Assessing the Context of Permanency and Reunification in the Foster Care System. 4.1 Best-Fitting Regression Model

12/01/2001

To examine interaction effects among the stronger predictors, many regression models (such as race by entry age, race by services, race by job skills, race by substance abuse problem, race by kinship placement, substance abuse by services, etc.) were run that included various combinations of those variables. However, none of the interaction terms maintained significance in combination with the remaining variables. In some cases, the addition of interaction terms reduced the size of the R2 . Thus, it was decided to focus on regression models that excluded interactions.

In order to determine the best-fitting logistic regression model, the top-ranking eight variables were entered in a stepwise sequence from stronger to weaker predictors. The resulting eight models are presented in Table 5. The top-ranking five variables (race, age at entry, job skills, caretaker services, substance abuse problem) are entered sequentially in Models 1-5. As each variable is entered, the others continue to be significant predictors of reunification. In Model 5, their combined effects accounted for a 20.5 percent variance (R2 ) in reunification with an F-value of 15.45.

However, when neglect was added in Model 6, it was not significantly related to reunification, while four of the five other predictors continued to be significant predictors. Similarly, when kin caretaker is added in Model 7, it no longer was significantly related to reunification, while all five of the other variables remained significant predictors. It is interesting that the strong correlation between race and parental reunification was not reduced when kinship placement was added to the model. Finally, when SES was added in Model 8, it also was no longer significantly related to reunification, while four of the five other variables continue to be significant predictors. Nor does the addition of SES reduce the influence of race. on reunification. Since time in foster care, abuse, and caretaker's education were found to be important predictors in past studies, additional regression models were run separately for each of these three variables. Moreover, a broader definition of neglect (including physical, emotional, medical, abandonment, lack of supervision and failure to thrive) was also included in a separate regression model. However, none of these additional variables continued to be significantly related to reunification when combined with the other five predictors. Consequently, Model 5 was considered to be the best-fitting model for predicting parental reunification, since it was the only model in which all of the predictors were able to maintain a significant relationship with reunification in combination with one another.

Table 5.
Steps To Obtain Best-Fitting Model For Predicting Reunification

Predictors

Model 1

Model 2

Model 3

Model 4

Intercept

-69 (.00)*

-.32 (.07)

-1.05 (.00)

-.55 (.06)

Race (black=1)

-1.61 (.00)*

-1.54 (.00)*

-1.50 (.00)*

-1.57 (.00)*

No Services

-----

-1.06 (.00)*

-1.07 (.00)*

-.97 (.01)*

Age at Entry

-----

-----

.09 (.00)*

.08 (.00)*

No Job Skills

-----

-----

-----

-1.40 (.00)*

(100)R 2 =

9.2%

12.7%

16.8%

20.1%

F-Value =

22.8 (.00)*

28.12 (.00)*

21.28 (.00)*

15.78 (.00)*

*Statistically significant relationships at p=< .05


Table 5. 
Steps To Obtain Best-Fitting Model For Predicting Reunification 
(Continued)

Predictors

Model 5

Model 6

Model 7

Model 8

Intercept

-.33 (.33) -.18 (.69) -.22 (.64) -.33 (.53)

Race (black = 1)

-1.46 (.00)* -1.47 (.00)* -1.49 (.00)* -.93 (.05)*

No Services

-1.04 (.00)* -1.04 (.00) -1.08 (.01) -.99 (.01)

Age at Entry

.08 (.01)* .08 (.02)* .09 (.02)* .11 (.01)*

No Job Skills

-1.36 (.00)* -1.36 (.00)* -1.36 (.00)* -1.07 (.01)

Substance Abuse Problem

-.48 (.04)* -.46 (.10) -.46 (.00) -.04 (.92)

Neglect Allegations

----- -.33 (.39) -.27 (.47) .04 (.93)

Kin Caretaker

----- ----- .10 (.79) .18 (.66)

Low SES

----- ----- ----- -.78 (.09)

(100)R2 =

20.5% 22.4% 22.1% 19.8%

F-Value =

15.45 (.00)* 18.77 (.00)* 14.77 (.00)* 7.08 (.00)*

*Statistically significant relationships at p=< .05