Studies have found that gender, age and race are associated with the risk of sexual abuse victimization, although questions remain. Research consistently indicates that girls are sexually abused at higher rates than boys at a ratio of 2.5 to 1 (Finkelhor, 1993; Putnam, 2003; Sedlak et al., 2010). However, sexual abuse is under reported by both genders (Basile, Chen, Black, & Saltzman, 2007). Risk for CSA is believed to increase around age 6 and again at age 10 (Finkelhor, 1993). However, data from the Fourth National Incidence Study of Child Abuse and Neglect (NIS-4) indicate a recent increase in rates of sexual abuse among children from birth to age 2 (Sedlak et al., 2010). If the first sexual abuse experience occurs at an early age this may increase the odds of sexual revictimization during adolescence or adulthood (Casey & Nurius, 2005). Although a review of the literature by Putnam (2003) indicated that race and ethnicity are not risk factors for CSA, data collected for NIS-4 found significantly higher rates of sexual abuse among Black children than White or Hispanic children (Sedlak et al., 2010).

Several maternal characteristics have been associated with risk for child maltreatment (Dubowitz et al., 2011; Lee & Goerge, 1999; Phillips & Erkanli, 2008). Lee and Goerge (1999) found that children of mothers who give birth at 17 or younger were nearly four times more likely to have a report of sexual abuse by age 5. Children of mothers aged 18 to 19 were at nearly three times higher risk, and those whose mothers were 20 to 21 at birth were at close to two and a half times higher risk than children born to older mothers (Lee & Goerge, 1999). Other studies support this finding (Mersky, Berger, Reynolds, & Gromoske, 2009). Also children whose mothers have not completed high school are at one and half times the risk of having a maltreatment report (Dubowitz et al., 2011). Additionally, when mothers have been arrested the risk of maltreatment increases (Phillips & Erkanli, 2008). However, the relationships between education and criminality, and CSA specifically, are not clear.

Parental mental illness is associated with increased risk of child maltreatment (Chaffin, Kelleher, & Hollenberg, 1996). Having a parent with a mental health disorder approximately doubles a child's risk of physical or sexual abuse (Walsh, MacMillan, & Jamieson, 2002). Moreover, some studies also indicate parental mental health issues can exacerbate adjustment problems for children following a report of sexual abuse (Deblinger, Steer, & Lippmann, 1999). While it is possible that parents with increased mental health problems are in treatment, there is little research that specifically addresses the effects of parental mental health treatment prior to a report of abuse on longer-term outcomes of the child. One exception was a study conducted in the United Kingdom that found that a maternal history of psychiatric illness increased the risk of child maltreatment by 2.34. If the mother was engaged in mental health treatment during her childhood the risk of maltreatment was 3.65 times higher (Sidebotham & Heron, 2006). It is unclear whether or not maternal participation in mental health treatment, as compared to need for treatment, is associated with the risk of CSA.

Family poverty significantly increases the risk of childhood maltreatment (Chaffin et al., 1996). Children in poor families are three to seven times more likely to experience maltreatment (Sedlak et al., 2010). This is likely due to higher levels of family and neighborhood risk factors for maltreatment among poor families (Jonson-Reid, Drake, & Kohl, 2009). The risk of CSA specifically has also been found to be associated with poverty (Hussey et al., 2005; Sedlak et al., 2010).

Various community characteristics have also been found to be associated with maltreatment (Coulton, Crampton, Irwin, Spilsbury, & Korbin, 2007). Chief among these is neighborhood poverty. The higher the poverty rate in a community, the higher the risk is of maltreatment (Coulton, Korbin, Su, & Chow, 1995; Paxson & Waldfogel, 2002). Neighborhoods with the highest poverty rates have sexual abuse reporting rates four times higher than neighborhoods with the lowest poverty rates (Drake & Pandey, 1996).

