The Incredibly Credible Prevalence of Child Protective Services Contact in New Zealand and the United States

In late 2014, I was invited to present some of my recent research estimating the cumulative prevalence of confirmed child maltreatment¹ and foster care placement² among US children at the Inequality and Social Policy Seminar at Harvard University. This was a huge moment for me. Rarely, if ever, had mainstream sociology seriously engaged with either of these social problems.

Unfortunately, the presentation did not go exactly how I had hoped. Within the first 10 minutes, one of the most prominent sociologists in the world—someone whose work I had poured over as a graduate student and who I still greatly admire—stopped me and said that he did not find the estimates credible. I must have done something wrong—a horrifying possibility considering the simplicity of the code—or something must be wrong with my data—another horrifying possibility because the data I was using were the only US population data on the topic. At first, his incredulous response surprised me. And then it dawned on me. The numbers themselves are, to put it simply, incredible: they are difficult (or impossible) to believe or extraordinary.

SIMPLY TOO HIGH

It is in fact quite difficult to believe that 6% of US children will ever be placed in foster care,² 12% will ever experience confirmed maltreatment,¹ and 37% will ever have a child protective services (CPS) investigation.³ This is to say nothing of the risks for African American and Native American children, for whom these events are even more ubiquitous. All these numbers are difficult or impossible to believe and, hence, incredulity absent further confirmation is reasonable.

The source of the incredulity about these estimates seems to revolve primarily around the concern that they are simply too high—that so many children cannot experience any of those events. To be fair, there is good reason for this skepticism because of the limitations of existing data. Research in this area uses two data sets: the Adoption and Foster Care Analysis and Reporting System (AFCARS) and the National Child Abuse and Neglect Data System (NCANDS). Neither of these data sets can track the cross-state movement of children because they contain only state-specific identifiers instead of national ones.^1–3 Thus, if the same child experiences CPS investigations in New York, Rhode Island, and Georgia, for instance, then that one child could incorrectly be counted as three children experiencing a first CPS investigation.

Previous estimates using the AFCARS and NCANDS data will thus be upwardly biased. I would argue that the level of bias here is likely small—not more than a 10% difference. But the reality is that this is just my own bias and that estimates using birth cohort data (in which children can only migrate out, hence biasing estimates down) would be a useful corrective here. This is precisely the sort of data that the excellent article by Rouland and Vaithianathan (p. 511) uses, although from New Zealand rather than the United States.

NEW ZEALAND VS UNITED STATES

Using these data, Rouland and Vaithianathan use a birth cohort life table to provide estimates of the cumulative prevalence of having a CPS referral (which is an earlier stage of CPS contact than can be estimated using the NCANDS), experiencing a confirmed maltreatment case, and being placed in foster care in New Zealand. Table 1 presents estimates from this birth cohort study in New Zealand and three previous studies using synthetic cohorts in the United States to provide an easy comparison. (Estimates by race/ethnicity are not presented because Rouland and Vaithianathan do not present them, the lone limitation of their excellent article.) US children are 25% to 50% more likely to experience confirmed maltreatment than are children from New Zealand; they are also 50% to 100% more likely to ever be placed in foster care. Differences earlier in the system are also pronounced: US children are 50% more likely to ever have a CPS investigation than New Zealand children are to have a complaint made to CPS.

TABLE 1—

Percentage of Children Experiencing a CPS Referral, a CPS Investigation, Confirmed Maltreatment, and Foster Care Placement Birth to 5 Years and Birth to 18 Years: United States (Various Years, 2000–2014) and New Zealand (1998–2015)

Age and Location	CPS Referral, %	CPS Investigation, %	Confirmed Maltreatment, %	Foster Care Placement, %
By 18 y
United States1–3	. . .	37.0	12.0–15.1	4.8–5.9
New Zealandb	23.5	. . .	9.7	3.1
By 5 y
United States1–3	. . .	15.9	6.6	2.6
California birth cohort4	8.3	14.0	5.3	2.2
NCANDSa	. . .	. . .	5.9	2.5
New Zealandb	. . .	. . .	2.7	1.2

Note. CPS = Child Protective Services; NCANDS = National Child Abuse and Neglect Monitoring System.

^aAuthor’s unpublished analyses of Adoption and Foster Care Analysis and Reporting System and NCANDS data from California.

^bFrom Rouland and Vaithianathan (p. 511).

On the basis of these estimates, which come from another country and use a method for estimating the cumulative prevalence of CPS contact that is downwardly biased, we could still reasonably conclude that CPS contact is quite common for children growing up in New Zealand. And, indeed, this is precisely the argument that the authors of this article make when they note that more children in New Zealand experience confirmed maltreatment than are obese.

Of course, knowing that CPS contact is highly prevalent in other developed nations, although important, provides little insight into whether estimates of the cumulative risk of CPS contact among US children are incredible and credible or incredible and horribly biased. To provide preliminary leverage on this question, Table 1 also shows additional analyses to age five years using (1) the New Zealand birth cohort data, (2) synthetic cohort estimates using the AFCARS and NCANDS data in the United States (in 2005), (3) birth cohort estimates using the 2002 birth cohort of children in California, and (4) synthetic cohort estimates using the AFCARS and NCANDS for California only (in 2005). By presenting these California-specific estimates using the NCANDS data and the birth cohort data, I provide insight into how large differences in the cumulative prevalence of CPS contact may be across these different types of data sets.

As Table 1 shows, the differences to age five years in California using linked administrative data (and birth cohort estimation) and the NCANDS (and synthetic cohort estimation) are small but perceptible. Linked administrative data show cumulative prevalences of confirmed maltreatment and foster care placement of 5.3% and 2.2%, respectively; NCANDS data estimate cumulative risks of 5.9% and 2.5%, respectively. Thus, NCANDS estimates are 11% higher for confirmed maltreatment and 14% higher for foster care placement than for birth cohort estimates. Although this is, to reiterate, a real difference, if the magnitude of the difference between these birth cohort and synthetic cohort estimates persists throughout all of childhood, which is certainly a distinct possibility, that would suggest that the incredibly high cumulative risks of CPS contact that previous research has documented are almost certainly credible and must be treated as such.

POPULATION DATA

Of course, instructive as I believe the results presented in Table 1 are, the reality is that we cannot know for sure how many US children will ever experience CPS contact until we have data like those of Rouland and Vaithianathan: population data on all children in a nation, including national identification numbers, that is neither upwardly biased (as our synthetic cohort estimates of NCANDS are) nor downwardly biased (as our birth cohort estimates of single-state data are). Absent these data, the best those of us preoccupied with estimating the prevalence of CPS contact for US children can do is provide a range of upwardly and downwardly biased estimates and hope that the differences between them are, as they are in Table 1, small.