 

LIBRARY
Michigan State
University

 

 

 

PLACE IN REfURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

DCT 0 2 2005

 

LREB3111 5007

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 c:/CIRCIDateDua.p6&p.15

TOWARD THE PREVENTION OF CHILD MALTREATMENT THROUGH RISK
ASSESSMENT: EVALUATION OF AN ECOLOGICAL, PROSPECTIVE MODEL OF
RISK FOR CHILD ABUSE POTENTIAL
By

Lisa B. Galasso

A DISSERTATION

Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Department of Psychology

2001

ABSTRACT
TOWARD THE PREVENTION OF CHILD MALTREATMENT THROUGH RISK
ASSESSMENT: EVALUATION OF AN ECOLOGICAL, PROSPECTIVE MODEL OF
RISK FOR CHILD ABUSE POTENTIAL
By

Lisa B. Galasso

Secondary prevention, the predominant model of prevention of child abuse and
neglect, involves identiﬁcation of individuals considered to be “at risk” for engaging in
child maltreatment. While theories of child maltreatment etiology recognize the complex
interplay of factors at numerous ecological levels, most risk assessment practices
primarily assess risk at the level of the mother-infant dyad. This strategy neglects the
inﬂuence of broader ecological forces on parental functioning, and has led to narrow and
incomplete conceptions of risk. New methodologies need to be created that apply an
ecologically based integrative understanding of the etiology of child maltreatment. The
current investigation tested a risk assessment strategy that was intended to improve upon
current methods by employing a theoretically-based ecological approach, and by
predicting risk across a twelve month span of time. Structural equation modeling
methods were applied to 12-month longitudinal data from 125 new parents labeled “at
risk” for inadequate parenting. Results indicate that person-level risk in the current
model, as measured by a traditional, unidimensional checklist approach to risk
assessment, was unrelated to parenting outcome as measured by parental attitudes, but
Signiﬁcantly predicted child development. Findings further reveal that community-level

factors had a signiﬁcant direct and stress-mediated effect upon potentially abusive

parenting attitudes; notably, these are factors which tend to be overlooked by most current
risk assessment practices. Despite the signiﬁcant relationships between community-level
risk, stress, and parenting outcome, this model of multiple inﬂuences of risk still only
accounted for 10 percent of the variance in child abuse potential. Results are discussed in
terms of their impact on the prevalent secondary prevention paradigm, and suggestions
are made for developing more complete assessment strategies on which to base

intervention decisions.

ACKNOWLEDGMENTS

The popular saying “it takes a village to raise a child” can also be applied to the
efforts that were involved in completing this research project. Without the participation
of the families enrolled in the Kent County Healthy Start Program it would not have been
possible. I wish to thank them for sharing their time and for their courageousness in
allowing us to glimpse the richness of their personal lives. To the family support workers
and program supervisors at KCHS, I offer my gratitude for the valuable time and effort
they spent carefully collecting the data used in this project. The undergraduate research
support staff who were responsible for data management proved to be a dedicated team
and they are greatly appreciated for their meticulous and hard work. I offer my gratitude
to Dr. Anne Bogat who provided invaluable advice, feedback, and support through all
phases of this project. Her mentoring has not only affected my academic and professional
development, but has inﬂuenced my personal growth as well. Likewise, I thank my
dissertation committee members Drs. Alytia Levendosky and Gershen Kaufman for their
helpful comments and support. Above all, I wish to thank my dissertation advisor, Dr.
Robert Caldwell, for the ideas he inspired, the patience he bestowed, and the vast amount
of support he provided throughout this endeavor. His conﬁdence in me has been
unwavering, even when I have been plagued with doubts, and for that I am truly grateful.
Lastly, my family has played an integral role in my ability to complete this research
project, and for their endless support, I am eternally grateful. The efforts of many were
involved in this project, and it is my hope that the culmination of these efforts will result

in taking one small step closer to understanding how to prevent child maltreatment.

iv

TABLE OF CONTENTS

List of Tables ...................................................................................................................... vi
List of Figures ..................................................................................................................... vii
Introduction .......................................................................................................................... 1
Overview of Theoretical Issues ........................................................................................... 5
Conceptual Issues Related to the Assessment of Risk ....................................................... 15
Practical Issues Related to the Assessment of Risk ........................................................... 50
Construction of the Ecological, Prospective Model of Risk to be Tested ......................... 62
Hypotheses ......................................................................................................................... 67
Methods .............................................................................................................................. 69
Results ................................................................................................................................ 86
Discussion ........................................................................................................................ 102
References ........................................................................................................................ l 12
Appendix A ...................................................................................................................... 126
Appendix B ...................................................................................................................... 127
Appendix C ...................................................................................................................... 131
Appendix D ...................................................................................................................... 138
Appendix E ...................................................................................................................... 142

LIST OF TABLES

Table 1: Group differences between research sample and excluded subjects .................... 70
Table 2: Frequency table of all risk factors from KCHS Risk Screen ............................... 73
Table 3: Frequency table of intra- and inter- personal level risk indicators ..................... 76
Table 4: Descriptive statistics for study variables ............................................................. 90
Table 5: Standardized parameter estimates for Model I - CAP outcome .......................... 92
Table 6: Standardized parameter estimates for Model 11 - Denver outcome .................... 94

Table 7: Alternative models in post hoc model-ﬁtting process compared to original

hypothesized models ......................................................................................................... 99

vi

LIST OF FIGURES
Figure 1: Theoretical model ............................................................................................... 65
Figure 2: Measurement Model ........................................................................................... 66
Figure 3: Model of multiple inﬂuences of risk on child abuse potential (CAP subscales
measuring outcome) .......................................................................................................... 93
Figure 4: Model of multiple inﬂuences of risk on child abuse potential (Denver
Developmental Screening Test measuring outcome) ........................................................ 96
Figure 5: Model III - Inna-personal level risk in relation to model variables (CAP abuse scale
measuring outcome) ......................................................................................................... 100
Figure 6: Model IV - Intra—personal level risk in relation to model variables (Denver

measuring outcome) ......................................................................................................... 101

vii

INTRODUCTION

Risk assessment is essentially the practice of predicting future events. In the ﬁeld
of child abuse prevention, this phenomenon has been overwhelmingly adopted by child
protective service (CPS) agencies as a way to make service delivery more effective. Risk
assessment refers to a decision-making process that determines the degree to which a
child is likely to be maltreated in the near future, and resources are allocated accordingly
(English & Pecora, 1994). In the practical realm, most clinicians use risk assessment
aﬁer substantiation of abuse or neglect to identify those children at greatest risk for
continued maltreatment, and to aid in case planning and intervention. This use of risk
assessment misses the mark of prevention, and more accurately addresses intervention
needs. However, in the ﬁeld of prevention, practitioners and researchers alike have
recognized that the process of determining risk status is complicated by the absence of
clear-cut guidelines for identiﬁcation procedures, a lack of understanding of risk
processes, and the relatively low incidence of child maltreatment in the general
population (Wolf, 1993). As a result, a lively debate has ensued among those concerned
with prevention. Some have argued for the abandonment of risk assessment in favor of
primary prevention, in which general indicators such as ﬁrst-time parenthood, not risk
indicators, should serve as entry criteria into prevention programs (cf. Cohen, 1996;
Wolfe, 1993). Others, however, have argued for the development of more valid risk
assessment models (cf. Bavolek, 1989) that have the ability to identify individuals for
inclusion into secondary prevention programs. The current research is concerned with the

latter side of the debate and accordingly will focus on the development and use of

predictive models of abuse to inform preventive interventions.

Despite numerous problems associated with risk assessment --many of which will
be discussed-- it is a practice that is widely utilized in the public sector. As English and
Pecora (1994) point out, risk models are presented as guidelines, but they nevertheless
serve as decision-making models. This practice calls for renewed emphasis on improving
the quality of research on risk prediction, because the lack of empirical guidance has not
diminished the use of such models. The over-utilization of risk models in CPS practice
can also be attributed to the fact that they improve communication among workers, make
information more readily available to practitioners, and lead to more systematic collection
of information (Cicchinelli & Keller, 1990).

Although risk models are integral parts of CPS casework practice, a study by the
National Child Welfare Resource Center for Management and Administration (NCWRC)
found that there is a general lack of agreement in the ﬁeld about what factors are to be
used when assessing risk, and that less than 50% of the risk factors employed by risk
models have been empirically tested (English & Pecora, 1994). This lack of consistency
epitomizes the piecemeal approach that is characteristic of current risk assessment
practice. Another main problem associated with risk assessment concerns the criteria
against which to measure predictive validity. If the only purpose of risk assessment is to
predict who will re-enter the CPS system in the future, then pragmatic approaches that are
limited to empirical correlations between risk factors and outcome, regardless of
theoretical explanations of those relationships, are appropriate (English & Pecora, 1994).

This avenue of investigation, however, can only address questions of recidivism.

Alternatively, if the purpose of risk assessment is to guide true prevention services - that
is, working with families who have not yet entered the CPS system for intervention — then
research on risk models must be theoretically-based. Such risk models have the potential
to answer questions regarding future parental behavior and the effects of the child’s
environment, rather than simply restating a documented history of child maltreatment
(Brissett-Chapman, 1997). This goal has not been realized due to the lack of an overall
coherent framework for theory and practice in the ﬁeld of risk assessment (DePanﬁlis,
1988). In an effort to address this lack of theoretical base, the current investigation was
concerned with applying an ecological theoretical framework to the practice of risk
assessment to determine if this strategy improves the efﬁcacy of prediction. In addition,
multiple stage screening was used in which risk was measured at birth, and at six months
of age, to predict outcome at one year. The use of prospective methods was employed to
determine whether predictive validity can be enhanced by increasing the assessment
period beyond a single encounter.

The following analysis of risk assessment has three primary foci: theoretical
issues, conceptual matters, and current practical strategies. The empirical scope of this
investigation was to test the feasibility of a theoretically-based model of risk assessment.
However, a discussion of some of the theoretical issues associated with risk assessment
are presented to provide a context for the current research. Likewise, consideration of
some of the conceptual issues that convolute the practice of assessing risk illuminate the
inherent difﬁculties in conducting this type of research, and translating ﬁndings into

practice. Finally, current risk assessment procedures are critiqued in an effort to identify

both the strengths and inadequacies in the practice of risk assessment. Lastly, in an effort
to address some of the issues raised, the current model of risk assessment tests the ability
of multiple ecological levels, including person and community level variables, to predict
indicators of child maltreatment. The direct and mediating effects of parenting stress are
also evaluated in the model. In addition, by assessing risk at birth and at six months of

age, this method provides information about the temporal stability of risk factors.

OVERVIEW OF THEORETICAL ISSUES
The Risk-Driven Model of Prevention

The universal application of preventive services is not widespread in this country.
Universal prevention has been applied, for the most part, in the ﬁeld of infectious
diseases. The most noted example of this prevention strategy is childhood
immunizations, which are administered to all children before they can attend public
schools as a means of preventing diseases of early childhood. As history will afﬁrm, this
strategy has been overwhelmingly successful in wiping out diseases such as polio,
measles, and diphtheria, which signiﬁcantly increased mortality rates among children
mere decades ago. Conversely, children continue to suffer from the effects of
maltreatment because successful prevention strategies have not yet been universally
applied to the social problem of child abuse and neglect. The prevention model that has
most often been applied to child maltreatment involves the identiﬁcation of at-risk
individuals and consequent intervention with these individuals aimed at preventing future
harmful parenting practices. Because child maltreatment still occurs with alarming
ﬁ'equency in our country, it is unclear whether the risk-driven model of prevention has
been poorly applied to the problem of child maltreatment, or whether it lacks fundamental
applicability when utilized in this capacity.

The Science of Prevention

Three Models of Prevention

There are three models of prevention that were coined by Caplan (1964) to

distinguish between different prevention goals, population foci, and timing of service

delivery: primary prevention, secondary prevention, and tertiary prevention. This
terminology has become the standard for describing the different prevention paradigms.
The ﬁrst, primary prevention, refers to prevention in its truest sense. It involves
intervening before the development of an unwanted outcome, and is directed at total
populations to reduce overall incidence rates. An example would be putting ﬂuoride in
the water to prevent tooth decay. Secondary prevention involves the early identiﬁcation
of target individuals who are at risk for an unwanted outcome, and is aimed at reducing
overall prevalence rates of the unwanted condition. An example of this type of
prevention would be conducting sex education classes with 9‘h grade students who were
thought to be sexually active. In this case, the goal would be to target individuals who
were at increased risk of contracting a sexually transmitted disease or becoming pregnant
unintentionally, and to teach them how to protect themselves effectively from these
unwanted conditions. This differs from primary prevention in terms of the scope of the
population it attempts to reach — it only targets individuals considered to be most at risk
for the unwanted outcome. This model can also be thought of as a method of early
intervention. Lastly, tertiary prevention is targeted at individuals who have already been
diagnosed with a given condition, and is intended to reduce the adverse consequences
associated with a disease state or disorder, or to prevent the individual from becoming
incapacitated (Willis, Holden, & Rosenberg, 1992). An example of this type of
prevention would be support groups for Alzheimers patients and their spouses that
provide strategies for managing the disease at home in an effort to prevent or postpone

institutionalization. Although more like rehabilitation, this type of prevention attempts to

reduce the negative sequelae of a given disorder and thus represents prevention in its
narrowest form.
Utilization of the Different Prevention Models: Costs and Beneﬁts

