Cognitive and behavioral risk factors for child physical abuse among Chinese children: a multiple-informant study

The present study used secondary data collected from the Wave II of the Jintan Child Cohort Study, which is an ongoing prospective longitudinal study. The cohort study recruited 1385 children aged 3–5 years old from upper grade (i.e. mean age about 5 years old), middle grade (i.e. mean age about 4 years old), and bottom grade (i.e. mean age about 3 years old) in preschools in Jintan, China in 2004–2005, which was a representative sample of children in the city in terms of gender, age, and residential locations. The cohort study design was described elsewhere [26‐28].

The children from upper grade, middle grade, and bottom grade were followed up with during the Wave II to assess behavioral problems (reported by children, mothers, and teachers) and IQ in 2010–2011, 2011–2012, and 2013, respectively. All of the children were also invited to participate in a child abuse questionnaire survey in 2013 when children were 6th, 7th and 8th graders. In order to maintain temporal order to test the association of IQ and behavioral problems in earlier life and later child abuse, we included the 7th and 8th graders whose behavioral problems and IQ were assessed in 2010–2011 and 2010–2012, and child physical abuse was assessed in 2013. We obtained complete data from 265 children (47.2 % boys). The temporal design of the parent cohort study and the present study is shown in Fig. 1. Compared with those who did not have complete data, these children did not show significant differences in age, verbal IQ (VIQ), performance IQ (PIQ), or externalizing and internalizing behaviors (regardless of the reporters), or minor or severe physical abuse (regardless of the perpetrators). There were slightly more girls, more children from better socioeconomic background, and fewer children from the rural areas in the retained sample (Additional file 1: Table S1).

We obtained written informed consent from both mothers and teachers and verbal consent from children during the Wave II of data collection. Two trained research assistants distributed and collected the questionnaires, explained the objectives and confidentiality of the study and the principle of voluntary participation and participations’ right of withdrawing the study at any time point, and answered any of the respondents’ questions. All questionnaire surveys for the children took place in classrooms during school hours. Children completed the IQ test at Jintan Hospital and, in the meantime, parents rated their children’s behavior in the waiting rooms. Teachers rated child behavior in their offices after understanding the study. We obtained approval from the Institutional Review Board of the University of Pennsylvania and the Ethical Committee for Research at Jintan Hospital, China.

We first ran descriptive analyses for all variables. We described the prevalence of child physical abuse by mothers and fathers, respectively, and compared the intelligence and behavioral characteristics of children with a specific type of abuse to those without it. We then ran variance inflated factor (VIF) analysis to determine the multicollinearity of the independent variables. The result showed that VIF of the six behavior variables ranged from 2.07 to 2.32. Therefore, multicollinearity of behavioral variables was not a severe concern. Age and grade were highly correlated and, thus, only grade was controlled in multivariate analyses. In order to illustrate meaningful odds ratios, we rescaled VIQ, PIQ, and behavior variables by dividing each of them by 10. Therefore, the OR and 95 % confidence intervals indicate a change in the risk of being maltreated with a 10-point increase in VIQ, PIQ, or behavior scores. Using the rescaled IQ and behavior scores as independent variables, we constructed four generalized linear models with binomial family and logit link function to test the association of IQ and behavioral problems with the risk of child physical abuse 1 or 2 years later, controlling for the co-variables. Minor or severe physical abuse by mothers and fathers were treated as dependent variables in the four models, respectively. Next, we constructed GLMs with binomial family and log link to obtain the risk ratio (RR) for the significant cognitive and behavioral factors associated with physical abuse to estimate the effect sizes of their associations with physical abuse. In order to get convergent GLMs with log link, one case with the highest predicted value obtained from the GLMS with logit link was removed for each GLM model with log link. We set the significance level at α = 0.05/4 = 0.125 using the Bonferroni correction of four outcomes, and regarded a p value less than 0.05 but higher than 0.0125 as marginally significant or a trend of significance. We performed all the analyses using STATA 13.0 for Windows (College Station, TX).

