Assessing the Context of Permanency and Reunification in the Foster Care System. 3.1.3 Individual State Models of Reunification and Family Exits

12/01/2001

Tables IIC.3 and IIC.4 report results obtained by applying the multivariate logistic regression techniques to each state separately. In Tables IIC.3 (reunification models) and IIC.4(family exit models), each column presents the odds ratios estimated by fitting an independent logistic regression model for a separate state. The model from the nine-state analysis is modified in two ways for use with state data. First, the variable for state is removed from the model. Second, a separate category for kinship care is added to the placement type variable because kinship foster care can be distinguished from foster care with unrelated caregivers in six of these states. For the three states where kinship care is not identifiable, the odds ratios for this category are not estimated.

Table IIC.3 
Logistic Regression Models of Reunification from Substitute Care. Nine States, entry cohorts 1990-1994. Relative Odds of Reunification Within 36 Months of Entry to Care.

Predictor Variable

Category AL CA IL MD MI MO NM NY WI 9-State

Age at Entry

< 3 months 1.06 0.42 0.43 0.42 0.42 0.30 0.35 0.30 0.40 0.40
3-11 months 0.92 0.81 0.77 0.81 0.75 0.61 0.75 0.72 0.72 0.77
1 - 2 years 1.08 0.96 0.89 0.85 0.85 0.84 0.97 0.99 0.89 0.93
3 - 5 years 0.99 1.02 0.95 0.92 0.85 0.87 0.93 1.06 0.88 0.97
6 - 8 years 0.97 1.01 0.95 1.00 0.93 1.03 1.00 0.99 0.90 0.98
9-11 years * 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
12-14 years 1.52 0.73 0.78 1.00 0.93 0.88 0.55 0.85 0.87 0.83
15-17 years 1.59 0.36 0.49 0.57 0.66 0.72 0.26 0.67 0.64 0.57

Gender

Male 1.03 0.99 0.99 1.00 0.97 1.03 0.94 1.04 1.05 1.02
Female * 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00

Race/Ethnic

African American 0.68 0.64 0.55 0.72 0.84 0.78 0.53 0.74 0.85 0.61
Hispanic - - 1.00 1.21 1.13 1.00 0.86 0.97 0.98 1.05 1.11
White, Other * 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00

Region

Primary urban county 1.64 0.88 0.25 0.58 0.82 0.83 0.75 0.62 0.58 0.65
Remainder of State * 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00

Sequence

First spell in care * 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
Reentry Spell 0.60 0.67 0.84 0.81 0.84 0.66 0.65 0.65 0.69 0.71

Care Type

Kinship Care - - 0.71 0.85 0.38 - - 0.52 0.75 0.77 - - 0.76
Congregate Care 1.38 0.87 1.28 1.52 0.90 0.93 0.94 1.07 1.19 1.05
Foster Care * 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00

Stability

One placement * 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
2+ placements 0.45 0.30 0.49 0.39 0.36 0.50 0.34 0.48 0.36 0.37

Year of Entry

1990 1.15 1.04 1.60 1.14 1.03 0.90 1.00 1.18 0.89 1.12
1991 1.14 1.04 1.23 1.32 1.02 0.89 0.94 1.05 1.00 1.05
1992 * 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
1993 1.07 0.98 0.91 1.08 0.89 1.00 1.10 0.98 1.05 0.99
1994 0.95 0.93 0.89 1.09 0.96 0.90 0.98 0.88 1.03 0.93
  Intercept -0.89 1.18 0.87 0.35 0.71 0.83 1.52 0.71 1.11 0.91
exp(intercept) 0.41 3.25 2.39 1.42 2.03 2.29 4.57 2.03 3.03 2.48
Predictive Concordance 0.66 0.70 0.76 0.69 0.66 0.64 0.70 0.68 0.66 0.69
N Entrants 9,224 143,973 56,354 14,003 31,537 22,629 5,889 94,758 26,049 404,416
N Reunifications 2,369 72,838 15,825 4,849 13,823 10,248 3,300 37,921 15,180 176,353
Prop.Reunifications 0.26 0.51 0.28 0.35 0.44 0.45 0.56 0.40 0.58 0.44

Note: Predictors noted with "*" are not contained in model, but are the "excluded" category for their variable. Predetermined odds ratios of 1.00 are assigned to these categories. Odds ratios for other categories of the same variable express effects in relation to that of the excluded category.


Table IIC.4
Logistic Regression Model of Family Exit from Substitute Care. Nine States, Entry Cohorts 1990-1994.Relative odds of Reunification or Exit to Care of Relative Within 36 Months of Entry to Care.
Predictor variable Category AL CA IL MD MI MO NM NY WI 9-State

Age at Entry

< 3 months 0.76 0.41 0.43 0.44 0.42 0.34 0.34 0.30 0.36 0.39
3 -11 months 1.10 0.80 0.76 0.96 0.79 0.64 0.68 0.75 0.72 0.78
1 - 2 years 1.19 0.95 0.87 1.02 0.89 0.86 0.95 1.05 0.92 0.95
3 - 5 years 1.14 1.01 0.93 1.01 0.90 0.91 0.83 1.11 0.93 0.99
6 - 8 years 1.03 1.00 0.94 1.01 0.95 1.08 0.90 1.03 0.92 0.99
9 -11 years * 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
12-14 years 1.11 0.72 0.77 0.97 0.91 0.86 0.49 0.76 0.81 0.79
15-17 years 0.99 0.36 0.49 0.46 0.54 0.72 0.23 0.58 0.52 0.51

