Table 4.

Logistic regression models showing predictors of having perpetrated or being a victim of a major crime against a person over 12·4 years among 27,585 patients with a history of psychiatric hospitalization^c.

Criterion Variable		Perpetrated at least one crime against a persona					Victim of at least one crime against a personb
Criterion Variable		bd	SE	P	Exp (b)e	95% CI	bd	SE	P	Exp (b)e	95% CI
Model 1: “CTO” and “Treatment Day Effect” variables with stepwise control for: **
Structural & Bias Indicators**	CTO	−.138	.04	.002	.871	.799–.949	−.091	.04	.040	.913	.837–.996
	Treatment days per community care episode	−.006	.00	<.001	.994	.991–.997	−.002	.00	.083	.998	.996–1.000
Plus Demographic, Diagnostic, and Service Indicators***	CTO	−.166	.05	<.001	.847	.772–.930	−.110	.05	.017	.896	.818–.981
	Treatment days per community care episode	−.008	.00	<.001	.992	.989–.995	−.003	.00	.013	.997	.995–.999
Plus Behavioral and Social Circumstance Indicators**	CTO	−.183	.05	<.001	.833	.757–.915	−.120	.05	.011	.887	.809–.972
	Treatment days per community care episode	−.008	.00	<.001	.993	.990–.995	−.002	.00	.040	.998	.996–.999
Model 2: “CTO-initiated re-hospitalization” and “Treatment Day Effect” Variables with stepwise control for: **
Structural & Bias Control Indicators**	CTO-initiated-re-hospitalization	.063	.05	.228	1.065	.961–1.179	.007	.05	.899	1.007	.907–1.118
	Treatment days per community care episode	−.006	.00	<.001	.994	.991–.997	−.002	.00	.085	.998	.996–1.00
Plus Demographic, Diagnostic, and Service Indicators**	CTO-initiated-re-hospitalization	−.066	.06	.235	.936	.839–1.044	−.089	.06	.110	.914	.820–1.020
	Treatment days per community care episode	−.008	.00	<.001	.992	.989–.995	−.003	.00	.015	.997	.995–.999
Plus Behavioral and Social Circumstance Indicators**	CTO-initiated-re-hospitalization	−.141	.06	.016	.868	.774–.974	−.183	.06	.002	.833	.743–.933
	Treatment days per community care episode	−.007	.00	<.001	.993	.990–.996	−.002	.00	.047	.998	.996–.999

^aModel 1: When crime against a person is the dependent variable and the “CTO effect” is the primary independent variable and all three groups of control variables are included in the model, χ² = 4737.18, df = 46,25,969, p < .001; N for regression = 26,015. Model 2: When crime against a person is the dependent variable and “CTO terminated with hospitalization” is the primary independent variable and all three groups of control variables are included in the model, χ² = 4728.59, df = 46, 25,969, p < .001; N for regression = 26,015.

^bModel 1: When victimization by a crime against a person is the dependent variable and the “CTO effect” is the primary independent variable and all three groups of control variables are included in the model, χ² = 3162.98, df = 46, 25,969, p < .001; N for regression = 26,015.

Model 2: When victimization by a crime against a person is the dependent variable and “CTO terminated with hospitalization” is the primary independent variable and all three groups of control variables are included in the model, χ² = 3166.44, df = 46, 25,969, p < .001; N for regression = 26,015.

^cMany “b”, and “SE” coefficients appear as 0.00 or 0.99 as they have not been extended to more than two decimal places in order to allow all of the data reporting in a single table. The zeros 0.00 or 0.99 refer to small and sometimes significant effects of the adjustment variables.

^d“b” is a multiple-partial slope indicating how much change can be expected for a unit change in a predictor variable associated with a unit change in the criterion variable, i.e. after all other predictors have been given causal priority in explaining the change.

^eExp(b) is an odds ratio providing an indication of relative risk. When below one it indicates a reduced risk of one minus the odds ratio, when above one it is an indicator of the percent of increased risk.

^**In addition to the two factors noted in the table, a propensity score for selection into the CTO cohort and involvement in crime/victimization prior to the study, the models adjust for three groups of control variables by adding the additional controls to the models in steps:

^1.Structural and potential bias indicators: The lowest (most disadvantaged) SEIFA rank of a neighborhood in which the person resided during the study period, Aboriginal or Torres Strait Island heritage, required an interpreter in tribunal hearings, not born in Australia or New Zealand, whether their preferred language was not English, and the days spent in prison during the study period.

^2.Demographic, Diagnostic, and Service Indicators: gender, age, unemployment status, age at entry to the mental health system, time in the study, diagnosis (schizophrenia, affective disorder, dementia, paranoia), duration of inpatient care, episode start year, < 11^th grade education.

^3.Indicators of behavior and circumstance: 24 HoNOS psychosocial assessments of the patient’s life situation, 12 completed at admission and 12 at discharge from inpatient care over the study period.