Risk factors for child abuse and neglect:
A meta-analytic review
Master thesis, Forensic Child and Youth Care Sciences
Author: T. M. Mulder (10504028) Primary reader: Dhr. Drs. M. Assink.
Secondary reader: Mw. Dr. C. E. Van der Put Date: 01-08-2014
2 Abstract
Introduction: Knowledge of the relative effect of risk factors for child abuse and neglect is vital for improving policies, (preventive) interventions and risk assessments. The current study aimed to update contemporary knowledge regarding the effect of (domains of) risk factors, and to determine if, and how, these effects are moderated (e.g., by characteristics of the studies). Method: A meta-analysis was conducted on 34 primary studies, which reported on 468 effect sizes. A multilevel random-effects model was used to conduct moderator analyses, and to calculate an overall effect. Results: A small to moderate overall effect was found. Effect sizes proved heterogeneous, several moderators were found. The largest mean effect sizes were found for five parental domains (intimate partner violence, personality/temperament, mental health issues, substance use and intergenerational continuity), one child domain (age), and one familial domain (SES). Discussion: These findings indicate that emphasis should be placed on targeting dynamic parental risk factors. Keywords:child abuse, child neglect, risk factors, meta-analysis
Samenvatting
Inleiding: Kennis over het relatieve effect van risicofactoren voor kindermishandeling is van groot belang voor het verbeteren van beleid, (preventieve) interventies en risicotaxaties. Deze studie richtte zich op het actualiseren van de huidige kennis over het effect van (domeinen van) risicofactoren, en om te bepalen of, en hoe, deze effecten worden gemodereerd (bijvoorbeeld door kenmerken van de studies). Methode: Een meta-analyse werd uitgevoerd over 34 primaire studies, die rapporteerden over 468 effectgroottes. Een ‘multilevel random-effects’ model werd gebruikt voor het uitvoeren van moderatoranalyses, en voor het berekenen van een overkoepelend effect. Resultaten: Een klein tot middelgroot overkoepelend effect werd gevonden. Effectgroottes bleken heterogeen, verschillende moderatoreffecten werden aangetoond. De grootste gemiddelde effectgroottes werden gevonden voor vijf ouderlijke domeinen (huiselijk geweld, persoonlijkheid/temperament, geestelijke gezondheid, middelengebruik en intergenerationele overdracht), één kind domein (leeftijd), en één gezinsdomein (SES). Discussie: Deze resultaten tonen aan dat nadruk dient te worden gelegd op het veranderen van dynamische ouderlijke risicofactoren.
3 Since the appearance of Kempe, Silverman, Steele, Droegemueller, and Silver’s (1962) groundbreaking article, detailing the first medically reported cases of physical abuse inflicted upon children, a wide variety of research has been conducted towards this phenomenon alongside the development of relevant laws and policies in a broad range of child-related fields (Leventhal & Krugman, 2012). The contemporary notion regarding child abuse has been expanded, to include neglect and forms of abuse other than physical, and is defined as: “any act or series of acts of commission or omission by a parent or other caregiver that results in harm, potential for harm, or threat of harm to a child” (Leeb, Paulozzi, Melanson, Simon, & Arias, 2008, p. 11). In this context, and for the purpose of this thesis, a child is defined as “... every human being below the age of eighteen years unless under the law applicable to the child, majority is attained earlier” (United Nations [UN], 1989, article 1).
In concordance with the aforementioned definition, child abuse and neglect can be divided into six distinct categories, which can be subdivided into acts of commission, and forms of omission (Leeb et al., 2008). In adherence to the subdivision made in the most recent national incidence studies of both the United States (NIS-4; Sedlak et al., 2010) and The Netherlands (NPM-2010; Alink et al., 2011), acts of commission entail sexual abuse, physical abuse, and emotional abuse. Forms of omission entail physical neglect, emotional neglect, and educational neglect. A detailed description of these six categories falls outside the scope of this thesis. For an in-depth elaboration on these categories, both the NPM-2010 (Alink et al., 2011) and NIS-4 (Sedlak et al., 2010) offer excellent information.
Research consistently showed that experiencing abuse and neglect during childhood is linked to an increased chance for the development of a wide range of adverse health effects, as well as detrimental psychosocial outcomes, during the course of life. These include, but are not limited to: attempted suicide (Dube et al., 2001; Felitti et al., 1998), heavy and earlier smoking initiation (Anda et al., 1999; Ramiro, Madrid, & Brown, 2010), alcoholism (Dube et al., 2002; Felitti et al., 1998; Ramiro et al., 2010), illicit drug use (Dube et al., 2003; Felitti et al., 1998), risky sexual behavior (Felitti et al., 1998; Ramiro et al., 2010), depression (Chapman et al., 2004; Felitti et al., 1998), poor educational achievement (Gilbert et al., 2009), adolescent offending (Van der Put, Asscher, Wissink, & Stams, 2013), ischemic heart disease and cancer (Dong et al., 2004b; Felitti et al., 1998; Yang et al., 2013), diabetes, malnutrition, vision problems and chronic lung disease (Widom, Czaja, Bentley, & Johnson, 2012; Felitti et al., 1998), and skeletal fractures and liver disease (Felitti et al., 1998). The odds of developing these psychological and medical problems increase when the child has been exposed to more than one form of abuse and neglect (e.g., Chartier, Walker, & Naimark, 2010). Having
4 experienced one such form, increases the chance of being subjected to an additional form of child abuse and neglect by over 17 times (Dong et al., 2004a). The resulting coexistence of multiple forms of child abuse and neglect is referred to as comorbidity, or multi-type maltreatment (Higgins & McCabe, 2001).
Aside from these negative effects on the individual level, child abuse and neglect also places a large (economic) burden on society. Children who are hospitalized in relation to abuse and neglect, have significantly longer hospital stays than children who are hospitalized for unrelated medical issues, and their total medical charges are significantly increased (Rovi, Chen, & Johnson, 2004; Florence, Brown, Fang, & Thompson, 2013). An association was also found between being subjected to adverse childhood experiences, and high health care utilization in adulthood (Chartier et al., 2010).
Contrary to these well-studied health effects, studies conducted to determine the prevalence of child abuse and neglect are relatively scarce (World Health Organization [WHO], 2014). Examples of recently reported nation-wide prevalence rates, expressed as the proportion of victims of child abuse and neglect, are: one in every 34 children in The Netherlands (Alink et al., 2011), one in every 25 – 58 children in the United States (Sedlak et al., 2010), and one in every 35 children in Canada (Public Health Agency of Canada, 2010). Meta-analytic estimates of the worldwide prevalence rates of different forms of child abuse and neglect have only very recently appeared, and vary widely. Concerning child abuse, the reported prevalence rates, expressed as a percentage of the global population, are: 0.3-36.3% for emotional abuse (Stoltenborgh, Bakermans-Kranenburg, Alink, & Van IJzendoorn, 2012), 0.3%-22.6% for physical abuse (Stoltenborgh, Bakermans-Kranenburg, Alink, & Van IJzendoorn, 2013), and 0.4%-12.7% (Stoltenborgh, Van IJzendoorn, Euser, & Bakermans-Kranenburg, 2011), to 3%-31% for sexual abuse (Barth, Bermetz, Heim, Trelle, & Tonia, 2013). Meta-analytic research regarding worldwide prevalence rates of child neglect is nearly non-existent. Only one such study was conducted, which reported global prevalence rates of 16.3% for physical neglect, and 18.4% for emotional neglect (Stoltenborgh, Bakermans-Kranenburg, & Van IJzendoorn, 2013). For an in-depth elaboration regarding the substantial differences between global prevalence rates, see Stoltenborgh (2012).
