Reply to Putnam-Hornstein et al.: On honest mistakes and raceless children

We appreciate the interest Putnam-Hornstein et al. (1) have shown in our work (2) in their letter, “Contact with the child protection system is pervasive, but are recent estimates correct?”

We have three points in response.

First, as Putnam-Hornstein et al. (1) are now aware but were not when they submitted their letter, there was a coding error that led our investigation estimates to be positively biased. We thank them for raising concerns that led to a rapid correction (2). Our confirmed maltreatment, foster care placement, and termination of parental rights estimates were correct and are unchanged.

Second, as Fig. 1 shows, the differences between our estimates and theirs for the share of all children who have an investigation are miniscule, with the exception of a few counties where the differences are small rather than miniscule. As such, we see our corrected estimates, when compared to their estimates, as demonstrating that the National Child Abuse and Neglect Data System data accurately reflect the cumulative prevalence of having an investigation.

Fig. 1.

Estimation methods and missing data for California counties. Each row has its own x axis scale; counties are ordered by mean estimate within rows. AI/AN, American Indian/Alaska Native; PI, Pacific Islander.

Finally, although we appreciate Putnam-Hornstein et al. (1) generating these estimates so quickly, they made a decision in how they dealt with race that leads to substantial bias in their race-specific estimates. This bias virtually explains any differences between their race-specific estimates and ours. As Fig. 1 shows, rates of missingness on race are high in most counties and are especially so in Los Angeles and Orange. In such situations, researchers can do one of two things: impute information on race or assume race is missing completely at random (MCAR). Putnam-Hornstein et al. instead opt to not count children missing on race as having a race—the equivalent of removing >13% of the children in their data from the numerator but, questionably, not from the denominator in their race-specific estimates (although these children are included in the numerator in their total estimates).

Fig. 1 includes our corrected estimates, their estimates, and our MCAR estimates using their data. It shows their investigation estimates are far too low for Black and Hispanic children in several counties. This method of handling missing data also largely—although not completely—explains the divergence in our foster care estimates for Black children in Los Angeles.

And so, to return to the estimate they end with, the debate is not whether 72% (our error) or 46% (their missing children not counted) of Black children in Los Angeles ever have an investigation. It is, instead, whether 58% (our corrected estimate) or 56% (their missing children counted) do.