Return to John Carter's Home Page

Evidence of a Link Between Maltreatment and Delinquency

A Literature Review


For over three decades, researchers have asked whether violence breeds violence. Many researchers have studied the links between child maltreatment and juvenile delinquency to evaluate the hypothesized "cycle of violence". Some studies demonstrate support for a linkage between the two phenomena, showing that abused and neglected children are at elevated risk of delinquency. Other studies have failed to confirm any relationship between the two phenomena. Some studies even suggest a negative correlation between maltreatment and delinquency. Much of the variation in study findings probably results from methodological and research design differences in the studies. (Widom & Maxfield 1996; Widom 1988; Widom 1989a)

Clinical Studies

The earliest investigations were basically case reports and case series. Most of these studies were conducted by clinicians using psychiatric patients or incarcerated individuals. Eason and Steinhilber examined the histories of eight boys who had committed murder. Five of these eight had been brutalized in childhood. (See Kakar's 1996 review) Duncan and Duncan analyzed the histories of six adult murderers and found that four of them had been severely beaten by both parents. The other two were psychotic and thus no histories could be ascertained. (See Kakar's 1996 review) In the 1970's and 1980's, researchers examined the histories of psychiatric patients and frequently concluded that the violent patients were much more likely to have been abused as children than the non-violent patients. (See, Kakar's 1996 review of Climent & Ervin 1972, Blount & Chandler 1979, and Monane et. al. 1984)

While these clinical studies were insightful, they did not establish a causal link between child maltreatment and delinquency. In addition, given the nature of their study populations, the results of these studies are not generalizable to the general populace. (Kakar 1996) Recognizing the limitations of the clinical studies, a number of researchers began conducting observational studies, examining the behavioral impacts of maltreatment.

Observational Studies

Some observational studies demonstrated that abused children exhibit more problem behaviors at earlier ages than did non-abused children. Abused infants more often ignored or refused maternal distractions. Abused toddlers assaulted their peers and harassed their caregivers more often than did non-abused toddlers. Abused children were found to be more aggressive in their fantasy and play and they were more emotionally maladjusted, especially in the development of their self-concepts. These findings offer support for the hypothesis that violence is a learned behavior passed on from one generation to the next. (See Kakar's 1996 review of Wassermann & Allen 1983, George & Main 1979, Reidy 1977, and Kinard 1980) On the other hand, other observational studies failed to discover any significant behavioral differences between abused and non-abused children. (See Kakar's 1996 review of Wassermann & Allen 1983, Friedreich et. al. 1983, & Straker and Jacobson 1981)

Cross-Sectional Survey Design Studies

In addition to observational studies, a number of researchers began conducting cross-sectional surveys of various groups of youths in the 1980's. Some of these studies considered only adjudicated delinquents, while others investigated youths from the general population. Some of the studies also surveyed the children's parents. All surveys gathered data about the youths' involvement in delinquency and their histories of maltreatment. Many of these studies found significant correlations between abuse and delinquency.

Eighty-six percent of the respondents in Mouzakitis' study of female adjudicated delinquents reported being physically punished with hands, objects or belts. Fifty-one percent recalled bruises, 25% recalled scars and 38% recalled bleeding from those punishments. (See Kakar's 1996 review) Geller and Ford-Somma also found significant abuse among the histories of delinquents. Sixty-six percent reported being beaten with a belt or extension cord; 32% reported repeated beatings; and 20% reported being threatened with a knife or gun. Thirty-three percent reported bruises from the beatings, while 29% reported bleeding and 8% required hospitalization. (See Kakar's 1996 review) Hotalling's general population based study also found that abused children engaged in more violent behavior than did non-abused children. Abused children more frequently assaulted their siblings, their parents and persons outside their families. (See Kakar's 1996 review)

Retrospective Case-Control Studies

Retrospective case-control studies also suggest a link between maltreatment and delinquency. Glueck and Glueck compared 500 institutionalized delinquents with 500 non-delinquents and found that 85% of the families of delinquents were abusive compared to 44% of those of non-delinquents. Lewis and Shanok compared 109 delinquents with 109 controls and concluded that 9% of the delinquents had received hospital treatment for abuse related injuries while only 1% of the non-delinquents had received such treatment. (See Kakar's 1996 review) Like the cross-sectional studies, these retrospective studies suggest that maltreatment and delinquency are related.

Although the findings of the survey design and retrospective studies add evidence supporting the association between maltreatment and delinquency, they cannot establish causality. The cross-sectional nature of the data collected in the surveys precludes the establishment of the temporal sequence. These data cannot elucidate whether maltreatment caused delinquency, delinquency caused maltreatment, or maltreatment and delinquency were both caused by a third factor. (Kakar 1996, Howing et. al. 1990) The retrospective studies go farther in establishing the temporal connection, but the connection may be an artifact of after-the-fact analysis or recall bias. (Kakar 1996; Widom 1988) Additionally, many of the retrospective studies investigate only those who became delinquent rather than a representative sample of all maltreated children. Therefore, these studies may estimate the probability of abuse, given delinquency, without accurately estimating the risk of delinquency, given abuse. (Kakar 1996)

Prospective Case-Control Studies

More recently, researchers have begun to utilize longitudinal prospective studies that follow individuals forward in time from the occurrence of maltreatment. Alfaro studied 4,465 children referred to New York State protection agencies for maltreatment between 1952 and 1953. He followed their juvenile court histories through 1967 and found that 42% to 64%(1) of the families reported for maltreatment had at least one child who was subsequently taken to court as delinquent or ungovernable. Although he did not utilize a control group, in one county Alfaro was able to compare the rate of juvenile delinquency and ungovernability in maltreated children with the rate in the general population. The rate for maltreated children was five times greater, 10% versus 2%. (Alfaro 1981)

Over the past decade, Widom has conducted a well-designed prospective study. (Widom & Maxfield 1996; Maxfield & Widom 1996; Widom & Kuhns 1996; Widom 1995; Luntz & Widom 1994; Widom & Ames 1994; Widom 1994; Widom 1992; Widom 1991a; Widom 1991b; Widom 1991c; Widom 1989a; Widom 1989b) She began with a cohort of 908 abused or neglected children who were processed through the juvenile or adult courts of a Midwestern metropolitan area between 1967 and 1971. Widom selected a matched control group for these children from the community. For school aged children, controls were drawn from the same school and were matched for sex, race, date of birth (plus or minus 6 months). For preschool children, controlswere drawn from the hospital of birth and matched for sex, race, date of birth (plus or minus one week). The goal of matching the children with controls from their schools or hospitals of birth was to control for socioeconomic status. Widom contends that, while not perfect, this procedure served as a good proxy. The community in question did not engage in busing and thus public schools in the community were relatively homogenous. Widom assumed that persons born in the same urban hospital during the 1950's through 1970's were relatively similar with respect to family income. The matching procedures produced 667 controls for a total of 1,575 study subjects.

After constructing the study population, Widom examined local, state and federal law enforcement records to determine which of the subjects had subsequently been arrested, either as juveniles or adults.(2) Widom categorized arrest information into juvenile arrests, adult arrests and violent offenses. Adult offenses included all non-traffic offenses committed by a person 18 years of age or older. Juvenile offenses referred to those committed by a person under 18 years of age. Violent offenses included the following crimes, whether committed by an adult or juvenile: assault, battery, robbery, manslaughter, murder, rape, and burglary with injury.