As indicated by these figures, child abuse and neglect is both a world-wide problem, and highly prevalent. It is referred to as “... that portion of harm to children that results from human action that is proscribed, proximate and preventable” (Korbin, 1988, p. 4). When this notion is combined with the high prevalence rates on both a national and global level, alongside the aforementioned deleterious effects on psychosocial development and general
5 health, it is imperative from a preventive point of view, to systematically identify which children are at increased risk for abuse and/or neglect.
Situations in which children are at increased risk, are systematically charted through the use of risk factors, which indicate “... a probability of an outcome within a population of subjects” (Kraemer et al., 1997, p. 337). In general, risk factors are used to signify an increased chance of the occurrence of a specific event (e.g., child abuse and neglect), whilst protective factors function as a buffer (moderator effect) that prevent the occurrence during times of high risk (Folger & Wright, 2013; Kraemer et al., 1997).
Extensive research has been conducted regarding risk factors for child abuse and neglect, with many studies reporting on one, or several, risk factors. These risk factors can be categorized as being either static or dynamic (Douglas & Skeem, 2005), or as being part of the different ecological contexts in which a child exists, e.g., the child itself, the child’s family, and the overarching community, media, government and timeframe (Bronfenbrenner, 2000). An additional differentiation can be made between risk factors specific for the initial occurrence, and risk factors for possible subsequent recurrence of abuse and neglect (Lyons, Doueck, & Wodarski, 1996).
Static risk factors are factors that do not change over time, e.g., gender and ethnicity, and are therefore unlikely targets for intervention. However, due to this unchangeable nature, these factors play an important role in risk assessment (e.g., Lodewijks & Van Domburgh, 2012). Examples of static risk factors for child abuse and neglect are: prior maltreatment incidences (Connel, Vanderploeg, Katz, Caron, Saunders, & Tebes, 2009), low birth weight (Wu et al., 2004), and unintended pregnancy (Sidebotham, Heron, & The ALSPAC Study Team, 2003). Contrary to static factors, dynamic factors posses a capability of change, and can potentially improve due to an intervention (Douglas & Skeem, 2005). Examples of dynamic risk factors for child abuse and neglect are: parental alcohol abuse (Widom & Hiller-Sturmhöfel, 2001), parental drug abuse (Hurme, Alanko, Antilla, Juven, & Svedström, 2008), parental depression (Lee, Taylor, & Bellamy, 2012), and an adverse housing situation (Palusci, 2011).
Examples of ecological risk factors, in correspondence with a categorization according to Bronfenbrenner’s (2000) theory are: prematurity and overactivity of the child (individual child factor; Hurme et al., 2008), a low parental educational level (microsystem; Kotch, Brown, Dufort, & Winsor, 1999), a mother who smoked during pregnancy, families with three or more children and becoming a mother under twenty years of age (microsystem; Zhou, Hallisey, & Freymann, 2006), low levels of involvement in (in)formal community agencies
6 (exosystem; Sidebotham, Heron, Golding, & The ALSPAC Study Team, 2002), and the manner in which the media reports on child abuse and neglect (macrosystem; Ayre, 2001). Examples of other risk factors associated with the occurrence of child abuse and neglect, in line with the categorization as proposed by Lyons et al., (1996), are: maternal depressive symptoms, child’s low performance on a standardized developmental assessment, and low maternal education (Dubowitz et al., 2011). Examples of risk factors associated with subsequent recurrence are: caregiver alcohol abuse, depression, and lack of social support (Proctor et al., 2012).
This wide range of risk factors and varying classification methods, signifies that single risk factors are incapable of fully predicting either the occurrence, or recurrence, of child abuse and neglect. Instead, focusing on the accumulation of different risk factors proves to be far more useful (MacKenzie, Kotch, & Lee, 2011). An accurate prediction of which children are most likely to be subjected to abuse and neglect, is exactly what risk assessment instruments hope to achieve (Chan, 2012). Completely accurate predictions have (thus far) proven impossible, causing the contemporary focus to be on “imperfect, but better-than-chance predictive validity” (Johnson, 2011, p. 18).
Recently developed risk assessment instruments use the presence and absence of risk factors, combined with specific methods to aggregate the resulting scores, to predict the chance that child abuse and neglect will occur in a specific situation (e.g., Johnson, 2011). Examples of recently developed risk assessment instruments are: the Juhnke, Henderson, Juhnke Child Abuse and Neglect Risk Assessment scale, which predicts risk based on factors in the familial environment, alongside factors in the physical, demographic, and emotional/behavioral domains (Juhnke, Henderson, & Juhnke, 2013); the Child Abuse Risk Assessment Scale (CARAS), which predicts risk based on parental factors (Chan, 2012); the California Family Risk Assessment (CFRA), which predicts risk based on child, parental, and familial factors (Ten Berge, 2008; Johnson, 2011); the Child Abuse Potential Inventory (CAPI), which focuses solely on risk factors for physical abuse (Walker & Davies, 2010); and the Child Abuse Risk Evaluation – Nederland (CARE-NL), which focuses on parental, parent-child, child, and familial factors (De Ruiter, Hildebrand, & Van der Hoorn, 2012). The methods mentioned above do not incorporate the “risk-needs-responsivity” (RNR) model, which is widely used in penal law interventions (e.g., Andrews, Bonta, & Wormith, 2011; Ward, Melser, & Yates, 2007). According to the RNR model, risk factors are crucial in risk assessment and treatment of young offenders (Ward et al., 2007). To increase the validity and reliability of the aforementioned risk assessment instruments for child abuse and neglect,
7 an incorporation of the RNR model could lead to the design of more fruitful (preventive) intervention models, based on civil law.
The RNR model emphasizes three principles to guide intervention for criminal offenders. The “risk” principle suggests that, based on either a high or low risk for reoffending, the intervention offered should either be high or low in intensity. The “needs” principle states that only specific dynamic factors, which are associated with reductions in recidivism, should be targeted in interventions. Finally, the “responsivity” principle proposes that interventions should be matched with characteristics of the offender, to make optimal use of the offender’s personal and interpersonal circumstances (Andrews & Bonta, 2010; Ward et al., 2007). In (preventive) interventions regarding child abuse and neglect, the RNR model can potentially be incorporated to determine (1) in which situations a high intensity intervention is appropriate (“risk” principle), and (2) which dynamic risk factors should be treated (“needs” principle).
For the continuing improvement of risk assessment instruments, possibly based on the RNR model, and the improvement of (preventive) interventions regarding child abuse and neglect, systematic knowledge regarding the relative importance of different risk factors is urgently needed (Lodewijks & Van Domburgh, 2012). Until recently, no systematic meta-analytic review was available in which studies reporting on risk factors for child abuse and neglect were combined into an overview, detailing the relative importance of the different risk factors. Stith et al. (2009), were the first to provide such an overview. The publication highlighted the importance of examining child abuse and neglect from a multifactorial perspective, in adherence to Bronfenbrenner’s (2000) ecological theory. Also of importance were the findings that parent-child relationships and parental perception of the child, proved the two strongest risk factors for child neglect, whilst parental anger/hyper-reactivity, high family conflict and low family cohesion, yielded the largest effect sizes for the occurrence of physical child abuse (Stith et al., 2009).
