# Understanding Foster Parenting: Using Administrative Data to Explore Retention. Multivariate Models of Length of Service

For Oklahoma and Oregon, Cox regression models were used to examine the relationship between length of service and foster parents'characteristics and activity.(5)  As noted earlier, the dependent variable for these models is a hazard ratio that represents the likelihood of exit from foster parenting, thus identifying variations in length of service.

For most variables, the largest stratum was used as the reference group against which the relative likelihood of exiting foster parenting was estimated for other groups. In the Oregon model, high levels of missing data for income and employment status limited the number of cases available for analysis. Because a model excluding these variables yielded similar findings, the model with all variables is shown here.

Findings shown in Tables 4-4 (for Oklahoma) and 4-5 (for Oregon) were generally consistent with the bivariate analyses described in the previous section. In reading these tables, the key statistic is the hazard ratio in the third column. Hazard ratios less than one indicate reduced likelihood of leaving foster parenting, or greater length of service. Hazard ratios greater than one indicate increased likelihood of exit, or shorter length of service. The fourth column indicates the statistical significance of the hazard ratio as compared to the reference category for each variable.

Table 4-4.
Cox Proportional Hazards Model of Length of Service in Oklahoma (n = 2,765)
Variable B S.E Hazard Ratio p
License type (vs. regular foster care)
Restricted foster care -0.0922 0.05865 0.912 0.1161
Age (vs. all foster parents between 30 and 55 years)
At least one foster parent aged > 18 and < 30 years 0.2232 0.0632 1.2500** 0.0004
At least one foster parent over age 55 -0.0759 0.0737 0.9270 0.3030
Race (vs. all foster parents white)
At least one foster parent Native American 0.0570 0.0943 1.0590 0.5453
At least one foster parent black 0.1414 0.0678 1.1520* 0.0371
Location (vs. metropolitan)
Nonmetropolitan 0.1464 0.0506 1.1580** 0.0038
Foster home composition (vs. two parents)
Single parent -0.0121 0.0582 0.9880 0.8358
Occupancy (vs. lowest quartile)
2nd quartile -0.2306 0.0994 0.7940* 0.0203
3rd quartile -0.2877 0.0704 0.7500** <0.0001
4th quartile -0.7651 0.0883 0.4650** <0.0001
Percent of placements that were infants (vs. none)
Between 0 and 100 -0.6778 0.0696 0.5080** <0.0001
100 -0.3781 0.1220 0.6850** 0.0019
Percent of placements that were adolescents (vs. none)
Between 0 and 100 -0.3501 0.0711 0.7050** <0.0001
100 -0.2014 0.0679 0.8180** 0.0030
Percent of placements that had special needs
Between 0 and 100 -0.2259 0.0687 0.7980** 0.0010
100 0.1319 0.0715 1.1410 0.0651
Notes: Model Chi-Square (Wald) 503.4187 with 16 DF (p < .0001)
*p <.05
**p<.01
Table 4-5.
Cox Proportional Hazards Model of Length of Service in Oregon (n = 7,908)
Variable B S.E Hazard Ratio p
License type (vs. regular foster care)
Restricted foster care 0.13141 0.02867 1.14** <0.0001
Age (vs. all foster parents between 30 and 55 years)
At least one foster parent aged > 18 and < 30 years 0.1244 0.03017 1.132** <0.0001
At least one foster parent over age 55 -0.01764 0.04217 0.983 0.6758
Race (vs. all foster parents white)
At least one foster parent Native American -0.01306 0.06043 0.987 0.8289
At least one foster parent black -0.03668 0.04667 0.964 0.432
Location (vs. metropolitan)
Nonmetropolitan 0.25489 0.02642 1.29** <0.0001
Foster home composition (vs. two parents)
Single parent -0.02839 0.03307 0.972 0.3906
Income (vs. less than or equal to median)
Greater than median -0.05915 0.02637 0.943* 0.0249
Employment (vs. one at-home foster parent)
All foster parents at work full time -0.02133 0.02642 0.979 0.4194
All foster parents at home -0.0358 0.04993 0.965 0.4734
Occupancy rate (vs. < 1)
Occupancy rate > 1 and < 2 -0.425 0.03027 0.654** <0.0001
Occupancy rate > 2 -0.86162 0.04252 0.422** <0.0001
Percent of placements that were infants (vs. none)
Between 0 and 100 -0.57879 0.03683 0.561** <0.0001
100 -0.16414 0.05556 0.849** 0.0031
Percent of placements that were adolescents (vs. none)
Between 0 and 100 -0.75801 0.03696 0.469** <0.0001
100 -0.06861 0.03126 0.934* 0.0282
Notes: Model Chi-Square (Wald) 2250.9879 with 19 DF (p < .0001)
* p <.05
** p<.01

Table 4-6 summarizes the models for Oklahoma and Oregon. In both states, younger foster parents had significantly higher hazard ratios, indicating a higher likelihood of exit from foster parenting, or shorter length of service. Foster parents in metropolitan areas and those caring for infants, adolescents, or (in Oklahoma) children with special needs all had longer lengths of service. In Oregon, higher income was associated with longer length of service, but length of service did not vary by employment status.

Table 4-6.
Summary of Factors Associated With
Greater Length of Service in Multivariate Models
Characteristic Oklahoma Oregon
License type n.s Regular foster care
Age Aged 30 or greater Aged 30 or greater
Race n.s n.s
Location Metropolitan Metropolitan
Foster home composition n.s n.s
Income Greater than median
Employment n.s
Occupancy rate Higher occupancy Higher occupancy
Care for infants Some or all Some or all
Care for adolescents Some or all Some or all
Care for children with special needs Some

Some of the length of service pattern variations seen in the bivariate analyses in Table 4-2 were eliminated when controlling for all variables. The increased length of service for foster parents with restricted licenses in Oklahoma was not apparent after controlling for factors such as age. A linear test of the impact of each set of variables on the overall model found that race was not significant in either Oklahoma or Oregon (p = 0.1057 and 0.9283). Foster home composition was not significant in the multivariate model for either state, although two-parent homes appeared to have substantially higher length of service in bivariate analyses.

Perhaps the most striking finding related to length of service is a pattern seen in bivariate analyses for all three states and persisting in multivariate analyses for Oklahoma and Oregon. Higher occupancy was consistently associated with increased length of service. In addition, care for children who might be considered more demanding  infants, adolescents, and children with special needs  was also associated with longer length of service.

#### View full report

"report.pdf" (pdf, 528.13Kb)