116 Chapter 4 mid-age and late-age adults) [15, 34]. Our results show that assessors are stricter with a specific case type (i.e. the excluded subgroup) if they are also stricter with all other case types (Table 2, column 2)9. Table 2. First-stage coefficients of residualised leniency for the entire population and by subgroup. Full population (main) (1) Reverse-sample (2) Relative likelihood§ (3) N (4) coef.* (se) coef.** (se) All population 0.978 (0.048) n.a. n.a. 1.00 7,953 Male 0.992 (0.050) 0.478 (0.075) 1.01 5,612 Female 0.935 (0.093) 0.682 (0.116) 0.95 2,341 Young Adults (18-30) 1.152 (0.074) 0.848 (0.114) 1.18 3,250 Mid-age adults (31-50) 0.964 (0.074) 0.662 (0.104) 0.99 2,898 Late-age adults (51-79) 0.689 (0.107) 0.402 (0.093) 0.70 1,557 Dutch background 0.982 (0.056) 0.412 (0.065) 1.00 5,784 Western migration background 0.880 (0.157) 0.671 (0.174) 0.89 768 Non-western migration background 0.968 (0.150) 0.666 (0.152) 0.98 1,401 Living alone 0.968 (0.072) 0.574 (0.092) 0.98 3,619 Living with partner 0.794 (0.131) 0.599 (0.128) 0.81 560 Living with parents 1.088 (0.099) 0.815 (0.132) 1.11 1,825 Living in an institution 0.863 (0.105) 0.515 (0.116) 0.88 1,588 Lowest quintile of income 1.111 (0.080) 0.934 (0.121) 1.13 1,595 Highest quintile of income 0.887 (0.099) 0.614 (0.104) 0.90 1,585 Previous work 1.135 (0.106) 0.119 (0.034) 1.16 2,027 Previous use of specialist mental health care 0.949 (0.056) 0.445 (0.083) 0.97 5,291 Previous use of home care 0.726 (0.101) 0.548 (0.101) 0.74 1,213 Lowest quintile of heath care expenditure 1.092 (0.113) 0.852 (0.125) 1.11 1,491 Highest quintile of health care expenditure 0.798 (0.121) 0.527 (0.131) 0.81 1,848 Substance use disorder 1.101 (0.102) 0.800 (0.124) 1.12 1,874 Psychotic disorder 0.570 (0.080) 0.391 (0.076) 0.58 1,531 Disorder diagnosed in the childhood 0.851 (0.128) 0.647 (0.149) 0.87 1,064 Mood disorder 0.902 (0.159) 0.693 (0.189) 0.92 663 Personality disorder 1.270 (0.167) 1.102 (0.192) 1.29 641 9 Monotonicity at the individual-level (strict monotonicity) is a requirement to estimate marginal treatment effects (MTEs). We use the test of Frandsen et al., (2019), that jointly checks for the exclusion and strict monotonicity assumptions. As suggested by the authors, and given that our institutional context supports excludability, we interpret Frandsen’s test as a test for strict monotonicity only. The estimated test statistics indicate a violation of strict monotonicity (Table A5 of the appendix). This prevents estimation of MTEs, but has no implications on the average monotonicity condition, under which the instrumental variables estimator converges to a proper weighted average of the LATEs.
RkJQdWJsaXNoZXIy MjY0ODMw