Comparing Physically Abused, Neglected, and Nonmaltreated Children During Interactions with their Parents: A Meta-Analysis of Observational Studies

Abstract

A meta-analysis of 30 observational studies compares abused, neglected, and nonmaltreated children's behavior during interactions with their parents. Drawing on the relational communication literature, children's behaviors from various coding schemes were grouped into those communicating positivity (e.g., affection, approval), aversiveness (e.g., anger, resistance), and involvement (e.g., attention, interest). Results reveal that abused and neglected children are distinguished from nonmaltreated children on all three behavioral clusters, with overall mean weighted effect sizes ranging from d=.29 to .55. Several moderators qualify the magnitude though not the direction of these differences, including maltreatment type, child/parent age, and observation length and setting. These findings have implications for understanding the etiology and outcomes of child maltreatment as well as for intervention and prevention efforts.

Keywords:

• Child Abuse and Neglect
• Relational Communication
• Parent-Child Interaction
• Meta-Analysis
• Observational Research

Acknowledgements

The authors would like to thank Mike Allen and three anonymous reviewers for helpful feedback.

Notes

1. “Child maltreatment” is an umbrella term that typically incorporates the physical abuse, sexual abuse, neglect, and psychological maltreatment of children (see Barnett, Miller-Perrin, & Perrin, 2005; Morgan & Wilson, 2005). Virtually all observational studies of parent–child interaction have compared families in which parents have a documented history of either child physical abuse or neglect with nonmaltreating families, and hence that is our focus here.

2. There are complexities involved in defining child physical abuse and neglect. The terms “abuse” and “neglect” imply parenting practices that not only are simply less than ideal but actually less than what is minimally adequate (i.e., harmful) and hence practices that require intervention by actors outside of the family (e.g., the state). Ideas about what constitutes minimally adequate parenting or when the state should intervene in the lives of families vary across culture and also have changed over time in the US, leading to changes in policy or legal definitions (Barnett et al., 2005). Research (conceptual) definitions need not be identical to legal ones. For example, many researchers define child physical abuse using an “endangerment” standard (i.e., acts that have a strong potential to injure a child are defined as abusive, regardless of whether injury actually occurs in any specific case; Straus et al., 1998) rather than the “harm” standard in most legal definitions (i.e.,, acts which actually result in injury). Yet researchers are not immune to the society in which they live, and hence conceptual definitions also have evolved over time (e.g., the ongoing debate about whether corporal punishment should be defined as abusive). Although each of the 50 states in the US sets its own legal definitions of child abuse and neglect, the definitions we present reflect themes currently present in most legal definitions.

3. Virtually all of the studies in this meta-analysis use as case comparison method; that is, they purposefully recruit nonmaltreating families who are sociodemographically similar to the maltreating families (e.g., parent and child age, family income) to serve as a comparison group (see inclusion criterion 2). Some studies exploring child abuse potential also have used matching, such as by comparing 10 families in which a parent scored high on the CAPI with 10 sociodemographically similar families where parents scored low, with both groups being drawn from a larger sample of parents who completed the CAPI (Dolz et al., 1997). Most research on child abuse potential, however, has correlated parents’ CAPI scores with parent and child behaviors without matching, relying instead on statistical controls, after the fact, to account for sociodemographic factors (e.g., Wilson et al., 2004). Because this design is different from the design used in studies comparing families with and without a documented history of child maltreatment, studies of child abuse potential were not included in the meta-analysis.

4. For maltreatment type, “abuse” means that these children were recruited into the study because they came from families with a documented history of physical abuse (e.g., substantiated by CPS agencies). Because CPS agencies may give greater priority to “active” forms of abuse (sexual or physical abuse) than to “passive” forms (neglect or emotional maltreatment; see Lau et al., 2005), it is possible some children may have also been neglected, but their family's classification with the CPS agency was physical abuse. “Neglect” means that these children came from a family with a history of neglect. “Mixed” means that the “maltreatment” group includes some children with a documented history of being physically abused but also other children with a documented history of neglect (or in two cases, either physical abuse or emotional maltreatment); hence, “mixed” indicates that these children were included based on varied but not necessarily multiple forms of maltreatment.