The arrest records were searched in 1987, 1988 and again in 1994. In 1988, the average age of subjects was 26 years. At that time, only 65% of the study population had passed through the peak years of committing violent offenses. By 1994, when the average age of subjects was 32.5 years, 99% of the population had passed through those years. (Maxfield & Widom 1996)

At all data collection points, the maltreated children were at elevated risk of arrest, both juvenile and adult. In 1988, 29% of the abused and neglected children had been arrested in comparison to 21% of the controls. By 1994, the percentages of arrests had risen for both groups. Forty-nine percent of the maltreated children had been arrested while 38% of the controls had been arrested. (Maxfield & Widom 1996) Abused and neglected children were also at elevated risk for violent offenses. Eighteen percent of the maltreated group had been arrested for a violent offense, in comparison to 14% of the control group. (Maxfield & Widom 1996) The odds that an abused and neglected child would be arrested as a juvenile were 1.8 times higher than for the matched controls. The odds for adult arrest and for arrest for a violent offense were 1.57 and 1.35 times higher, respectively, for the maltreated group. (Widom & Maxfield 1996)

The abused and neglected children also committed more offenses (mean of 2.43 versus 1.41) and were first arrested at earlier ages (mean of 16.48 versus 17.29). (Widom 1989b) However, maltreated children were no more likely than controls to continue criminal activities once they had begun. Unfortunately, substantial proportions of both maltreated and non-maltreated children who had a juvenile arrest went on to have an adult arrest (71% and 66%, respectively). (Widom & Maxfield 1996) These findings suggest that abuse and neglect increase a child's propensity to become involved in criminal activities, but that they do not increase the likelihood of continuing a life of crime once it has begun. (Widom & Maxfield 1996)

Although maltreatment appeared as a risk factor for subsequent arrest, the pattern of risk varied according to gender, race and type of maltreatment. For example, the increased risk for offenses was often more substantial for females than for males. The odds that maltreated males would be arrested for any offense, as an adult or a juvenile, were 1.57 times those of control males, whereas the odds of any arrest for maltreated women were 2.13 times those of control women. The odds that maltreated females would be arrested for a juvenile offense were 1.94 times greater than the odds that female controls would be arrested for such an offense, whereas the odds for maltreated males were 1.8 times that of control males. For females, the risk increased 82% (20% of maltreated females had juvenile arrests versus 11% of the control females). For males, the increase was still significant, 52%, but less pronounced (35% of maltreated males had juvenile arrests versus 23% of the control males). (Maxfield & Widom 1996)

The pattern of risk for violent offenses also varied according to gender. The 1988 data suggested that maltreated males were at elevated risk of violence, while females were not. When the 1994 data were collected, however, these findings were reversed. By that date, abused and neglected females were significantly more likely to be arrested for violent offenses than control females (Odds Ratio = 2.38 for any violent offense, as an adult or a juvenile). Maltreated males, however, did not significantly differ from male controls regarding the risk of violence. (Maxfield & Widom 1996)

Risk of violence also varied according to race. Thirty-four percent of the abused and neglected African-Americans were arrested for violent offenses, in comparison to 22% of the African-American controls (Odds Ratio = 1.81). For Whites, the percentages were not significantly different, 11% and 10% respectively. Thus, maltreatment was a significant risk for violence only among African-Americans, not among Whites. (Maxfield & Widom 1996)

Patterns of risk also varied in accordance with the type of maltreatment to which a child was exposed. Those who experienced physical abuse were most likely to be arrested for violent offenses at some point in their lives, either as an adult or a juvenile. Twenty-one percent of the victims of physical abuse were subsequently arrested for a violent offense (Odds Ratio = 1.9). Nearly as many neglected children (20%) were arrested for such offenses (Odds Ratio = 1.6). Victims of sexual abuse were the least likely (9%) to be arrested for violent offenses. In fact, compared to controls, sexual abuse victims were not at increased risk of violence.(3) (Widom & Maxfield 1996; Maxfield & Widom 1996)

The pattern of risk for juvenile violence was slightly different from that of any violent offense. For all juveniles, neglected children were the most at risk of violence. However, for females, the risk of juvenile violent offenses was greatest for those who had suffered physical abuse. (Widom 1994)

The patterns of risk also varied according to the category or type of offense. For example, overall, abused and neglected children were at increased risk for sex crimes, but the significance of the risk differed according to gender, type of maltreatment and type of offense. Sexually abused children were 4.7 times more likely than controls to be arrested for a sex crime. Physically abused children were 4.1 times more likely and neglected children were 2.2 times more likely than controls to be arrested for such offenses.

Within the category of sex crimes, the patterns were even more varying and complex. Sexually abused children and neglected children had significantly elevated risks for prostitution in both bivariate and multivariate analyses (Odds Ratios = 27.7 and 10.2, respectively). Physically abused children experienced significantly increased risk in bivariate analysis, but when demographic characteristics and approximate socioeconomic status were controlled for, the relationship became insignificant. (Widom 1995;Widom & Kuhns 1996) When the analysis was done separately for each gender, the pattern of increased risk for prostitution remained for females, but not for males. (Widom & Kuhns 1996) For rape and sodomy a different pattern appears. For these crimes, only those children(4) who suffered physical abuse were at significantly increased risk of arrest. These children were 7.6 times more likely than controls to be arrested for such offenses. (Widom 1995) These findings suggest that while all maltreated children are at increased risk of arrest for sex crimes, sexually abused and neglected children are more at risk for prostitution and physically abused children are more at risk for rape and sodomy arrests. (Widom & Ames 1994)

Within the category of juvenile offenses, the patterns of risk also varied. The risk of arrest was not significant for all types of offenses. Maltreatment was most strongly associated with property crimes and status offenses, such as truancy, runaway and ungovernability. Only these types of juvenile offenses (as opposed to violent, drug, alcohol, sex, and order(5) offenses) were significant at the .05 level. (Widom 1991)

Although Widom's findings support the hypothesis that maltreatment leads to subsequent arrests, it is worth noting that race and gender were stronger predictors of violence than childhood abuse and neglect. To assess the relationship between demographic variables, type of maltreatment, and arrests for violent offenses, Widom performed multivariate analysis. This analysis demonstrated that the odds ratio for physically abused children was 1.9 and for those who suffered neglect, it was 1.6. The odds ratios for male gender and African-American race were substantially higher; 6.4 and 4.1 respectively.(6) (Maxfield & Widom 1996)

Zingraff et. al. also used a prospective design to examine the connection between maltreatment and delinquency. These researchers randomly sampled one in three substantiated cases from the child abuse and neglect registry of an urban North Carolina county during the period 1983-1989. They compared this maltreated group with two random comparison samples, one drawn from the county's public school population and one from the county's welfare caseload. Unlike Widom, Zingraff et. al. did not rely upon court records to determine maltreatment status. Instead, they used substantiated social service agency records. Thus, they used a broader definition of maltreatment and included incidents that may never have reached the court's attention. (Zingraff et. al. 1993) Zingraff and his colleagues also used a measure of delinquency different from that selected by Widom. Widom relied upon arrest records to assess delinquency whereas Zingraff et. al. relied upon court records. Specifically, delinquency was identified via complaints filed with the juvenile court. Since 30% to 40% of police-juvenile encounters are never reported to juvenile court, Zingraff and associates probably utilized a more restrictive definition of delinquency than did Widom. Their definition did allow for delinquent acts that came to the attention of the court via means other than police contact, however, whereas Widom's measure would not have captured such offenses. (Zingraff et. al. 1993)

Zingraff et. al. conducted separate analysis for three types of maltreatment: physical abuse, sexual abuse and neglect. In addition, they attempted to assess the impact frequency of maltreatment had on subsequent delinquency. The frequency measure they selected was dichotomous; one substantiated case or more than one substantiated case of maltreatment. Zingraff and associates believed this was an adequate measure since only four children had three substantiated cases of maltreatment.(7) In their analyses, these researchers also controlled for age, sex, race and family structure.(8)

Zingraff et. al. concluded that in comparison to the school based sample, maltreated children were at elevated risk of delinquency (14% of maltreated children had juvenile court complaints versus 5% of school children). However, this rate of delinquency is less than a third the rate Widom detected in her study. (Maxfield & Widom 1996) Additionally, Zingraff and his associates note that most of the difference between maltreated children and the school children was explained by an increased risk for status offenses. There were no significant differences between the groups for either violent or property offenses; whereas Widom found significant differences for status offenses and property crimes. (Widom 1991) When Zingraff et. al. compared the maltreated children to the impoverished children, they found a similar pattern but one of less magnitude. Overall, 9% of the impoverished children were referred to the juvenile court on delinquency complaints. After controlling for age, sex, race and family structure, this difference became insignificant. A significant relationship remained only for status offenses. (Zingraff et. al. 1993)

Zingraff and his associates' assessment of the variations in the risk of delinquency according to maltreatment type revealed that only neglect had an independent effect on risk of delinquency. When frequency (as defined by Zingraff et. al.) was controlled for, however, neglect was no longer statistically significant as a risk factor. Thus, unlike Widom, Zingraff et. al. concluded that type of maltreatment was not particularly predictive of any offense. (Zingraff et. al. 1993)

Zingraff and his colleagues conclude that prior research, including Widom's study, overestimated the strength of the relationship between maltreatment and delinquency. Perhaps they are right. Perhaps, however, these differences are due, at least in part, to the differences in definitions of maltreatment and delinquency. Widom used a narrower definition of maltreatment and a broader definition of delinquency. Thus, it is not surprising that she found a more significant relationship between the two phenomena. The differing findings may also be due to the varying lengths of follow-up. Consider that Widom's findings regarding the connection between maltreatment and violence changed over the course of the follow-up period. For juveniles, the relationship was not significant. It was only in the adult years, after the youths had passed through the peak years for committing violent crime, that the relationship became significant, and then, only for females. (Maxfield & Widom 1996; Widom 1991) Perhaps the results of Zingraff et. al would look different if they had followed the children until age 32.5, as did Widom, rather than until age 15.