One of the beneﬁts of the primary prevention model is that it does not adhere to
a “person-blame” ideology. Community psychologists have long argued that traditional
social service practices tend to place blame on troubled people for their own problems,
without recognizing that broader levels of society such as families, neighborhoods,
communities, and social and cultural institutions also shoulder some of the responsibility
(Mitchell, Davidson, Chodakowski, & McVeigh, 1985). However, targeting entire
1 populations for preventive services becomes a social leveling tool that meets with
economic, social, and political resistence (Costin, Karger, & Stoesz, 1996). For instance,
it is a widely accepted belief that primary prevention of child maltreatment is
unwarranted because of the excessive ﬁnancial cost involved, and that secondary
prevention is a far more efficient strategy. However, Caldwell, Bogat and Davidson
(1988) seriously challenged this assumption when they demonstrated that identifying
“populations in need of preventive services currently yields very small increases in
programming efﬁciency” (p. 620) due to the low base rate of child abuse and neglect in
our society, and the lack of sensitivity and speciﬁcity of available measurement tools.
Another beneﬁt of the primary prevention model is that it does not involve the labeling of
individuals as “at-risk,” which may result in iatrogenic effects (Pillow, Sandler, Brave,
Wolchik, & Gersten, 1991). Research suggests that labeling can have powerful, negative

consequences (Broskowski & Baker, 1974), although the actual effects of inaccurately

labeling someone as a potential child abuser have yet to be determined. Lastly, secondary
prevention strategies are more socially acceptable because they are exclusive -- that is,
risk notions of prevention exclude most people (Cowen, 1996). Alternatively, primary
prevention is founded upon the premise that everyone in a given population could beneﬁt
from some form of the preventive intervention, and in the case of child maltreatment,
these strategies implicitly suggest that all parents may be capable of inadequate parenting,
if not some form of Child maltreatment. Moreover, the intensity of the preventive
services delivered depends on how the problem is deﬁned, which makes it necessary to
differentiate between levels of abuse, and to distinguish maltreatment from violence
(Emery & Laumann-Billings, 1998). In contrast, programs that are aimed at individuals
already predetermined to be at risk for perpetrating violence can be more homogeneous
because they are based on the assumption that everyone entering the program has a
similar “problem” warranting similar treatment. In any case, with such sensitive social
and political issues, as well as practical considerations, weighing upon the utilization of
the different prevention ideologies, it is not surprising that secondary prevention is the
most commonly used model in the ﬁeld of child abuse prevention. Cowen (1996)
summarized this best when he stated that the “policy-shaping, fund-grabbing deﬁnition of
prevention that has risen to the top is a notion based on risk-reduction as the gateway to
disease prevention” (p. 244).
The Pervasive Model: Secondary Prevention of Child Maltreatment
The secondary prevention model involves identifying parents who are likely to

maltreat their children; then, effective methods of preventive intervention must be

initiated in an effort to derail the process of maltreatment before it begins. Despite the
ﬁeld’s general lack of progress developing empirically-validated theories of child
maltreatment and empirically-derived identiﬁcation of causal mechanisms, there has been
a burgeoning of work in the area of risk assessment and program development and
implementation. Although well-intentioned, the approach that characterizes the majority
of this work can be likened to building an expensive, modern new home atop an
unﬁnished foundation. For the most part, resources have been directed at developing
methods to identify and treat “at-risk” individuals and families without a deﬁnitive
understanding of the etiology of child maltreatment on which to base them. In the ﬁeld of
child abuse prevention, methods of prediction (i.e., risk assessment) have come to be
treated as if they were the proverbial holy grail — researchers and practitioners have
overwhelmingly subscribed to a paradigm that has yet to prove its legitimacy. This
phenomenon of using unchallenged theories to guide practice is not an uncommon event.
As history has demonstrated, incomplete and even incorrect theories of disease etiology
can sometimes lead to successful prevention. Bloom (1965) eloquently illustrated this
phenomenon with the example of miasma theory to explain malaria. It was originally
believed that malaria was caused by vapors (or miasmas) emanating from rotting material
in swamps. Although this was completely wrong, the solution that was implemented was
to drain the swamps, therefore eliminating the breeding ground for the tse-tse ﬂy which
was responsible for transmitting the disease. Hence, the intervention, based on an
incorrect understanding of the disease etiology, had the desired effect of reducing the

incidence of malaria. While this example demonstrates that false etiological theories can

sometimes lead to effective treatments, the ﬁeld of child maltreatment has yet to
demonstrate a proven method of prevention. Etiological theories of child maltreatment
have evolved over the past two decades but an equivocal theoretical explanation of this
complicated process has yet to be asserted. Hence, risk assessment may be the “miasma
theory” of the ﬁeld of child maltreatment (i.e., it is a solution based on an incomplete
understanding of the problem), but it has not yielded such positive results as those
described by Bloom.
Temporal Decay of Prediction

One of the main problems with the prediction of future events, regardless of the
ﬁeld of investigation, is that accuracy decreases over time. For example, in meteorology,
it is relatively easy to predict tomorrow’s weather, but accuracy signiﬁcantly decreases as
the window of time increases to next week or next month. Despite the sophistication of
current satellite and radar technology, the extant methods of prediction are still very
limited in their capacity to predict future atmospheric events. For example, one Japanese
meteorologic study reported that “by using 9 years of observational data in the middle and
the west part of Japan to predict whether the weather of Tokyo 6 hours ahead would be
rainy or not, an accuracy of [only] 87.2% was achieved” (Mohri & Tanaka, 1995). This
example illustrates that prediction, even in a technologically advanced ﬁeld such as
meteorology, is inexact and suffers from decay of accuracy over time —-—- even as short a
time-span as six hours. In the social and behavioral sciences, researchers and
practitioners don’t need to predict when abuse will occur (alas, weather forecasters would

be pretty accurate if all they had to do was to predict that it would rain in the future),

10

rather they need to predict who will be most likely to abuse their children. Returning to
the weather analogy, we know that rain (child abuse) will occur -- the question is whether
we can predict with some accuracy where (who) it might be more likely to rain (abuse
their children). Whereas weather forecasters can be reasonably accurate in predicting that
it is much more likely to rain in the rainforest than the desert, social scientists and
practitioners are still severely limited in their ability to predict which families are more
like rainforests and which ones are more like deserts. Despite an inadequate arsenal of
methodologies, most of the ﬁeld of child abuse prevention hinges upon risk assessment
procedures that make a practice of predicting future behavior, typically over an extensive
time span. Some of the complexities of this practice will be discussed presently.
Relationship Between Risk and Outcome

Because prediction of future behavior is founded upon the identiﬁcation of risk
factors, it is essential to consider the relationship between risk factors and outcome. In
the ﬁeld of child abuse prevention there has been the tendency to assume that direct
relationships exist between risk factors and abuse. The literature on prediction has
focused on identifying characteristics that are commonly found among abusing parents,
abused children, and the environments in which they live (Agathonos-Georgopoulou &
Browne, 1997). Some of the parental “risk variables” described in this literature include
young maternal age at birth (Connelly & Strauss, 1992), mothers’ indifference to the
child’s needs (Baranowski, Schilmoeller, & Higgins, 1990), and a childhood history of
abuse (Caliso & Milner, 1992; Milner, Robertson & Rogers, 1990). Child variables such

as age and temperament have also been associated with abuse (Stratton & Swaffer, 1988).

11

In addition, factors related to the adequacy of social support, situational stress, and
parental coping strategies have been linked with abusive parenting practices (Cantos,
Neale, O'Leary, & Gaines, 1997). Still others argue that social policies that fail to foster
positive social interactions, and leave communities without adequate social, health, or
educational infrastructures create environments ripe for abusive and violent behavior
(Garborino & Sherman, 1980). The majority of these studies make assumptions about the
relationship between risk and outcome that are linear, additive, and direct. These simple
assumptions reﬂect traditional perpetrator-victim perspectives of abuse that retain an
exclusive focus on the individual and individual concepts of risk. Although some
researchers acknowledge the contribution of exosystemic and macrosystemic variables
(c.f., Garborino & Sherman, 1980), it is most often the case that risk is assessed at the
ecological level at which preventive services are delivered (Caldwell, Bogat, & Davidson,
1988)

Due to the individually-focused clinical tradition that dominates this ﬁeld, most
interventions are aimed at individuals and families. This maintains the status quo of
viewing child maltreatment as a phenomena that is caused by factors within the
individual. Even when community level risk is evaluated and services are delivered to
entire communities, such interventions are low-intensity and thus lack effectiveness in
treating individuals who will actually abuse their children. For example, a billboard
advertisement that warns against the dangers of shaken baby syndrome will be generally
informative to the community, but will not provide an intense enough intervention for

individuals who are actually using violence (Carty & Ratcliffe, 1995; Showers, 1992).

12

Therefore, the practice of exclusively assessing risk at the ecological level at which
prevention services are delivered maintains a direct, simplistic relationship between risk
and outcome, which does not account for the complex combination of ecological contexts
that may lead to maltreatment.

Incongruence Between Etiological Theories of Child Maltreatment and Risk Assessment

Most current etiological models of child maltreatment have adopted a
multifactorial, ecological, transactional approach in which abuse is the result of a process
that involves the accumulation of risks at various ecological levels, combined with a
deﬁciency in compensatory or protective mechanisms (Belsky, 1980; Bogenschneider,
1996; Bronfenbrenner, 1979; Cicchetti, 1993). From this perspective, parental behavior
is seen as strongly inﬂuenced by character traits, but is also affected by, and has effects
upon, familial, neighborhood, social, and cultural environments. In addition, the
importance of protective processes that enhance coping and promote adaptation and
competence in the face of adversity are recognized. Such protective processes can reside
within the individual as well as the environment (Garmezy, 1983).

Given the advances in theorizing about the causes of child maltreatment, why are
risk assessment models lagging so far behind? It is unlikely that the narrow, linear scope
of most investigations into the prediction of risk is due to a general lack of sophistication
among researchers in this ﬁeld. In fact, it is the practice of most researchers to refer to
issues of ecology and methodological inadequacy when explaining their ﬁndings
(Hobfoll, 1998). Instead, what appears to be lacking in the ﬁeld is the use of ecological

theories to create new methodologies and avenues of exploration. It is unlikely that

13

newer ecologically-valid models of risk are a panacea and will lead to the obliteration of
child maltreatment. They may, however, provide the impetus for the creation of new
methods of risk assessment that more closely reﬂect the processes that determine parental
functioning, which in turn may lead to more efficacious preventive interventions. As
Guterman (1997) keenly points out, “in the context of an ecologically based integrative
understanding of the etiology of physical child abuse and neglect, the existing knowledge
base sheds little light on the importance of broadening the target of intervention beyond
the mother-infant dyad onto familial, meso-, exo-, and macrosystemic considerations”(p.

31).

14

CONCEPTUAL ISSUES RELATED TO THE ASSESSMENT OF RISK

The practice of risk assessment is severely hindered by a lack of conceptual
clarity. A discussion of some of the conceptual shortcomings associated with assessment
of risk for child maltreatment will be explored in the following section.

The Deﬁnitional Dilemma: What Is Child Maltreatment?

Dijﬁring Agendas for Deﬁnitions

The question of what constitutes child maltreatment seems at ﬁrst glance to be a
rather straightforward question. However, the way this phenomenon is operationalized
depends entirely on the purpose for which it is being deﬁned, whether it is to arrive at a
statutory deﬁnition, implement investigation policies, or set criteria for data collection
procedures. Not only is there disagreement between professions, but there is also a great
deal of disagreement among experts within the same ﬁeld concerning how broadly or
how narrowly to deﬁne child abuse. This disagreement is exempliﬁed by the discrepant
ﬁndings among epidemiological studies concerned with reporting national incidence rates
of child abuse and neglect. Such studies generate statistics on reported and conﬁrmed
cases of child maltreatment, and are illustrative of the dramatic differences in who and
what gets counted as an instance of abuse. For example, the National Committee to
Prevent Child Abuse (N CPCA)’S SO-state survey claims that approximately 3 million
children were reported to Child Protective Services (CPS) in 1995, of which about one-
third were substantiated as cases of child maltreatment. However, higher rates were
reported by the National Incidence Studies of Child Abuse and Neglect (Sedlak &

Broadhurst, 1996), estimating that between 1.8 and 2.8 million children were maltreated

15

in 1993, which translates into 23 to 42 per 1,000 children being abused and/or neglected
annually in the United States. Moreover, population surveys such as the Gallup poll
(Gallup et al., 1995), which surveys a random representative sample of 1,000 families
across the United States, reports even higher rates, estimating that 5% of American
children are physically abused and 2% are sexually abused. These substantial differences
in numbers are due in part to the lack of a consensual deﬁnition of child abuse among the
experts working in epidemiological research, as well as differing sources of data and
different methods used to identify cases (Marsh & Wolfe, 1991). Another example of
discrepant deﬁnitions comes from law; speciﬁcally, there is no universal deﬁnition of
child abuse that is applied in all states. Instead each state provides its own deﬁnition of
child abuse and neglect that establishes the grounds for state intervention in the protection
of children’s well-being. Moreover, there is a great deal of variance between states, with
some states deﬁning child abuse and neglect as a single concept while others provide
separate deﬁnitions for abuse, neglect, sexual abuse, and emotional abuse (National
Clearinghouse on Child Abuse and Neglect, 1998). The breadth of deﬁnitions varies
also, with most states deﬁning abuse as involving serious physical injury such as
disﬁgurement, impairment of bodily functioning, or death, while others specify that any
injury, serious or not, is abuse. In states with broader deﬁnitions, it is usually left up to
the discretion of Child Protective Services to determine what constitutes abuse. Neglect
is more difﬁcult to operationalize, but is frequently deﬁned in terms of deprivation of
adequate food, shelter, clothing, or medical care, while taking into account the ﬁnancial

ability of the family to provide these necessities. In addition to deﬁning the acts or

16

omissions that constitute maltreatment, states also vary in terms of their inclusion or
exclusion of deﬁnitions of perpetrators of abuse, as well as exceptions of abuse based on
religious beliefs and cultural practices. Still further, each state differentiates between
civil statutory deﬁnitions and criminal statutes, which are punitive in nature and deﬁne
acts that are criminally punishable. Thus, in each of the ﬁfty states, there are different
statutory deﬁnitions of child maltreatment that are based solely on the individual social
and political climates of that particular state.

A ﬁnal illustration of the deﬁnitional inadequacy plaguing the ﬁeld of child
maltreatment prevention comes from research. Two decades ago, Besharov (1981)
pointed out that deﬁnitions of child abuse and neglect are the building blocks of research
studies, and that existing deﬁnitions failed to meet research needs because they lacked
comparability, reliability, and taxonomic delineation. Ten years later, Marsh and Wolfe
(1991) explained that differing research agendas (i.e., social work, medical, legal,
psychological, etc.) determined how child maltreatment is deﬁned and measured, and
these vary from study to study, leading to the inability to generalize ﬁndings beyond the
speciﬁc scope of individual studies. In fact, the lack of consensus among researchers
regarding how to deﬁne child maltreatment has led investigators to develop their own
idiosyncratic measures and variables, which has resulted in as many deﬁnitions as there
are research studies (Besharov, 1981). Besharov claimed that the only characteristic that
all deﬁnitions share is their imprecision. For instance, deﬁnitions often contain such
phrases as “a child who lacks proper parental care” or “a child whose environment is

injurious to his welfare” (Besharov, 1981). In addition, popular deﬁnitions often refer to

17

child maltreatment as if it were a single behavioral entity rather than a variety of different
forms of parental conduct that are harmful to children (Besharov, 1981). As Zigler
(1976) pointed out,

“the nature of child abuse is in need of a more differentiated and conceptually-

based classiﬁcatory system. Child abuse is a phenotypic event having a variety of

expressions and causes, and we will make little headway so long as we insist on

viewing every act of child abuse as the equivalent of every other” (p. 171).

This homogeneous view, according to which abuse is seen as a single behavioral entity,
stems from the medical framework which has deﬁned abuse as a syndrome with an
identiﬁable cause and predictable outcome (Kempe et al., 1962), a view that has
dominated the ﬁeld since the inception of the term “battered child syndrome.” This
framework, although helpﬁrl in early medical research on child abuse, has been overly
utilized by differing research agendas, and has perpetuated a deﬁnition that has outgrown .
its usefulness.

These examples from epidemiology, law, and research domains illustrate that
child maltreatment is a multifaceted phenomenon that does not lend itself to simple
description or operationalization. One reason for this problem is that conceptualizations
of child abuse are inherently driven by sOcial judgement and not by empirical science
(Emery, 1989). However, as we have seen, the problem of deﬁning abuse is not only a
scientiﬁc issue, but is also political, social, and medical, and hence is not likely to be
resolved because of the differing strategies and agendas employed by each profession.

