Adverse Impacts of Mental Health Needs Assessment on Jail Outcomes

Notes: Each column represents a linear probability model with the binary outcome of interest being the mental health classification of moderate/severe as opposed to none/mild. Clustered (month–year) standard errors are in parentheses. ^*p < 0.10, ^**p < 0.05, ^***p < 0.01.

Notes: Each column represents a linear probability model with the binary outcome of interest being the mental health classification of moderate/severe as opposed to none/mild. Clustered (month–year) standard errors are in parentheses. ^*p < 0.10, ^**p < 0.05, ^***p < 0.01.

Notes: We report the first-stage results of a linear probability model stratified by age group. The binary outcome of interest is the initial assessment of an inmate’s mental health needs being either none/low or moderate/severe. The propensity to assign the most severe score is estimated using data from other cases assigned to the clinician following the procedure described in the text. Columns 1 and 2 limit the sample to transition age youth, whereas Columns 3 and 4 limit the sample to adults. Columns 1 and 3 show the results by controlling only for month–year fixed effects, whereas Columns 2 and 4 also include the inmate baseline controls as shown in Table 1. Each column gives the corresponding clinician and inmate robust two-way clustered standard errors in parentheses. Robust (Kleibergen–Paap) first-stage F-statistic is reported. Note this is equivalent to the effective F-statistic of Montiel Olea and Pflueger (2013) in this case of a single instrument. ^*p < 0.10, ^**p < 0.05, ^***p < 0.01.

Notes: We report the first stage results of a linear probability model stratified by age group. The binary outcome of interest is the initial assessment of an inmate’s mental health needs being either none/low or moderate/severe. The propensity to assign the most severe score is estimated using data from other cases assigned to the clinician following the procedure described in the text. Columns 1 and 2 limit the sample to transition age youth, whereas Columns 3 and 4 limit the sample to adults. Columns 1 and 3 show the results by controlling only for month–year fixed effects, whereas Columns 2 and 4 also include the inmate baseline controls as shown in Table 1. Each column gives the corresponding clinician and inmate robust two-way clustered standard errors in parentheses. Robust (Kleibergen–Paap) first-stage F-statistic is reported. Note this is equivalent to the effective F-statistic of Olea and Pflueger (2013) in this case of a single instrument. ^*p < 0.10, ^**p < 0.05, ^***p < 0.01.

Notes: These linear probability models control for the baseline characteristics used in the instrumental variables analyses. The binary dependent variable in Columns 1 and 3 is being assigned a moderate to severe mental illness score at initial assessment. The dependent variable in Columns 2 and 4 is the propensity to assign a high or low score to inmates. Time fixed effects include month–year fixed effects. Clinician and inmate two-way clustered standard errors shown in parentheses. ^*p < 0.10, ^**p < 0.05, ^***p < 0.01.

Notes: These linear probability models control for the baseline characteristics used in the instrumental variables analyses. The binary dependent variable in Columns 1 and 3 is being assigned a moderate to severe mental illness score at initial assessment. The dependent variable in Columns 2 and 4 is the propensity to assign a high or low score to inmates. Time fixed effects include month–year fixed effects. Clinician and inmate two-way clustered standard errors shown in parentheses. ^*p < 0.10, ^**p < 0.05, ^***p < 0.01.

Notes: Data are from a large county correctional complex. Time fixed effects include month–year fixed effects. Clinician and inmate two-way clustered standard errors shown in parentheses. ^*p < 0.10, ^**p < 0.05, ^***p < 0.01.

Notes: Data are from a large county correctional complex. Time fixed effects include month–year fixed effects. Clinician and inmate two-way clustered standard errors shown in parentheses. ^*p < 0.10, ^**p < 0.05, ^***p < 0.01.

Notes: This table presents a test for satisfying average monotonicity as proposed in Frandsen, Lefgren, and Leslie (2020), where they show that average monotonicity can suffice in lieu of strict monotonicity if the average treatment propensities move in the same direction as their potential treatment decisions. In the context of our paper, we can relax strict monotonicity for any given clinician if the individual monotonically complies with enough other judges. Thus, the coefficients should all be significant and the same direction to support average monotonicity.

Notes: This table presents a test for satisfying average monotonicity as proposed in Frandsen, Lefgren, and Leslie (2020), where they show that average monotonicity can suffice in lieu of strict monotonicity if the average treatment propensities move in the same direction as their potential treatment decisions. In the context of our paper, we can relax strict monotonicity for any given clinician if the individual monotonically complies with enough other judges. Thus, the coefficients should all be significant and the same direction to support average monotonicity.

Notes: This table reports the ordinary least squares (OLS), two-stage least squares (2SLS), and the instrumental variables LASSO (IVLASSO) estimates of the impact of being assigned a moderate/severe mental health needs rating. The dependent variable is listed in each row. The recidivism within one year sample is truncated to one year prior to the last date our data set. Similarly, the recidivism within 18 months and two years samples are truncated to 18 months and two years prior to the last date. The 2SLS and IVLASSO specifications instrument for assignation of a high mental health needs score using a clinician leniency measure that is estimated using data from other cases assigned to a clinician as described in the text. We include month–year fixed effects and baseline controls for all specifications. The clinician and inmate robust two-way clustered standard errors are shown in parentheses. For the 2SLS estimates, confidence intervals based on the inversion of the Anderson–Rubin test are shown in brackets. ^*p < 0.10, ^**p < 0.05, ^***p < 0.01.