Criterion Variables When using binary Logistic Regression all the criterion (dependent) variables are dichoto- mous at the nominal level of measurement. The first model used participants entering treat- ment following SAL services (0 = No, 1 = Yes) as its criterion variable. The second model used having children removed from their home while participating in SAL services (0 = No, 1 = Yes) as its criterion variable. The third model used having children reunified while par- ticipating in SAL services (0 = No, 1 = Yes) as its criterion variable. All these variables were based on data collected by the evaluation team.
Control Variables Control variables also are known as covariates. Participant level control variables included
scores from SOCRATES taking steps scales from the alcohol and drug forms at pretest and during the last meeting with SALs5, and
whether parents needed inpatient treatment (0 = No, 1 = Yes), based on American Society of Addiction Medicine (ASAM, 2001) criteria6 . Data were available to include one system level control variable of any prior substantiated CPS referrals related to participants at any time preceding PTP services7 (0 = No, 1 = Yes). SOCRATES variables were entered first in a single block, followed by a block with only the ASAM inpatient determination, and a block with only the CPS variable.
Predictor Variables For the first model, predicting professional AOD treatment entry, the sole predictor variable was participation intensity in SAL services. This variable was created through two steps; (1) the number of service contacts with SALs was placed in ratio to the number of weeks that parents participated in services, (2) participants were placed into low, moderate, and high participation groups based on this ratio. Participants who had less than one weekly service contact were placed in the low intensity group (value = 1), those who had between one and three weekly contacts were placed in the moderate intensity group (value = 2), and those who had more than three weekly contacts were placed in the high intensity group (value = 3). This procedure resulted in an ordinal level variable that allowed for a meaningful interpretation of incremental odds ratios (Hosmer & Lemeshow, 1989). In the second and third models, pre- dicting child welfare outcomes, an additional variable of entering professional AOD treatment subsequent to SAL services8 was added. In all the models the single or two-predictor variables entered the model last in a single block.
Model Evaluation The criteria used to evaluate each model included overall improvements in classification, specific improvements in sensitivity and selectivity9 of classification, variance in the criterion variable explained by the model, and goodness of fit between the data and an ideal model. Additionally, each model was evaluated for the presence of outlying cases based on standard- ized residuals. These criteria allowed for a thorough evaluation of the quality of the models
(Hosmer & Lemeshow, 1989).
This variable was measured at the ordinal level. All the client level control variables were derived from the evaluation database specific to this project. This variable was derived from the Idaho Department of Health and Welfare database. This was the criterion variable from the first model. Sensitivity is the ability for a model to correctly classify cases as having the occurrence of the criterion variable, and selectivity is the ability for a model to correctly classify cases as having its non-occurrence.
Idaho Pre-Treatment Program