Diﬂiculties Related to Inclusive Definitions

18

Cicchetti and Barnett (1991) have advocated for the development of consensus
deﬁnitions similar to those used in the DSM system (APA, 1994) to classify mental
illnesses. Such deﬁnitions would provide a common language for different professionals
to use in research and practice, and would allow for greater generalizability of empirical
ﬁndings. However, deﬁnitions of abuse would be classiﬁcations whose utility depended
on the speciﬁcation of operational criteria for assigning cases to categories and the
inclusion of non-overlapping categories (Giovannoni, 1989). This would require that
child abuse and neglect be precisely categorized according to some pre-speciﬁed criteria
and would necessarily preserve the conceptualization of child maltreatment as a discrete
phenomenon (i.e., physically abused versus not physically abused). There is little
evidence, however, that child abuse is a discrete phenomenon (Marsh & Wolfe, 1991).
Instead, it is more likely that there are degrees of child abuse and in most cases there is no
threshold that deﬁnitively determines at what point a parent is to be labeled abusive.
Moreover, it is a judgement regarding a pattern of parenting that deﬁnes abuse rather than
a speciﬁc action (Marsh & Wolfe, 1991). However, discrete instances of abuse and
neglect certainly occur in the context of otherwise sound parenting, such as a parent who
brieﬂy leaves her child unattended in a car in hot weather while she runs into a store.
Such exceptions to the rule further complicate our ability to arrive at precise deﬁnitions
of abuse — whether it is a discrete phenomenon or a pattern of parenting that signiﬁes
abuse. Hence, a categorical system of classiﬁcation is most likely a poor ﬁt with the
phenomenon of child maltreatment. More speciﬁcally, as Marsh and Wolfe (1991) point

out, there are currently no agreed-on criteria that can reliably differentiate physical abuse

19

from other forms of maltreatment, differentiate combined forms of maltreatment from
one another, or differentiate subtypes of physical abuse based on possible dimensions
such as the type of act, its severity, chronicity, and age of onset. Moreover, the few
studies that have attempted to categorize maltreating or abusive parents using
empirically-derived approaches have tended to reveal constructs based on broad patterns
of care-giving such as intrusiveness, physical abusiveness and hostility, psychological
unavailability, marginal maltreatment, etc. (Oldshaw, Walters, & Shaw, 1989). On the
other hand, extreme types of physical or sexual abuse that often form the basis for legal
action are relatively easy to identify. It is not yet known, however, whether the primary
risks to the child stem from actively hostile, cruel or punitive parenting, a lack of
affectionate care, social and economic impoverishment, or some combination of all of
these (Marsh & Wolfe, 1991). Hence, it may be a poor use of time and resources to
endeavor to categorize subtypes of child maltreatment much the same way subtypes of
mental illness are classiﬁed, a strategy that exclusively employs a medical model
framework. It may be more productive to investigate abuse in relation to the overall
quality of care that is received by the child (Marsh & Wolfe, 1991). Until the causal
mechanisms and the full impact of abusive parenting processes are understood, a broader,
more expansive deﬁnition of abuse may be most appropriate in child maltreatment
research.
The State of the Art: How Can Abuse be Defined and Measured as an Outcome?

Given the previous argument that child maltreatment cannot be precisely deﬁned,

the question that arises is how any deﬁnition can be justiﬁed as an outcome measure in

20

prevention research. To date, most of the research on risk assessment has dealt with
identifying substantiated cases of child maltreatment. Substantiation is a ﬁnding based
on the presence of observable physical evidence, although this is subject to bias in
reporting, and is affected by policy, law, and practitioner workload (English & Pecora,
1994). In a meta-analytical study, MacMillan and colleagues (1993) found that there
were numerous different outcomes reported across studies, making it difﬁcult to select for
analyses those outcomes considered to be most representative of child abuse and neglect.
Included among the many outcomes described in their review were reports of child abuse
and neglect, child health measures such as immunizations and developmental
assessments, aspects of the parent-child relationship, parenting capacity, and parental
competency. Their solution was to select three “proxy” measures that they considered
most closely related to episodes of child physical abuse and neglect: l) hospitalizations,
2) rates of visitation to the emergency room, and 3) injury or accident rates.

In addition to behavioral outcomes, attitudinal markers have also been employed
in prevention research, the most notable being the construct of child abuse potential. One
of the main advantages of using child abuse potential as an outcome measure is that it is
not limited by deﬁnitions of what constitutes an actual act of child maltreatment. It is
also consistent with conceptions of inadequate parenting as falling on a continuum, and
does not restrict investigators by only allowing consideration of speciﬁc acts or
behaviors. This construct is limited, however, because it was derived from a review of
the literature of traits that distinguish abusers from non-abusers -- a body of research that

is plagued with signiﬁcant problems. In addition, the reliability and validity of this

21

construct have not yet been unequivocally established. Milner (1979, 1980, 1984, 1986,
1990) and his colleagues have developed a research program directed at validating the
construct of child abuse potential.

Milner and Wimberley (1979) originally set out to “construct and analyze a test
instrument that could provide a quick, client-administered screening device for assessing
an individual’s child abuse potential” (p. 95). They conducted a review of the literature
on child abuse and neglect from which they developed the 160 items that make up the
Child Abuse Potential Inventory (CAP). They then tested the instrument on 19 abusing,
and 19 non-abusing parents matched on key demographic variables. They found that the
factors of rigidity, problems, and control best discriminated abusers from non-abusers in
their sample. In their next study (Milner & Wimberley, 1980), these researchers set out to
replicate and extend their original ﬁndings. They administered the CAP to a matched
sample of 130 abusing and non-abusing parents and found that 77 of the 160 items
signiﬁcantly discriminated between abusers and non-abusers, correctly classifying 96% of
the participants. They further found that rigidity, distress, and unhappiness were the most
signiﬁcant factors in distinguishing the two groups. In a similar study, Milner and Ayoub
(1980) tested the ability of the CAP to distinguish “at-risk” participants, rather than
individuals known to have committed abuse. The CAP was given to 67 parents in a
program for parents at risk for inadequate parenting and their scores were compared to a
standardization group of non-abusing parents. They found that the at-risk parents scored
signiﬁcantly higher than the control group, with 92% scoring above the mean of the

normative sample. In a further investigation to provide cross-validation data for the CAP,

22

Milner, Gold and Wimberley (1986) administered the instrument to 220 child abusers and
matched control subjects. Using discriminant analysis, they found that the Abuse Scale
correctly classiﬁed 85.4% of the subjects, with 82.7% of the abusers and 88.2% of the
control subjects correctly identiﬁed. In later research, the CAP was found to be positively
related to measures of family conﬂict and marital dissatisfaction (Mollerstrom, Patchner,
& Milner, 1992), and childhood physical abuse (Caliso & Milner, 1992, 1994; Milner,
Robertson, & Rogers, 1990).

Finally, the CAP has been used in several studies as an outcome measure (cf.,
Burrell et al., 1994; Kolko et al., 1993; Stringer & LaGreca, 1985). One such study by
Dukewich, Borkowski, and Whitman (1996) examined psychological adjustment, child
characteristics, social support, and parenting orientation in relation to child abuse
potential among a sample of adolescent mothers. They found that parenting preparation
and child temperament had a signiﬁcant relationship with abuse potential, while, contrary
to other risk studies, social support and maternal psychological adjustment were unrelated
to abuse potential.

Overall, the choice of an outcome measure is an inherently ambiguous decision
that ultimately rests on the shoulders of the individual researcher. Substantiated cases of
abuse and neglect may represent the most clear-cut dependent variables. However, only
about 40% of abuse allegations are substantiated (English & Pecora, 1994) due to the fact
that substantiation is an institutional variable that differs according to site, policy, and
law. Hence, many children may be abused according to some deﬁnitions without a legal

ﬁnding of substantiation. Another problem associated with using substantiation as an

23

outcome measure is that it is a poor ﬁt with prevention research. By deﬁnition,
substantiated cases of abuse warrant intervention services, not prevention services,
because the abuse has already occurred. In prevention research, where the goal is for risk
assessment to be used to identify individuals who are at-risk, but who have not yet
crossed the behavioral threshold signifying abuse, the outcome measure must be
indicative of potential to abuse. Taking these factors into consideration, the construct of
child abuse potential, as measured by some combination of attitudinal and behavioral
indicators, provides an adequate assessment of risk.
The Entanglement of Risk with Process

Similar to the lack of clarity deﬁning what constitutes child abuse, there is also no
clear-cut distinction between risk and outcome processes in this ﬁeld of investigation.
Risk variables have historically been conceptualized as factors (i.e., behaviors, attitudes,
events) that temporally precede outcome (abuse). AS will be discussed, this assumption,
based on medical model conceptions of disease processes and outcomes, may not be well
suited to the risk assessment paradigm.
Utilization of the Medical Model

Historically, the use of the medical model to deﬁne, identify, and treat child
maltreatment has led to a limited view of the relationship between risk factors for abuse
and outcome. Investigations have traditionally focused on the description and
measurement of physical violence, injuries to the child, and health status outcomes, and
these medical indices have been the metric by which abuse was gauged. Likening child

maltreatment to a disease entity has afforded the luxury of Viewing risk factors as

24

discernibly distinct from abuse outcomes. For instance, if hitting is considered child
abuse, then authoritarian child-rearing attitudes (i.e., “spare the rod, spoil the child”)
would be a “risk factor” for that particular outcome. However, parents who adhere to
authoritarian parenting practices may also engage in psychologically controlling
behaviors such as intimidation, rejection, or degradation, practices that may be used in
addition to corporal punishment. These practices, in and of themselves, are considered by
many to be psychologically abusive (McGee & Wolfe, 1991). According to this example,
the risk factor —— authoritarian parenting -—— can also be seen as the abusive outcome.
Hence, where the line is drawn distinguishing risk factors from outcome factors is an
inherently arbitrary practice that is completely dependent upon how abuse is deﬁned.

In theory, any parenting practices that have either physically injurious or
psychopathogenic effects could be considered abuse. But such a broad conceptualization
would render the distinction between risk factors and abusive practices virtually
meaningless. Baumrind (1995) points out that most developmentally psychopathogenic
acts of rejecting, degrading, mis-socializing, exploiting, or being emotionally
unresponsive are manifestations of inadequate care-giving rather than legally actionable
abuse. AS such, these behaviors can generate psychopathology in the child, but such
parenting practices are not clearly outside the norms of parental conduct, which would
make the identiﬁcation of risk factors for such behavior impossible. This argument
brings back into question the validity of assessing risk for child abuse potential because it
demonstrates that all parenting practices are merely points along a continuum, with

arbitrary demarcations differentiating adequate parenting from inadequate parenting.

25

Suitability of the Medical Model Framework

The medical model framework my not be suited to risk assessment for CAN
because parenting characteristics can be considered both risk variables and abuse
outcomes, depending entirely upon where the line is drawn between adequate and
inadequate parenting. Speciﬁcally, the medical model was originally applied to physical
conditions in which there is usually a clear-cut and consistent pattern of symptoms and
signs in the patient (Graham, Dingwall, & Wolkind, 1985). Even in most disease states,
however, arbitrary criteria are used to deﬁne what does and does not constitute the
presence of a speciﬁed disease. As Graham and colleagues point out,

even in apparently well-deﬁned conditions like cancer, the presence of

precancerous states means that cut-off points are often difﬁcult to apply. Further,

in most “diseases,” e. g., obesity and hypertension, the cut-off points that are

applied are highly arbitrary because there is, in fact, no more than ﬂimsy

justiﬁcation for any particular criterion (p. 1218).
The main disadvantage of applying this model to psychological and behavioral
conditions, such as child maltreatment, is that “the condition becomes ‘reiﬁed’ or given
the status of a disease, when in fact, there is little justiﬁcation for this view to be taken”
(Graham et al., 1985, p. 1218). Whereas disease proCesses imply some underlying
change in physical structures or functioning, child maltreatment does not necessarily
involve the presence of some reliable and valid behavioral criteria that establishes the
presence or absence of the condition.

With disease states, there is often (but not always) a clear distinction between risk

26

variables for the disease and symptoms of that disease. Speciﬁcally, risk variables are
those factors that bring about subclinical changes in physical structures or functioning,
while symptoms represent the development of the changes to clinically signiﬁcant levels.
For example, high blood pressure, smoking, and obesity are risk factors for heart disease,
whereas chest pain and shortness of breath are symptoms of the disease (Oliver, Ashley-
Miller, & Wood, 1987). One would not say that chest pain is a risk factor for heart
disease. Such distinctions between risk factors and symptoms cannot be easily made in
the case of child maltreatment. For instance, a parent who regularly slaps her child may
be at risk for engaging in more violent abuse such as punching or kicking her child (in
this case, slapping is the “risky” behavior); however, another parent may only slap her
child when she gets in a ﬁght with her spouse (in this case, marital discord is the risk
factor, and slapping is the symptom of abuse). This interchangeability of risk factors and
abuse symptoms brings into serious question the reliability of risk assessment for child
maltreatment because the criteria for “risk” can change with each individual case.
The Conundrum of Comorbidity

Differentiating risk and outcome factors is somewhat akin to the proverbial
chicken or egg conundrum. In the medical sciences, the risk driven approach has
contributed to the successful prevention of disease states. Epidemiologists have
successfully decreased the number of cases of heart and lung disease by targeting risk
factors that increase the risk for these diseases such as smoking, lack of exercise, and high
fat diets (Maccoby & Altman, 1988). The essence of this risk-focused approach is that

problems can be prevented by identifying the factors or processes that increase the risk of

27

these processes and then eliminating them or mitigating their effects. Similarly, in the
social sciences, risk processes have been deﬁned as individual or environmental hazards
that increase individual vulnerability to future negative outcomes (Bogenschneider,
1996). Inherent in this deﬁnition is the temporal requirement of the presence of risk
before the occurrence of outcome. But in the case of child maltreatment, this is often not
the case. For instance, alcohol abuse is considered to be a risk factor for child
maltreatment; maltreating parents are 18 to 38% more likely to abuse alcohol than non-
maltreating parents (Widom, 1992). It is unclear in this example of comorbidity whether
alcohol abuse is a risk factor for child maltreatment, or whether it is part of the process of
substandard parental care-giving. The association between alcohol abuse and child
maltreatment is further complicated by the interplay of other variables such as personality
style, the family composition, the degree of family conﬂict, etc. It is also unclear whether
alcohol use disinhibits aggression, interacting with personality type to predict Violence, or
whether aggressive personality types act violently whether or not they have been drinking
(Emery & Laumann-Billings, 1998).

The issue of comorbidity of other types of abuse with child maltreatment further
blurs the boundary between risk factors and process factors. For example, marital
violence has been found in 40 to 75% of cases of child abuse (Layzer, Goodson, &
DeLange, 1986), and children living with a sexually abused mother are 12 to 14 times
more likely to be sexually abused themselves (McCloskey, F iguerdo, & Koss, 1995).
History of childhood abuse and domestic violence are considered to be risk factors for

child abuse, but it is possible that child maltreatment is actually an extension of these

28

complicated behavioral processes. It seems more likely that comorbid conditions
contribute to abusive outcomes through a cumulative effect, much like lead poisoning.
Overall, the risk-focused approach to prevention, when applied to child

maltreatment, suffers from a serious blurring of the boundary between risk and process
factors. The comorbidity of different forms of abuse make it difﬁcult to determine the
causes and consequences of violence in terms of what is a risk factor and what is a
process factor. If risk factors cannot be reliably distinguished from process factors, the
assessment of risk is rendered virtually meaningless. The issue of comorbidity
exempliﬁes the complex relationship between risk factors and process factors, as well as
highlights the importance of looking at other variables at different ecological levels that
may contribute to violent outcomes. An ecological approach to understanding child
maltreatment will be discussed subsequently.