The parent study sample was drawn from a Midwestern metropolitan area in 1993 and 1994 and included three groups: poor children with no child maltreatment reports or child welfare services (AFDC Only), poor children with investigated child abuse/neglect reports (CAN & AFDC), and non-poor children with investigated child abuse/neglect reports (CAN Only). All CAN subjects were reported due to alleged maltreatment for the first time in 1993 to 1994 from birth through 11 years. CAN subjects were matched to income maintenance data (then Aid to Families with Dependent Children or AFDC) to obtain recent poverty status. Those CAN cases without AFDC records became the CAN Only group. To obtain the poverty only comparison, children were matched by birth year and area of residence. Due to the difficulty in trying to match entire families and to ensure independence of observations, one child per family was randomly selected to be followed for the study. Subjects were followed through multiple administrative datasets with exact dates allowing for prospective analyses. Although sampling occurred in 1993 and 1994 based on the date of first maltreatment report, several of the datasets (e.g., birth records) contain information prior to the sampling period. Subjects were followed through 2009 and were age 16 through 28 at the end of the follow-up period. For this paper, data were limited to children with CAN reports who were at least age 18 at the end of 2009 (n = 5,344) and then further limited to children whose first maltreatment report contained allegations of CSA (n = 555). Prior analyses and studies have shown consistent differences between those with maltreatment reports and those without (e.g., Jonson-Reid et al., 2009) that are consistent with the extant literature regarding maltreatment and later outcomes.

Data include tract-level data from the 1990 U.S. census, birth and death records (for censoring), child welfare (specific subtypes of maltreatment reported, reporter source, investigation conclusions, in-home and foster care services), Department of Mental Health Medicaid and non-Medicaid programs, emergency room and hospital records, income maintenance data (AFDC and later TANF), juvenile corrections and juvenile court records, statewide Medicaid data including health hospitalization (ER) and inpatient and outpatient mental health treatment, and special education eligibility records. Data were received several times throughout the study allowing for tracking of multiple contacts within and across service systems. Mental health treatment records as well as crime and demographic characteristics were available for parents as well as the child subjects.

Linkage between most of the state level data sets was simplified by the presence of a unique common ID number. Other data were matched according to a combination of individual identifiers that were then reviewed manually. Coding decisions were made based on the relevant extant literature, such as attention to the need to aggregate at a sufficient level to protect confidentiality and a thorough understanding of the practices and policies related to the recording of data elements. Since data are administrative rather than survey or interview record, it is not possible to confirm whether data are missing relative to presence or absence in a given agency dataset or particular service within an agency. Basic descriptive information regarding system overlap and services within systems were checked with agency staff as well as extant literature to ascertain they were in the expected range. Children with missing age, initial location, or demographic information were excluded from the sample due to the need to link across data systems and geocode to link to census information. Less than 5% of the sample pool was lost due to this type of missing data.

For research question one, the dependent variable was a re-report of maltreatment measured by an alleged report made at least one week following the index report and prior to age 18. For research question two, the dependent variables were adolescent negative behaviors (i.e., runaway shelter use, problematic substance use as defined by a juvenile court petition, arrest or health treatment for drug or alcohol use prior to age 18, hospital care for a suicide attempt, a juvenile court petition or arrest for violent delinquency, and treatment for a STI not associated with initial report). Treatment for a STI as an outcome, used as a proxy for sexual risk behaviors, was limited to later treatment not coinciding with a report of CSA. Due to a small subsample size for adolescent outcomes by gender, these outcomes had to be collapsed into a variable called “any negative” in the multivariate analysis for research question two.

Data management and analyses were done in Statistical Analysis Software (SAS) 9.2. The main focus of the present analysis was how maltreatment recurrence and select outcomes examined varied according to gender among children from birth to age 11 reported for CSA. Analyses included bivariate chi-square, survival, and logistic regression analyses. Cox regression was used for multivariate analyses of a repeat report of maltreatment because of the capacity to model a dichotomous outcome, control for differing amounts of time at risk, potential clustering by census tract, and to censor cases lost due to death (Allison, 2010). While it is possible cases could have been lost due to movement, a large portion of the datasets used are statewide, allowing for greater coverage. Due to overlap in outcomes (some children experienced more than one) a competing risks approach was not used.

