How We Analyzed Child Welfare Investigations

A yearlong investigation by ProPublica and NBC News has explored inequities across the U.S. child welfare system, looking at mandatory reporting requirements, frequency of investigations and more.

By some estimates, the likelihood of Black youths experiencing an investigation by a child protective services agency is far higher than their likelihood of being stopped by police.

And in Maricopa County, a study from last year estimated that 63% of Black children will experience an investigation before they turn 18, the highest rate among the 20 largest counties in the country.

That study was based on an analysis of child protective services cases in two databases obtained from the Department of Health and Human Services’ National Data Archive on Child Abuse and Neglect: the National Child Abuse and Neglect Data System, which provides information on child maltreatment reports and investigations, and the Adoption and Foster Care Analysis and Reporting System, which provides information on removals of children from home, terminations of parental rights and adoptions.

We obtained both datasets to broaden the scope of the study’s county-level analysis and dive deeper into why families were being investigated. NDACAN and the Department of Health and Human Services’ Children’s Bureau work together to make this data available to researchers. They do not endorse the independent findings of researchers, and bear no responsibility for the analyses or interpretations presented here.

Our analysis confirmed that Maricopa County, where Phoenix is located, had one the highest rates of investigation for Black children among the nation’s largest counties. The rate there was nearly three times as high as the county’s rate for white children.

The analysis, which took more than a year to complete and included counties of all sizes, also found that Maricopa County isn’t much of an outlier nationwide, as dozens of counties had similar or higher rates of investigations for Black children.

How We Analyzed the Databases

The NCANDS database required steps to clean and deduplicate before we could make comparisons across counties and states.

For our analysis of investigations, we merged the separate fiscal year files for the NCANDS database between 2015 and 2020 and deduplicated according to the unique child IDs provided in the dataset. For race and ethnicity information, we took the information from the most recent report for each child ID for which the race and ethnicity was known. Then we filtered this list to the first investigation by county for each child that occurred between the calendar years 2015 and 2019, the latest full year of available data, based on the date the investigation started.

We grouped this list by county and counted the number of entries by race. For this count, we excluded children for which multiple races were indicated to match data from the Census Bureau’s American Community Survey. For our count of white children, we included only the entries in which the ethnicity was marked as “non-Hispanic.” The final rate calculation took the number of children investigated by race over the five-year time period and divided it by the under-18 population from the 2015-2019 ACS survey.

For Maricopa County, we found that 38% of Black children had their families investigated by a child welfare agency during the period analyzed, the sixth-highest rate among the 20 largest counties in the country. Due to changes in the underlying population over the five-year period, such as births, deaths and moving in and out of the county, our rate figures should not be interpreted as the likelihood that a child living there would be involved in an investigation. Rates could not be calculated for many smaller counties because the data archive masks what county an investigation took place in if that county has less than 1,000 entries in a fiscal year.

We used a similar deduplication method to analyze the types of allegations in each case and whether they were substantiated, but instead of limiting it to the first investigation for each child, we looked at all investigations that started between 2015 and 2019. Using this list, we counted how many children were either confirmed or suspected of being victims of maltreatment and how many of those cases were for allegations of physical or sexual abuse.

Differences Between Our Analysis and Other Methods

While our analysis used the same dataset as the study that found 63% of Black children will experience a CPS investigation during their childhoods, there are some important differences in how we analyzed the data.

The biggest difference is that the study used the number of children who experienced their first CPS investigation ever during a five-year period (2014 to 2018) to estimate the likelihood that a child would experience an investigation before they turn 18. To ensure that the estimate was as accurate as possible, the researchers used statistical methods to impute what the races would likely be for children whose races were marked as unknown. Furthermore, the study included cases for children with multiple races.

Because we chose not to impute the missing race values or include cases involving children with multiple races, our counts of investigations by race could be lower than the true number.

Analyzing Termination of Parental Rights Using Foster Care Data

We also analyzed AFCARS data to determine how quickly states have been moving to terminate parental rights. The analysis found that West Virginia severed the parental rights of 2% of the state’s children between 2015 and 2019, the highest rate in the country. The typical case lasted less than a year — nine months shorter than the national median.

Before we could compare termination of parental rights across states, the AFCARS data required cleaning steps similar to the NCANDS data.

This database had unique identifiers for children called an AFCARS ID, which we used to remove duplicates. We found the latest report containing each AFCARS ID and pulled the information for that entry into our database. For cases in which the most recent report had multiple entries for the same unique identifier, we kept the earlier “first removal date,” or the date on which the child was first placed in foster care, and the latest termination of parental rights date.

We then filtered the dataset to the entries in which rights for both parents were severed between 2015 and 2019. Then we grouped this dataset by state and counted the number of terminations, the median time between a child’s first removal from the home and the termination, and the reasons why the child was removed, including physical or sexual abuse, parental drug use or neglect. To find the rate of terminations per child in each state, we divided our count by the Census child population figures described above.