Gender

Male 0.94 0.98 0.99 0.97 0.92 1.04 0.93 1.01 1.02 1.00
Female * 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00

Race/Ethnic

African American 1.10 0.61 0.56 0.81 1.03 0.80 0.50 0.83 0.84 0.66
Hispanic 0.98 0.99 1.20 1.07 1.24 0.79 0.82 1.07 1.01 1.06
White, Other * 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00

Region

Primary urban county 1.92 0.91 0.26 0.81 0.75 1.04 0.71 0.55 0.40 0.65
Remainder of State * 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00

Sequence

First spell in care * 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
Reentry Spell 0.54 0.65 0.83 0.72 0.77 0.65 0.60 0.60 0.64 0.69

Care Type

Kinship Care - - 0.73 0.90 0.41 - - 0.85 1.04 0.74 - - 0.66
Congregate Care 1.31 0.84 1.25 1.78 1.07 0.89 0.98 0.95 1.05 1.04
Foster Care * 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00

Stability

One placement * 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
2+ placements 0.35 0.29 0.48 0.33 0.25 0.49 0.26 0.39 0.30 0.32

Year of Entry

1990 0.99 1.06 1.58 1.09 0.97 0.88 1.28 1.20 0.87 1.14
1991 1.17 1.05 1.20 1.30 1.00 0.88 1.04 1.07 0.98 1.06
1992 * 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
1993 1.12 0.98 0.95 1.00 0.83 1.02 1.16 0.99 1.02 0.98
1994 0.99 0.92 0.94 1.09 0.92 0.89 1.11 0.90 1.01 0.92
  Intercept 0.49 1.30 0.87 0.92 1.31 0.89 2.06 1.25 1.71 1.26
exp(intercept) 1.63 3.67 2.39 2.51 3.71 2.44 7.85 3.49 5.53 3.53
Predictive Concordance 0.67 0.70 0.75 0.69 0.70 0.63 0.73 0.70 0.69 0.70
N Entrants 9,224 143,973 56,354 14,003 31,537 22,629 5,889 94,758 26,049 404,416
N Family Exits 4,942 75,482 16,151 7,108 16,839 11,118 3,724 45,608 16,710 197,682
Prop. Family Exits 0.54 0.52 0.29 0.51 0.53 0.49 0.63 0.48 0.64 0.49

Note: Predictors noted with "*" are not contained in model, but are the "excluded" category for their variable. Predetermined odds ratios of 1.00 are assigned to these categories. Odds ratios for other categories of the same variable express effects in relation to that of the excluded category.


The findings reported from each of the separate state models resemble one another and the pooled nine-state result. This suggests that there is substantial similarity in the nature of the reunification process across states. Certain results are remarkably stable. These include the findings that gender has little influence on reunification and that reentry spells are much less likely to result in reunification than are first spells.

For other findings, the direction of the results is fairly consistent, but the magnitude of the effects varies widely among states. For example, white children are more likely to be reunified from substitute care than are African American children in all states. In Illinois and New Mexico, the likelihood of reunification is almost twice as high for whites as for African Americans (ratio of 100:55), while race has less predictive importance in Michigan and Wisconsin as shown by odds ratios that are near 100:85. The pattern for Hispanics varies, but they usually are equally likely or somewhat more likely to be reunified than whites.

The placement type variable is more useful with state-level data because kinship care can be separately identified for six of the nine states. In each of these states, reunification within three years is much less likely from kinship foster care than from nonrelative foster care. In most states, the odds of reunification from congregate care placements and nonrelative foster care placements are rather similar, although in Alabama and Maryland the odds of reunification from congregate placements are much higher.

Alabama appears to be the state that is most systematically different in the relationships between these predictive variables and reunification. The patterns of influence by age and by regional influence on reunification observed for Alabama are directly reversed from what is seen in the other eight states.

There is also variability by state in the apparent trends in the likelihood of reunification over time, as seen through the experiences of cohorts defined by year of entry to care. These trends differ from those presented in earlier sections because the multivariate models introduce factors that are related to the changing composition of each state's caseload. For example, if a state foster care caseload has an increasing share of infant cases, we would expect reunification rates to decrease, even if all other practice factors remained stable. In this case, the likelihood of reunification for any given individual would not be changing, but the statewide reunification rates would drop solely because the caseload contains a greater proportion of children that have a lower likelihood of reunification. So, these measures of cohort effects can be interpreted as changes in the likelihood of reunification across years, with some control for shifts in composition based on age, race, gender, region, etc.

The nine-state total shows the odds of reunification decreasing steadily between the 1990 and 1994 cohorts, a total 17 percent reduction [I - (0.93/1.12)] across the five years. For individual states, a clear and direct downward trend for reunification is apparent only for California, Illinois, and New York. For Wisconsin, the odds of reunification increase across the five annual cohorts, and the trends over times in the other states are inconsistent or absent.

The fact that trends observed primarily in California, Illinois and New York have strong influence on the nine-state results bears some thought. Together, these three states contain 73 percent of the foster care spells in the universe being examined, yet they represent only three of the nine separate child welfare systems. To the extent that we are trying to understand national trends and patterns, consideration of pooled data (which gives weight to the larger caseloads) seems useful. If we are trying to understand the behavior of child welfare systems, then examination of individual state models might be more appropriate.