Time at risk (in years) was created by starting at the child’s date of birth. Children were censored (no longer at risk) at the time of an outcome, death, or when they turned age 18. Analyses were adjusted using robust standard errors due to potential intra-group correlation of subjects within a census tract. Bivariate survival analyses between each independent variable and dependent variable were utilized (not shown) to test the equality of survivor functions (log-rank test for categorical independent variables and simple Cox model for continuous independent variables). The estimated survivor functions from the bivariate tests were graphed to assess the assumption of proportionality. If a violation was found, a time interaction was created. The interaction term was tested in the multivariate model and was only retained if significant or if it altered the significance of other variables or the overall model fit. The Cox regression output includes a magnitude of effect called hazard ratios (exponentiated parameter coefficients). A hazard ratio greater than one and statistically significant indicates an increased risk, while a statistically significant hazard ratio between zero and one reflects decreased risk. For continuous variables, the hazard ratio is interpreted as the change per unit of measurement.

For any negative outcome other than recurrence, a logistic regression model was used since all subjects were at risk until age 18 and because some individuals had more than one negative outcome, but the sample size was insufficient to break up by type or timing. Further, some outcomes are more common for males or females (e.g., males have a greater likelihood of receiving a juvenile delinquency petition, but females were more likely to run away). Proc Survey logistic was used to adjust standard errors due to potential intra-group correlation of subjects within a census tract. Odds ratios are reported with the practical interpretation being the same as the hazard ratios.

Among children who had two or more reports (n = 259; 46.8% of sample), males were more likely to have a second report for maltreatment that did not include sexual abuse than females (75.7% versus 67.0%), but this difference was not significant. Most subsequent reports were for physical abuse or neglect. Initial CSA reports were far more common for females (n = 426) than males (n = 129), but the average number of maltreatment reports over time for females (1 to 14) was similar to the males (1 to 12) in this sample.

Table 2 illustrates the results of a Cox regression model of recurrence (Wald Chi-square = 119.85, df = 15, p < .0001). Children from a family receiving AFDC/TANF at baseline were over four times more likely to have a second report (HR = 4.34, p < .001). In the multivariate model, females were less likely to have a second report (HR = .64, p < .05). While the main effect for race indicated a higher re-report risk for Black children (HR = 1.94, p < .05), an interaction term indicates that this is only true for Black children who did not live in poor families at baseline (HR = .34, p < .01). If the alleged perpetrator was the caregiver’s paramour, the rate of recurrence was higher when compared to other perpetrator types (HR = 1.71, p < .01). Children whose alleged perpetrator was male were also more likely to have a subsequent report (HR = 1.57, p < .05). Children in families receiving family centered services only, compared to no services, were less likely to have a second report (HR = .64, p < .01). Children who were eligible for special education were also less likely to have a subsequent report (HR = .54, p < .001).

Among those initially reported for sexual abuse, 6.5% had at least one emergency room visit prior to age 18 due to violence (i.e., gunshot wound, assault, or rape), but not maltreatment. Among the sexually abused children in this sample, 2.5% spent time in a runaway shelter, 8.5% had a juvenile court record or a treatment record for substance abuse, and 4.5% had records of treatment for an STI (reported at a separate time than the CSA). Children whose first report of CSA was classified as severe (i.e., included penetration, STI, or sexual injury) were more likely to have at least one of these negative outcomes (30.6% versus 21.4%, p < .05). Males were more likely to have records of arrest or juvenile court petitions for a violent offense (13.8% versus 7.8%, p < .05), but only females had records of hospital treatment for suicide. There was no statistically significant difference in later runaway shelter use, substance abuse, or STI by gender.