The Application of Etiological Theories to Risk Assessment: Ecological Levels of

Analysis
Historically, a focus on treatment has taken precedence over the search for causes

of child maltreatment, severely limiting the knowledge base from which model building
could take place (Azar, 1991). Theory building has also been hampered by domination in
this ﬁeld by law and medical professionals whose efforts have been concentrated on
aspects of injury, crime, and punishment, and whose primary methods have consisted of
intensive case studies. Such a concentration on the extreme and bizarre cases has not
generated an adequate empirical knowledge base from which to derive well-articulated

and validated theories of child maltreatment (Azar, 1991). Fortunately, the ﬁeld has

29

recently beneﬁtted from the attention of psychologists who come from strong empirical
traditions and who have brought rich theoretical backgrounds to bear on the problem of
family violence (Azar, 1991). This has led to an expansion of theoretical models, most
signiﬁcantly moving beyond single causal models to searches for causes at multiple levels
of analysis, as well as broader conceptions of child maltreatment as a heterogeneous
phenomenon.

Advances in etiological theories of child abuse and neglect, however, have
outdistanced theorizing in the domain of risk assessment. While it is widely accepted that
family violence is the product of various, interacting ecologically distinct processes, most
risk assessment procedures continue to utilize unidimensional models that do not reﬂect
the ecological approach. Ecological models are speciﬁcally concerned with identifying
the effects of causal agents at multiple levels of environmental contexts. Such models
assume multivariate causality and posit interactional effects between components at
different levels of the same ecological context (Holden, Willis, & Corcoran, 1992).
Bronfenbrenner’s (1979) ecological model, which was developed to provide a framework
for understanding child development, is most often used to describe the levels of analysis
involved in human functioning. This model posits four primary ecological levels: the
ontogenic level is concerned with individual characteristics such as psychological
functioning and personality traits; the microsystemic level involves family system
characteristics such as the marital relationship, parent-child interactions, as well as stress
and social support; the exosystemic level deals with community variables such as poverty

rates, proportion of female headed households, and transitory neighborhoods; and the

30

macrosystemic level is concerned with the overarching social context that includes
cultural and religious values and belief systems. This framework has helped to integrate
the research on child abuse leading to more complete etiological models that recognize
the important balance of potentiating factors and compensatory mechanisms at multiple
ecological levels (Belsky, 1993; Cicchetti & Lynch, 1993; Garborino, 1977).

The scope of the current investigation is to apply an ecological framework to the
process of risk assessment because there are currently no risk assessment procedures that
adequately incorporate etiological variables at all of the ecological levels (Caldwell et al.,
1988). To this end, the ontogenic (individual), microsystemic (family), exosystemic
(community), and macrosystemic (cultural) levels of analysis will be discussed, and the
evidence linking these variables to child maltreatment etiology will be elucidated. This
will provide a framework for a subsequent review of extant risk assessment procedures in
terms of their dimensionality and their ecological focus.

The Ontogenic Level of Analysis

This level of analysis is concerned with characteristics of the individual, and
refers to parent and child factors that contribute to maltreatment. Attempts have been
made to identify a single personality type that ﬁts all abusive parents, but this task has
proven impossible and has led to the realization that people with very different
personality characteristics exhibit abusive behavior towards their children (Frances,
Hughes, & Hitz, 1992). The typology approach has met with greater success in
describing abusive parents, although conclusions from this research support the idea that

characteristics identiﬁed do not apply to all abusers, but rather to several different types

31

of abusers (Sloan & Meier, 1983). Frances and colleagues (1992) conducted a typology
study with 82 conﬁrmed child abusers using psychometrically obtained personality data
rather than clinical observations and demographic information, which has traditionally
formed the basis for this type of research. They found that the physically abusive parents
in their sample reliably ﬁt into one of ﬁve psychological typologies: 1) abusive parents
who tend to be shy and withdrawn, apprehensive, sober, restrained, and who have the
least amount of education and the greatest number of children compared to other abusive
parents; 2) abusive parents who present as being “normal” in terms of personality
features, tend to have relatively high educational levels, and fewer children compared to
other abusers; 3) abusive parents who are compulsive, bold, dominant, assertive, and who
tend to be highly manipulative; 4) abusive parents who tend to be compulsive and who
are relatively passive and submissive in their interpersonal relationships; and 5) abusive
parents who are isolated and withdrawn, suspicious of others, tense and apprehensive,
and who tend toward emotional instability. These typologies are strikingly dissimilar
from one another and highlight the heterogeneity of features that characterize parents who
maltreat their children. Not only do they fail to identify certain types of individuals prone
to abuse, they are so broad that they don’t rule anyone out either. Current approaches to
risk assessment at this level of analysis often fail to take into account this diversity of
personality characteristics, and this may be reﬂected in the high recidivism rates
associated with current treatment approaches (Gabinet, 1983). In other words,
intervention programs that are poorly suited to the personality characteristics of the

individuals in treatment are surely destined to fail. And from Francis et al.’s study, it

32

seems that the range of personality characteristics of abusing parents spans the continuum
from apparently “normal” to obviously “unstable,” suggesting that targeting of
individuals for intervention should be all inclusive.

Aside from the typology approach, many parental personality characteristics have
been implicated in child maltreatment; a summary of this literature was provided by
Pianta and colleagues (1989). These researchers reported that factors such as low self-
esteem, poor impulse control, external locus of control, negative affectivity, and
decreased ability to cope with stress were found to increase the likelihood that an
individual will commit child abuse (Pianta, Egeland, & Erickson, 1989). Parental age
has also been investigated as a contributing factor, with the common assumption that
children of adolescent mothers are at greater risk for maltreatment due to their mother’s
immaturity. Buchholz and Korn-Busztyn (1993) conducted a meta-analysis in this
domain and found that adolescent parents were not signiﬁcantly different from their adult
counterparts in terms of abusive practices. In this same vein, Olds (1982) conducted a
child abuse prevention program for ﬁrst time mothers who were either teenagers,
unmarried, or of low socioeconomic status. Low SES, however, was the only variable
that, when considered by itself, was consistently related to child abuse (Holden, Willis, &
Corcoran, 1992). This illustrates that individual level variables such as age or marital
status may only be indirectly related to maltreatment, and do not provide sufﬁcient
information about abuse potential.

In addition to parent characteristics, Belsky (1993) contends that children who are

abused may have characteristics that contribute to their own victimization such as age,

33

sex, poor health, behavioral deviance, and difficult temperament. Herrenkohl,
Herronkohl, and Egolf (1983) found that physical abuse, as opposed to emotional cruelty
and neglect, was signiﬁcantly associated with child behaviors, whereas the latter types of
abuse were associated with adult conﬂict and insufﬁcient knowledge of children’s needs,
respectively. These researchers suggest that the child’s role in his or her own physical
abuse is as a source of frustration that blocks the parent’s efforts to achieve desired goals.
For instance, in the case of physically or mentally handicapped children, the parent may
feel unduly burdened by the time required to care for the child, and the resulting
ﬁ'ustration can lead the parent to strike out at the child (Zirpoli, 1986).

Another important dimension of this level of analysis is the developmental history
of the parent and the much debated process of intergenerational transmission of abusive
parenting practices (Kauﬁnan & Zigler, 1989; Widom, 1989). Research has consistently
shown that there are differences in developmental history between adults who do and do
not abuse their children. However, the rates of intergenerational continuity of abuse that
are reported range from 7% (Gil, 1973) to 70% (Egeland, Jacobvitz, & Papatola, 1987).
Some of the problems with this database leading to such discrepant ﬁndings are
retrospective accounts of abuse, under- and over- reporting of abusive experiences, and
inadequate control procedures. Belsky (1993) concluded from his review of this literature
that about one-third of individuals who were abused or neglected during childhood will
maltreat their own children. However, Kaufman and Zigler (1989) add that although
“being maltreated puts one at risk for becoming abusive, the path between these points is

far from direct or inevitable” (p. 190).

34

An important mediating factor that may affect whether or not abusive practices are
passed on from one generation to the next is attachment, which provides a theoretical
basis ﬁom which to View resilient and nonresilient outcomes. Attachment theory posits
that the mother, or primary care-giver, provides a secure base for the infant who in turn
feels that she is a stable, steady, dependable force in his or her life and is emotionally
available. Bowlby (1969) claimed that the quality of attachment, not merely its presence,
was central to healthy developmental outcomes. Bowlby further believed that the primary
attachment relationship was a prototype for later social relationships. The importance of
attachment to the intergenerational effects of abuse is that the way parents organize their
own childhood experiences is a powerful predictor of how they will parent their own
children. Several studies have shown that parental attachment history, as inferred from
interviews, is related to the quality of attachment in the next generation (Main, Kaplan, &
Cassidy, 1985; Morris, 1980; Ricks, 1985).

Attachment theory also provides a theoretical basis for the discontinuity of
patterns of abuse across generations. For example, Egeland, Jacobvitz, and Sroufe (1988)
followed the child care practices (for three years) of women who had been abused
themselves as children. They found that a common denominator among those who did
not abuse their own offspring was the presence of a warm, supportive relationship with
either another adult during childhood, a therapeutic relationship at any point in life, or a
partner relationship in adulthood — such relationships seemed to play an emotionally
corrective role. These researchers interpreted their ﬁndings from an attachment

perspective, claiming that positive relationships can ameliorate the effects of poor

35

attachment representations resulting from negative childhood experiences. This ﬁnding
is consistent with Caliso and Milner’s (1992) report that among mothers at risk for
committing abuse due to an abusive developmental history, those who did not were
involved in more satisfying interpersonal relationships than those who did perpetuate
abuse (Belsky, 1993). Such relationships are postulated by attachment theory to modify
internal working models and interpersonal expectations among individuals with seriously
troubled parent-child relationships, and enable them to care for their own offspring in a
way that is different from the way they themselves were raised (Belsky, 1993). In this
same vein, Ettema and Caldwell (submitted) found that mothers with secure attachment
styles showed the least degree of parenting problems and child problems, whereas
mothers with dismissing styles (positive self, negative other) demonstrated the greatest
parenting problems such as increased stress, attitudes that were indicative of child abuse
potential, and perceived child behavior problems.

In summary, within the ontogenic level of analysis, there is a vast literature that
describes many individual personality factors of the parent and child that are associated
with child maltreatment, only a few of which have been outlined here. The characteristics
at this level of analysis that are of the most interest to the current investigation are adult
personality characteristics, psychological resources, developmental history, attachment
style, and child characteristics, all of which are considered aspects of person-level risk in
the current model of risk assessment. Individual characteristics are no doubt an important
etiological component, but they must be viewed in relation to the other ecological levels,

of which the family context will be discussed next.

36

The Microsystemic Level of Analysis

This level of analysis is mainly concerned with familial ﬁrnctioning as it relates to
the perpetuation of abusive practices. Studies in this area have examined such factors as
parent-child interactions, family structure and size, and characteristic styles of resolving
conﬂicts (Emery & Laumann-Billings, 1998). Belsky’s (1993) literature review of
parent-child interactions as they relate to abuse found overall that abusive parents are less
positive in their parenting and more punitive in their discipline. Vasta (1982) argues that
what happens in the abusive event is that “an aggressive act of physical punishment that
has a functional goal of inﬂuencing behavior gets out of control and turns into an irritably
aggressive act that is more intense, severe, and repetitive than the perpetrator ever
intended” (quoted in Belsky, 1993, p. 421). Belsky further recognizes that the
interpersonal context of abuse is shaped by the interaction of individual characteristics of
the parent-Child dyad. He states that:

abuse appears to emerge...when a parent with a predisposition toward anxiety,

depression, and hostility becomes irritated with a child, attempts to physically and

instrumentally control the child, but becomes so aroused as to lose control of him

or herself and overdoes what was initially intended to be an act of discipline. It is

not difﬁcult to imagine how this process could be very much shaped by a

childhood history of mistreatment and exacerbated by features and actions of the

victimized child (p. 421).

An important dimension of the parent-child dyad that contributes to maltreatment

involves the attitudes and parenting style of abusive parents. Cicchetti and Lynch (1993)

37

reviewed this literature and concluded that maltreating parents, compared to non-
maltreating parents, were less satisﬁed with their children, perceived their child rearing as
more difﬁcult and less enjoyable, used more controlling disciplinary techniques, and
generally did not encourage independence and autonomy in their children. Robataille
and colleagues (1985), however, found that parental authoritarianism did not predict
abuse potential, and was only predictive of rigid parenting practices. Other evidence
suggests that maltreating parents are more likely to “parentify” their children, resulting in
the child expressing more nurturing and sensitivity toward the parent than vice versa
(Dean, Malik, Richards, & Stringer, 1986).

Marital disharmony is another variable at this level of analysis that has been
described in the child abuse etiology literature (Herrenkohl et al., 1983). Herrenkohl and
colleagues (1983) found that marital discord was signiﬁcantly related to emotional
cruelty, although not to physical abuse or neglect. They suggest that the anger and
frustration associated with marital conﬂict is often displaced toward the child in the form
of anger and hostility. A link between child maltreatment and marital violence has also
been reported in the literature (Kempe et al., 1962; Layzer et al., 1986; Straus, Gelles, &
Steinmetz, 1980). Overall, variables at this ecological level are entwined with those at
the previous level because each person in the family system brings with them their own
personal developmental experiences, physical and emotional health, tolerance for
frustration, coping skills, and self-image, among numerous other factors. These in turn
affect marital stability, familial interactions, parenting style, and social resources and

supports. These two ecological tiers are further embedded in a community ecology which

38

will be discussed next.
The Exosystemic Level of Analysis

This ecological tier has recently been recognized as playing a signiﬁcant role in
the etiology of child maltreatment. Researchers have found that violence in the family is
related to qualities of the community in which the family is embedded such as poverty,
absence of family services, social isolation, lack of community cohesion, high levels of
unemployment, inadequate housing, and community violence (Emery & Laumann-
Billings, 1998). Poverty has been most extensively studied in relation to child
maltreatment, but primarily at the level of the individual and the family, with attention
only recently being directed at the effects of poverty on neighborhood and community
conditions that inﬂuence child care practices (cf. Coulton, Korbin, Su, & Chow, 1995;
Garborino & Crouter, 1978; Garborino & Kostelny, 1992; Zuravin, 1989). Most families
living in impoverished communities, however, do not abuse their children, which
suggests that the relationship between community level factors and child maltreatment is
not simple or direct. Causal relationships have also been disputed by researchers who
disagree on whether community impoverishment leads to higher rates of maltreatment or
whether maltreating families are drawn together in impoverished neighborhoods but
would behave similarly no matter where they lived (Polansky, Gaudin, Ammons, &
Davis, 1985).

