Recruitment of the States
WRMA engaged in a single recruitment effort for both the Key Informant Survey and the Prevalence Study, approaching all 50 States, the District of Columbia, and Puerto Rico. Primary contacts in each State, usually State child welfare directors, were sent recruitment packets, including letters describing the purpose of the study, the nature of the data requested, and the estimated time required to respond. Copies of the surveys were also provided in the recruitment packet. Trained staff followed up with a telephone call approximately 1 week later to confirm receipt of the packet, to make an appointment to discuss the project, and to seek each State’s participation.
Primary contacts who agreed to have their State participate supplied the names and contact information for respondents for each of the surveys and the prevalence study. These respondents were contacted by email with a packet of information about the data they were asked to supply and a timeframe for their response (2 weeks for the surveys, 1 month for the prevalence study data). Detailed instructions for accessing and completing the online KIS surveys or preparing and submitting prevalence data, as appropriate, were also supplied.
Verbal and written reminders were sent to all contacts shortly before their data were due. Additional reminders were sent after the initial due date to those who had not completed their data submission. A total of 38 States (including Puerto Rico) completed one or more of the three KIS surveys, with 36 States completing each of the surveys. Twenty-two States provided data for the prevalence study.
Key Informant Survey
For the Key Informant Survey, three stand-alone online questionnaires were fielded: (1) Current Legal and Policy Requirements Regarding Sharing Information on Child Maltreatment Perpetrators (Legal/Policy Questionnaire), (2) Current Practices on Sharing Information on Child Maltreatment Perpetrators (Current Practices Questionnaire), and (3) Technical Information on Data Repositories of Child Maltreatment Perpetrators (Technical Questionnaire). Each of the questionnaires was self-administered online by State respondents using Survey Monkey. &™ In a small number of States, more than one respondent filled out separate versions of the same questionnaire. In such cases, the answers were reviewed and consolidated into a single set of answers by WRMA staff. In the few instances in which answers differed, respondents were contacted to resolve the inconsistencies prior to consolidation.
A review of current State law in the States that did not participate in the Legal/Policy Questionnaire was conducted to provide a more comprehensive status of State law. Eleven questions were identified for follow-up based on the likelihood that State law would address the issue. These included questions 1, 5, 6, 9, 13, 20, 26, 34, 35, 36, and 37. In addition, information for question 4 was obtained from Child Maltreatment 2009. States for which responses were added included Alaska, California, District of Columbia, Hawaii, Idaho, Iowa, Kentucky, Massachusetts, Mississippi, Montana, Pennsylvania, Utah, Vermont, Washington, West Virginia, and Wisconsin.
Data were downloaded into SPSS for data checking and analysis. Unless otherwise indicated in text or tables, all percentages were calculated based on 36 States. For some of the tables, the percentages were calculated based on 52 States, and this is noted. These are tables for which additional data were obtained for nonparticipating States, based on a review of existing State law.
Prevalence Study: Creating National Estimates
WRMA’s core data set for creating national estimates of interstate perpetrators of child maltreatment consists of individual perpetrator and child data extracted from the 2005-2009 NCANDS microdata child files, supplemented with name and date of birth information for each substantiated perpetrator to facilitate interstate matching. Last names were encoded to protect confidentiality and to facilitate matching. Perpetrators ages 18 and older at the time of the report were included. The ultimate goal was to model a national estimate of the number of substantiated perpetrators in 2009 that had prior substantiations in other States within the previous 5 years, providing an estimate of the number of matches that might be expected over the course of a year from a fully functioning national registry of substantiated maltreatment perpetrators.
Twenty-two States supplied encoded name and date of birth data for the prevalence study, covering 54 percent of the U.S. population. For each participating State, the number of matches found with records from other participating States was adjusted upwards to account for interstate perpetrators from nonparticipating States based on 5-year interstate migration rates from the 2000 Decennial Census. For non-participating States, WRMA modeled estimates of interstate perpetrators using State aggregate totals of unique substantiated perpetrators, applying an interstate perpetration rate derived from the prevalence data, and adjusted for each State using interstate migration rates from the 2000 Census. Finally, when estimates had been produced for all 50 States and the District of Columbia using these methods, a national estimate was produced.
The Prevalence Database
The prevalence database was created to facilitate the matching of perpetrator records across States, and to support analysis of selected interstate perpetrators’ characteristics. Individual records were produced for each substantiated perpetrator age 18 and older based on NCANDS child/perpetrator records for 2005–2009. Each record contains the encrypted CaseID, selected demographic characteristics of the perpetrator, as well as maltreatment type and date of birth for each child related to that CaseID.
NCANDS data are necessary but not sufficient to support matching of perpetrators across States. WRMA approached all States for additional information on substantiated perpetrators that would allow it to make a reasonable match across States. A major constraint was that States, for reasons of confidentiality, would likely not provide information such as a social security number or complete names that would allow for the unique identification of perpetrators. Instead, WRMA asked States for the date of birth, an encoded version of the perpetrator’s last name, and first initial of the first name. This information, combined with the data extracted from the NCANDS records, WRMA felt would support a high probability match while maintaining confidentiality.
The Interstate Record Matching Process
To determine the optimal matching algorithm, a progressively restrictive set of matching criteria was estimated. The first algorithm relied only on matching the encoded last name and first initial of each perpetrator. This is the only algorithm that would be possible under the current law, which allows only name and maltreatment type case-record information. This resulted in one or more interstate matches for 88.6 percent of all perpetrators in 2009 among participating States. Of those that matched, matches were found in three or more States in 85.1 percent of the cases. (See table 23 and 24 in Section II of this report.)