The individual outcomes were collapsed into a variable called “any negative outcome” for multivariate analyses due to the small subsample sizes for some of the outcomes. Initial model building analyses indicated several interaction terms by gender, therefore, logistic regression models were constructed for males and females separately. Both models had good model fit statistics. The variables tested included those in the overall model of recurrence above as well as measures of recurrence such as type of maltreatment and number of subsequent reports.

The model for males had a statistically significant Wald Chi-square with a c statistic above .8, indicating that the model fit the data relatively well (Wald Chi-square = 44.6147, df = 8, p < .0001, c = .85). Among males, those first reported to CPS prior to age 5 were about four times less likely to have a negative outcome (OR = .25, p < .01). Males who were Black (OR = 4.09, p < .01) or had records of special education due to a learning disability or emotional disturbance (OR = 4.73, p < .01) had a higher likelihood of having a negative outcome. Finally, males who had a subsequent report for physical abuse or neglect were at greater risk of a negative outcome (OR = 4.02, p < .05) than those with no subsequent report, the coefficient for a re-report of CSA was not significant. There was no significant relationship between the characteristics of the initial report and later negative outcomes. Table 3 reports odds ratios and confidence limits.

The model fit statistics for females were similar to the male model (Wald Chi-square = 76.2017, df = 9, p < .0001, c = .80). While having had a record of severe sexual abuse was significant in bivariate analyses, once recurrence was controlled this was not significant and was dropped from the final model. Similar to males, Black females experienced a higher likelihood of a later poor outcome (OR = 3.30, p < .001). The main effect for having a report made prior to age 5 (OR = .23, p < .01) was similar to males, but for females there was an interaction term indicating that among poor females there was little variation according to the age of the first report (OR = 4.67, p < .05). The likelihood of a negative outcome increased with the number of reports (OR = 1.63, p < .001) rather than the type of subsequent report. There was no relationship between special education eligibility and later negative outcome, but females with histories of mental health treatment had a higher risk of negative outcomes (OR = 3.21, p < .001). Females born to older caregivers had a lower likelihood of later negative outcomes (4% reduction in risk for each additional year of increase in caregiver age). Although family size and having had a parent's paramour as the perpetrator were not statistically significant, their presence added to the model fit and also altered the effect of parent age. Interactions with parent age at birth were not significant, suggesting that the two variables served as suppressors. Table 4 presents odds ratios and confidence limits.

Typically CSA with increased physical intrusiveness (e.g., penetration) or injury (e.g., genital bleeding) is categorized as more severe than other forms of CSA (Young, Riggs, & Robinson, 2011). We found no association between severity and subsequent revictimization or later outcomes, once we controlled for general maltreatment recurrence. Regarding longer term outcomes, our findings differ from prior studies of suicidal behavior and STI (Dube et al., 2005; Senn et al., 2007), but this may be partly attributed to the need to collapse these outcomes along with others in the present study. Given the likelihood of negative outcomes in our study, however, our findings suggest that caution be used in assuming that such categorizations of severity should be focal criteria for intervening to prevent other negative outcomes.

While reports for sexual abuse are less common among males generally (Putnam, 2003), the revictimization patterns and subsequent negative outcomes were similar to females in this sample. While types of negative outcomes were collapsed in the multivariate model, only hospitalization for suicide had a bivariate association with gender.