In their seminal paper, Garborino and Sherman (1980a) used a mapping technique
to screen neighborhoods for the proximal environmental factors of families that were

associated with child maltreatment. In accord with their “social impoverishment”

39

hypothesis, they found that “high risk” families -- families who were most in need of
resources -- were clustered together in settings where they had to struggle to meet those
needs. They described “high risk” neighborhoods as areas made up of very needy
families competing for scarce social resources. These families’ problems were
compounded by the community context of tenuous informal helping networks and
pervasive estrangement and withdrawal from neighbors. They emphasize that community
risk operated independently from conventional socioeconomic variables, and instead it
was associated with aspects of family stresses and supports. Moreover, neighborhood
risk scores were found to be signiﬁcantly and negatively correlated with the overall rating
of the neighborhood as a place to raise children made by parents living in that
neighborhood, even after controlling for family income and structure (Garborino &
Sherman, 1980). Hence, parents were found to be sensitive to the way neighborhood
factors establish a particular climate for raising children that is unrelated to conventional
socioeconomic and demographic indices. Overall, Garborino and Sherman (1980b)
concluded that high-risk neighborhoods compound internal familial problems, increasing
their potential for child abuse, whereas low-risk neighborhoods help support families by
overcoming their internal weaknesses thereby decreasing the likelihood of child
maltreatment.

The identiﬁcation of high and low risk neighborhoods was accomplished by
Garborino and Sherman through a prOcess they developed called “ecological mapping”
(Garborino & Sherman, 1980a, 1980b) First, these researchers plotted reported cases of

child maltreatment on a map of neighborhoods within a city district. They then

40

superimposed on this map a map plotting facts about the local ecology to determine
which social conditions were associated with reported cases of child maltreatment. The
social conditions they included in their ecological map consisted of two economic
variables and three variables selected to represent neighborhood social resources. The
former category consisted of the percent of households with incomes less than $8,000 a
year (in 1975 dollars), which was considered to be economically “struggling,” and the
percent of households with incomes over $15,000 a year, which was considered to be
economically “comfortable.” The latter category consisted of indicators of social
resources of the neighborhood and included percent of female headed households, percent
of married mothers with children under the age of eighteen in the labor force, and percent
of residents who had resided in the neighborhood for less than one year. These variables
were chosen to represent social resources because they believed that single parent
families, working mothers of dependent children, and transients were all drains on the
neighborhood because people in these situations tend to be concerned with meeting their
own needs and unable to participate in natural helping networks. In matching these two
maps, these researchers found that the ﬁve indicators of social conditions accounted for
eighty percent of the variance in child maltreatment among the neighborhoods. They
concluded from their data that a neighborhood’s economic and social character provides a
very accurate predictor of child maltreatment.

Since Garborino and Sherman’s (1980) initial study of “social impoverishment”
as it relates to child maltreatment, many researchers have embarked on their own

explorations of the community context of child maltreatment. Deccio, Homer, and

41

Wilson (1994) attempted to replicate Garborino and Sherman’s ﬁndings by mapping child
maltreatment reports according to census tracts, as well as conducting interviews with
parents in high and low risk areas, while controlling for family income. These
researchers found that neighborhood characteristics other than income differentiated two
neighborhoods in terms of rate of reports to Child Protective Services. These differences
were: a) families in the low risk neighborhood were more likely to have lived in the
neighborhood for more than 5 years, b) families in the low risk neighborhood were more
likely to have a telephone, and c) there were fewer vacant housing units in the low risk
neighborhood. Contrary to Garborino and Sherman’s ﬁndings, these researchers did not
ﬁnd that parents in the high and low risk neighborhoods differed in their perception of
available social support. They did, however, ﬁnd that the low risk neighborhood in their
study was more socially integrated than its high risk counterpart, as demonstrated by its
higher employment rate, more stable history of residence, and connectedness of
individuals to friends, neighbors, and relatives by telephone. They attributed the
differences in ﬁndings to the conceptual differences between the constructs of “social
support” and “social integration.” They argued that social integration connotes
membership, participation, and belonging, while social support emphasizes the quantity
and quality of interactions among individuals. Based on their ﬁndings, they suggest that
“the quantity and quality of interpersonal contacts per se may be less important in
preventing child abuse and neglect than the sense of having a stake and a role, however
small, in the community” (p. 136).

In another study, Zuravin (1989) used the same “social impoverishment”

42

indicators as Garborino and Sherman (1980) (see above) to determine the relationship
between community characteristics and maltreatment rates. However, this study looked
at child abuse and child neglect as separate dependent variables to determine whether
areas in which physical abuse predominated had different community characteristics than
those areas in which child neglect was most prevalent. She found that higher rates of
child neglect were signiﬁcantly related to the community characteristics of extreme
poverty (200% below the National poverty line), single family dwellings, vacant homes,
and transiency. Contrary to Garborino and Sherman, she found that female headed
households and working mothers with young children were actually associated with lower
rates of child neglect and physical child abuse. Signiﬁcant neighborhood predictors of
physical child abuse were extreme poverty, vacant homes in the neighborhood, and single
family dwellings. From these ﬁndings, this researcher concluded that several
neighborhood characteristics were signiﬁcant, independent predictors of each type of
child maltreatment, which is consistent with the ecological hypothesis. However, she
suggests that alternative explanations such as selection (neglecting families elect to live in
a particular type of neighborhood) or aggregation (substantial stress directly inﬂuences a
large proportion of neighborhood residents) cannot be ruled out.

Overall, a summary of the literature on neighborhood characteristics associated
with child maltreatment reveals that twelve neighborhood characteristics (transience or
population change, economic status, unemployment, neighborhood development,
educational development of residents, stresses on mothers, housing status, residential

density, population density, desirability of neighborhood, ethnic composition, and degree

43

of urbanization), operationalized by twenty-seven different indicators, have been
examined (Zuravin, 1989). Studies generally agree that reported incidence rates of child
maltreatment vary as a function of community characteristics. However, due to the vastly
different methodologies employed to calculate child abuse rates and operationally deﬁne
community characteristics, child maltreatment, etc., this literature cannot be interpreted as
deﬁnitively supporting the ecological hypothesis. However, this is a promising area for
future research into the causes of child maltreatment, and warrants substantial
consideration in our efforts to identify risk factors for child maltreatment. To this end,
the community characteristics under scrutiny in the current investigation include
neighborhood poverty rates, percentage of female headed households, percentage of
working mothers, and transiency rates. These speciﬁc neighborhood characteristics were
chosen because they have been the most extensively studied (see above), yet with
equivocal results. Maintaining consistency across studies will allow easier comparison of
ﬁndings. Furthermore, they have not been studied in relation to child abuse potential, but
rather in relation to substantiated cases of abuse and neglect. They were also chosen
because they are publicly available via US. census tract data.
The Macrosystemic Level of Analysis

Although this level of analysis is concerned with the broader social and cultural
context in which child maltreatment occurs, the current investigation is primarily
concerned with cultural differences in sanctions of violence as a contributor to child
maltreatment. Physical punishment as a child-rearing technique is endorsed in American

culture more so than in other cultures (Parke, 1982), which may account for the greater

44

rates of child maltreatment in this country compared to rates in many Asian countries. A
literature review of risk factors for child maltreatment associated with different cultures
indicates that some risk factors may be present in families regardless of culture (English
& Pecora, 1994). These include parent characteristics such as mental illness, substance
abuse, and history of maltreatment (English et al., 1993). Cross-cultural differences, on
the other hand, have been found in terms of family values, physical discipline,
supervision, and provision of medical care (English & Pecora, 1994). English and
Pecora (1994) point out that culture affects family’s concepts regarding contextual
interpretation of behavior, and their response to individuals outside of the family network.
Brissett-Chapman (1993), in her work with African American adolescents, stated that
understanding risk as it relates to this population must take into account “attitudes toward
self, speech and language, interpersonal relations, sexuality, anxiety and patterns of
defense, coping and adaptive behavior, family structure and roles, socioeconomic status
and living arrangements, degree of integration-acculturation, social support system,
communication patterns, help-seeking behaviors, social environment, peer relationships,
and community assessment” (p. 55). This researcher argues for a synthesized approach to
risk assessment that brings together multifaceted information in order to deliver more
appropriate care-giving responses by the CPS system. The small body of research dealing
with cultural differences in risk status highlights the need for additional research that
controls for income and socioeconomic factors, as well as captures substantive and
interactive factors. Overall, it appears that culture plays a dynamic role in risk for child

maltreatment, but the intricacies of cultural factors have yet to be explored in this domain.

45

To this end, the ethno-cultural make-up of the neighborhoods in the current investigation
will be explored in relation to risk status.
The Role of Stress as a Mediating Variable

Stress is a variable that is not easily classiﬁed in the ecological hierarchy. Stress
operates at all of the ecological levels (i.e., job-elicited tension, marital disputes,
disobedient children, housing shortages, etc.) with various stress factors having additive
and interactive relationships with other contextual variables to produce their effects
(Parke, 1982). Thus, stress mediates many of the relationships between ecological
variables and abuse outcomes. For instance, Tietjen (1980) explains that poverty is a
source of stress on individuals and families that in turn promotes child maltreatment.
However, not all families who live in poverty maltreat their children, and abuse and
neglect are by no means a lower-class phenomena (Egelend et al., 1980). Garborino’s
research program has demonstrated that community stressors, exacerbated by lack of
social resources, is related to child maltreatment rates. The role of stress at the individual
and interpersonal level in the expression of abuse is further illustrated by Egeland and
colleagues (1981). These researchers found that number of stressful life events
distinguished families who were reported for child maltreatment. However, the majority
of highly stressed families in their study did not abuse their children. These researchers
found that abusers were more aggressive, anxious, and defensive than their non-
maltreating counterparts. In this study, individual characteristics interacted with stress,
producing violent behavior in some cases. In a further study of the contribution of stress

to child maltreatment, Justice and Calvert (1990) found that abusing and non-abusing

46

families who were matched on mean stress levels exhibited differences in perception of
stress. Perception of stress was mediated by cohesion and independence, with abusing
families reporting signiﬁcantly less family cohesion as well as less independence of its
members from one another. Taken together, these studies highlight the inappropriateness
of viewing stress as simplistically and directly related to child maltreatment, as it often is
viewed when it is used as a checklist item for risk of potential abuse (cf., Agathonos-
Georgopoulou & Browne, 1997; Brayden, Altemeier, Deitrich, Tucker, Christensen et al.,
1993). Moreover, they illustrate the necessity of viewing stress as a moderating variable
that interacts with factors on all levels of the ecological hierarchy. Accordingly, in the
current investigation, the person-level and community-level characteristics that are
mediated by stress will be explored in relation to their effects on child abuse potential.
Stress in this model will be operationalized using the Parenting Stress Index (Abidin,
1990) which'measures the relative magnitude of stress in the parent-child system. The
two major sources of stress in the parent-child relationship that are measured by the PSI
are parent characteristics and child characteristics. Speciﬁc sources of stress within each
of these domains include the degree of adaptability of the child, the child’s reinforcing
qualities, the demandingness of the child, the child’s activity level, the child’s mood, the
parent’s subjective experience of feeling trapped by the parenting role, the parent’s social
isolation, the parent’s level of depression, the parent’s feeling of competence, the parent’s
physical health complaints, parental attachment to the child, the spousal relationship, and
parental social support (Lloyd & Abidin, 1985). This measure yields an aggregate score

that was designed to identify parent-child systems that are under stress and are at-risk for

47

the development of dysfunctional parenting behavior or behavior problems in the child
(Abidin, 1983).

Finally, English and Pecora (1994) speculate that one reason there has been such
inconsistent implementation of risk models concerns the lack of clarity regarding the
effects of interactions among risk factors. Important questions they raise include how
individual risk factors are weighted, how they interact with one another, and how they
interact with strengths. The absence of empirical evidence to answer these questions
represents a signiﬁcant breach in our knowledge base regarding risk processes.
Exploration of the mediating role of stress in the current model may help address some of
these questions. As a mediator variable, stress in the current model is speciﬁed to be part
of a causal chain that is affected by prior variables (person-level and community-level
risk, measured at Time 0) and in turn affects a subsequent one (child abuse outcome,
measured at Time 2).

Summary of Conceptual Issues Related to the Assessment of Risk for Child Maltreatment

The assessment of risk is an inexact science that suffers from several conceptual
shortcomings. First, there are no agreed upon criteria that deﬁnitively deﬁne child abuse
and neglect due to the connection of deﬁnitions to differing political agendas. Second,
evidence suggests that child maltreatment is not a discrete phenomenon, but rather a
pattern of parenting that shows great variation over timeand across situations. This
makes it difﬁcult to distinguish risk factors from factors that are part of the process of
parenting, and any distinction is arbitrarily determined based upon the speciﬁc deﬁnition

of maltreatment being used. Third, recent etiological models of child abuse adhere to an

48

ecological framework that is dynamic in nature and recognizes that multiple pathways
lead to abusive situations. However, risk assessment is conducted unidimensionally
which does not accurately reﬂect our current understanding of causal mechanisms. These
conceptual shortcomings do not necessitate the abandonment of risk assessment for child
maltreatment. Rather, they emphasize the need for more ﬁne-tuned, theoretically sound
procedures. The assessment of risk is a practice that has been beneﬁcial to medicine
despite its imperfections. Hence, it may also have signiﬁcant utility in the behavioral
sciences if it is more accurately applied. The current investigation attempts to determine
whether risk assessment, when theoretically applied, has value as a predictor of future
parenting behavior. The remainder of this analysis will focus on the practical issues
related to risk assessment. This analysis will pave the way for the construction of the

model of risk assessment to be empirically tested in this investigation.

49

PRACTICAL ISSUES RELATED TO THE ASSESSMENT OF RISK

As previously mentioned, risk assessment procedures have two primary
applications. The ﬁrst is for practical purposes such as screening participants for
inclusion into secondary prevention programs or identiﬁcation by child protective
services, and the second is for research purposes such as determining the reliability and
validity of assessment procedures. Measurement systems vary in terms of the type,
content, and scope of the information that is utilized, as well as differ in terms of the
structure of the format, from subjective clinical interviews, to qualitative open-ended
questions, to more discrete, objective, and empirically-derived scales or checklists
(Murphy-Berman, 1994). It is questionable, however, the extent to which these
assessment procedures are founded upon our current empirical knowledge base (of.
Besharov, 1981). Keller, Ciccinelli, and Gardner (1988) argue that typical approaches to
risk assessment are developed by consensus. This involves selecting risk factors by
consensus from a group of factors associated with child maltreatment in the research
literature. However, as already noted, child maltreatment research lacks deﬁnitional
precision, delineation, and comparability across studies, making this approach similar to
combining apples with oranges.

In a comprehensive review of the literature, Pecora (1991) found that there are
four basic types of risk assessment systems used in CPS practice. The ﬁrst system, the
matrix approach, uses a table comprised of factors that are rated in terms of their severity
of risk to the child. English and Pecora (1994) state that “although the factors in these

matrices have some face “validity,” there is little empirical validation of the predictive

50

accuracy of various sets of items in any given matrix” (p. 454). The second system
involves identifying a small set of risk factors most predictive of child maltreatment.
With this approach, CPS workers are provided with risk factors that assist them in
identifying the likelihood of substantiation (English & Pecora, 1994). This method uses
risk indicators from substantiated cases of abuse — a method that is based on the more
severe cases and may not be predictive of the full spectrum of risk. The third system
involves the use of family assessment scales. This method uses behaviorally-anchored
scales to assess the levels of parent, child, family, and household functioning in order to
identify areas of abuse (English & Pecora, 1994). This approach to risk assessment is
focused on child and family functioning more so than identiﬁcation of risk factors. A
major beneﬁt of this method is that the family is rated at multiple points during the
casework process, and family strengths and resources are assessed in addition to risk.
There are no predictive validity studies to date on this system, although concurrent
validity has been demonstrated (English & Pecora, 1994). The fourth and ﬁnal system of
risk evaluation is termed the Child at Risk Field (CARP). This system, organized around
child, parent, and family characteristics, is the only method that uses an ecological
approach to identify risk factors. This system does not have proven predictive validity,
although preliminary studies of this method have indicated high inter-rater reliability and
practice improvement (English & Pecora, 1994). Although these systems serve to
structure the practice of risk assessment, predictive accuracy of these methods have yet to
be empirically documented. In addition, only one system reviewed incorporated a

multiple stage screening procedure, and none of the systems were theoretically-driven.