However, the meta-analytic overview of Stith et al. (2009) has some important limitations. First and foremost, the authors did not perform a moderator analysis on their data. A moderator analysis offers insight into variables (e.g., characteristics of the study or sample) that potentially influence the strength or direction of a relationship between a predictor variable (e.g., risk factor), and a criterion variable (e.g., the occurrence of child abuse and neglect) (Rose, Holmbeck, Coackley, & Franks, 2004). An additional limitation is the lack of a multi-level analysis, which does not rely on averaging effect sizes, as is the case in the methodology used by Stith et al. (2009). A multi-level meta-analysis allows for the inclusion
8 of each individual effect size as reported in the primary studies, and therefore, an optimal statistical power of the (moderator) analyses is achieved. In addition, including all reported effect sizes will yield a more accurate estimate of the effect of risk factors [De Leeuw & Meijer, 2008]. Furthermore, as indicated by Stith et al. (2009), literature searches have only been conducted in the PsychINFO database, most likely causing the meta-analysis to be incomplete. In addition, several relevant publications were excluded from the meta-analysis, no correction was made for the potential “file drawer bias” (e.g., Rosenberg, 2005), and several of the largest effect sizes were computed from a relatively small number of studies (Stith et al., 2009).
Considering the importance of systematic, accurate knowledge regarding risk factors for the occurrence of child abuse and neglect, combined with the shortcomings in the study of Stith et al. (2009), a replication and update of the analysis is needed. The current meta-analysis aims to achieve this through incorporation of the methodologies mentioned above, whilst simultaneously including recently published primary studies on risk factors for child abuse and neglect. Since individual risk factors are incapable of fully predicting child abuse and neglect, and to counteract statistical power problems which often arise when comparing individual risk factors, risk factors were clustered into domains, consisting of highly related risk factors (see Appendix A). In sum, the aim of the present study was to meta-analytically determine the effect of (domains of) risk factors for child abuse and neglect. A second aim was to determine how effects of risk factors are moderated by risk factor, sample and study characteristics.
Method Literature Search
Literature searches were conducted for published and unpublished articles, doctoral dissertations, and book(chapters), written in either English or Dutch, within the PsycINFO, ERIC, Sociological abstracts, Pubmed, and Google Scholar databases. Combinations of the following search terms were used: risk*, factor*, child*, maltreatment*, abuse*, neglect*. When the literature found was deemed potentially eligible for inclusion in the meta-analysis based on their respective titles, the abstracts and full-texts were read for a further assessment of relevance, and to verify if the inclusion criteria were met.
9 Inclusion Criteria
Firstly, the article had to have been published after 2003, to achieve a correct update of the meta-analysis of Stith et al. (2009), who used 2003 as the ultimate date for study inclusion. Secondly, studies had to make use of both a control group of non-abused and non-neglected children, and an experimental group, consisting of children that did experience a form of child abuse and/or neglect. Thirdly, the article had to report sufficient statistical data to allow for the calculation of Cohen’s d as effect size (Cohen, 1988). Finally, the article had to report univariate analysis of the data, to allow for the determination of the bivariate association between risk factor(s) and the occurrence of child abuse and neglect.
Coding of Studies
Coding of the studies was done according to a specifically developed coding form, which was based on the coding form of Lipsey & Wilson (2001). With the final version of the coding form, the following variables were coded: characteristics of the children included in the studies (e.g., age, gender, socioeconomic status), characteristics of the parents or caregivers (e.g., educational level, age at child birth), the specific forms of child abuse and neglect that were examined in the studies (e.g., physical abuse, emotional neglect), methodological characteristics of the studies (e.g., design, drop-out rate, sample size), reported risk factors (e.g., parental, demographic, child domains) and effect size data for conversion to Cohen’s d (e.g., standard deviation, odds-ratio, chi-squared values).
Although the majority of domain titles are self-explanatory (e.g., ‘child age’), several domains require a brief description. ‘Child static factors not otherwise classified’ comprises, among others, risk factors related to a history of foster care, low birth weight, and APGAR score. ‘Parental static factors’ comprises, among others, a parental criminal history, unwanted pregnancy, and parental age. ‘Parental intergenerational continuity’ comprises a parental childhood history of maltreatment. ‘Family immigration factors’ comprises a recent migratory status and the attitude towards acculturation. ‘Family environment’ comprises, among others,
the number of social family and community connections, and feelings of isolation.
Four studies were coded separately by the author and a co-student who had a significant contribution in the collection and analysis of the included data, and who wrote a Master’s thesis using the same database (Van Stokkom, 2014). This allowed for the calculation of interrater reliability, which ranged from 66.67% on variables related to the educational level of the sample, to 100% on variables related to the type of child abuse and neglect. The averaged interrater reliability proved to be high at 78.88% agreement (Leary,
10
2008). The final dataset was extensively checked by both coders before data analysis commenced, to further increase interrater reliability. Any remaining ambiguous items were extensively discussed, until consensus was reached.
One study (Sledjeski et al., 2008) included two very distinct research samples. To prevent possible confounding influences, both samples were coded as separate studies.
Data Analysis
Statistical analysis of the data was done using MLwiN, version 2.3.0, made suitable for a meta-analysis by using settings in adherence to Hox’ (2002) proposition. For the calculation of an overall effect and the conduction of moderator analyses, a multilevel random-effects model was used (Hox, 2002; Van den Noortgate & Onghema, 2003). This model takes into account the hierarchical structure of the data, in which the effect sizes (or study results), are nested within studies (De Leeuw & Meijer, 2008). The random-effects model can be expanded through inclusion of moderators. Estimates of the unknown parameters can be obtained through the use of iterative maximum-likelihood-procedures, in which iterative estimates are made of every coefficient in the model (Cole, Chu, & Greenland, 2014). This method increases the accuracy of the estimations, until the data, as predicted by the model, fits the dataset (e.g., Hoeve, Stams, Van der Put, Dubas, Van der Laan, & Gerris, 2012). A test for homogeneity was conducted, to assess whether effect sizes were constant between studies. Since the data proved heterogeneous (see results section), moderator analyses were conducted, in which discrete and continuous moderator variables were used to explain differences in effect sizes (Van der Put, Assink, Bindels, Stams, & De Vries, 2013). See Hox (2002), and Van den Noortgate & Onghena (2003), for an in-depth elaboration of the multilevel random-effects model.
Since individual studies reported on more than a single risk factor, Cohen’s d effect sizes were calculated for every relevant measure within a study. This was done using formulas proposed by Lipsey & Wilson (2001) and Mullen (1989). Cohen (1992), considers an effect size of >0.20 to be small, >0.50 to be moderate, and >0.80 to be large.
In the context of a meta-analysis, it is important to consider the knock-on effects of journals and researchers accepting or submitting only those studies which report significant associations between variables. This may result in unpublished results, which, when these results would have been published and included in the meta-analysis, could impact the conclusions (Reed, 2009). This is referred to as the “file drawer problem” (Rosenthal, 1979). To estimate the number of studies averaging null results, which would reduce the overall
11 significant effect size to non-significance, a fail-safe number is computed (Rosenthal, 1979; Reed, 2009). The fail safe number is generally considered to be robust when it exceeds 5 * k + 10, where k is the total amount of effect sizes included in the analysis (Rosenthal, 1991). Although the fail-safe method is often used in meta-analyses, applicability in multilevel meta-analysis is statistically less valid. This is due to a multilevel model resulting in multiple effect sizes from a single study, whilst the fail-safe method assumes a single averaged effect size per study. For a more robust conclusion, and in line with Rothstein’s (2008) proposition, the fail-safe method was used in conjunction with Egger’s regression test (1997; Hoeve et al., 2012). This method places the distribution of each individual effect size on the horizontal axis, against the standard error on the vertical axis. The presence of publication bias is then determined using a funnel plot, with a violation of funnel plot symmetry reflecting publication bias (Sutton et al., 2000).