There was some indication, however, that there were gender differences in risk factors for later negative outcomes. Similar to studies of the risk of CSA, our study found that poverty continues to influence risk through revictimization for both males and females. Among females, poverty persisted as a predictor of poor outcomes for those with first CSA reported prior to the age of five. For males the type of subsequent report was significant, but not the total number of reports, while the opposite was true for females. Among males indication of special education for a learning disability or emotional disturbance was associated with worse outcomes. It is not clear based on existing literature why the predictor variables varied by gender, but some speculation as to differences by child service system and maternal age follow. Regarding the findings relative to special education, it may be that these particular disability types are proxies for experiencing problems in school which may contribute to later negative outcomes. While special education status was not significant for females, contact with the mental health system was. It is possible that females are having contact with the mental health system rather than special education because of differences in how difficulties coping with the longer term consequences of CSA manifest for female and male victims. In other words, it is not clear whether there are gender differences in the type of system that responds to problems rather than the underlying need. Nationally, about 75% of children receiving special education for LD or ED are male (National Education Association of the United States & National Association of School Psychologists, 2007). In a study of children reported for maltreatment, mental health need did not vary by gender but access to mental health treatment did (Lindsey et al., 2012). Although Lindsey and colleagues (2012) found that girls were less likely to access treatment, their sample included all maltreatment types rather than being restricted to CSA. Additionally, other studies with high risk adolescent populations have found that female adolescents are significantly more likely to receive mental health treatment than males (Herz, 2001; Lopez-Williams, Stoep, Kuo, & Stewart, 2006).

Regarding maternal age, gender differences were found in studies with children of teen parents: with boys showing greater risk of multiple untoward outcomes, while girls’ risks were only related to early parenthood (Pogarsky, Thornberry, & Lizotte, 2006). It may be that such variation appears between males and females while another relationship is found within models for males and females. Further such studies are not focused on children reported for CSA. More research is needed to better understand the dynamics associated with CSA, gender, and other adolescent outcomes.

There are several limitations to this study. Due to the nature of administrative data, it is impossible to assess the need for services as compared to services received. On the other hand, it is of value to understand the known costs (as measured by system contacts) associated with CSA. Similarly, it is possible that some of the children lacking later reports, experienced unreported maltreatment. Children with repeat reports, however, did differ in terms of longer term outcomes. The subsample of males was relatively small, so it is not known whether the gender differences in later negative outcomes would hold in a larger sample. We were also unable to model the various negative outcomes separately due to sample size, however, in bivariate analyses only hospitalization for attempted suicide appeared to vary by gender. Because we could not assess child functioning prior to the first report of CSA, it is not possible to rule out other risk factors associated with the later outcomes that existed independent of CSA. More research with a larger sample and replications in other sites is needed.

Our findings indicate that recurrence among children initially reported for CSA is common and highly likely to include other forms of maltreatment. Many studies have found strong associations between recurrent maltreatment and worse later outcomes (Balsam, Lehavot, & Beadnell, 2011; Jasinski et al., 2000; Jonson-Reid et al., 2012; Lang, Stein, Kennedy, & Foy, 2004). This suggests caution in applying interventions that are narrowly focused on preventing CSA recurrence. Rather interventions should be aimed at preventing subsequent maltreatment of any type.

While younger age at first report of maltreatment is often thought to be a signal for greater concern (Finkelhor, 1993), our findings suggest this may vary by gender and poverty. Poverty also played a role in the form of later maltreatment recurrence. Families receiving AFDC at baseline in our data also lived in lower income neighborhoods. It may be that poverty among CSA victims both confers additional risks and restricts access to treatment resources. Both primary prevention and intervention planning should include a careful assessment of the resources available and target filling gaps associated with income.

While special education due to a learning disability (for males) and mental health treatment (for females) were risk factors in our models, this should not be taken to mean that these systems are playing a causal role in later negative outcomes. It was impossible to assess the level and quality of services provided to the children in the sample. Nor was it possible to assess the presence of cognitive or mental health needs prior to the first report of maltreatment. It is possible, however, that the addition of these other system contacts could serve as a signal for more intensive intervention for the individual child.

While much research attention has been paid to the subject of CSA, our study is one of the few to examine gender differences in outcomes prior to adulthood. While there is a tendency to equate sexual abuse with female gender, our analyses indicate that more research and intervention attention should be paid to males reported for CSA. While preventing initial maltreatment and any recurrence is always preferable, understanding such trajectories may be helpful to treatment providers who later interface with children with long histories of maltreatment.