51

As previously noted, next to its practical application, risk assessment is also the
focus of substantial empirical literature. Extant risk assessment research will be critiqued
with a focus on study designs (retrospective vs. prospective), ﬁndings of predictive versus
concurrent validity, measurement sensitivity and speciﬁcity, and ecological
dimensionality of assessment instruments.

Retrospective versus Prospective Study Designs

In a comprehensive review of the literature, Rodwell and Chambers (1992)
summarized the state of the art of risk assessment for child abuse and neglect. They
began with 63 published works, and through a process of selection, they found that only
nineteen studies met their qualitative criteria for inclusion in their review. Notably, the
selected studies differed greatly in their choice of variables related to child abuse, with no
two studies considering the same variables. Other overall methodological limitations
they reported were the systematic exclusion of fathers from almost all studies, and the
conducting of most studies in public hospital settings that reﬂect a lower socioeconomic
strata than would be found in private hospitals or non-medical settings. Most studies also
lacked control procedures such as the use of comparison groups. In addition, potential
treatment effects undermined the results of many studies because data were gathered
simultaneously with the provision of services to abusing families (Rodwell & Chambers,
1992)

These researchers report that of the nineteen studies reviewed, seventeen were
retrospective, meaning that the samples consisted of individuals who had already been

identiﬁed as engaging in abuse. One of the two prospective studies reviewed by these

52

investigators was conducted by Altemeier, O’Connor, Vietze, Sandler, and Sherrod
(1984) who attempted to identify abuse risk prenatally among 1400 expectant mothers.
Using a structured interview format during the mother’s ﬁrst prenatal appointment, these
researchers predicted that 273 of the participants were high risk; subsequent abuse was
substantiated for 6% of these, versus 1% of the remaining mothers. Although there was a
high false positive rate (94%), the true incidence of maltreatment was almost certainly
higher because only severe physical abuse was considered, while milder forms of
maltreatment went unmeasured. Through the evaluation of child protection records
annually for 3 years following the prenatal interview, they found that accuracy of
prediction decreased over time with prediction only being valid for the ﬁrst 24 months
following the interview. Moreover, using regression analyses, these researchers identiﬁed
several statistically signiﬁcant (r > .44) predictors of abuse including subjective
impressions of the interviewers, residency transience, untruthfulness in the interview,
disturbed childhood nurturance, and unwanted pregnancy, most of which have been
identiﬁed in the retrospective literature as risk factors for abuse. Conversely, several risk
factors that have been identiﬁed in the retrospective literature (i.e., lack of social support,
social isolation, alcohol and drug abuse, excessive expectations of child development) did
not materialize in this investigation. Perhaps the open-ended format of their interview
and their reliance on subjective impressions of the interviewers account for some of these
differences. The researchers recognized the limitations of conducting structured
interviews and reliance on subsequent subjective clinical impressions as predictors of

abuse, stating that these are not practical strategies for predicting abuse. In an attempt to

53

extend the original study, several of these researchers conducted a second investigation to
determine whether a preventive intervention with mothers deemed to be high risk would
decrease the number of substantiated cases of child abuse and neglect (Brayden,
Altemeier, Dietrich, Tucker, Christensen, McLaughlin, & Sherrod, 1993). They
conducted a structured interview (Maternal History Interview -2) with 1154 women
prenatally, of which 314 were considered to be high risk (this criteria was met if the
woman scored in' the high-risk percentile of any sub-scale of the measure). High risk
women were then assigned to either an intervention group or a standard care comparison
group and outcomes were measured when the children were two years of age.
Substantive physical abuse reports were made on 5.1% of the families in the high risk
group, a number that was lower than the predictive percentage attained in the original
study. In terms of prevention, no effect was demonstrated for physical abuse by the
intervention program, and in fact more reports were substantiated for mothers in the
intervention group (9.2%) than the standard care comparison group (6.6%). The authors
suggest that bringing high risk mothers together for support groups may be a poor choice
for preventive intervention because it may have worsened parenting practices through
unintended social facilitation of abnormal parenting. Such groups may have also made
mothers more anxious by introducing new expectations of them that they felt they could
not meet. These authors conclude that prevention of child abuse appears to require “a
different, more intensive approach to psychosocial intervention than was offered...and
leaves questions about whether secondary prevention is a practical approach to

maltreatment prevention” (p. 516).

54

Other prospective investigations report diverse ﬁndings. Murphy, Orkow, and
Nicola (1985) found that 52% of a group of mothers screened to be high risk prenatally
were subsequently reported for abuse and/or neglect, compared to only 2% of the
comparison group mothers. These researchers argue that their screening instrument, the
Family Stress Checklist, was remarkably accurate (predicting 20 out of 25 of all families
identiﬁed as abusive/ neglectful) and provides evidence that prenatal prediction can be
employed to select individuals into prevention programs. However, these authors do not
use any control procedures which may weaken the conclusions of their study. A further
prospective study, conducted by Egeland, Breitenbucher, and Rosenberg (1980),
attempted to differentiate between mothers who mistreated their children and those who
provided adequate care. Mothers were interviewed in their third trimester of pregnancy
and then at three month intervals for one year. Although they report a 15% false positive
rate, they found highly signiﬁcant differences between life event scores of mothers
providing adequate care and those providing inadequate care, supporting their hypothesis
that environmental stress was a strong predictor of child maltreatment. They further
report that certain mother and infant characteristics differentiated high stress mothers who
did, and did not, mistreat their children; these included anxiety, aggression, defendence,
and succorance. Lastly, Milner, Gold, Ayoub, and Jacewitz (1984) conducted a
prospective investigation into the predictive validity of the Child Abuse Potential
Inventory. They administered this instrument to 200 parents who were at risk for
problems in parenting, but who had not engaged in child abuse or neglect at the time of

screening. Parents were then followed to determine if those with the most elevated abuse

55

scores would be those most frequently reported and conﬁrmed for abuse. They found a I
signiﬁcant correlation of .34 between abuse scores and subsequent conﬁrmed reports of
abuse and neglect (N=42). One hundred percent of the conﬁrmed abuse cases had scores
on the CAP above the cutoff for abuse potential, however, there was an 89.3% false
positive rate. A major weakness of this study, however, was the absence of control
subjects, producing a truncated range of relatively high CAP scores.
Predictive versus Concurrent Validity

Predictive validity can only be established using a prospective methodology. As
the previously reviewed prospective studies illustrated, predictive validity has established
a poor track record, with reported ranges from 5% to 52%. As stated above, Rodwell and
Chambers (1992) found that the majority of studies in their review assessed concurrent as
opposed to predictive validity. These researchers report an overall predictive accuracy
rate of 71% when all studies were combined. However, the average false positive rate
was over 50%, demonstrating that “on average, the [predictors] used in these studies will
falsely identify half of those it accuses of child abuse, misidentifying one child as abused
for every child correctly predicted to be abused” (Rodwell & Chambers, 1992, p. 169).
There was a substantial amount of variation among studies in terms of false positive rates,
with the lowest being 20%, while the highest was 95%. In terms of missed cases, the
studies reviewed had an overall false negative rate of 35%, which means that on average
they failed to identify one in three children who were actually being abused. The lowest
rate of missed cases was 16% while the highest rate was 63%. These researchers

attempted to identify which methodologies were associated with the worst predictive

56

efﬁciency rates and found that the variations in predictive efﬁciency could not be
explained by obvious features of the research designs such as sample size, method,
population, and instrumentation. They stated that “there are high and low overall
accuracy, false positives and missed cases rates among studies with large samples
(N=1400) and small samples (N=39)...and among studies using matched controls...some
version of random assignment...and prospective and retrospective designs” (p. 170).
From these ﬁndings, these researchers concluded that, based on existing empirical data,
there is no reason to believe that child abuse can be efﬁciently predicted, even
concurrently, much less in advance, without an alarmingly large proportion of mislabeling
of non-abusers or of missed cases (Rodwell & Chambers, 1992). They claim that
efﬁcient prediction is lacking for two primary reasons. First, technical and
methodological shortcomings in the research, such as errors in instrumentation, research
design, and sampling methods, have obscured our ability to accurately predict child
maltreatment. Second, due to the relatively low base rate of child abuse, efﬁcient
predictions cannot be made without extraordinarily accurate instruments. Speciﬁcally,
screening instruments for child abuse must be more than 99% accurate in order to be
acceptable (Light & Nagi, 1977).
Measurement Sensitivity and Speciﬁcity

Sensitivity refers to a measurement instrument’s ability to correctly identify
potentially abusive families, while speciﬁcity refers to its ability to correctly identify non-
abusing families (Caldwell et al., 1988). Caldwell and colleagues (1988) provide an

illustration of these concepts based on Starr’s (1979) hypothetical example that it would

57

be possible to design a risk assessment measure with a sensitivity rate of 85% and a
speciﬁcity rate of 82.5%. They explain that

“in a sample of 100,000 with an actual abuse rate of 1%, this hypothetical

instrument would identify 18,175 families at risk (850 valid positives and 17,325

false positives for a selection rate of approximately 18%)...If preventive services

were delivered to people identiﬁed as high risk, over 95% of the families would

be receiving services unnecessarily” (p. 618).

These authors further illustrate that program efﬁciency increases slowly even with
dramatic increases in sensitivity and speciﬁcity (Caldwell et al., 1988). Program
efﬁciency depends primarily on the speciﬁcity coefﬁcient, but an assessment instrument
with as high a sensitivity and speciﬁcity coefﬁcient as .99 will still misclassify 50% of the
population as high-risk.

The sensitivity and speciﬁcity rates of actual — as opposed to hypothetical — risk
assessment instruments often go unreported. The ranges that have been reported in the
literature vary signiﬁcantly. Agathonas-Georgopoulou and Browne (1997) report that
their checklist of risk factors achieved 93% sensitivity and 93.5% speciﬁcity. Murphy et
a1. (1985) report that the Family Stress Checklist used in their study reached 80%
sensitivity and 89.4% speciﬁcity. Brayden et al. (1993) report that their measure, the
Maternal History Interview-2, reached a sensitivity rate of 55.6%.

Ecological Dimensionality of Risk Assessment Strategies
In their review of the literature, Rodwell and Chambers (1992) found that

investigators generally measure three conceptual categories of risk: attitudinal/personality

58

factors, interactions between parent and child characteristics, and environmental
variables. As discussed above, they found very poor overall predictive efﬁciency rates
among the studies they reviewed. They further found that none of the predictor variables
employed by the various studies stood out as more efﬁcient predictors than any others.
They reported that regardless of what instruments or predictor variables were used,
overall accuracy rates, false positives, and missed cases varied substantially.

The most notable deﬁciency in the literature to date are investigations that
measure different categories of risk simultaneously. A survey of this literature subsequent
to the Rodwell and Chambers (1992) review, reveals that most unidimensional
investigations look primarily at characteristics of the mother-infant dyad such as
attachment (Morton & Browne, 1998), maternal psychosocial history (Brayden et al.,
1993; Caliso & Milner, 1994), maternal stress (Agathonas-Georgopoulou & Browne,
1997), maternal coping strategies (Cantos, Neale, O’Leary, & Gaines, 1997), parental
attitudes toward child care (Huxley & Warner, 1993), dissociative processes (Egeland &
Susan-Stillman, 1996), maternal age (Buchholz & Korn-Bursztyn, 1993; Connelly &
Straus, 1992), family environment (Justice & Calvert, 1990), and child antisocial
behavior (Kolko, Kazdin, Thomas, & Day, 1993). When multiple levels are identiﬁed,
they almost exclusively look at social support variables in addition to maternal and child
characteristics (Burell, Thompson, & Sexton, 1994; Dukewich, Borkowski, & Whitman,
1996). Lastly, there is a growing body of literature that examines system level variables
in relation to child maltreatment. For instance, Krishnan and Morrison (1995) found that

population change (migration, fertility, and mortality), unemployment rate, and percent

59

Native Canadian residents positively predicted child maltreatment rates. Drake and
Pandy (1996) found that neighborhood poverty was more strongly associated with child
neglect than child physical or sexual abuse. Gillham, Tanner, Cheyne, Freeman, Rooney,
and Lambie (1998) found that of the ecological variables examined in their study (child
population, social worker ratio, unemployment rates, single parent density), male
unemployment rates alone accounted for two-thirds of the variance in total abuse and
neglect rates in their sample.

While all of these ﬁndings are intriguing, this body of literature is characterized
by a piecemeal approach. The plethora of variables that have been studied in relation to
child maltreatment have yet to be integrated into a comprehensive ecological assessment
of risk. This strategy may hold promise for improved predictive efﬁciency.

Summary of Practical Issues Related to the Assessment of Risk for Child Maltreatment

The literature demonstrates that efﬁcient prediction of who will mistreat their
children is impossible for any practical professional purpose (Rodwell & Chambers,
1992). According to the literature reviewed here, instrumentation has not yet been
perfected to the degree that child abuse can be predicted without simultaneously
overlooking or incorrectly classifying a signiﬁcant number of individuals. No set of
variables, or combination of variables, does a good enough job at identifying those at risk
to accurately target people for prevention programs. Hence, the state of the art of risk
assessment is that all existing methods will intervene in many more homes than are
necessary at best, and at worst run the risk of having unwanted iotragenic effects.

Wren in Rome...

60

Given the aforementioned shortcomings of risk assessment, it can be argued that
this practice has no merit and should be abandoned in favor of universal prevention.
Alternatively, it can be argued that this practice has yet to be applied properly to the
problem of child abuse and neglect. Although extant strategies have failed to empirically
document their usefulness, the utility of theoretically-based, ecologically sound
assessment tools have yet to be evaluated. The plethora of variables, at numerous
ecological levels, that have been studied in relation to child maltreatment have yet to be
integrated into a comprehensive ecological assessment tool. The purpose of the current
investigation is to test whether such screening yields more accurate prediction than
unidimensional means. This type of ecological screening may prove to be a promising
advance in our efforts to predict accurately who is at risk for dysfunctional parenting
practices. Whereas the risk assessment paradigm has been successfully used in the
medical sciences, it has yet to be determined whether it is an appropriate model for
prevention in the behavioral sciences. The current research attempts to work within the
existing paradigm, but apply a more theoretically and ecologically sound approach to this

practice.