Most recently, Smith and Thornberry conducted a prospective study on the relationship between maltreatment and delinquency. Unlike Widom and Zingraff, these investigators did not start with a group of maltreated children and construct a comparison group. Instead, they interviewed 1,000 students in a general school population, as well as their primary caretakers, over a period of four and one-half years. The interviews began when the youths were attending seventh or eighth grade (average age 14) and continued at six month intervals until they were in eleventh or twelfth grade (average age 17). These interviews elicited self-reported prevalence and frequency of delinquent acts. The investigators categorized the offenses as: general delinquency (an index of both minor and serious offenses); serious delinquency (an index of violent and property offenses, including armed robbery and burglary); moderate delinquency (an index of moderately serious offenses such as joy-riding and simple assault); minor delinquency (an index including minor theft and being loud and rowdy in public); and violent delinquency (an index of violent offenses). In addition, Smith and Thornberry reviewed records from the schools, the police department, and the department of social services. These records served as the basis for determining maltreatment and official delinquency. (Kelley, Smith & Thornberry 1997; Smith & Thornberry 1995).

Smith and Thornberry, like Widom and Zingraff, found a significant relationship between maltreatment and the prevalence of official delinquency. Forty-five percent of the maltreated students had an arrest compared to 31.7% of the non-maltreated students. This relationship remained significant after controlling for sex, race/ethnicity, underclass status, family structure and mobility.(9) With regard to the prevalence of self-reported delinquency, the investigators found significant bivariate relationships between maltreatment and all five categories of delinquency. However, after controlling for the variables described above, only the relationships between maltreatment and moderate and violent delinquency remained significant. (Smith & Thornberry 1995)

Smith and Thornberry also examined the relationship between maltreatment and the frequency of delinquent offenses. As with the prevalence of official delinquency, the researchers found a significant relationship between maltreatment and the frequency of official delinquency, a relationship that remained when the control variables were held constant. The analysis of the frequency of self-reported delinquency showed a more complex pattern. All bivariate relationships were significant, except the relationship between maltreatment and minor delinquency. For example, maltreated students reported nearly twice as many violent and serious offenses as did the non-maltreated students. When the control variables were held constant, maltreatment continued to exert a significant impact on serious, moderate and violent delinquency, though not on general or minor delinquency. These findings suggest that maltreatment is a consistent predictor of both the prevalence and the frequency of official, moderate self-reported and violent self-reported delinquency. It is also a significant predictor of the frequency, though not the prevalence, of serious self-reported delinquency. It does not appear to be related to self-reported minor or general delinquency. (Smith & Thornberry 1995)

Although Smith and Thornberry did not separate their analyses according to type of maltreatment, they did examine the relationship between severity of maltreatment and delinquency. To assess whether delinquency increased with an increase in severity of maltreatment, the investigators divided the sample of maltreated children into two groups (the top one-third and the bottom two-thirds) along four severity measures; number of maltreatment incidents, severity of all incidents, duration of incidents, and number of subtypes of maltreatment across all incidents. On each of these measures, there were significant relationships between severity and official delinquency. The most-maltreated groups had significantly more arrests than did the less-maltreated groups and the less-maltreated groups had significantly more arrests than did the non-maltreated groups. For self-reported delinquency, the results are less robust. There was an incremental increase of serious and violent crimes between non-maltreated and less-maltreated students and between less-maltreated and most-maltreated students. However, the only significant increases were between the non-maltreated and the most-maltreated, not between the less-maltreated and more maltreated students. (Smith and Thornberry 1996; Kelley, Smith and Thornberry 1997) These findings offer some support for the contention that the greater the exposure to maltreatment in childhood, the greater is the risk of delinquency. However, the support is somewhat equivocal.


Reviewing the literature on the link between maltreatment and delinquency suggests that the relationship is complex. Most of the studies, and all of the prospective, case-control studies, demonstrate a significant relationship between childhood maltreatment and the subsequent development of delinquency. However, the strength and patterns of this relationship vary according to gender, race/ethnicity, socioeconomic status, type of maltreatment, and type of delinquency. It also worth noting that many maltreated children do not go on to engage in delinquent or criminal acts. In Widom's analyses, which showed the strongest relationship between maltreatment and delinquency, close to half of these children ultimately experienced some arrest, but just over half did not. Thus, delinquency is not an inevitable consequence of maltreatment. The key may be to ascertain what factors or interventions offer protection against the development of delinquency.

Methodological Considerations For The Current Study

The complexity of the relationship suggested by the research to date, may also be, at least in part, a reflection of the methodological differences in these studies. For example, each researcher defines maltreatment and delinquency differently. Some use social service agency or court records of maltreatment, while others rely upon unsubstantiated, self-reports of maltreatment. Substantiated reports may be more reliable, but they are likely to underestimate the true prevalence of maltreatment since much maltreatment goes unreported. (Widom 1988) Researchers also vary in their decisions to combine or disaggregate the types of abuse. Those who choose to aggregate all types of abuse and neglect into one category of maltreatment may be masking the true relationship between certain types of maltreatment and delinquency. (Howing et. al. 1990; Widom 1988) Consider the variations that Widom discovered in her research.

Investigators also differ in their definitions and measurements of delinquency. Some rely upon official contacts with police, while others are even more limited, utilizing only arrest or court records. Still other investigators utilize self-reported delinquency measures. The significance of these varying definitions is suggested by the fact that researchers who utilized more than one measure of delinquency found different results depending on which definition they used. (See, Smith and Thornberry 1996; Kelley, Smith & Thornberry 1997)

The variation in existing studies may also be a reflection of the differing lengths of follow-up the studies employed. Recall that the results Widom found changed over the 27 years of her investigation.

Given the existing research, what methodological strategies should we undertake? There are three categories of decisions we need to make. The first category deals with decision about definitions we should select. The second category deals with decisions about what control variables to utilize. The third category deals with decisions about length of follow-up.


In selecting our definition of maltreatment there are several issues we should address. It may be that our data source limits us to relying upon substantiated court records. If so, we need to proceed with the knowledge that our definition of abuse will miss many cases of maltreatment that never come to the attention of authorities or that never get referred to court by social service agencies or police departments. (Zingraff et. al. 1993) We must also keep in mind that the cases we do include based on court records will be biased towards the more serious end of the continuum of maltreatment. (Widom 1988; Zingraff et. al. 1993)

Whatever our source of data, we must decide how to categorize it. Will we combine all types of maltreatment together and compare children with any form of abuse or neglect only with children with no history of maltreatment? Alternatively, will we separate the maltreatment into subgroups along the lines of Widom and Zingraff? Widom found distinctly different risk patterns for each type of maltreatment, whereas Zingraff and his colleagues did not. (See also, Brown 1984 finding that emotional abuse and neglect were positively correlated with self-reported delinquency in ninth grade students but that physical abuse was not) If different patterns do exist, combining all types of abuse into a single maltreatment variable may lead to an underestimation of the relationship of maltreatment and delinquency by combining types of abuse that have no relationship to delinquency with those that have a strong relationship to delinquency. (See, Howing et. al. 1990 for a discussion of the problems of lumping all types of maltreatment into a single variable)