Results Study Descriptors
Literature searches resulted in a total of 211 studies (excluding duplicate findings), of which the title indicated potential inclusion eligibility. Abstracts and full-texts were read to verify relevance, after which 82 studies were removed from the selection. The remaining 129 studies were thoroughly examined to verify whether the inclusion criteria were met: 39 studies did not meet the criteria of reporting on a control group, 21 studies did not report sufficient statistical data, and 35 studies reported only multivariate results. The authors of relevant studies were contacted when insufficient statistical information was reported to calculate a Cohen’s d value for each individual risk factor. However, no additional information was received. This procedure led to a final inclusion of 34 studies which met all the inclusion criteria, yielding a total of 468 effect sizes. Included studies are marked with an * in the reference list.
The total sample size of all studies combined was N = 639,624, consisting of n = 89,618 children in the maltreatment group, and n = 550,006 children in the control group. Study sample sizes ranged from 48 to 530,832 participants. The overall mean child age at start of the studies was 5.42 years (SD = 4.67), and the overall mean parental age was 33.06 years (SD = 8.10).
12 Publication Bias
Based on the aggregated effect sizes, the fail-safe number was 4,909,461. Since this figure is larger than the critical value of 2,350 (5 * 468 + 10), it is unlikely that the final results were biased due to the file-drawer problem (Rosenthal, 1991).
By contrast, Egger’s regression test yielded strong evidence for the presence of publication bias. Regression of the standard normal deviate against the estimate’s precision revealed a strong indication of funnel plot asymmetry (t = 6.648, p < 0.001), see appendix B. Overall Effect
The overall effect of the risk factors was small to moderate (Cohen, 1992), and highly significant (mean d = 0.432, p < 0.001, see Table 1). Results of the homogeneity test revealed that the results were heterogeneous (Z = 4.058, p < 0.001), which meant that although a significant overall effect was found, this effect was not consistent across all studies. This implied that the results of the included studies were likely influenced by characteristics of those studies. Subsequently, moderator analyses were conducted to investigate the origin of this influence.
Table 1 shows the results from analyses of the discrete moderator variables and Table 2 shows the results from the analyses of the continuous moderator variables. Only moderating variables that significantly improved model fit (p < 0.05) were included in the tables, and are described in the text. Full tables, including results where model fit did not improve significantly, are given in Appendix C for discrete moderator variables, and in Appendix D for continuous moderator variables.
Risk Factor Characteristics
Firstly, a significant moderating effect was found for the type of parent who is associated with risk factors. Compared to risk factors that applied to any parental figure (i.e., parent unspecified), risk factors that applied to both parents showed larger effect sizes (Z = 9.371, p < 0.001). By contrast, risk factors that applied solely to the father figure (Z = -8.355, p < 0.001), or solely to the mother figure (Z = -64.456, p < 0.001), reduced the effect.
As for the type of maltreatment, the effect of risk factors proved to be highly different between different maltreatment types. Compared to the unspecified maltreatment type, smaller effect sizes were found for sexual abuse (Z = -1.976, p < 0.05), unspecified neglect (Z = -2.292, p < 0.05), and multiple forms of maltreatment (Z = -3.655, p < 0.001).
Table 1
Overview of the Overall Mean Effect Sizes and Discrete Moderator Variables (Bivariate Models)
Moderator variables #Studies #ES β1 (SE) Z1 Mean. d (SD) Z0 Heterogeneity Δ Fit
Overall effect 34 468 0.432 (0.057) 7.573*** 4.048*** -
Risk factor characteristics
Parent associated with risk factor 4.042*** 4428.003***
Unspecified (RC) 30 281 0.459 (0.056) 8.171***
Both Parents 16 32 0.085 (0.009) 9.371*** 0.544 (0.057) 9.585***
Mother(figure) 18 106 -0.117 (0.002) -64.456*** 0.341 (0.056) 6.085*** Father(figure) 9 49 -0.105 (0.013) -8.355*** 0.354 (0.057) 6.178***
Maltreatment type 4.054*** 74.643***
Unspecified maltreatment type (RC) 13 167 0.581 (0.088) 6.636*** Physical Abuse 12 135 -0.190 (0.114) -1.669+ 0.391 (0.073) 5.374***
Sexual Abuse 5 26 -0.227 (0.115) -1.976* 0.354 (0.074) 4.779***
Abuse Type Unspecified 2 15 -0.324 (0.246) -1.317 0.257 (0.230) 1.117 Neglect Type Unspecified 7 104 -0.262 (0.114) -2.292* 0.319 (0.073) 4.350*** Multiple Forms Of Maltreatment 3 21 -0.426 (0.117) -3.655*** 0.155 (0.077) 2.008*
Domain 4.022*** 33135.617***
Child Static Factors 12 20 0.253 (0.051) 4.940***
Not Otherwise Classified (RC)
Child Age 8 13 0.447 (0.020) 22.905*** 0.700 (0.054) 12.909***
Child Gender 16 16 -0.186 (0.005) -34.531*** 0.067 (0.051) 1.314
Child Ethnicity 6 8 0.041 (0.024) 1.729+ 0.294 (0.056) 5.275***
Child Internalizing Problems 8 10 -0.048 (0.022) -2.161* 0.204 (0.055) 3.712*** Child Externalizing Problems 4 5 0.017 (0.049) 0.343 0.270 (0.070) 3.852*** Child Physical Health 5 5 0.017 (0.024) 0.721 0.270 (0.056) 4.847***
14 Parental Static Factors 13 21 0.247 (0.004) 58.045*** 0.499 (0.051) 9.782***
Parental Ethnicity 8 19 -0.179 (0.005) -39.690*** 0.074 (0.051) 1.448 Parental Age at Childbirth 8 10 0.033 (0.005) 7.370*** 0.286 (0.051) 5.595*** Parental Personality/Temperament 9 27 0.501 (0.013) 39.123*** 0.754 (0.052) 14.471*** Parental Mental Health Issues 15 23 0.422 (0.024) 17.675*** 0.675 (0.055) 12.182*** Parental Unemployment 11 16 0.109 (0.011) 9.695*** 0.362 (0.052) 6.992*** Parental Education 17 42 0.158 (0.005) 33.374*** 0.411 (0.051) 8.038*** Parental Upbringing Skills 14 21 0.201 (0.027) 7.482*** 0.453 (0.057) 7.987*** Parental Substance Use 14 29 0.269 (0.010) 25.827*** 0.522 (0.052) 10.127*** Parental Intimate Partner Violence 9 15 0.568 (0.019) 30.663*** 0.820 (0.054) 15.322*** Parental Intergenerational Continuity 9 14 0.339 (0.014) 23.734*** 0.592 (0.052) 11.295*** Family Static Factors 15 25 -0.034 (0.005) -7.166*** 0.219 (0.051) 4.281*** Family Marital Status 19 37 0.131 (0.012) 11.122*** 0.384 (0.052) 7.422*** Family SES Factors 17 38 0.380 (0.011) 34.832*** 0.633 (0.052) 12.247*** Family Environment 13 28 0.232 (0.017) 13.799*** 0.485 (0.053) 9.143*** Family Immigration Factors 4 9 -0.068 (0.023) -2.896** 0.185 (0.056) 3.325*** Child Parent Interaction 5 6 -0.050 (0.005) -9.077*** 0.203 (0.051) 3.968*** Parental Interaction Factors 4 11 0.183 (0.013) 13.774*** 0.436 (0.052) 8.341***
Risk factor type 4.048*** 41.698***
Static (RC) 34 318 0.423 (0.057) 7.413***
Dynamic 25 150 0.037 (0.006) 6.457*** 0.460 (0.057) 8.041***
Note. # Studies = number of independent studies; # ES = number of effect sizes; Mean d = mean effect size; β1 = regression coefficient; SD =
standard deviation; SE = standard error; Z0 = significance of intercept / mean d; Z1 = significance of moderator; Heterogeneity = within class
heterogeneity (Z); Δ Fit = difference with model without moderators ( 2); RC = reference category; SES = social economic status.