61

CONSTRUCTION OF THE ECOLOGICAL PROSPECTIVE MODEL
OF RISK TO BE TESTED

The ecological levels to be tested in relation to child abuse potential include
person-level characteristics, and neighborhood-level descriptors. Each of these variables
were measured at the birth of the child (Time 0). Parenting stress was measured when the
child was six months of age (Time 1). Direct and mediated effects of the ecological risk
indicators and stress are tested in relation to two indicators of child maltreatment after the
child’s ﬁrst birthday (Time 2): child abuse potential, and child developmental delays.
Figure 1 illustrates this theoretical model, while Figure 2 provides the measurement
properties of this model.
Rationale for paths in model

The literature supporting the theoretical underpinnings of the current investigation
have been discussed in detail throughout this analysis. Here, it is important to summarize
this literature, thereby explicating the construction of the speciﬁc paths in this model.

Path 1: Person level risk == child abuse potential. There is a vast literature
documenting the connection between person-level risk factors and child abuse. As
discussed previously, this relationship has been empirically well established. In
summary, maternal characteristics including age (Connelly & Strauss, 1992), parenting
style (Baranowsky et al., 1990), history of abuse (Caliso & Milner, 1992), and personality
factors (self-esteem, locus of control, impulse control, coping style, and affect quality)
(Pianta et al., 1989) all play a role in abusive parenting practices.

ﬁlth 2: Person level risk .4 stress. According to the research program of Egelend

62

and colleagues (1981; 1987; 1988; 1996), stress at the intra- and inter-personal levels
plays a signiﬁcant role in the expression of abusive parenting practices. These
researchers have demonstrated that individuals considered to be highly stressed (as
measured by number of stressﬁrl life events) are more likely to abuse their children if
their personality styles are characterized by aggressiveness, anxiety, and defensiveness.
Subjective perception of stress, among individuals matched on mean levels of stress, has
also been linked with child maltreatment (Justice & Calvert, 1990). Based on this
literature, the current model assumes that person level risk factors will affect parenting
stress.

Path 3: Stress =~ child abuse potential. Parenting stress has been described in the
literature as contributing to dysfunctional parenting practices (of, Abidin, 1983). Stress
is conceptualized in this model as the magnitude of strain within the parent-child system.
In a review of the literature, Cicchetti and Lynch (1993) found that maltreating parents
are less satisﬁed with their children, perceive child rearing as difﬁcult and unenjoyable,
use controlling disciplinary techniques, and do not encourage autonomy in their children.
Such qualities of the parent-child dyad are conceptualized in the current model as
parenting stress and are assumed to be directly related to abuse potential.

Path 4: Communig level risk == person level risk. Community risk factors, as
described by Garbarino and Sherman (1980), are conceptualized as independent from
conventional socioeconomic variables, and are associated with aspects of family stresses
and supports. In their research, these authors found that high-risk neighborhoods

compound internal familial problems, increasing their potential for child abuse, whereas

63

low-risk neighborhoods help support individuals and families by overcoming their
internal weaknesses thereby decreasing their likelihood of maltreatment. Hence, in the
current model, it is posited that neighborhood or community level strain will affect
person-level risk characteristics.

Path 5: Communig level risk =4 child abuse potential. For the past two decades,
Garbarino and Sherman (cf., 1980a & 1980b) have been developing a research program
that elucidates the community variables that may contribute to child maltreatment. This
link in the model represents a recent burgeoning of work in the ﬁeld of risk assessment,
as well as provides an exclusively ecological foci. Zuravin (1989) reviewed this
literature and reports that there is general consensus among studies that incidence rates of
child maltreatment vary as a function of community characteristics.

Path 6: Community level risk = stress. Community level risk, conceptualized as
social strain on the neighborhood, is hypothesized to affect parenting stress in the current
model. For instance, people living in socially strained conditions (e.g., below the poverty
line, female-headed households, working mothers) tend to be concerned with meeting
their own basic needs and are less able to participate in natural helping networks.
Garbarino and Sherman (1980) have demonstrated that community stressors, exacerbated

by a lack of social resources, are related to parenting stress and maltreatment rates.

64

Figure 1: Theoretical Model

Time 0 Time 1 Time 2

ﬂ

' person-level >
nsk

at-

a.“
,r

,_.—— —-.—_.__,

A

< ,,;.>—~-—e<

 

.-

,— 4V

 

,.,-/~---/
(community leve - ”
risk

V/’"

65

 

 

 

 

 

 

 

 

 

 

 

Figure 2: Measurement Model

 

Time 0 Time 1
Sum of Intra-
and Inter-
Personal Risk
(Risk Screening Form)
1
2
parent
domain 1
0
parent
d0ma'” 2 Parenting
Stress (P l)
child
domain 1
4
child
domain 2
6
5
0.
Community-level
Risk
(1990 Census Data)
per cent percent ’
poverty ""9 households 9

 

 

 

 

Time 2

 

parental
distress

 

 

 

 

 

problems
with family

 

 

Parenting
Outcome

(CAP and Denver)

 

 

problems
with others

 

 

parental
rigidity

 

 

 

 

 

unhappiness

 

 

 

problems with
self and child

 

 

 

develo mental
de ays

 

 

 

developmental

 

cautions

 

 

 

 

developmental
advances

 

 

 

neighbor-
hood
transiency

 

 

66

 

HYPOTHESES

The hypotheses of this study are stated as follows and speciﬁed by the numbered
paths in the model presented in Figure 2.
l. Person-level variables (maternal age, maternal physical health, maternal mental
illness, maternal history of abuse, maternal substance abuse, unwanted pregnancy, child
physical health, child temperament, child accidents, family crisis, family violence,
inadequate social support, 4 or more preschool children at home, mother-child
separations), measured at Time 0, will have a direct effect on outcome (child abuse
potential and developmental delays), measured at Time 2 (path 1).
2. Person-level variables, measured at Time 0, will be mediated by stress, measured at
Time 1, in relation to outcome, measured at Time 2 (paths 2 and 3).
3. Community-level variables (neighborhood percentage of non-white residents,
neighborhood income level, neighborhood percentage of female headed households,
neighborhood percentage of working mothers) will have a direct effect on person-level
variables, measured at Time 0 (path 4).
4. Community-level variables, measured at Time 0, will have a direct effect on outcome,
measured at Time 2 (path 5).
5. Community-level variables, measured at Time 0, will be mediated by stress, measured
at Time 1, in relation to outcome, measured at Time 2 (paths 6 and 3)
6. This model as a whole, will account for a signiﬁcant proportion of the variance in
outcome (indicators of potential child maltreatment). This model will provide empirical

validation for the predictive accuracy of risk factors at multiple ecological levels. As

67

such, this model will contribute to our current knowledge base by: a) demonstrating
predictive validity through the use of a prospective design, b) documenting empirical
relationships between risk factors and potential to abuse, c) providing empirical evidence
of the mediating role of stress in relation to child abuse potential, and (1) providing
empirical evidence of the interrelationships among ecological levels of risk in relation to

outcome.

68

METHODS

Participants

Data collected for the Kent County Healthy Start Evaluation Project were used for
this study. The participants include mothers and infants who were screened into the
Home Visitor Component of the evaluation, and does not include mothers judged to be at
very low or no risk. The mean number of risk factors for each participant at screening
was 3, with a range of 0 to 9. Members from both the intervention and control groups are
represented in the current analyses because no effects were found due to the intervention
(See Appendix A for summary of intervention effects). No initial group differences were
found on number and type of risk factors from the screening instrument. Data were
collected at three time periods: Time 0 was collected at the birth of the baby (while
mother and child were still in the hospital), Time 1 was collected during the child’s ﬁrst
year (mean infant age = 4.3 months, range 1 to 15 months), and Time 2 was collected
during the child’s second year (mean infant age = 17.8 months, range 12 to 30 months).
A subset of the total study sample of 277 were used in these analyses; namely, only those
participants who had complete Time 0, Time 1 and Time 2 data were included (N=125).
The loss of subjects (from 277 to 125) was not due to attrition, but to the lack of complete
Time 0 data. Speciﬁcally, only 125 subjects from the entire sample (N =277) had
available screening data. Although there were no signiﬁcant differences between the
current and the excluded subjects on Time 1 demographic variables, or Time 2 outcome
measures (CAP scores and developmental measure), signiﬁcant group differences existed

on the Time 1 stress indicator (See Table 1), suggesting that excluded subjects had more

69

overall stress related to parenting than those subjects included in the analyses. The reason
for this difference stems from the way in which subjects were recruited for study
participation and will be discussed in detail as a procedural issue.

Table 1: Group differences between research sample (N =125) and excluded subjects

(N=152)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Group
Variable Included in sample Excluded from sample
T1 PSI Parent Domain
mean (sd) 128.4 (26.0) 137.4 (26.1)
F 748*
T1 PSI Total Score
mean (sd) 231.6 (43.2) 246.7 (40.3)
F 8.15*
"' p < .05

The mean age of the mothers in this sample was 24 years old (s.d. = 5.7), with a
mean educational level of 11.3 years (s.d. = 2.2). The majority of the women in this study
were Caucasian (47%), while 18% were African-American, 12% were Hispanic, 3% were
either American Indian or Asian, 7% identiﬁed themselves as belonging to other ethnic
minority backgrounds, and 13% chose not to identify their racial background. The
majority of the mothers in this sample were single (56%), while 30% were married, 2%
were either divorced or separated, and 15% did not provide their marital status. Most of
the women were unemployed (53%), and received Medicaid (58%). F ifry percent lived
on an income under $15,000 per year (only 5% reported an annual income over $50,000),

and they had an average of 2 children (s.d. = 1.1) at Time 1.

70

Measures

Independent Variables

In an effort to provide a fuller picture of parental functioning based on ecological
theory, the current model examines the relationship between risk at distinct ecological
levels and child abuse potential. The ecological contexts being examined are person-
level risk, community-level risk, and stress related to parenting. The variables at each of
these levels came from three sources, the KCHS Risk Screening Instrument, the Parenting
Stress Index (PSI), and 1990 US. Census Data.
The KCHS Risk Screening Instrument

This measure consists of a checklist of 46 “risk factors.” This measure was used
by KCHS screeners either at Butterworth Hospital in Grand Rapids, Michigan or by Kent
County Health Department public health nurses in order to assign a risk status to
participants. In the original study, this checklist was used as an aggregate measure — that
is, participants were assigned a number of risk factors which then determined their group
assignment, “at-risk” or “low-risk.” Subjects who endorsed two or more risk factors were
considered at-risk for the purposes of study group assignment, and were offered the
home-visitor intervention. The aggregation of multiple concurrent risks can be
conceptualized as an individual’s risk load, which has been positively associated with
poor outcomes (Sarneroff & Seifer, 1990). However, this conceptualization of risk does
not provide information about the differential contribution of risk at distinct ecological
levels.

In keeping with ecological theory, an attempt was ﬁrst made to divide this

71

measure into intra-personal and interpersonal level risk factors so that the differential
contribution of these levels of risk could be measured in relation to outcome. This was
accomplished by obtaining inter-rater agreement from 20 independent raters who
assigned each risk variable to an ecological tier. Raters, all of whom were advanced level
graduate students in a Clinical Psychology Ph.D. program, were ﬁrst asked to read brief
descriptions of the ecological levels being examined in this study. They were then given
the list of the 46 risk factors from the screening instrument and they were asked to assign
each item to an ecological context (See Appendix B for the narrative description of each
ecological context and the total list of risk factors). Percentages of agreement were then
calculated for each risk factor. Of the 46 risk factors, 13 were discarded due to lack of
agreement into which ecological context they “ﬁt.” The cutoff point that was chosen for
exclusion was a rate of less than 75% agreement. Average inter-rater agreement for the
items retained was 90%. The risk factors falling into the intra-personal and interpersonal
categories were the only variables from this measure used in the current model. Variables
from this measure that were judged to fall within the community and cultural contexts
were discarded from the analyses because these ecological levels are being measured with
US. Census tract data. In total, there were 15 risk indicators (33%) discarded from this
measure. These items included lead poisoning, frequent atypical accidents, immunization
concerns, unplanned pregnancy, mother was less than 18 years old with inadequate
support, adolescent mother, single parent, no prenatal care, mother smokes, language
barrier, no medical insurance, inadequate health care, ﬁnancial need, unemployed, and

homeless. An illustration of the frequency of all 46 risk factors endorsed by participants

72

is provided in Table 2 in order to examine the risk trends of the sample upon screening
into the program, before risk items were discarded due to poor inter-rater agreement
and/or assignment to the community/cultural ecological tier.

Table 2: Frequency table of all risk factors from KCHS Risk Screen (discarded items
marked with an asterisk)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Risk item Percent endorsement
Frequent atypical accidents“ 0%
Child bronchiopulmonary dysplasia 0%
Child excessive irritability 0%
Infant positive drug screen 0%
Intrauterine growth retardation 0%
Infant lead poisoning“ 0%
Maternal immunization concerns“ 0%
Severe perinatal complications 0%
Child asphyxia 1%
Maternal severe chronic physical illness 1%
Homelessness* 1%
Congenital abnormalities ‘ 1%
Child organic failure to thrive 1%
Infant small for gestational age 1%
Mother/child separations 1%
Other risk 1%
Severe prenatal complications 2%
Family violence 2%
Child chronic ear infections 2%
Child meconium aspiration 2%
Other concerns 3%

 

 

 

 

73

 

Acute family crisis

4%

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Maternal language barrier* 5%
No prenatal care“ 5%
Four or more preschoolers in the home 5%
Inadequate health care" 5%
Maternal poor nutrition 6%
Maternal medical or genetic problems 6%
Infant feeding problems 6%
Child jaundice 7%
Premature or low birth weight infant 7%
No medical insurance" 8%
Maternal substance abuse 8%
Isolation and lack of support 10%
Maternal mental illness/low cognitive functioning 12%
Maternal history of abuse or neglect 13%
Maternal history of reproductive problems 13%
Mother less than 18 with inadequate support“ 13%
Maternal preexisting medical condition 13%
Parenting concerns 15%
Adolescent mother" 15%
Maternal overweight/underweight 25%
Maternal unemployment“ 34%
Maternal smoking“ 34%
Unplanned pregnancy* 43%
Poor/ﬁnancial need“ 53%
Single parent" 58%

 

74

 

The ﬁrst ecological tier extracted from the KCHS Screening Instrument consists
of maternal and infant variables. Maternal variables include physical health problems
(history of medical or genetic problems, preexisting medical condition, over- or under-
weight, prenatal complications, perinatal complications, poor nutrition, history of
reproductive problems, severe chronic physical illness), mental illness, history of having
been abused as a child, and substance abuse. Child variables include physical health
problems (asphyxia, bronchopulmonary dysphagia, congenital anomalies, organic failure
to thrive, feeding problems, infra-uterine growth retardation, jaundice, low birth weight or
prematurity, meconium aspiration, frequent ear infections, positive drug screen at birth,
and small size for gestational age), and temperament (excessive irritability).

The second ecological tier that was constructed from the KCHS Screening
Instrument involves interpersonal and familial contextual variables. This level is
concerned with characteristics of parent-child interactions, and the familial context. The
variables that exist at this level of analysis include acute family crisis, family violence,
the presence of four or more preschool children in the home, and mother-child
separations.

Although this multidimensional conceptualization of risk is theoretically sound, it
posed several methodological problems in the current model. First, there was virtually no
variance in the distribution of scores for the intra-personal and inter-personal levels (see
Table 3). Because most subjects endorsed so few risk factors, this measure did not lend
itself to dimensionality when scored this way. In order to address this problem, a

principle components factor analysis was conducted on this measure, yielding a four

75

factor solution.