If we decide to deconstruct the category of maltreatment into separate maltreatment types, we will need to decide how to do so. Will we be able to tell from our data source the nature of the maltreatment each child suffered? Some researchers, such as Smith and Thornberry, were able to make their classifications by reviewing social service agency records and rankings them with Cicchetti and Barnett's classification scheme. This scheme employs seven types of maltreatment; physical abuse, sexual abuse, emotional maltreatment, moral/legal maltreatment, educational maltreatment, physical neglect, and lack of supervision. Each of these categories is further assessed for severity on a five point scale.(10) (Smith & Thornberry 1995)

If our data source permits disaggregation of the maltreatment category and we decide to separate the children into subtypes of maltreatment, we will need to decide how to classify those children who experience more than one type of abuse. Should we create one category of "multiple types" of abuse or different combinations such as, "sexual abuse and neglect," "physical abuse and neglect," "physical abuse and sexual abuse"? To some extent these decisions may be dictated by the data, but it is worth considering because there may be different risks of delinquency for different combinations of maltreatment. Widom, for example, found that in comparison to other types of maltreatment, children who were sexually abused in addition to some other form of maltreatment were at greatest risk for running away. (Widom 1995, Widom & Ames 1994)

We also need to decide whether to include maltreatment at the hands of someone other than the child's primary caregiver. Some researchers have restricted their definitions of abuse and neglect to that suffered in the child's home environment. (See, Widom's 1988 review of methodological variations) There may be different impacts on a child abused at a day care center or school, a child abused at home and a child abused by a stranger. Some researchers suggest, for example, that the effect of sexual abuse by a family member is greater than sexual abuse by a stranger, while others have found no significant differences. (See, Widom & Ames' 1994 review)(11) On the other hand, if we limit our definition of maltreatment to that suffered at the hands of primary caregivers, will our results be generalizable to all maltreated children? (Widom 1988) If we do include maltreatment by anyone in our definition, will we separately analyze maltreatment according to the child's relationship to the perpetrator?

Finally, we need to decide whether to segregate maltreatment according to severity and frequency. If we combine all levels of severity and frequency, we may underestimate the relationship between maltreatment and delinquency. The insignificant effects of "minor" or infrequent maltreatment may negate the significant effects of severe, frequent maltreatment. (Smith & Thornberry 1995) Smith and Thornberry found some evidence to suggest that the magnitude of risk for official delinquency increased with the severity and frequency of the maltreatment, but the results were equivocal for self-reported delinquency. (Smith & Thornberry 1995; Kelley, Smith & Thornberry 1997) Widom compared two groups of sexually abused children, those whose abuse involved penetration and those whose abuse did not, since prior research suggested that more severe symptomatology was related, among other factors, to vaginal or rectal penetration. Widom's research found no significant differences between the two groups in terms of later criminal consequences, however. (Widom & Ames 1994) The benefit of assessing severity and frequency of maltreatment, if our data allow such an appraisal, is that separate analyses of differing levels of severity and frequency will allow us to evaluate whether there is a dose-response type relationship between maltreatment and delinquency.

In addition to definitional decisions about maltreatment, we have a number of decisions to make about delinquency. Will our sole source of data be court records or will we be able to assess official contacts with police or social service agencies, arrest records or self-reported delinquency?

If we rely only upon court records, we will need to remember that this is a relatively conservative estimate of subsequent delinquency. (Zingraff et. al. 1993, Smith & Thornberry 1995) What court records are available to us? Do we have records only of adjudicated delinquents or do we have access to referrals, as did Smith and Thornberry? If so, these records would provide a broader definition of delinquency than adjudication records.

As with maltreatment, we need to decide whether or not to disaggregate the offenses within the category of delinquency. Should we combine status offenses, property offenses and violent offenses into a single variable? Perhaps we would like to be able to tell whether maltreatment leads to truancy and runaway types of offenses or to armed robberies, assaults and rapes. Howing and her colleagues argue that including status offenses in the overall measure of delinquency increases the likelihood of misclassification. They point out that there is a great deal of discretion in the juvenile justice system and that status offenses such as running away may actually be markers of maltreatment rather than delinquency. (Howing et. al 1990; see also, Smith, Black & Weir 1980 for a discussion of the lack of guidelines and the resulting inconsistencies in the juvenile justice system) If we conduct separate analyses for each type of offense, we may reveal a truer picture of the relationship between maltreatment and delinquency.

We also need to decide whether to segregate delinquency on the basis of frequency. Will we characterize delinquents who have committed a single offense the same way we categorize delinquents with 25 offenses? Recall that Smith and Thornberry found that maltreatment predicted the prevalence and the frequency of self-reported delinquency differently.

In addition to carefully considering and making explicit our definitional choices, we need to be concerned with several potential areas of misclassification. First, we need to deal with the possibility that we will misclassify children, either in terms of delinquency or in terms of maltreatment, because they have moved into or out of the metropolitan area. (Kakar 1996; Smith & Thornberry 1995) If for example, a ten year-old boy appears in Fulton County Juvenile Court because he has been neglected, we will classify him, correctly, as maltreated. We will then follow him in the court records, looking for delinquency. If this child moves to South Carolina during the follow-up and commits armed robbery, we will misclassify him as non-delinquent because we are relying solely on local court records for evidence of delinquency. Similarly, we may select a potential control and search our five county record base to be sure she is not maltreated. If this child moved from Alabama where she was repeatedly sexually assaulted by her mother's former boyfriend, we will misclassify her as non-maltreated.

Smith & Thornberry tried to assess the level of potential misclassification resulting from immigration into the community by creating a dummy variable of mobility for those children who had not been in the community public schools since first grade. Their analysis did not differ when they controlled for this variable and they concluded that the problem was not significant. (Smith & Thornberry 1995) However, this variable does not seem to account for abuse that may have occurred to children who were mobile before the first grade. Smith & Thornberry also tracked and interviewed students who moved to another school district, to minimize misclassification due to subsequent relocation. (Smith & Thornberry 1995)

We also need to develop mechanisms to minimize the risk of misclassification resulting from both immigration to and emigration from the metropolitan Atlanta area. We might exclude from the study all children who have not lived in the area their entire lives, if our data source has information that would allow us to determine who meets that criterion. On the other hand, with the mobility of the Atlanta population, such exclusion criteria may make it difficult to construct an adequate sample population. We also need some mechanism for ascertaining when study subjects have moved out of the metropolitan area.

A second form of misclassification may occur due to mistaken assessment of the temporal order of maltreatment and delinquency. For example, a child who has been abused or neglected for a number of years may not come to the attention of authorities until he or she gets into trouble with the law.(12) If we are relying solely on official records to determine maltreatment and delinquency status, we will misclassify this child as delinquent, with no history of maltreatment. Misclassification between maltreatment and delinquency may also occur in the opposite direction. A child's delinquency may precede maltreatment. (See, Howing et. al. for a theory postulating that both delinquency and maltreatment arise from the interaction of the child's personality, parental inadequacies, and environmental factors) If the delinquency is relatively minor, it may not come to the attention of authorities until the parents respond with abusive behavior. If we rely on official records for classification purposes, we will erroneously identify this child as maltreated and subsequently delinquent. To minimize the risk of classifying a child as maltreated when he or she is delinquent first, many researchers take the approach used by Widom and require the maltreatment to occur before a certain age (11 for Widom, 12 for Smith & Thornberry, and 9 for Zingraff et. al.). (Widom 1989a) The assumption seems to be that delinquency cannot begin before a certain age.

Widom's approach does little to address the first temporal mistake, however. We need to minimize the number of children we misidentify as delinquent, without a history of maltreatment, who are actually children who were maltreated without detection of the authorities. If resources permit, we could interview all children in the study to gather self-reported assessment of maltreatment. We could then use that data to classify the children. Alternatively, we could exclude or separately analyze any children whose self-reports differ from official records.