Table 2
Results for the Continuous Moderator Variables (Bivariate Models)
Note. #Studies = number of independent studies; #ES = number of effect sizes; β0 = intercept; β1= regression coefficient; SE = standard error; Z0
= significance of intercept; Z1 = significance of moderator; Heterogeneity = within class heterogeneity (Z); Δ Fit = difference with model without
moderators ( 2); RC = reference category.
+ p < .10; * p < .05; ** p < .01; *** p < .001.
Moderator variables #Studies #ES β0 (SE ) Z0 β1 (SE ) Z1 Heterogeneity Δ Fit
Sample Descriptors
Parental educational level high school (%) Abuse Type Unspecified (n) 7.038* 8 153 0.258 (0.072) 3.583*** -0.011622 (0.009638) -1.206 2.037* 6.254* 1 24 0.692 (0.061) 11.344*** -0.034700 (0.009427) -3.681*** 0
A significant and highly varying moderating effect was found for the domains of risk factors, in which strongly related risk factors were clustered. The results revealed that the domain of risk factors associated with intimate partner violence among parents, yielded a large mean effect size (d = 0.820, Z = 15.322, p < 0.001). Six domains yielded moderate mean effect sizes: ‘child age’ (d = 0.700, Z = 12.909, p < 0.001); ‘parental personality/temperament’ (d = 0.754, Z = 14.471, p < 0.001); ‘parental mental health issues’ (d = 0.675, Z = 12.182, p < 0.001); ‘parental substance use’ (d = 0.522, Z = 10.127, p < 0.001); ‘parental intergenerational continuity’ (d = 0.592, Z = 11.295, p < 0.001); and ‘family SES factors’ (d = 0.633, Z = 12.247, p < 0.001). ‘Child externalizing problems’ (Z = 0.343, p > 0.10), and ‘child physical health’ (Z = 0.721, p > 0.10), revealed no significant effects. The remaining sixteen domains all yielded small effect sizes.
Moderator analysis of the reported risk factor type revealed a significant difference between the effect of static and dynamic risk factors. Dynamic risk factors yielded significantly higher effect sizes (Z = 6.457, p < 0.001).
Sample Descriptors
No moderator effect was found for ‘parental educational level high school (%)’ (Z = -1.206, p > 0.10), although delta fit improved significantly when this variable was included as a moderator (p < 0.05). The variable indicating study sample sizes of children subjected to unspecified abuse, revealed a significant effect when added to the model. The corresponding regression coefficient was highly significant (Z = -3.681, p < 0.001), whilst heterogeneity was exactly zero, indicating full consistency in study results on this variable.
Discussion
Child abuse and neglect has a profound impact on everyone involved, and places a large personal and financial burden on the individual, societal and global level (e.g., Felliti et al., 1998; Florence et al., 2013). These personal and financial burdens however, can be prevented to a large extent (Korbin, 1988). The quest for increasingly effective intervention and prevention strategies, requires a systematic focus on risk factors through both an improvement in contemporary theoretical knowledge, and a subsequent practical implementation of this knowledge.
Risk factors for child abuse and neglect have been studied in a wide variety of situations, and in a wide array of populations across the globe (e.g., Hurme et al., 2008; Chan et al., 2011; Mikaeili, Barahmand, & Abdi 2013). Although these studies were very important
17 for improving our understanding of this widespread and highly prevalent phenomenon, their conclusions only hold for the targeted research group, and thus lack a wider applicability and generalizability to other populations. A systematic integration of all available research on the effect of risk factors for the occurrence of child abuse and neglect, is therefore vital. To the best of the author’s knowledge, only one such study is currently available (Stith et al., 2009), which has several (previously explained) shortcomings. Therefore, this study sought to expand and improve the current body of knowledge, through a statistically sound meta-analytic integration of risk factors for child abuse and neglect. This study is the first to conduct a moderator analysis on the included data, using a multilevel model.
The results of the current study showed that the overall effect of risk factors for child abuse and neglect was small to moderate according to the criteria set by Cohen (1992), and highly significant. This indicates the importance of examining risk factors, since children are significantly more likely to be subjected to abuse and neglect when these factors are present in their environment. The results revealed a varying effect for the different domains of risk factors, which signifies that substantial differences exist between the predictive power of the domains. The largest mean effect size, and therefore the domain with the highest predictive power, was found for the domain concerning ‘parental intimate partner violence’. The strong connection between intimate partner violence and child abuse, has long been established (e.g., Appel & Holden, 1998). Chan (2011) points out that children are likely to be involved in intimate partner violence, either indirectly, through being a witness of the violence, or directly, by being targeted themselves.
Moderate mean effect sizes were found for six domains: ‘child age’ (younger children are at increased risk), ‘parental personality/temperament’ (e.g., violence approval), ‘parental mental health issues’ (e.g., being depressed), ‘parental substance use’ (e.g. illicit drug use), ‘parental intergenerational continuity’ (e.g., having a history of child abuse and neglect), and ‘family SES factors’ (e.g., relying on governmental financial assistance). The remaining eighteen domains yielded small effect sizes.
Overall, it is remarkable that five out of seven domains with the highest mean effect sizes are related to the parental figure(s). Although the majority of child and familial factors also show consistent and statistically relevant mean effect sizes, these findings implicate that (preventive) interventions and risk assessment instruments should focus most on changing dynamic parental factors. This is broadly in line with the findings from Stith et al. (2009), who found eight risk factors with large mean effect sizes, six of which concerned parental risk factors.
18 Reducing the influence of risk factors is not possible where static factors are concerned, since these are by definition unchangeable (e.g., Lofthouse, Totsika, Hastings, Lindsay, Hogue, & Taylor, 2014), although they remain highly important for risk assessment (Lodewijks & Van Domburgh, 2012). Therefore, an important and logical aspect of any effective (preventive) intervention, is a strong focus on dynamic risk factors. The present results indicated that effects of dynamic risk factors were slightly but significantly larger than effects of static risk factors. This implicates that although the risk for child abuse and neglect to occur can never be completely ruled out when both static and dynamic risk factors are present, there is ample room for improvement in the situation of at-risk children.
Considering that risk factors are crucial in risk assessment (Ward et al., 2007), and based upon the RNR model (Andrews et al., 2011), the current findings also have potentially important implications for risk assessment instruments used in civil law practice. Related to the “needs” principle, emphasis should be placed on targeting dynamic parental factors which exert the largest influence on child abuse and neglect. Incorporation of the “risk” principle is then straightforward: when the number of risk factors belonging to the aforementioned domains with the largest mean effect sizes are present, a higher intensity (preventive) intervention is needed.
However, as indicated by the test of homogeneity, the effect was not consistent between studies and risk factors, signifying that the results were impacted by characteristics of the samples, studies, and/or risk factors. Moderator analyses revealed that the overall effect was significantly moderated by characteristics of the risk factors. Regarding the type of parent associated with risk factors, the strongest effect was found for risk factors that applied to both parents. A possible explanation is that when a risk factor such as substance abuse applies to both parents, the involved child is worse off, compared to a situation in which only one parent is associated with the risk factor. When one of the parental figures is intoxicated, the other parent may still be able to provide adequate care and attention to the child.