Among the 265 children, almost half of them experienced minor physical abuse by either their mothers or fathers, and about one-fourth of children experienced severe physical abuse from either their mothers or fathers. Boys were more likely to report physical abuse from their fathers than girls (χ ² = 6.944, p = 0.008). There is no significant difference between physically maltreated children and their non-maltreated counterparts in terms of age, location, and socioeconomic status. See Table 1.

Children who experienced maternal minor physical abuse in the preceding year had higher scores of externalizing behaviors as rated by their respective mothers (51.31 ± 9.36 vs. 48.35 ± 7.97, p = 0.006, Cohen’s d = 0.34) and themselves (50.92 ± 10.25 vs. 47.85 ± 8.76, p = 0.009, Cohen’s d = 0.32). Children with paternal minor (51.69 ± 9.52 vs. 48.50 ± 8.00, p = 0.003, Cohen’s d = 0.37) or severe physical abuse (52.32 ± 9.69 vs. 49.03 ± 8.37, p = 0.009, Cohen’s d = 0.38) scored higher on externalizing behaviors as rated by their mothers in the past. The effect sizes of these differences are small to medium. Children with an experience of maternal severe physical abuse showed a trend of lower PIQ scores, higher externalizing behaviors scores rated by their teachers, and higher self-reported internalizing behavior scores, while children with paternal severe physical abuse showed a trend of higher scores on teacher-rated externalizing behavior (Table 2). However, these results did not reach the significance level at 0.0125.

Table 3 illustrates the adjusted associations of child physical abuse with IQ and behavior problems. After adjusting for other variables in the model, the risk of maternal severe physical abuse increased with the increase in the scores of mother—[OR = 1.38 (1.09, 1.74), p = 0.007, RR = 1.28] or teacher—[OR = 1.47 (1.29, 1.69), p = 0.009, RR = 1.22] rated externalizing behavior, while such risk decreased with the increase in the score of mother-rated internalizing behavior [OR = 0.77 (0.63, 0.95), p = 0.011, RR = 0.79]. Similarly, the risk of paternal severe physical abuse grew with the increase in the scores of mother—[OR = 1.47 (1.29, 1.69), p < 0.001, RR = 1.31] or teacher-rated externalizing behavior [OR = 1.61 (1.44–1.81), p < 0.001, RR = 1.32]. Although a higher score of VIQ was related to increased risk of maternal minor physical abuse [OR = 1.06 (1.02–1.13), p = 0.006, RR = 0.04], the effect size was very small. Notably, the ORs change with the increase in behavior scores. For example, with an increase of 20 points in mother-rated externalizing behavior, the odds of maternal severe physical abuse increases from 1.38 to 1.90 (RR increases from 1.28 to 1.64), compared to the odds of not experiencing such abuse. Neither IQ nor behavioral problems rated by different informants were significantly associated with the risk of paternal minor physical abuse.

The present study found that mothers’ and teachers’ reports on externalizing behavior were associated with both maternal and paternal severe physical abuse. This is consistent with the finding from a longitudinal Chinese study that children with high externalizing behavior experienced more physical abuse 6 months later after controlling for the previous physical abuse experience [37]. Similarly, Stith et al. conducted a meta-analysis and reported that child externalizing behavior is a risk factor for child abuse [12]. In terms of the effect size of the association between externalizing behavior and severe physical abuse, the odds ratios are comparable to the estimates from a meta-analysis of 68 Chinese studies treating child abuse as a risk factor for behavioral outcomes. This meta-analysis found that the effect sizes of the associations between child abuse and behavioral outcomes (e.g. mental health disorders, depression, anxiety, drug use, etc.) range from 1.40 to 1.98 [4]. Taken together, the findings indicate that externalizing behaviors perceived by parents or teachers may increase parents’ negative attributions of child behavior that directly increases parenting stress [38] and the tendency of practicing harsh disciplining strategy to correct children’s misbehavior or to reduce their distress.

Interestingly, child self-report externalizing behavior decreased the risk of paternal severe physical abuse. It is possible that Chinese fathers may regard child self-reported externalizing behavior as normal extroversion, and therefore, are less likely to practice severe physical discipline when their mothers’ and teachers’ perceptions of child externalizing behavior are adjusted. Very few studies have attempted to examine the association between child behavior and paternal physical abuse, and more studies are needed.