Control Variables


The first variable for which we need to control is age. Research consistently demonstrates a positive relationship between age and delinquency throughout adolescence. The relationship exists whether the measurement of delinquency is drawn from official arrest records, social service agency processing records, or self-reports. (Zingraff et. al. 1993; Smith & Thornberry 1995; Widom 1991c)


Sex and delinquency also consistently appear to be related. In both official and self-reported data, males have higher incidence and prevalence rates than females, especially for more serious offenses. (Zingraff et. al. 1993; Smith & Thornberry 1995; Widom 1991c)


The third variable for which many researchers control is race and/or ethnicity. (Widom 1991c) The relationship between delinquency and this variable is more questionable than that of age and sex. Most official delinquency measures show a strong relationship between race/ethnicity; specifically, that non-whites have higher rates of offending. However, this disparity does not appear in most self-reported measures of delinquency, which suggests that the difference may be more of an artifact of reporting than a true difference in incidence and prevalence. (Zingraff et. al 1993; Tracy, Wolfgang & Figlio 1990) Nonetheless, given that we will be utilizing official records as our data source, it will be important to control for race/ethnicity.

Socioeconomic Status

In addition to controlling for age, sex, and race/ethnicity, most researchers try to control for socioeconomic status (SES). SES has been associated both with maltreatment rates (Brown 1984; Smith & Thornberry 1995; Abrams 1995; Pelton 1978; Pillitteri et. al. 1992) and delinquency rates (Brown 1984; Yoshikawa 1994; but see, Loeber & Dishion's meta-analysis concluding that SES is not a strong predictor of delinquency). The association with either phenomena has been marked by controversy, however. Many contend, as with race/ethnicity, that the discrepancies noted according to class are artifacts of reporting. They contend that lower SES individuals are under closer scrutiny of public authorities and thus are more likely to be reflected in official records of either maltreatment or delinquency.(13) Others contend that the differences are true differences that result from the increased stresses of living in impoverished environments.

Brown attempted to disentangle the effects of SES and maltreatment on the development of delinquency. His cross-sectional survey of ninth grade students in small southeastern town (population of 5,000 to 10,000) assessed six measures of social class (mother's education, father's education, mother's occupation, father's occupation, home ownership, and receipt of welfare benefits), three types of maltreatment (child's self-report of neglect, physical abuse and emotional abuse), and four types of delinquency (child's self-report of offenses against persons, property offenses, status offenses, and traffic offenses). Brown found significant negative correlations between maltreatment and SES, significant positive correlations between maltreatment and delinquency, and weak, though significant, negative correlations between class and delinquency. (Brown 1984)

Zingraff and his colleagues controlled for SES by selecting two control groups, one drawn from the school system in the community and one from a population receiving services (other than child protective services) from social service agencies in the community. Ninety percent of the latter group were receiving poverty-based services. Thus, the second control group represented a sample of impoverished, non-maltreated children. While Zingraff and associates found significant differences in the overall delinquency rates of maltreated children and non-maltreated, school children, when other background factors were held constant, the differences in overall delinquency between the maltreated children and the non-maltreated, impoverished children were insignificant. Only rates of status offending differed.(14) (Zingraff et. al. 1993)

Widom used matched samples drawn from the same neighborhood schools and hospitals to minimize the potential confounding of SES. (Widom & Maxfield 1996) Smith & Thornberry controlled for SES via a dichotomous variable of "underclass status." A child was classified as underclass if the principal wage earner in his/her family was unemployed, his/her family received welfare benefits or his/her family had an income below the federal poverty guidelines. (Smith & Thornberry 1995). Neither Widom nor Smith and Thornberry reported separate results according to class. Instead they held SES constant as they analyzed the effects of other variables.

Despite the controversial nature of the role of SES in maltreatment and delinquency, we should try to control for this variable. If poverty contributes significantly to the development of delinquency (or to maltreatment), policy makers need to know that. We cannot do much to reduce these problems if we are not addressing the actual contributing factors. Additionally, even if SES is not related to the occurrence of maltreatment or delinquency, but only to the reporting of these phenomena, we need to control for this variable since we are utilizing official records as our data source.

Family Functioning

A fifth control variable commonly employed in research investigating the link between maltreatment and delinquency involves family functioning. Researchers postulate that components of family life, besides the maltreatment itself, may influence a child's decisions regarding delinquency. Prior research consistently demonstrates that delinquency and poor parental supervision and discipline are related. (Patterson & Stouthamer-Loeber 1984; Van Voorhis et. al.; Loeber & Dishion 1983; Hoge, Andrews & Leschied 1996; Buka & Earls 1993) Research also indicates that rejecting or hostile parental-child relationships are associated with delinquency. (Hoge, Andrews & Leschied 1996; see also Youshikawa's 1994 review and Buka & Earls' 1993 review) Other research finds significant relationships between a families' involvement in criminal activities or experience with alcohol, drug and mental health problems and a child's risk of delinquency. (Dembo et. al. 1992; see also, Loeber & Dishion's 1983 review of early predictors of male delinquency; Buka & Earls 1993 review of early determinants of delinquency and violence and Widom's 1994 discussion of how substance abuse and maltreatment are intertwined; but see, Hoge, Andrews & Leschied 1996 finding that parental criminal history, mental health and substance abuse problems are not significantly related to delinquency) Still other investigators find relationships between family structure and delinquency. For example, large family size has been associated with delinquency. (Loeber & Dishion 1983; see also, Buka & Earls' 1993 review) Some researchers also postulate a relationship between broken homes and delinquency, but more detailed analysis suggests that marital/family conflict is the actual predictor of delinquency in such families. (Yoshikawa 1994; see also, Buka & Earls' 1993 review and the discussion of Van Voorhis et. al. infra)

On the whole, the research to date indicates that family functioning plays a significant role in the development of delinquent behavior. Unfortunately, just how this occurs is unclear. Thus far, the best predictors of delinquency have not been single components of family interaction, but composite measures of family functioning. (Loeber & Dishion 1983; Van Voorhis et. al. 1988) Despite the evidence that family functioning is a complex variable that needs composite measurement, many investigations of the link between maltreatment and delinquency have relied solely upon family structure measures to assess and control for family functioning.

Smith & Thornberry, for instance, employed a dichotomous variable differentiating between children who lived with both biological parents and all others. (Smith & Thornberry 1995) They utilized this information to control for family situation but did not report separately on the relationship, if any, between family structure and delinquency.

Zingraff et. al. developed four categories of family structure: children living with both biological parents; children living with one biological parent and a stepparent; children living with only one biological parent; and children living in any other family situation. They found that children living in blended families (those with a biological parent and stepparent) and those living in "other" family arrangements were more at risk of status offenses than those living in biologically intact families. Children living in single parent homes were not at increased risk for any offense. (Zingraff et. al. 1993)

In her primary analysis, Widom did not control for family functioning. However, in a separate analysis assessing the avoidance of criminality among maltreated children, she compared maltreated children who became delinquent with those who did not. In this analysis, Widom attempted to assess the contribution of family functioning to the development of delinquency and criminality. Most of the measures she utilized were structural, however. By abstracting juvenile court records, she identified whether the study population's parent(s): had criminal histories, had alcohol or mental health problems, were employed, were dead, were separated, and whether their whereabouts were unknown. She also determined whether the child's family: moved frequently, was separated, was composed of children of different fathers and whether the child was illegitimate. Widom did not attempt to assess the quality of family interactions. For example, she had no way of assessing whether a mother's employment, mental illness or criminal history influenced her parenting skills.(15)

Few of the characteristics Widom was able to measure differentiated the two groups. In bivariate analysis, only having a mother with a criminal history or a mother who was employed were factors significantly associated with arrest. In multivariate analysis, four variables were significant: mother dead, mother employed, father alcoholic, and living with neither parent. Having an employed mother or an alcoholic father increased the risk of subsequent arrest, whereas having a deceased mother or living with neither parent decreased the risk of criminality.(16) (Widom 1991c)

Perhaps the reason Zingraff et. al. and Widom did not find family functioning to be a substantial predictor of delinquency has to do with the measures they selected. Research indicates that structural variables often fail to adequately measure family dysfunction. (Van Voorhis et. al. 1988)