The effect of risk factors between mother(figures) and father(figures) proved almost similar, as signified by the close similarity between the respective mean d values and regression coefficients. Prior findings have indicated the existence of a relation between parental gender and an increased risk for the perpetration of specific forms of child maltreatment (e.g., abuse by father(figures), and neglect by mother(figures)) (McCoy & Keen, 2009). Although the current study did not specifically focus on this question, no indication for this specific relation between parental gender and increased perpetrator risk was found. The current result might be explained by the incorporation of every form of child abuse and
19 neglect into the moderator analysis, thereby possibly creating a confounding influence. A different explanation stems from the overrepresentation of the unspecified category of parental(figures), to whom the risk factor applies. The disparity between parental(figure) group sizes in the current sample, might have skewed the results. If the included data allowed for a more clear determination which risk factors applied to mother(figures) or father(figures), an effect between parental gender and increased perpetrator risk might have been found, due to increased statistical power to test for differences between parental categories.
Although a total of thirteen included studies did not make a clear differentiation between maltreatment types, making the unspecified maltreatment type the largest category, varying effects of risk factors were found for the different maltreatment types. Sexual abuse, unspecified neglect, and especially multiple forms of maltreatment yielded the lowest effect sizes, which implies that these forms of maltreatment have the lowest predictive power. These relatively low effects might be explained by differences in the degree of visibility and secrecy surrounding these maltreatment types. Whilst physical abuse is a relatively overt type of maltreatment, with an inherent chance to leave a visible mark on children, sexual abuse is often painstakingly kept secret (e.g., McCoy & Keen, 2009), making it harder to accurately predict this latter form of maltreatment. The low effect found for child neglect, and the inherent difficulty in predicting this type of maltreatment, is likely caused by a combined lack of societal consensus as to what constitutes neglect, and the finding that “neglect rarely contains enough visual impact for social services to consider these children as being in serious harm or to be very needy” (McSherry, 2007, p. 609). A second explanation for the low effects of the aforementioned maltreatment types, is the potential importance of other (currently unknown) risk factors, into which no, or very little, research has been conducted. Future research is needed to increase the contemporary knowledge of risk factors for child abuse and neglect, and to expand upon the currently included (domains of) risk factors.
When interpreting these results, several limitations should be taken into account. Firstly, based on the current research findings, no causal inferences can be made between the presence of a (domain of) risk factor(s), and the actual occurrence of child abuse and neglect. The primary studies included in the current meta-analysis aimed to determine potential differences between a maltreated and a maltreated group of children, using a non-experimental study design. It is both ethically unallowable, and practically unachievable, to conduct experimental research in the field of child maltreatment, which would rely on randomly assigning children to situations in which one or more risk factors are present, and subsequently monitoring if child abuse and neglect does, or does not, occur.
20 Secondly, although data was gathered from a wide variety of sources and databases to provide a meta-analytic overview that is as comprehensive as possible, the chosen timeframe caused several relevant publications to be excluded from the current study. This timeframe was chosen to (1) allow for an update of the study conducted by Stith et al. (2009), and (2) related to the limited available time in the present graduate research project. It is highly likely that this decision has influenced the results, since excluded studies (e.g., Kotch et al., 1999), might have reported data which is not in concordance with the current findings. On the other hand, the advantage of including only recently published studies is that it has provided a recent and valuable insight into risk factors which are most relevant in contemporary society.
A third limitation of this study is the presence of publication bias, as was indicated by results from the Egger’s regression test (1997). Although the fail-safe method yielded no evidence for publication bias, Egger’s regression test is better suited for a multilevel model, and evidence for publication bias is therefore based on the latter method. The reported funnel plot asymmetry implicates a selective inclusion of studies showing positive or negative results (Sutton et al., 2000), thereby possibly skewing the reported results of the current study. A fourth limitation of this study, is the missing data for children subjected to (solely) emotional abuse, physical neglect, emotional neglect, and educational neglect. This is due to the fact that none of the included studies reported on data of children subjected to only one of these different maltreatment types. It is therefore not possible to determine whether the current results are also applicable to these maltreatment types. Especially striking is that no data was found for three different types of neglect, whilst data was missing for only a single type of abuse. This scarcity of data is a logical consequence of a systematic lack of scientific attention, and hence, a lack of research concerning child neglect, which has been referred to as ‘the neglect of neglect’ (e.g., Dubowitz, 2007).
Lastly, there was a large number of studies that could not be included due to a diversity of methodological deficiencies. Of the 129 potentially suitable studies that looked into the relation between risk factors and the occurrence of child abuse and neglect, the vast majority (95 studies) could ultimately not be included in the current meta-analysis, due to reasons that could have been fully prevented by the primary researchers. For example, many studies were excluded from inclusion in the current meta-analysis due to a lack of statistical data reported, whilst this data must have been available to the researchers, based on the reported conclusions in those publications. A similar effect occurred in studies reporting on multivariate results, since univariate results were often not described in these studies. The final reason for the large gap between potentially eligible data, and the amount of data that
21 was ultimately included in the current meta-analysis, is induced by a frequently recurring methodological deficiency: the absence of comparisons between maltreated and non-maltreated groups. Although the methodological design of studies in this field is subject to several ethical limitations, researchers designing new studies are urged to design and carry out studies that take these principles into account as much as possible, to further increase knowledge in this field.
Although the aforementioned limitations may potentially have impacted the results, the current study offers a unique insight into the relation between risk factors and the occurrence of child abuse and neglect, based on recently published studies. This is the first multilevel meta-analytic study of risk factors for child abuse and neglect in which moderator analyses were conducted.
In sum, the results revealed that the domain related to ‘parental intimate partner violence’ yielded a large mean effect size, whilst four domains related to parental risk factors yielded moderate mean effect sizes: ‘parental intimate partner violence’, ‘parental personality/temperament’, ‘parental mental health issues, ‘parental substance use’, and ‘parental intergenerational continuity’. Moderate mean effect sizes were also found for ‘child age’ indicating younger children to be at increased risk, and for ‘family SES’, signifying the importance of taking into account the socio-economic status of the family in which the child grows up.
Professionals in the field of child protective services, employed in either a theoretical or practical position, are highly recommended to familiarize themselves with these findings, and to incorporate this knowledge in the development and improvement of (preventive) interventions, policies and risk assessment instruments. This in turn will, hopefully, decrease the number of children subjected to any form of maltreatment, and simultaneously reduce the individual, societal, and global burden associated with child abuse and neglect.
References
References marked with an asterisk were included in the meta-analysis
*Alexandre, G. C., Nadanovsky, P., Moraes, C. L., & Reichenheim, M. (2010). The presence of a stepfather and child physical abuse, as reported by a sample of Brazilian mothers in Rio de Janeiro. Child Abuse & Neglect, 34(12), 959-966. doi: 10.1016/j.chiabu.2010
.06.005
Alink, L., Van IJzendoorn, R., Bakermans-Kranenburg, M., Pannebakker, F., Vogels, T., & Euser, S. (2011). De tweede nationale prevalentiestudie mishandeling van kinderen en jeugdigen (NPM-2010). Leiden Attachment Research Program, Algemene en Gezinspedagogiek, Centrum voor Gezinsstudies, Universiteit Leiden, TNO.
