Rereporting and Recurrence of Child Maltreatment: Findings from NCANDS. Methodology


This section describes the process used for data construction of a multiyear, multistate NCANDS data set that spans the time period from 1998 to 2002. It also describes the analytic procedures used to develop the study findings.

NCANDS case-level data consist of CPS investigation events at the child level. Only reports that receive an investigation or assessment response from a CPS agency are included. Each record in the data file is referred to as a report-child pair, which indicates that there is a record for each child in each report who receives an investigation or assessment. Each report has a unique ID and more than one child can share the same report ID. Each child has a unique ID, thus the report-child pair is uniquely identified by the combination of its report and child IDs.

The number of States that voluntarily submit these data to the NCANDS increased from 11 States during 1993 to 42 States during 2002. For each investigation, CPS makes a disposition decision which involves determining whether or not a child or children have experienced or are at risk of maltreatment. A child is considered to be a victim of maltreatment if he or she has at least one maltreatment type that is coded as substantiated, indicated, or alternative response.

Compiling the data set for this study consisted of two stages  evaluating the quality of State submissions to determine which States data met the analytic requirements, and constructing a database with all the variables needed for analyses.(1) More detailed information concerning these selection criteria is located in appendix A.

State Selection

Nine States were included in this study based on an examination of the following characteristics of their data submissions to NCANDS.

  • Years of Submissions  States were considered for inclusion if they had submitted case-level data to the NCANDS for calendar years 19982002. Twenty States met this requirement.
  • Unique Child Identifiers  Child identifier data were examined to ensure that the data were of adequate integrity to support the analytic goals of the project. Twelve of the 20 States met this criterion for inclusion in the analysis.
  • Prior Victimization  Nine of the 12 States met the criterion of having at least 2 percent of records indicating that data on prior victimization were included.

The characteristics of the population in the resulting nine States were comparable to the national population on a range of demographic characteristics including age, race distribution, and poverty level. (See table 1.) However, the findings in this study should not be construed as representative of all reporting States or the entire nation.

Table 1.
Demographic and Reporting Characteristics of the Population in Nine States in the Longitudinal Data Set Compared with National Data 2002(2)
Demographic Indicators All States Nine States
Total Population 285,230,516 40,746,575
Percentage of U.S. Population 100% 14%
Population younger than 18 years: Total 26% 27%
Families Percent of total population living in family settings 80% 82%
Married-couple family, with own children younger than 18 years (percent of families) 9% 9%
Families with female householder, no husband present, with own children younger than 18 years (percent of families) 3% 3%
Average Child Poverty Rate 16.7% 17.0%
Race White (not Hispanic), percent of total population 68% 65%
Black or African-American (not Hispanic), percent of total population 12% 12%
American Indian and Alaska Native (not Hispanic), percent of total population 1% 1%
Asian or Pacific Islander (not Hispanic), percent of total population 4% 2%
Other or Multiple Race (not Hispanic), percent of total population 2% 1%
Hispanic or Latino, percent of total population 14% 19%
Sex Male, percent of total population 49% 49%
Female, percent of total population 51% 51%
Child Abuse and Neglect(3) Rate of children investigated per 1,000 children in population 43.8 45.3
Rate of victims of child maltreatment per 1,000 children in population 12.3 10.7

Data Construction

The process for creating the data construction consisted of the following steps.

  • Combinedata for all years  A single file including the data submitted for calendar years 19982002 was created for each State.
  • Create derived independent variables  Certain variables were derived from standard fields used by the States when submitting data to NCANDS. The derived variables included:
    • Source of report;
    • Maltreatment type;
    • Race/ethnicity;
    • Child age category;
    • Victim status (yes/no);
    • Prior victimization (before the initial report in the data set);
    • Postinvestigation service provision (yes/no);
    • Removal from home (yes/no);
    • Child disability (yes/no);
    • Caregiver alcohol abuse (yes/no); and
    • Caregiver substance abuse (yes/no).
  • Derive rereport and recurrence dependent variables  Data from multiple reports involving the same child were combined to develop variables indicating one or more subsequent reports or subsequent victimizations.
  • Children whose records indicated that they had been victims of maltreatment prior to the study period were excluded.

Following this process, data from all States were combined into a single file with all data at the level of the unique child. Finally, data extracts were developed to support specific analyses. Further examinations of the data were conducted to address potential compatibility issues for specific analyses, and States may have been excluded accordingly.

Analytic Techniques

The study variables were designed to use event history analytical techniques  including survival analysis and trajectory analysis. Such techniques are used when the objective of the study is to examine the time between events for the subjects, or the numbers of subsequent events. In this case the subjects were children who were reported. Once a child was reported, he or she was considered "at risk" for a subsequent report and the probability of the event occurring was considered the hazard probability. Event history analysis and survival analysis adjust for the bias associated with estimating hazards with differing lengths of observation periods.

A survival analysis technique, referred to as life tables, was used to analyze the length of time to rereports and recurrence events. Cox regression (proportional hazards analysis), a form of multivariate survival analysis, was used to study the factors that were associated with single rereports and recurrences. A type of event history analysis that focuses on counts of multiple repeated events, referred to as trajectory analysis, was used to address patterns of multiple rereports and recurrences.

View full report


"report.pdf" (pdf, 396.74Kb)

Note: Documents in PDF format require the Adobe Acrobat Reader®. If you experience problems with PDF documents, please download the latest version of the Reader®