PROTECTING THE SACRED: AN ANALYSIS OF INDIGENOUS CHILDREN’S EXPERIENCES IN THE OKLAHOMA CHILD WELFARE SYSTEM

Abstract

Introduction: Given what is known about the historical mistreatment of Indigenous children, the current study used Oklahoma data to determine if Indigenous identity was associated with specific child welfare outcomes (e.g., substantiation, removal, and service referrals) when controlling for public assistance, caregiver substance use, domestic violence, child demographics, and maltreatment allegation types. Methods: Oklahoma data from federal fiscal year 2019’s National Child Abuse and Neglect Data System’s child file was used. A simple random sample of 1000 children was used to create a calibration sample (n = 500), and validation sample (n = 500) while the remaining population level data (n = 65,200) was used as the holdout sample. Children were classified as either non-Hispanic White, Indigenous alone, Indigenous and White, Indigenous and other ethnoracial minority, or non-Indigenous ethnoracial minority for the purpose of analysis. Except for child age and number of children within the report, covariates were coded as “Yes” (1) or “No” (0). Descriptive statistics as well as hierarchical logistic and linear regressions were conducted using STATA (18) with p < .05 indicating statistical significance. Results: When assessing for associations between identity and substantiation the addition of identity enhanced model performance across the calibration (Pseudo-R2 for step 1 = .2562; ΔPseudo-R2 for step 2 = .0017; X2(15) = 102.26, p = .0000), validation (Pseudo-R2 for step 1 = .2861; ΔPseudo-R2 for step 2 = .0142; X2(14) = 104.59, p = .0000), and holdout (Pseudo-R2 for step 1 = .2475; ΔPseudo-R2 for step 2 = .0001; X2(15) = 9891.41, p = .0000) samples. Despite this none of the Indigenous identity categories were statistically associated with higher odds of substantiation within the calibration and validation model findings. This changed for the holdout model where Indigenous and White children had lower odds of substantiation (AOR: .91, 95%CI: .85, .97) compared to their non-Hispanic White peers. When assessing associations between identity and child removal, the addition of identity enhanced model performance across the calibration (Pseudo-R2 for step 1 = .1397; ΔPseudo-R2 for step 2 = .0197; X2(14) = 45.22, p = .0000), validation (Pseudo-R2 for step 1 = .1884; ΔPseudo-R2 for step 2 = .032; X2(14) = 49.05, p = .0000), and holdout (Pseudo-R2 for step 1 = .1461.; ΔPseudo-R2 for step 2 = .0009; X2(15) = 4190.31, p = .0000) samples. Across both the calibration (AOR: 3.42, 95%CI: 1.03 - 11.41) and validation (AOR: 4.36, 95%CI:1.36 - 13.98) samples, children who were Indigenous alone had higher odds of removal compared to their non-Hispanic White peers. Despite the consistency across the two samples, the relationship was not observed in the holdout sample. Instead, the Indigenous and White (AOR: 1.11, 95%CI: 1.01, 1.22) and Indigenous and other ethnoracial minority (AOR: 1.35, 95%CI:1.20, 1.52) subgroups were found to be associated with higher odds of child removal compared to their non-Hispanic White peers. Model performance assessing the relationship between Indigenous identity and service referrals was inconsistent across the calibration, validation, and holdout samples. The calibration model was not statistically significant even after the inclusion of identity (R2 for step 1 = .0171; ΔR2 for step 2 = .012; F(15, 469) = 1.11, p > .05). In comparison, the validation model was statistically significant (R2 for step 1 = .0661; ΔR2 for step 2 = .0106; F(15, 458) = 2.74, p = .0005); however, none of the Indigenous ethnoracial groups were associated with service referrals. When assessing the remaining population using the holdout sample (R2 for step 1 = .0286; ΔR2 for step 2 = .2714; F(15, 52,681) = 125.25, p = .0000) – both the Indigenous and White (p = .000) and Indigenous and other ethnoracial minority (p = .000) groups were positively associated with service referrals. Despite being statistically significant the validation and holdout models accounted for low overall variance. Conclusion: By grouping multiracial Indigenous children in their own ethnoracial categories – a better picture of Indigenous child welfare experiences was obtained. This is critically important as multiracial White and Indigenous children were found to have lower odds of substantiation. Further, higher odds of removal were observed in population data for both the Indigenous and White and Indigenous and other ethnoracial minority groups. These findings further support the need to not only evaluate Indigenous children in child welfare research, but to consider different approaches to the use of multiracial identity, particularly for Indigenous children. Further research is needed to explore the percentage of multiracial Indigenous children falling outside the protections afforded to citizens of federally recognized tribes via the Indian Child Welfare Act (ICWA). Additional research is needed to determine the intersection between Indigenous identity and family/community level factors that may influence child welfare outcomes.