Van Voorhis et. al. compared the predictive value of family structure with family functioning measures to assess the accuracy of family structure as a proxy for quality of family life. These investigators measured family structure in two ways. First, they compared children living in intact homes (those where both biological parents were present) with children living in broken homes (those where one or both biological parents were absent through death, desertion, divorce, separation, or prolonged institutional confinement). Van Voorhis and associates also conducted an analysis in which they defined intact homes as those with two biological parents or one biological parent and a stepparent. They measured family functioning by interviewing children on five scales: enjoyment of the home (pride in having friends over, enjoying home, and characterizing one's home life as "better than most kids"); supervision (parents have a say in what child does and child must have approval to go out at night); abuse of children (reports of parental physical abuse); conflict (arguing between parents, arguing between parent(s) and child, and tension in the home); affection (parents care for child, parents express love for child, child feels close to mother, child feels unloved at times, child does not feel love towards parents at times, parents know that child cares for them and child tells parent he/she cares for them). In addition to these five scales, Van Voorhis et. al. constructed a general family quality index consisting of all of the above items. (Van Voorhis et. al. 1988)

Van Voorhis and colleagues concluded that family structure variables were not correlated with the overall measure of family functioning and had only a weak relationship to one of the specific measures, supervision. Their analysis also demonstrated that the bivariate relationships between broken homes and delinquency were not significant, except for a modest relationship between such homes and status offenses. On the other hand, significant bivariate relationships were detected between all measures of family functioning and both overall delinquency and status offenses. The strongest relationships were noted for affection, supervision and the overall index of family quality. These family factors also had moderate relationships with drug and property offenses. Child maltreatment and overall family quality had weak, but significant, associations with violent offenses. When Van Voorhis and associates conducted multivariate analyses, they found that overall family functioning and gender, not family structure, were the best predictors of delinquency. No particular component of family functioning was more predictive than the index of overall functioning, however, so this research did not suggest that any particular segment of the familial interactions was crucial in the development (or non-development) of delinquency. (Van Voorhis et. al, 1988)

Given the state of the research on the relationship between family functioning and delinquency, we need to select a measure of functioning more sophisticated than one that merely assesses with whom the child lives. What this measure should be will depend, in part, on the data in the court records to which we have access. Perhaps we can measure variables such as the ones Widom analyzed and then construct a composite measure from those variables that better assesses family life than the individual variables do.

Childhood Problem Behaviors

In addition to age, sex, race/ethnicity, socioeconomic status and family functioning, we need to control for behavioral problems in the child. The exhibition of behavior problems at an early age has consistently been one of the best predictors of subsequent delinquency. Children who are aggressive, who steal and who are dishonest in late childhood and early adolescence (the research in this area examines behaviors of children between 6 and 15) are at increased risk of subsequent delinquency. (Loeber & Dishion 1983; Buka & Earls 1993, Howing et. al. 1990) In Widom's comparison of maltreated children who subsequently engaged in criminal conduct and those who did not, demonstrating childhood behavior problems was the only child centered characteristic that significantly predicted criminality. This relationship was significant in both bivariate and multivariate analyses. Children who had behavior problems were 8.16 times less likely to avoid arrest than were children without behavior problems. (Widom 1991c) If possible, therefore, we need to measure and control for behavior problems in our sample population.

Cognitive Functioning

Cognitive functioning and IQ are also variables for which we should consider controlling. Low IQ, poor school performance, learning disabilities, attention deficit hyperactivity disorder and impaired cognitive functioning have all been associated with the development of delinquency. (Buka & Earls 1993; Loeber & Dishion 1983; Yoshikawa 1994) Widom attempted to assess the contribution of these factors in her avoidance of criminality study by comparing the rates of mental handicap among the maltreated children who were subsequently arrested and those who were not. She did not find significant differences between the two groups. (Widom 1991c) Nonetheless, given other studies' findings of a relationship between cognitive functioning and delinquency, we should consider controlling for this variable it we are able to gather data to do so.

Medical Conditions

We should also consider controlling for certain medical conditions. Several prenatal and perinatal complications have been associated with the development of delinquency. Prematurity, low birth weight, and anoxia have demonstrated relationships with problem behavior in childhood and subsequent delinquency. (Yoshikawa 1994; Buka & Earls 1993) Some theorists postulate that the prenatal and perinatal stresses impact the central nervous system and set the stage for cognitive problems and attention deficit disorder, which, in turn, are linked to delinquency. There is also evidence that head injuries are related to the development of delinquent behavior. However, the nature of the interaction of these medical conditions with subsequent delinquency has not been definitively established. The relationships have not been demonstrated consistently and they appear to be mediated by other variables, such as family functioning. (Buka & Earls 1993) In her analysis of the avoidance of criminality, Widom failed to uncover any significant relationship between early medical problems and subsequent criminality. (Widom 1991c)

If we decide to control for behavior problems, cognitive functioning and/or medical conditions, we will need to be especially concerned about temporal assessment. For example, we will need to make sure that observed behavioral problems, cognitive deficits and head injuries did not result from maltreatment. Recall the association between maltreatment and childhood behavior problems that has been demonstrated in observational studies. (Kakar 1996; Luntz & Widom 1994) If we fail to ensure accurate temporal classification, we may conclude that a portion of delinquency is explained by cognitive deficits or behavior problems without knowing whether these problems themselves are the result of abuse and neglect.

Out-Of-Home Placement

In addition to the control variables discussed so far, we should consider controlling for out-of-home placement (such as foster care or institutional care). There is some evidence that such placements increase a child's risk of delinquency. (See, review by Buka &Earls 1993) However, the evidence is inconsistent and most of the studies examining the effects of placement on maltreated children suffer from methodological flaws. (Widom 1991b) Widom attempted to overcome the limitations of prior studies when examining the avoidance of criminality among maltreated children. She categorized the placements of the maltreated children in terms of: location of placement; length of placement; age at first placement; and number of moves. She also considered the reasons for the placements.

Widom's findings demonstrated that, on the whole, children who were placed outside their homes for abuse and/or neglect were not at increased risk of delinquency or adult criminality. Children placed for maltreatment and delinquency were more likely to have subsequent arrests, however. These children were six times more likely to be arrested as a juvenile, three times more likely to be arrested as an adult, seven times more likely to be arrested as both a juvenile and an adult, and three times more likely to be arrested for a violent offense than were children placed for abuse and/or neglect only. This suggests that out-of-home placement itself does not contribute as much to the development of delinquency as does the reason for the placement. (Widom 1991b)

Other results of Widom's analysis support this conclusion. Children placed in foster homes had similar criminality rates to those not placed outside the home, whereas those placed in institutions had higher rates of arrest. Within the group of children placed outside the home, few children placed before one year of age had arrest records. None of them had arrests for violent offenses. Children who spent more than ten years in their initial placement had the lowest arrest rates, while children placed for 4-6 years had the highest rates of arrest. Children with more frequent moves also had more arrest records than did those with fewer moves. (Widom 1991b) Taken as a whole, these findings suggest that out-of-home placements, in and of themselves, do not increase the risks of delinquency. Perhaps it is the type of placement, institutional versus foster home, that explains the relationship between placement and delinquency found in some studies. Alternatively, it may be the stability of the placement that explains the sometimes observed relationship. Perhaps children who move more often or at critical times in their development, respond by engaging in delinquency. The nature of the relationship cannot be established from Widom's analysis, however. She notes that the children who moved most frequently had more behavior problems recorded in their court files. She could not determine which came first, though. Was it that children who were moved often or placed in institutions developed behavior problems and subsequent criminality or was it that children exhibiting behavior problems were difficult to place and thus had to be moved more often and placed more often in institutional settings? Without knowledge of the temporal relationship between these variables, it was impossible to determine causal direction. (Widom 1991b) Given the current state of the research regarding the relationship between placement and delinquency, we should control for placement if we are able to do so.

Summary of Control Variables

In summary, there are a number of variables for which we should control to strengthen the validity of our conclusions about the relationship between maltreatment and delinquency. We need to control for age, sex, race/ethnicity and socioeconomic status. We should also attempt to assess and control for family functioning, apart from the maltreatment. We should control for behavior problems, cognitive functioning and medical conditions if we can gather appropriate data to do so. We should also control for the placement experiences of the maltreated children. If we hold these variables constant, we can be more certain that any demonstrated relationship between maltreatment and delinquency reflects a true relationship and not the confounding of other variables.