The findings also suggest that there is a discrepancy in the perceptions of externalizing behavior between children and their parents and teachers. Research found that children usually report fewer behavior problems than their parents or other informants [39]. The disparate perspectives of externalizing behavior may be a source of conflict that triggers parental physical abuse. Hence, it may be effective to prevent child abuse by modifying parents’ and teachers’ perceptions of child behavior.

We found that mother-rated internalizing behavior was associated with less risk of maternal minor or severe physical abuse. Literature from western studies indicates that physically abusive mothers usually rated higher on child internalizing behavior [12, 40], an inconsistent result with the present finding. This inconsistency may indicate that Chinese parents tend not to use physical discipline when they perceive that their children are introverted. Previous research has argued that from the perspective of Chinese parents, the characteristics of internalizing problems may align with desired characteristics in Chinese culture, such as being quiet and sensitive [37, 41]. Therefore, mother-perceived internalizing behavior relates to less-frequent physical abuse.

Although the positive association between VIQ and maternal minor physical abuse (that is independent of behavioral problems and sociodemographic variables) was statistically significant, the effect size is very small. We did not find significant associations of VIQ with other types of physical abuse or significant associations of PIQ with all types of physical abuse. The previous findings of the association between IQ and child abuse under the assumption of the child effect are not conclusive. Breslau et al. conducted a longitudinal study and found that full-scale IQ lower than 115 at the age of 6 increased the risk of exposure to general assaultive violence at age 17, and they explained that children with low IQs might be more likely to interact with disruptive peers and, therefore, be exposed to assaultive violence [14]. In contrast, Brown et al. [42] and Young et al. [13] found that low IQ scores were associated with child neglect but not physical abuse, indicating that different types of child abuse may be associated with IQ differently. Further research can be conducted to examine the relationship between IQ and other forms of child abuse other than physical abuse in the Chinese context.

The absence of the significant association between IQ and physical abuse could also be because child behavior fully mediates the relationship between IQ and child physical abuse. Prior studies have suggested that children with intellectual disabilities are at higher risk of developing behavioral problems that may further make children more prone to physical abuse [43, 44]. Future research is warranted to explore the possible mediating role of behavioral problems in the relationship between IQ and child abuse.

The findings should be interpreted cautiously due to study limitations. First, a relatively small proportion of the original cohort children participated in the survey, and there were slightly more girls and less children from rural areas (Additional file 1: Table S1). Therefore, the present study’s generalizability is limited. Despite this, the present study does exhibit value in offering a new perspective to investigate the relationship between IQ, behavioral problems, and child abuse.

Second, we did not examine gender differences in the relationships of child physical abuse with IQ and behavioral problems concerning low statistical power. Prior studies suggest that there are gender differences in the predictive effect of externalizing and internalizing behavior on physical abuse among Chinese children. Specifically, compared with Chinese girls, Chinese boys with behavioral problems were more likely to experience physical abuse [37, 41]. Future studies are needed to explore whether the association between IQ and child abuse depends on child gender.

Third, we only collected information on child abuse once. The status of child abuse prior to the study was not assessed. It is possible that maltreated children in the present sample had also experienced abuse before the study, and such experience may serve as a confounder in the relationship between behavioral problems and child physical abuse. However, given the findings from the qualitative studies in the Chinese context that Chinese parents practice harsh discipline towards children because of their misbehavior, disobedience, and poor academic performance [18, 20, 45], as well as the bidirectional relationship between child abuse and behavioral problems revealed from the longitudinal studies [8, 9, 11], it is plausible to regard IQ and child externalizing and internalizing behavior as potential risk factors for child abuse. It is worth noting that the majority of studies regarding child abuse as a risk factor for behavioral problems failed to control for previous behavioral problems. Therefore, we suggest that future research further explore the reciprocal relationship of child abuse with cognition and behavior.

Lastly, some confounders that were not included in the present study need to be considered for future studies. For example, parental mental health status could be an important confounder that is related to both child abuse [46] and child behavioral problems [47]. However, very few Chinese researchers have attempted to examine the effect of parental mental health status on child abuse, and, therefore, this needs more attention.