Anda, R. F., Croft, J. B., Felitti, V. J., Nordenberg, D., Giles, W. H., Williamson, D. F., & Giovino, G. A. (1999). Adverse childhood experiences and smoking during adolescence and adulthood. Journal of the American Medical Association, 282(17), 1652-1658. doi: 10.1001/jama.282.17.1652
Andrews, D, A., & Bonta, J. (2010). Rehabilitating criminal justice policy and practice. Psychology, Public Policy, and Law, 16(1), 39-55. doi: 10.1037/a0018362 Andrews, D. A., Bonta, J., & Wormith, J. S. (2011). The risk-need-responsivity (RNR) model: Does adding the good lives model contribute to effective crime prevention? Criminal Justice and Behavior, 38(7), 735-755. doi: 10.1177/0093854811406356
*Annerbäck, E., Wingren, G., Svedin, C. G., & Gustafsson, P. A. (2010). Prevalence and characteristics of child physical abuse in Sweden - findings from a population-based youth survey. Acta Paediatrica, 99(8), 1229-1236. doi: 10.1111/j.1651-2227.2010
01792.x
Appel, A. E., & Holden, G. W. (1998). The co-occurrence of spouse and physical child abuse: A review and appraisal. Journal of Family Psychology, 12(4), 578-599. doi: 10.1037 /0893-3200.12.4.578
Ayre, P. (2001). Child protection and the media: Lessons from the last three decades. British Journal of Social Work, 31(6), 887-901. doi: 10.1093/bjsw/31.6.887
Barth, J., Bermetz, L., Heim, E., Trelle, S., & Tonia, T. (2013). The current prevalence of child sexual abuse worldwide: A systematic review and meta-analysis. International Journal of Public Health, 58(3), 469-483. doi: 10.1007/s00038-012-0426-1
*Bartlett, J. D. (2012). Young mothers, infant neglect and discontinuities in intergenerational cycles of maltreatment (Doctoral dissertation). Available from: http://gradworks.umi.com/34/95/3495286.html
23 *Bartlett, J. D., Raskin, M., Kotake, C., Nearing, K. D., & Eaterbrooks, M. A. (2014). An ecological analysis of infant neglect by adolescent mothers. Child Abuse & Neglect, 38(4), 723-734. doi: 10.1016/j.chiabu.2013.11.011
*Briscoe-Smith, A. M., & Hinshaw, S. P. (2006). Linkages between child abuse and attention- deficit/hyperactivity disorder in girls: Behavioral and social correlates. Child Abuse & Neglect, 30(11), 1239-1255. doi:10.1016/j.chiabu.2006.04.008
Bronfenbrenner, U. (2000). Ecological theory. In A. Kazdin (Ed.), Encyclopedia of psychology. Washington, DC: American Psychological Association and Oxford University Press.
*Butler, A. C. (2013). Child sexual assault: Risk factors for girls. Child Abuse & Neglect, 37(9), 643-652. doi: 1016/j.chiabu.2013.06.009
*Chan, K. L. (2011). Co-occurrence of intimate partner violence and child abuse in Hong Kong Chinese families. Journal of Interpersonal Violence, 26 (7), 1322-1342. doi: 10.1177/0886260510369136
*Chan, K. L. (2012). Evaluating the risk of child abuse: The child abuse risk assessment scale (CARAS). Journal of Interpersonal Violence, 27(5), 951-973. doi: 10.1177 /0886260511423252
Chapman, D. P., Whitfield, C. L., Felitti, V. J., Dube, S. R., Edwards, V. J., & Anda, R. F. (2004). Adverse childhood experiences and the risk of depressive disorders in adulthood. Journal of Affective Disorders, 82(2), 217-225. doi: 10.1016/j.jad.2003.12 .013
Chartier, M. J., Walker, J. R., & Naimark, B. (2010). Separate and cumulative effects of adverse childhood experiences in predicting adult health and health care utilization. Child Abuse & Neglect, 34(6), 454-464. doi: 10.1016/j.chiabu.2009.09.020 Cohen, J. (1988). Statistical power analysis for the behavioral sciences. New York: Academic Press.
Cohen, J. (1992). A power primer. Psychological Bulletin, 112(1), 155-159.
Cole, S. R., Chu, H., & Greenland, S. (2014). Maximum likelihood, profile likelihood, and penalized likelihood: A primer. American Journal of Epidemiology, 179(2), 252-260. doi: 10.1093/aje/kwt245
*Connel, C. M., Vanderploeg, J. J., Katz, K. H., Caron, C., Saunders, L., & Tebes (2009). Maltreatment following reunification: Predictors of subsequent child protective services contact after children return home. Child Abuse & Neglect, 33(4), 218-228. doi: 10.1016/j.chiabu.2008.07.005
24 De Leeuw, J., & Meijer, E. (2008). Handbook of Multilevel Analysis. New York, U.S.: Springer Science+Business Media, LLC.
De Ruiter, C., Hildebrand, M., & Van der Hoorn, S. (2012). Gestructureerde risicotaxatie bij kindermishandeling: De child abuse risk evaluation - Nederlandse versie (CARE-NL). GZ-Psychologie, 3, 10-17.
*Diderich, H. M., Fekkes, M., Verkerk, P. H., Pannebakker, F. D., Klein Velderman, M., Sorensen, P. J. G., Baeten, P., & Oudesluys-Murphy, A. M. (2013). A new protocol for screening adults presenting with their own medical problems at the emergency department to identify children at high risk for maltreatment. Child Abuse & Neglect, 37(12), 1122-1131. doi: 10.1016j.chiabu.2013.04.005
*Dixon, L., Browne, K., & Hamilton-Giachritsis, C. (2005). Risk factors of parents abused as children: a mediational analysis of the intergenerational continuity of child maltreatment (Part I). Journal of Child Psychology and Psychiatry, 46(1), 47-57. doi: 10.1111/j.1469-7610.2004.00339.x
*Dixon, L., Browne, K., Hamilton-Giachritsis, C. (2009. Patterns of risk and protective factors in intergenerational cycle of maltreatment. Journal of Family Violence, 24(2), 111-122. doi: 10.1007/s10896-008-9215-2
Dong, M., Anda, R., Felitti, V. J., Dube, S. R., Williamson, D. F., Thompson, T. J., . . . Giles, W. H. (2004). The interrelatedness of multiple forms of childhood abuse, neglect, and household dysfunction. Child Abuse & Neglect, 28(7), 771-784. doi: 10.1016/j.chiabu
.2004.01.008
Dong, M., Giles, W. H., Felitti, V. J., Dube, S. R., Williams, J. E., Chapman, D. P., & Anda, R. F. (2004). Insights into causal pathways for ischemic heart disease. Circulation, 110(13), 1761-1766. doi: 10.1161/01.CIR.0000143074.54995.7F
Douglas, K. S., & Skeem, J. L. (2005). Violence risk assessment: Getting specific about being dynamic. Psychology, Public Policy, and Law, 11(3), 347-383. doi: 10.1037/1076 -8971.11.3.347
Dube, S. R., Anda, R. F., Felitti, V. J., Chapman, D. P., Williamson, D. F., & Giles, W. H. (2001). Childhood abuse, household dysfunction, and the risk of attempted suicide throughout the life span: Findings from the adverse childhood experiences study. Journal of the American Medical Association, 286(24), 3089-3096. doi: 10.1001/jama .286.24.3089
25 childhood experiences and personal alcohol abuse as an adult. Addictive behaviors, 27(5), 713-725. doi: 10.1016/S0306-4603(01)00204-0
Dube, S. R., Felitti, V. J., Dong, M., Chapman, D. P., Giles, W. H., & Anda, R. F. (2003). Childhood abuse, neglect, and household dysfunction and the risk of illicit drug use: The adverse childhood experiences study. Pediatrics, 111(3), 564-572. doi: 10.1542 /peds.111.3.564
Dubowitz, H. (2007). Understanding and addressing the “neglect of neglect:” Digging into the molehill. Child Abuse & Neglect, 31, 603-606. doi: 10.1016/j.chiabu.2007.04.002 *Dubowitz, H., Kim, J., Black, M. M., Weisbart, C., Semiatin, J., & Magder, L. S. (2011). Identifying children at high risk for a child maltreatment report. Child Abuse & Neglect, 35(2), 96-104. doi: 10.1016/j.chiabu.2010.09.003
Egger, M. (1997). Bias in meta-analysis detected by a simple, graphical test. British Medical Journal, 315, 629. doi: 10.1136/bmj.315.7109.629
Felitti, V. J., Anda, R. F., Nordenberg, D., Williamson, D. F., Spitz, A. M., Edwards, V., . . . Marks, J. S. (1998). Relationship of childhood abuse and household dysfunction to many of the leading causes of death in adults: The adverse childhood experiences (ACE) study. American Journal of Preventive Medicine, 14(4), 245-258. doi: 10.1016/ S0749-3797(98)00017-8
Florence, C., Brown, D. S., Fang, X., & Thompson, H. F. (2013). Health care costs associated with child maltreatment: Impact on Medicaid. Pediatrics, 132(2), 312-318. doi: 10
.1542/peds.2012-2212
Folger, S. F., & Wright, M. O. (2013). Altering risk following child maltreatment: Family and friend support as protective factors. Journal of Family Violence, 28(4), 325-337. doi: 10.1007/s10896-013-9510-4
*Francis, K. J., & Wolfe, D. A. (2008). Cognitive and emotional differences between abusive and non-abusive fathers. Child Abuse & Neglect, 32, 1127-1137. doi:10.1016/j.chiabu
.2008.05.007.
