120 Chapter 4 Being eligible for supported housing increased total care spending with 20,017 euros (se: 5,006) in the following calendar year (average spending among the non-eligible of 17,578 euros). This is mostly due to the higher costs of supported housing (11,883 euros, se: 2,634) but also due to higher mental health care expenditure, for which the coefficient is less precisely estimated but large in magnitude (7,698 euros, se: 4,221). Findings on mental health care expenditure reported separately for the outpatient and inpatient settings suggest a higher consumption of outpatient care (1,213 euros, se: 643) by those eligible. Income from work (unconditional on working) is 2,149 euros (se: 717) lower, which is large compared to the non-eligible average of 3,904 euros. This is in line with results suggesting lower likelihood of in being in paid employment by 7.0 pp (se: 3.6), which corresponds to a 25% reduction compared to the non-eligible group (-7.0/28.0). Total personal income decreased by an average of 1,470 euros (se: 748, average among noneligible 15,458 euros) indicating that the income loss from work was partly compensated by other sources of income. (Income from) paid employment captures regular (competitive) jobs and sheltered employment which offers individuals with a disability a protected environment with extra guidance, and employees get paid at least the minimum wage. After 4 years, labour related effects are consistently more negative in absolute terms (Table 4). Table 4. Two-stage least squares estimates for the effects of being eligible for supported housing admission in the long-run: 4 calendar years after application 4-year outcomes All-cause mortality Working Income from work (€) Personal Income (€) Eligibility 0.050 -0.075 -3,410* -2,017*** (0.033) (0.055) (1,488) (747) First-stage 0.978*** 0.996*** 0.996*** 0.996*** (0.048) (0.049) (0.049) (0.049) Observations 7,953 7,218§ 7,218§ 7,218§ Mean dependent variable non-eligible 0.06 0.33 7,260 18,438 Notes: Robust standard errors in parentheses, clustered at the assessor level; *** p<0.01, ** p<0.05, * p<0.1. All regressions include the main specification controls described in section Empirical implementation. §Fewer observations due to missing data in the health insurance claims and tax returns databases. The analysis was replicated for the smallest sample available for all outcomes (N=7,218) and displayed similar results. The 2SLS results in Table 3 provide consistent estimates of the average treatment effects for the group of compliers (LATE). Ordinary-least squares estimates would, in theory, capture the average treatment effects for the population, but are very likely biased (Tables S6 of the appendix). A better understanding of the LATE can be achieved with further characterization of the compliers. The compliers comprise 41% of the population and the remaining individuals are always-takers. The absence of never-takers possibly reflects
RkJQdWJsaXNoZXIy MjY0ODMw