5. There are eight studies labeled as “M & F” (see Table 1) which designates children interacting with either both their mother and father or only their father. Specifically, Burgess and Conger (1978), Kavanagh, Youngblade, Reid, and Fagot (1988), Reid et al. (1981 ,1987), and Silber, Bermann, Henderson, and Lehman (1993) observed children interacting with their entire family. Whipple and Webster-Stratton (1991) had children interact with both parents, if available; some lacked fathers. Haskett, Ahern, Ward, and Allaire (2006) as well as Lau et al. (2006) had children interact with either their mother or their father, but in both studies the number of fathers was small and results were collapsed across parent sex.

6. In an unstructured task, the researcher does not provide rules about what task to perform, what order to do activities in, or what should be the result of the activity. The researcher may provide a standard set of toys or objects but does not dictate how they should be played with, in what order, or with what outcome. In a highly structured task, the researcher explicitly tells participants what tasks to do, what order to do them in, what objects to use, how to use those objects, and what the end result of the activity should look like. The more of these present, the more structured the task. Moderately structured tasks represent a middle ground between the researcher providing rules about what should be done so that everyone does the same thing, versus no rules being provided at all. The researcher might tell the participants what to do but not how to do it (e.g., clean-up, play with blocks) or how to do it but not what to do (e.g., teach the child something, follow the child's lead). Mixed structure indicates a study in which participants completed multiple tasks with differing levels of task structure. Most studies with mixed structure reported findings averaged across multiple tasks (e.g., Borrego et al., 2004; Coster, Gersten, Beeghly, & Cicchetti, 1989; Lau et al., 2006; Schindler & Arkowitz, 1986); hence, for those studies that did report separate comparisons for each task (e.g., Mash, Johnston, & Kovitz., 1983) effect sizes were computed for each task and then averaged to obtain an overall effect size for that study.

7. Nearly one-half of the studies in this meta-analysis reported insufficient information to correct for measurement error, which for observational studies would be estimates of inter-rater reliability for each pertinent category of child behavior. Three studies (Egeland et al., 1983; Givens, 1978; Reid et al., 1981) failed to report any information regarding inter-rater agreement. Ten studies reported only ranges for inter-rater reliability coefficients across multiple categories rather than for individual categories (Bousha & Twentyman, 1984; Brassard, Hart, & Hardy, 1993; Lahey et al., 1984; Lau et al., 2006; Mash et al., 1983; Reid et al., 1987; Schindler & Arkowitz, 1986; Valentino et al, 2006; Webster-Stratton, 1985; Whipple & Webster-Stratton, 1991). In many cases, these ranges varied widely across categories; for example, Bousha and Twentyman reported that correlations between two coders’ frequency counts of 11 different categories of child behavior ranged from .74 to .98, Lau et al. stated that inter-class correlations for a group of coders across five different child composite codes ranged from .43 to .89, and Schindler and Arkowitz reported that percentages of agreement between two coders who pushed one of six different buttons each time one of six types of child behavior occurred ranged from 61 percent to 94 percent. Because we could not determine which categories were rated more or less reliably for so many studies, we conducted only a “bare bones” meta-analysis. One consequence is that the mean effect sizes reported in the results section likely underestimate actual group differences.

8. The overall mean weighted effect size can be smaller when using “maximum” as opposed to “minimum” estimates if studies in which both estimates were calculated had results opposite in direction from what was predicted (i.e., the sign for some maximum estimates was negative).

9. Lau et al. (2006) not only compared two methods for assessing differences between maltreated and nonmaltreated children (observer ratings versus parental reports), but also assessed whether convergence between the two methods differed for maltreating versus nonmaltreating parents. For nonmaltreating parents, the authors found a statistically significant, positive association between an observer's ratings of a child's demandingness during the parent–child interaction and the parent's rating of that same child's externalizing behavioral problems. In contrast, the same correlation was not significant (and, in fact, was inverse in direction) for maltreating parents. Maltreating parents’ perceptions of their children's behavior converged less with the perceptions of other adults as compared to nonmaltreating parents, which also suggests biased perceptions on the part of maltreating parents.

Additional information

Notes on contributors

Steven R. Wilson

Steven R. Wilson is Professor of Communication at Purdue Univeristy

Alda M. Norris

Alda M. Norris is PhD student at Purdue Univeristy

Xiaowei Shi

Xiaowei Shi is an Assistant Professor in the Department of Speech and Theatre at Middle Tennessee State University

Jessica J. Rack

Jessica J. Rack is PhD student at Purdue Univeristy