*Fujiwara, T., Okuyama, M., Tsui, H., & Koenen, K. C. (2008). Perinatal factors associated with infant maltreatment. Clinical Medicine: Pediatrics, 1, 29-36.
Gilbert, R., Widom, C. S., Browne, K., Fergusson, D., Webb, E., & Janson, S. (2009). Burden and consequences of child maltreatment in high-income countries. Lancet, 373(9657), 68-81. doi: 10.1016/S0140-6736(08)61706-7
*Haskett, M. E., Scott, S. S., Grant, R., Ward, C. S., & Robinson, C. (2003). Child-related cognitions and affective functioning of physically abusive and comparison parents.
26 Child Abuse & Neglect, 27(6), 663-686. doi: 10.1016/S0145-2134(03)00103-0 Higgins, D. J., & McCabe, M. P. (2001). Multiple forms of child abuse and neglect: Adult retrospective reports. Aggression and Violent Behavior, 6(6), 547-578. doi: 10.1016 /S1359-1789(00)00030-6
Hoeve, M., Stams, G. J. J. M., Van der Put, C. E., Dubas, J. S., Van der Laan, P. H., & Gerris, J. R. M. (2012). A meta-analysis of attachment to parents and delinquency. Journal of Abnormal Child Psychology, 40(5), 771-785. doi: 10.1007/s10802-011-9608-1 Hox, J. (2002). Multilevel analysis: Techniques and applications. Mahwah: Lawrence Erlbaum Associates.
Hurme, T., Alanko, S., Antilla, P., Juven, T., & Svedström, E. (2008). Risk factors for physical child abuse in infants and toddlers. European Journal of Pediatric Surgery, 18(6), 387-391. doi: 10.1055/s-2008-1038922
*Jaffee, S. R., Caspi, A., Moffitt, T. E., Polo-Tomás, M., & Taylor, A. (2007). Individual, family, and neighborhood factors distinguish resilient from non-resilient maltreated children: A cumulative stressors model. Child Abuse & Neglect, 31(3), 231-253. doi: 10.1016/j.chiabu.2006.03.011
Johnson, W. L. (2011). The validity and utility of the California Family Risk Assessment under practice conditions in the field: A prospective study. Child Abuse & Neglect, 35(1), 18-28. doi: 10.1016/j.chiabu.2010.08.002
Juhnke, G. A., Henderson, K., & Juhnke, B. A. (2013). The Juhnke, Henderson, Juhnke child abuse and neglect risk assessment scale: A mnemonic instrument used to assess potential maltreatment of children. The Family Journal: Counseling and Therapy for Couples and Families, 21(1), 57-64. doi: 10.1177/1066480712456820
Kempe, C. H., Silverman, F. N., Steele, B. F., Droegemueller, W., & Silver, H. K. (1962). The battered-child syndrome. Journal of the American Medical Association, 181(1), 17-24. doi: 10.1001/jama.1962.03050270019004
Korbin, J. (1988). Child abuse as an international issue. Child Abuse & Neglect, 12(1), 3-23. doi: 10.1016/0145-2134(88)90003-8
Kotch, J. B., Browne, D. C., Dufort, V., & Winsor, J. (1999). Predicting child maltreatment in the first 4 years of life from characteristics assessed in the neonatal period. Child Abuse & Neglect, 23(4), 305-319. doi: 10.1016/S0145-2134(99)00003-4
Kraemer, H. C., Kazdin, A. E., Offord, D. R., Kessler, R. C., Jensen, P. S., & Kupfer, D. J. (1997). Coming to terms with the terms of risk. Archives of General Psychiatry, 54(4), 337-343. doi: 10.1001/archpsyc.1997.01830160065009
27 *Lacelle, C., Hebert, M., Lavoie, F., Vitaro, F., & Tremblay, R. E. (2012) Child sexual abuse and women's sexual health: the contribution of CSA severity and exposure to multiple forms of childhood victimization. Journal of Child Sexual Abuse, 21(5), 571-592. doi: 10.1080/10538712.2012.688932
Lamb, M. E. (2010). The role of the father in child development. New Jersey: John Wiley & Sons, Inc.
*Lau, A. S., Valeri, S. M., Mc Carty, C. A., & Weisz, J. R. (2006). Abusive parents' reports of child behavior problems: relationship to observed parent-child interactions. Child Abuse & Neglect, 30(6), 639-655. doi: 10.1016/j.chiabu.2005.11.009
Leary, M. R. (2008). Introduction to behavioral research methods. Boston, U.S.: Pearson Education, Inc.
*Lee, S. J. (2013). Paternal and household characteristics associated with child neglect and child protective services involvement. Journal of Social Service Research, 39(2), 171- 187. doi: 10.1080/01488376.2012.744618.
*Lee, S. J., Taylor, C. A., & Bellamy, J. L. (2012). Paternal depression and risk for child neglect in father-involved families of young children. Child Abuse & Neglect, 36(5), 461-469. doi: 10.1016/j.chiabu.2012.04.002
Leeb, R. T., Paulozzi, L., Melanson, C., Simon, T., & Arias, I. (2008). Child maltreatment surveillance: Uniform Definitions for public health and recommended data elements, Version 1.0. Atlanta (GA): Centers for Disease Control and Prevention, National Center for Injury Prevention and Control. Retrieved from:
http://www.cdc.gov/violenceprevention/pdf/cm_surveillance-a.pdf
Leventhal, J. M., & Krugman, R. D. (2012). “The battered-child syndrome” 50 years later. Journal of the American Medical Association, 308(1), 35-36. doi: 10.1001/jama .2012 .6416
Lipsey, M. W., & Wilson, D. B. (2001). Practical meta-analysis. Thousands Oaks: Sage. Lodewijks, H. P. B., & Van Domburgh, L. (2012). Instrumenten voor risicotaxatie: Kinderen en jeugdigen. Amsterdam: Pearson.
Lofthouse, R. E., Totsika, V., Hastings, R. P., Lindsay, W. R., Hogue, T. E., & Taylor, J. L. (2014). How do static and dynamic risk factors work together to predict violent behavior among offenders with an intellectual disability? Journal of Intellectual Disability Research, 58(2), 125-133. doi: 10.1111/j.1365-2788.2012.01645.x