We should also consider that risk factors may combine in multiplicative relationships rather than additive relationships. A particular combination of risk factors may be more predictive of delinquency than a separate analysis of each of those factors would suggest. For example, cognitive impairment, maltreatment and impoverished living conditions may combine synergistically to create a very high risk of delinquency, while age, SES and maltreatment may combine to produce a risk of delinquency not much greater than the risk of maltreatment alone. To account for these potentially interactive effects, we should consider using a model such as that used by Dembo et. al. (1992).(17)

Length of Follow-Up

In addition to decisions about definitions and control variables, we must make one further methodological decision. How long should we follow the cohorts we identify? The length of follow-up is likely to effect the results of our study. The longer we follow the children, the greater the chance they will engage in officially detected delinquency. Kakar compared the rates of delinquency in studies of varying lengths of follow-up and noted that studies with follow-up periods of several decades had significantly higher rates of delinquency than did those with follow-up periods of 2-5 years. (Kakar 1996)

The length of follow-up should be determined, in part, by the average age of the cohorts we select. If we choose cohorts of ten-year olds and follow them for two years, we will be ending the study before these children reach the peak years of delinquent activity. Longitudinal studies of delinquency suggest that delinquent behavior increases from age 10 to age 16 or 17, where it peaks and begins declining. (Tracy, Wolfgang & Figlio 1990) Thus, we need to follow our cohorts until they are at least 17 years old.

We may also want to follow them for longer periods. Remember that Widom found no significant increase in juvenile violence among maltreated children, but did find increased risk of adult violence. Consider also the gender reversal in Widom's analysis regarding risk of violent offending. In the 1988 data, males, but not females, were at significant risk of arrest for violent offenses. In the 1994 data, females, but not males, were at elevated risk of such arrests. Widom speculates that this reversal of findings resulted from the aging of the sample population so that by 1994, 99% of the subjects had passed through their peak years of violent offending. (Maxfield & Widom 1996) Thus, if we want to assess long-term criminal consequences of maltreatment, we may need to follow our cohorts into adulthood.


Much of the research on the relationship between maltreatment and delinquency demonstrates a significant relationship between the two phenomena. Although many of the early studies had methodological flaws, recent studies have been well designed. These studies consistently show a significant relationship between maltreatment and delinquency. However, the magnitude of the relationship varies substantially from study to study. Many of the variations are probably the result of the investigators' differing methodological and definitional decisions. Thus, we must thoughtfully consider these choices as we design our study.

First, we must be cognizant of our data sources and their influence on the breadth of our definitions of maltreatment and delinquency. Regardless of the data source, we need to decide whether to combine all types of maltreatment into a single maltreatment/non-maltreatment variable. We need to determine whether we will assess severity and frequency of maltreatment and whether we will differentiate maltreatment at the hands of a primary caregiver from maltreatment by a stranger. We need to make similar decisions regarding delinquency. Will we define delinquency according to referral or adjudication records? Will we combine property and status offenses with violent offenses? Will we separately analyze delinquents according to the severity and frequency of their offending?

Once we make these definitional decisions, we need to minimize the potential for misclassification due to mobility and temporal errors. We also need to determine which variables to hold constant in our analysis. If we decide to control for socioeconomic status, family functioning, behavior problems, cognitive impairment, medical condition, and placement environment, we will need to develop a measure of these variables given the data to which we have access. Finally, we will need to determine the appropriate length of follow-up for the study.

Prior research regarding the relationship between maltreatment and delinquency reveals that the relationship, while significant, is far from simple and straight forward. Thus, to accurately assess the nature of the relationship in our data, we will need to develop a well-designed study.

Return to John Carter's Home Page


Alfaro, J. D (1981). Report on The Relationship Between Child Abuse and Neglect and Later Socially Deviant Behavior. In Hunner, R. J. & Walker, Y. E. Exploring the Relationship Between Child Abuse and Delinquency. New Jersey: Allanheld, Osmun & Co. Publishers Inc.

Abrams, L. A. (1995). Strengthening the Fabric of Child and Family Policies: Interweaving the Threads of Research. In Baumrind, D. Child Maltreatment and Optimal Caregiving in Social Contexts. New York: Garland Publishing, Inc.

Baumrind, D. (1995). Child Maltreatment and Optimal Caregiving in Social Contexts. New York: Garland Publishing, Inc.

Brown, S. E. (1984). Social Class, Child Maltreatment, and Delinquent Behavior. Criminology, 22(2): 259-78.

Buka, S. & Earls, F. (1993). Early Determinants of Delinquency and Violence. Health Affairs, Winter 1993: 46-64.

Dembo, R., Williams, L., Wothke, W., Schmeidler, J., & Brown, C. H. (1992). The Role of Family Factors, Physical Abuse, and Sexual Victimization Experiences in High-Risk Youths' Alcohol and Other Drug Use and Delinquency: A Longitudinal Model. Violence and Victims, 7(3): 245-66.

Doerner, W. G. (1987). Child Maltreatment Seriousness and Juvenile Delinquency. Youth & Society, 19(2): 197-224.

Hoge, R. D., Andrews, D. A., & Leschied, A. W. (1996). An Investigation of Risk and Protective Factors in a Sample of Youthful Offenders. Journal of Child Psychology and Psychiatry, 37(4): 419-24.

Howing, P. T., Wodarski, J. S., Kurtz, P. D., Gaudin, J. M., & Herbst, E. N. (1990). Child Abuse and Delinquency: The Empirical and Theoretical Links. Social Work, May 1990: 244-49.

Kakar, S. (1996). Child Abuse and Delinquency. Maryland: University Press of America.

Kelley, B. T., Thornberry, T. P., & Smith, C. A. (1997). In the Wake of Childhood Maltreatment. In Juvenile Justice Bulletin. Washington D.C.: United States Department of Justice, Office of Juvenile Justice and Delinquency Prevention.

Luntz, B. K. & Widom, C. S. (1994). Antisocial Personality Disorder in Abused and Neglected Children Grown Up. American Journal of Psychiatry, 151(5): 670-74.

Loeber, R. & Dishion, T. (1983). Early Predictors of Male Delinquency: A Review. Psychological Bulletin, 94(1): 68-99.

Maxfield, M. G. & Widom, C. S. (1996). The Cycle of Violence: Revisited Six Years Later. Archives of Pediatric and Adolescent Medicine, 150: 390-95.

Patterson, G. R. & Stouthamer-Loeber, M. (1984). The Correlation of Family Management Practices and Delinquency. Child Development, 55:1299-1307.

Pelton, L. H. (1978). Child Abuse and Neglect: The Myth of Classlessness. American Journal of Orthopsychiatry, 48(4): 608-17.

Pillitteri, A., Seidl, A., Smith, C., & Stanton, M. (1992). Parent Gender, Victim Gender, and Family Socioeconomic Level Influences of the Potential Reporting by Nurses of Physical Child Abuse. Issues in Comprehensive Pediatric Nursing, 15: 239-47.

Roberts, D. E. (1991). Punishing Drug Addicts Who Have Babies: Women of Color, Equality, and the Right to Privacy. Harvard Law Review, 104(7): 1419-82.

Smith, C. & Thornberry, T. P. (1995). The Relationship Between Childhood Maltreatment and Adolescent Involvement in Delinquency. Criminology, 33(4): 451-81.

Smith, C. P., Black, T. E., & Campbell, F. R. (1980). A National Assessment of Case Disposition and Classification in the Juvenile Justice System: Inconsistent Labeling. Reports of the National Juvenile Justice Assessment Centers. Washington D.C.: United States Department of Justice, Office of Juvenile Justice and Delinquency Prevention.

Tracy, P. E., Wolfgang, M. E., & Figlio, R. M. (1990). Delinquency Careers in Two Birth Cohorts. New York: Plenum Press.

Van Voorhis, P., Cullen, F. T., Mathers, R. A., & Garner, C. C. (1988). The Impact of Family Structure and Quality on Delinquency: A Comparative Assessment of Structural and Functional Factors. Criminology, 26(2): 235-61.

Warner, J. E., & Hansen, D. J. (1994). The Identification and Reporting of Physical Abuse by Physicians: A Review and Implications for Research. Child Abuse & Neglect, 18: 11-25.

Widom, C. S. & Maxfield, M. G. (1996). A Prospective Examination of Risk for Violence Among Abused and Neglected Children. Annals of New York Academy of Sciences, 1996:224-37.

