application. While the spillover effects are likely explained by the fact that paid work and informal care-giving are substitutes, the mechanisms behind lower employment participation and earnings are for the applicants are less obvious. Labour participation determinants for individuals with severe mental illness include personal factors such as self-efficacy and self-stigma, or work related factors such as employers stigma and discrimination [38]; which might impact supported housing clients differently 14 UJIVE was proposed by Kólesar M (2013) as an alternative to the jackknife instrumental variables estimator (JIVE) (Angrist, Imbens and Krueger 1999). UJIVE is similar to JIVE in that it “also uses a “leave-one-out” jackknife-type predictor of the treatment in the first-stage”, but it differs from the latter in the way it deals with the covariates as JIVE is not robust to a large number of covariates. The UJIVE partials out the effect of the covariates on the treatment, guaranteeing that the prediction error will be uncorrelated with the outcome. 4
RkJQdWJsaXNoZXIy MjY0ODMw