The addition of the sex of the perpetrator to the algorithm reduced the percentage of perpetrators with one or more interstate matches only slightly to 83.4 percent. Adding the perpetrator’s date of birth to their sex and name information produced an overall match rate of 0.7 percent, or 2,022 matches among the participating States. Of those, only 44, or 2.2 percent, found matches in more than one State. (See table 25 in Section II of this report.)
The algorithm using encoded last name, first initial, sex, and date of birth produced a reasonable match rate that eliminated nearly all matches to more than one State. Further, it is based on very complete data, with 97.9 percent of all perpetrator records containing valid data for all three characteristics. Further, if one assumes that about 16.7 percent of all substantiated perpetrators will reoffend within 5 years, and that they move between States at a rate similar to the general population in 2000 (8.9 percent), then one might expect an interstate match rate of about 1.5 percent. When the 0.7 percent match rate among participating States is inflated to account for data from nonparticipating States, the rate doubles to roughly 1.5 percent.
One additional check on the final algorithm was performed, examining the percentage of those matches in which the birth date of at least one of the children that could be connected to the perpetrator in the 2009 reports matched the birth date of children in the matching report. The result was a match rate of 27 percent, or about one-quarter of the records that matched using the perpetrator’s name, sex, and date of birth alone. There are a number of reasons why one would not find a match on children’s dates of birth, even when it is, in fact, the same perpetrator. First, it is not uncommon for birth dates to be inaccurate. In addition, fewer than one-half the States regularly include all children in the household in a maltreatment report. So, while the true match rate is probably substantially higher than 27 percent, this nevertheless indicates that the algorithm using only the perpetrator’s name, sex, and date of birth included a significant proportion of false positives, and should be treated as an upper-end estimate.
Creating Estimates for Participating States
The number of matches identified for each participating State in the prevalence study included only those perpetrators from other States also participating in the prevalence study. An adjustment to that number was required to account for interstate perpetrators coming from nonparticipating States. This adjustment was derived from 5-year interstate migration rates taken from the 2000 Decennial Census. In 2000, respondents were asked where household members lived 5 years before, in 1995. From these data the U.S. Census Bureau produced counts of interstate migration flows from each State to every other State for all persons ages 5–64 in 2000. Using these data, for each State, WRMA calculated the percentage of all Census interstate migrants who originated in States represented in the prevalence study. This percentage was used to create an inflation factor, which was then applied to the number of interstate perpetrators calculated through the matching process, to produce an estimate of the total number of interstate perpetrators coming from all 50 States and the District of Columbia.
For example, say that the matching process identifies 450 interstate perpetrators in State X. Examining the census migration data reveals that 60 percent of all migrants into State X in 2000 came from one of the States that participated in the prevalence study. By taking the inverse of that 60 percent (1.0/0.60 = 1.66) one can estimate the total number of interstate perpetrators for State X to be 450 * 1.66 = 747.
This estimation was produced for each of the participating States in the prevalence study.
Creating Estimates for nonParticipating States
Estimates for nonparticipating States must, of course, be modeled differently due to the lack of any actual data on interstate child maltreatment perpetrators. One begins by calculating the average interstate perpetration rate among the participating States (IAPR). This is accomplished by summing the total number of interstate child maltreatment perpetrators calculated in the matching process, and dividing it by the total number of substantiated child maltreatment perpetrators for the most recent year.
Next, this average rate for each State was adjusted to reflect the fact that States can have very different rates of persons migrating to them from other States. Some popular destination States such as Arizona have relatively high rates while others, such as Michigan, have lower rates.
Individual States’ inmigration rates can differ substantially. Using the Census data described above, a Census interstate migration rate was calculated for each nonparticipating State (SCIMR) and divided by the overall national census inmigration rate (NCIMR), which is just the percentage of all persons in the U.S. ages 5 to 64 who lived in a different State 5 years before. The SCIMR/NCIMR = SMA, or the State migration adjustment. For example, if a State’s inmigration rate is 5 percent and the national rate is 4 percent, the SMA = 5/4 = 1.25. The SMA is applied to the IAPR in order produce an adjusted interstate perpetration (AIP) rate for each nonparticipating State: SMA * IAPR = AIP.
Finally, the total number of unique substantiated child maltreatment perpetrators in each nonparticipating State (taken from the most recent year of NCANDS data) was multiplied by the AIP to produce the estimated number of interstate child maltreatment perpetrators for each nonparticipating State. It must be noted that unique child maltreatment perpetrator counts were not available from any source for North Dakota, Oregon, and Georgia. NCANDS aggregate duplicated counts of substantiated maltreatment victims were available for these States. To develop an estimate of unique perpetrators for these States, the ratio of unique perpetrators to duplicate victims for all reporting States within NCANDS was calculated and multiplied by the duplicated victim count for each of the three States (North Dakota, Oregon, and Georgia) to estimate the number of unique perpetrators in each State.
Creating the National Estimate
The final step in the process of creating national estimates was straightforward. The estimated number of substantiated child maltreatment perpetrators for each of the 50 States and the District of Columbia, participating and nonparticipating, were summed to produce a national estimate for the total number of substantiated child maltreatment perpetrators. To produce a national rate, that total was divided by the total number of child maltreatment perpetrators across the 50 States and the District of Columbia. These numbers are all reported in the body of the report.
"ResearchReport.pdf" (pdf, 4.61Mb)
"ReportMethodology.pdf" (pdf, 738.21Kb)
"CaseLawReview.pdf" (pdf, 598.85Kb)