Widom, C. S. & Kuhns, J. B. (1996). Childhood Victimization and Subsequent Risk for Promiscuity, Prostitution, and Teenage Pregnancy: A Prospective Study. American Journal of Public Health, 86(11): 1607-1612.

Widom, C. S. (1995). Victims of Childhood Sexual Abuse - Later Criminal Consequences. In National Institute of Justice Research Brief, March 1995. Washington D.C.: United States Department of Justice, Office of Juvenile Justice and Delinquency Prevention.

Widom, C. S. & Ames, M. A. (1994). Criminal Consequences of Childhood Sexual Victimization. Child Abuse & Neglect, 18(4): 303-18.

Widom, C. S. (1994). Childhood Victimization and Adolescent Problem Behaviors. In Lamb, M. E. & Ketterlinus, R. Adolescent Problem Behaviors. New York: Erlbaum.

Widom, C. S. (1991a). Childhood Victimization: Risk Factor for Delinquency. In Colten, M. E. & Gore, S. Adolescent Stress. New York: Aldine De Gruyter.

Widom, C. S. (1991b). The Role of Placement Experiences in Mediating the Criminal Consequences of Early Childhood Victimization. American Journal of Orthopsychiatry, 61(2): 195-208.

Widom, C. S. (1991c). Avoidance of Criminality in Abused and Neglected Children. Psychiatry, 54: 162-73.

Widom, C. S. (1989a). Child Abuse, Neglect, and Adult Behavior: Research Design and Findings on Criminality, Violence, and Child Abuse. American Journal of Orthopsychiatry, 59 (3): 355-67.

Widom, C. S. (1989b). The Cycle of Violence. Science, 244:160-66.

Widom, C. S. (1988). Sampling Biases and Implications for Child Abuse Research. American Journal of Orthopsychiatry, 58 (2): 260-70.

Yoshikawa, H. (1994). Prevention as Cumulative Protection: Effects of Early Family Support and Education on Chronic Delinquency and Its Risks. Psychological Bulletin, 115(1): 28-54.

Zingraff, M. T., Leiter, J., Myers, K. A., Johnsen, M. C. (1993). Child Maltreatment and Youthful Problem Behavior. Criminology, 31(2): 173-202.

Other References

Bolton, F. G. & Reich, J. W. (1979). Delinquency patterns in Maltreated Children and Siblings. Victimology: An International Journal, 2: 349-59.

Dembo, R., Williams, L., Schmeidler, J., Berry, E., Wothke, W., Getreu, A., Wish, E. D., & Christensen, C. (1992). A Structural Model Examining the Relationship Between Physical Child Abuse, Sexual Victimization, and Marijuana/Hashish Use in Delinquent Youth: A Longitudinal Study. Violence and Victims, 7(1):41-59.

Fagan, J. & Wexler, S. (1987). Family Origins of Violent Delinquents. Criminology, 25(3): 643-69.

Haapasalo, J. & Hamalainen, T. (1996). Childhood Family Problems and Current Psychiatric Problems Among Young Violent and Property Offenders. Journal of the American Academy of Child and Adolescent Psychiatry, 34(10): 1394-1401.

Prentky, R. A., Knight, R. A., Sims-Knight, J. E., Straus, H., Rokous, F., & Cerce, D. (1989). Developmental Antecedents of Sexual Aggression. Development and Pathology 1: 153-69.

Ryan, G. (1989). Victim to Victimizer. Journal of Interpersonal Violence, 4(3): 325-41.

Stouthamer-Loeber, M., Loeber, R., Farrington, D. P., Zhang, Q., van Kammen, W., & Maguin, E. (1993). The Double Edge of Protective and Risk Factors for Delinquency: Interrelations and Developmental Patterns. Development and Psychopathology, 5: 683-701.

Wolfgang, M. E., Figlio, R. M. & Sellin, T. (1972). Delinquency in a Birth Cohort. Chicago: The University of Chicago Press.
Return to John Carter's Home Page


0 The variation was between differing counties. Those counties with the most complete records had the highest percentages of children subsequently taken to court as delinquent or ungovernable. (Alfaro 1981)


0 Widom began a second phase of the research in 1989. During this phase, she is attempting to interview as many of the subjects as possible to examine a number of long-term consequences of child maltreatment, including: social, emotional, cognitive and intellectual, occupational, psychiatric, substance use/abuse and physical health outcomes. The interviews will also assess what protective factors may have buffered the effects of maltreatment. The results of the interviews have not yet been reported. (Widom 1995)


0 This finding may be somewhat misleading, however, since the overwhelming majority (84%) of the sexual abuse victims were female and females less often have official records for violent offenses. (Widom & Maxfield 1996)


0 This analysis applied only to males since only males in the study population had arrests for these crimes. (Widom & Ames 1994)


0 Order offenses were defined as criminal mischief, vandalism, trespassing, disorderly conduct, resisting arrest, fleeing police officer, vagrancy and visiting a common nuisance. (Widom 1991)


0 The predictive strength of these two factors, especially race, may reflect bias in the criminal justice system since Widom's outcome data for this study were drawn exclusively from official records. There is some evidence that African-Americans and males are disproportionately processed through the criminal justice system. (Widom 1991; Widom 1988; Tracy, Wolfgang & Figlio 1990)


0 The adequacy of this measure seems suspect. The number of substantiated incidents may not reliably reflect the frequency with which the children were maltreated, but only the frequency with which social services investigated the incidents. Perhaps children who were frequently abused or neglected before coming to the attention of the authorities had only one substantiated incident of maltreatment because they were removed from their homes and placed in foster care. Perhaps children who were less severely or frequently abused were more often allowed to remain in their homes so that it was these children who had two substantiated cases of abuse.

8. 0 The family strtucture variable included four categories: children living with both biological parents; children living with a biological parent and stepparent; children living with a single parent; and children living with others (one parent and other relatives, other relatives or nonrelatives, institutional placement, or foster care).


0 Family structure was a dichotomous variable indicating whether the student lived in a household with both biological parents or not. The underclass status variable indicated whether or not the student lived in a home where the principal household wage earner was unemployed, the family received welfare benefits or the family's income was below the federal poverty guidelines. The mobility variable indicated whether the student began first grade in the community's public school system. It was designed to control for misclassification of students as non-maltreated because they had not lived in the community as young children and thus did not appear in official records of abuse and neglect. (Smith and Thornberry 1996)


0 Interestingly, Smith and Thornberry used this schema only to assess severity of maltreatment. They did not conduct separate analyses for the different types of maltreatment, only for the various levels of severity they defined.


0 Widom and Ames found no significant differences when they compared the juvenile arrest records of those sexually abused by relatives and those sexually abused by nonrelatives. (Widom & Ames 1994)


0 John, on a personal note, I remember a 10 year-old boy who told me that no one did anything about his mother's substance abuse, even after they became homeless, until he started getting in trouble at school. When he thought he was about to be returned to his mother's custody, he started failing all his classes. He told us he did this because he wanted to make sure he did not go back to his mother in her present state. He thought getting in trouble at school again might keep that from happening.


0 For example, many argue that lower SES individuals are more likely to use public clinics for health care. They contend that doctors and health care workers in these settings are more likely to report suspected abuse and neglect than are private doctors who are relying upon their patients' parents to provide their income. There is evidence to suggest that physicians assess vignettes of maltreatment differently when only the SES variable is manipulated. (Pelton 1978; Roberts 1991; Warner & Hansen 1994; Baumrind 1995)


0 Recall that a similar pattern, of greater magnitude, was found for the school-based control group. Most of the difference in overall delinquency rates between these groups was explained by different rates of status offending. (Zingraff et. al. 1993)


0 Of course, the nature of her data sources did not allow her to do this. Perhaps when she completes the second phase of her research, which involves interviewing the study participants, she will be able to comment on the quality of family life.


0 Widom speculates that while these traumatic events normally involve changed circumstances that increase a child's risk for delinquency, when the child is being maltreated, the changes may be protective.


0 John, I understand what Dembo and his colleagues were trying to do, but the "how" of their equations and modeling is beyond my quantitative abilities.

Return to John Carter's Home Page