Table 3: Frequency table of intra- and inter- personal level risk indicators

 

 

 

 

 

 

 

 

 

 

 

 

Number of intra-personal level Frequency (Percentage)
variables endorsed
0 42 (33%)
1 39 (31%)
2 25 (20%)
3 11 (9%)
4 6 (5%)
5 1 (1%)
6 1 (1%)
Number of interpersonal level
variables endorsed
0 111 (89%)
1 13 (10%)
2 1 (1%)

 

 

 

 

The resulting factors can be characterized as 1) poor, single adolescent mother, 2) isolated
non-adolescent mother with parenting concerns, 3) parent with physical problems, and 4)
mentally ill parent. Upon further analysis, however, it became apparent that this factor
solution was inappropriate for this measure because the four factors were uncorrelated
with one another and did not hang together as indicators of a single latent “risk”
construct. In this case, the expectation that the contributors of risk would be inter-
correlated was unjustiﬁed because this risk measure is not a scale made up of items

assumed to measure a single underlying “risk” construct. Instead, the risk factors all

76

contribute to risk in the same way, for example, one’s salary, yard sale proﬁts, and
birthday gifts might contribute to one’s net ﬁnancial worth, but are not necessarily
indicative of it. Moreover, Chin (1998) explains that the indicators used to measure a
latent variable must be reﬂective in nature, such that the correlations among indicators for
the particular latent variable are caused by that latent variable. For example, potential
reﬂective measures of metal inebriation (the latent variable) are blood alcohol, driving
ability, MRI, and performance on mental calculations. If truly reﬂective, an improvement
in the blood alcohol level measure would also imply an improvement in the MRI activity
and other measures since they are all meant to tap into the same concept (Chin, 1998).
Conversely, an example of formative measures of inebriation would be the amount of
beer, wine and hard liquor consumed, where an increase in beer consumption does not
imply similar increases in wine and hard liquor consumption (Chin, 1998). Thus,
formative indicators are measures that form or cause a change in a latent variable, but
need not be correlated nor have high internal consistency (Bollen & Lennox, 1991). In
the case of the current risk measure, the individual risk items are formative indicators
(i.e., a decrease in the mother’s physical health score does not mean there will be a
corresponding decrease in the child’s physical health) and will be used as an observed
variable in the model representing cumulative person-level risk. Only those risk factors
that met criteria for inclusion in the ecological analysis were used (i.e., the 33 items left
after inter-rater agreement was achieved) in order to maintain the theoretical assertion that
these risk factors measure person-level risk, albeit intra- or inter- personal. Each person

was given a score of 0 (no risk factors), 1 (only one risk factor present), or 2 (two or more

77

risk factors present) which resulted in a nearly equal distribution of scores (3 0%, 31%,
and 39%, respectively). Women could receive a score of 0 risk factors, even though the
original sample was deemed at-risk, because 33% of the original risk items were
discarded due to poor inter-rater agreement and/or assignment to the community/cultural
ecological level. Hence, the division of the risk screen into ecological tiers resulted in
one-third of the sample losing their original risk status.
Parenting Stress Index - Long Form

The Parenting Stress Index (Abidin, 1990) assesses the amount of stress related to
parenting as perceived by the respondent (See Appendix C). It consists of 101 items that
make up two primary domains, overall perceptions of the child and overall perceptions of
parenting. Each domain contains several subscales. The child domain subscales include:
child distractability (Cronbach alpha =.50), child’s reinforcement of parent (Cronbach
alpha =.46), child mood (Cronbach alpha =.65), parent’s acceptance of child (Cronbach
alpha =.83), child’s adaptability (Cronbach alpha =.65), and demandingness of child
(Cronbach alpha =.52). The parent domain subscales include: parent attachment to child
(Cronbach alpha =.44), restrictiveness of the parental role (Cronbach alpha =.80),
depression related to parenting (Cronbach alpha =.75), parental feelings of competence
(Cronbach alpha =.34), spousal relationship problems related to parenting (Cronbach
alpha =.68), parental health problems (Cronbach alpha =.65), and parental social isolation
(Cronbach alpha =.65). A major criticism of this instrument, however, is that it measures
too broad a spectrum of “stressors” related to the parenting role. The numerous sub-

scales represent such a diverse array of factors associated with parenting that are

78

construed by the author as “parenting stress,” that the face validity of this measure is
called into question.

Previous research on this measure has found a high degree of internal consistency,
with alpha coefﬁcients of .95 for the Total Stress Index, .93 and .89 for the Parent and
Child domains, respectively, and subscale coefﬁcients ranging from .55 to .80 (Lloyd &
Abidin, 1985). Burke (1978) reported test—retest reliabilities for the Parent and Child
domains of .71 and .82, respectively, after an interval of 3 weeks, while Hamilton (1980)
found that after one year, test-retest reliabilities were .70 and .55. The current analysis
employs the parent domain and child domain scale scores to measure stress related to
parenting. However, for the purposes of the current structural equation modeling
analyses, in which latent variables must have more than two indicators in order to be
identiﬁed in the model, each scale was broken in half to provide four indicators: parent
domain 1, parent domain 2, child domain 1, and child domain 2.

1990 US. Census Data

Garbarino and Sherman found that ﬁve indicators of community social conditions
accounted for eighty percent of the variance in child maltreatment among the
neighborhoods they studied. These ﬁve categories consisted of l) the percent of
households with incomes less than $8,000 a year (in 1975 dollars), which was considered
to be economically “struggling,” 2) the percent of households with incomes over $15,000
a year, which was considered to be economically “comfortable,” 3) the percent of female
headed households, 4) the percent of married mothers with children under the age of

eighteen in the labor force, and 5) the percent of residents who have resided in their

79

neighborhood for less than one year. These variables were found to represent
neighborhood social resources because they all put drains on the neighborhood and
because people in these situations tend to be concerned with meeting their own needs and
unable to participate in natural helping networks. Similarly, neighborhood variables
representing community risk indicators are utilized in the current model. To this end,
each participant was assigned a census tract number according to where their Time 0
address was located on the 1990 US. census tract map. Because Time 0 data was
collected in 1994, 1990 census data was deemed most appropriate for these analyses.
Neighborhood characteristics were then supplied for each census tract in terms of the
community indices outlined by Garbarino and Sherman (1980). Participants were
assigned scores that reﬂect their neighborhood’s income level (above or below the
poverty level in 1990 US. dollars), their neighborhood’s percentage of female headed
households, their neighborhood’s percentage of working mothers, and neighborhood
transiency rate as indicated by the percentage of households vacated within the past year.
An additional neighborhood characteristic that was evaluated was the percentage of non-
white persons residing in the neighborhood. This was done to assess the effects of
neighborhood ethno-cultural makeup on parental functioning.
W

Risk and outcome variables in the current model are conceptualized as points
along a continuum of parental functioning. Hence, the dependent variables in the current
model consist of indicators of parent and child functioning rather than substantiated cases

of child maltreatment. It has been argued throughout this analysis that the term “child

80

abuse” is a term that is arbitrarily deﬁned. Hence, attitudinal and behavioral measures of
parenting quality are measured in relation to risk factors.
Child Abuse Potential Inventory (CAP)

Abusive practices are conceptualized in the current model within the overall
context of parenting which is measured by the Child Abuse Potential Inventory (Milner,
1986) (See Appendix D). Although an indirect indicator of maltreatment, this measure
taps elements of parenting behavior and attitudes empirically linked to maltreatment risk
(Milner, 1986). Administered at Time 2, this is a 160-item self-report measure that is
answered in an agree/disagree forced choice format. This measure contains a 77-item
physical child abuse scale (Cronbach alpha = .90) with six subscales: Distress (Cronbach
alpha = .90), Rigidity (Cronbach alpha = .80), Problems from Others (Cronbach alpha =
.5 5), Problems with Family (Cronbach alpha = .51), Unhappiness (Cronbach alpha = .50),
and Problems with Child and Self (Cronbach alpha = .32). The CAP also contains three
validity scales (lie scale, random response scale, and inconsistency scale) and two
subscales that measure ego strength and loneliness, which will not be used in the current
analyses. The current analyses use the six CAP subscales to measure potentially abusive
attitudes: parental distress (items include “I am often angry inside,” “I often feel
rejected”), parental rigidity (“Little boys should never learn sissy games,” “Children
should never disobey”), problems with family (“My family has many problems,” “My
family ﬁghts a lot”), unhappiness (“I do not laugh very much,” “I do not have any friends
in my neighborhood”), problems with child and self (“I have a child who is bad,” “I have

a physical handicap”), and problems with others (“You cannot depend on others,”

81

“People have caused me a lot of pain”).

Milner (1986) reports reliability coefﬁcients for internal consistency for the CAP
that range from .75 to .94. This researcher also reports that test—retest reliabilities of .91,
.90, .83, and .75 for the general population were found one day, one week, one month,
and three months after initial administration, respectively. Additionally, in terms of
predictive validity, Milner (1994) has found a signiﬁcant relationship between CAP abuse
scores and subsequent child maltreatment, as measured by substantiated child abuse
reports among parents deemed at-risk by this measure.
Denver Developmental Screening Test 11

The Denver 11 is a 1990 revision of the Denver Developmental Screening Test
developed by Frankenburg and Dodds (1967). Age norms at which 25%, 50%, 75%, and
90% of the sample performed each item were updated from the 1967 norms of the
original test. This behavioral observation measure, administered at Time 2, is a
developmental screening test to measure delayed development in children between the
ages of 2 weeks and 6.4 years. It assesses achievement in four functional areas of
development: gross motor, ﬁne motor, language and personal/social. Scoring is based on
comparison of the child’s performance against population norms for attainment of various
developmental tasks. Based on age-speciﬁc developmental norms, scores are derived by
determining whether the child is advanced (can perform skills that the majority of
children that age cannot), cautioned (falls below the average performance for that age
level), or delayed (cannot perform a speciﬁc behavior that the majority of children can

perform at that age).

82

This measure has been shown to have adequate validity and reliability.
Frankenburg, Camp, and Van Natta (1971) compared the performance of 236 children
evaluated with the Denver to performance on several criterion tests (Stanford-Binet,
Revised Yale Developmental Scale, Cattell Infant Intelligence Scale, and Revised Bayley
Scale of Infant Tests) and found correlations ranging from .86 to .97. Scoring the Denver
as advanced, cautioned, and delayed also corresponded highly with IQs or developmental
quotients obtained from the criterion tests. In another study, these same researchers
(1971) evaluated tester-other agreement and test-retest stability of this measure with 76
and 157 children, respectively. They report coefﬁcients ranging from .66 to .93 with no
age trends displayed.

This measure is scored as a continuous variable, with each subject receiving a
score for the total number of advances, cautions, and delays achieved across the four
developmental domains.

Procedure

KCHS participants entered the program either through screening at the maternity
ward of Butterworth Hospital in Grand Rapids, Michigan or through referral by a Kent
County Health Department public health nurse. Families who were deemed to have two
or more risk factors present were offered admission into the home visitor program or the
control group. Goals of the KCHS Home Visitor Program were to promote positive
parent-child relationships, increase awareness of community resources, and foster child
physical and mental health and school readiness. Random group assignment was

attempted by alternating the days of the week that each group was screened for.

83

However, upon analysis of T1 data, it was discovered that the two groups differed
signiﬁcantly on several dimensions, which suggests that random group assignment was
not achieved (See Appendix E for summary of intervention-control group differences at
Time 1). Failure of random assignment was attributable to changes in the referral system
that were introduced midway through the investigation by KCHS administrative staff.
Originally, families could only enter KCHS through “front door referrals”thch meant
they had to give birth at Butterworth Hospital and be formally screened by a trained
KCHS screener, who then randomly assigned the participants into the intervention and
control groups. Over time, “side door referrals” occurred (referrals made by other
agencies), and these referrals were made only into the intervention group. This resulted
in unequal groups, with the intervention group exhibiting signiﬁcantly more distress than
the control group. This also contributed to the differences found between subjects
included in these analyses and those excluded due to lack of screening data. Namely,
only those subjects with “front door referrals” had available screening data because
screening forms from later “side door referrals” were not made available to the
researchers. For the purposes of the present investigation, initial group differences are
considered statistically irrelevant because the groups are being combined and the
intervention is being ignored because it had no signiﬁcant effect on parenting outcomes.
Moreover, these differences introduce greater initial variance into the total group which is
a strategy that counteracts the restrictiveness of the sample (i.e., participants deemed to
have no risk-indicators were excluded from the sample).

Women who participated in the home visitor program were assigned a trained

84

paraprofessional Family Support Worker (FSW) who collected data upon the family’s
entry into the program (T1) and at yearly intervals for three years. Intervention group
families were paid $10 for their completion of each data packet after Tl. Data from the
control group was collected by a paid social worker who met with the family after the
birth of the baby, and at yearly intervals for the length of the study period (three years).

Each control group family was reimbursed $25 for each packet that they completed.

85

RESULTS

Analyses

The method of data analysis used in this study was structural equation modeling
with latent and observed variables. The statistical program AMOS 4.0 (Arbuckle, 1999)
was used to perform the structural modeling analyses. Structural Equation Modeling
(SEM) is a conﬁrmatory approach to data analysis (Byme, 1994), which is highly
appropriate in the present context. A primary advantage of using SEM is that it can test
the measurement model and the path model simultaneously. Chi-square is the most
common method of evaluating ﬁt, however, this ﬁt index is highly sensitive to sample
size and the signiﬁcance test can be misleading. Speciﬁcally, when evaluating the chi-
square statistic, non-signiﬁcance indicates that the actual observed covariance matrix is
not considerably different from the estimated covariance matrix. Thus, a low chi-square
value that is non-signiﬁcant represents a good ﬁt. In the current analyses, model ﬁt was
evaluated using several ﬁt indices considered measures of practical goodness of ﬁt, even
when signiﬁcant chi-square values are obtained (Bentley, 1989). First, the Comparative
Fit Index (CPI) is a noncentrality parameter-based index that has an advantage over other
ﬁt indices in that it avoids the underestimation of ﬁt in small samples (Bentler, 1990). A
CFI value of over 0.90 is desirable and indicates an acceptable ﬁt of the model to the data
(Bentler, 1992). The Tucker-Lewis index (TLI), also known as the non-normed ﬁt index
(NNFI), compares a proposed model’s ﬁt to a null model, as well as measures parsimony
by assessing the degrees of freedom from the proposed model to the degrees of freedom

of the null model (Garver & Mentzer, 1999). The TLI is also resilient to variations in

86

sample size. An acceptable threshold for this index is .90 or greater. Another ﬁt index,
the Bentler-Bonett Normed Fit Index (N F 1), measures the discrepancy between the
independence model and the saturated model, with values over .90 considered acceptable.
Bentler and Bonett (1980) assert that "since the scale of the ﬁt indices is not necessarily
easy to interpret (e.g., the indices are not squared multiple correlations), experience will
be required to establish values of the indices that are associated with various degrees of
meaningfulness of results. In our experience, models with overall ﬁt indices of less than
.90 can usually be improved substantially” (p. 600). Finally, the root mean square error
of approximation (RMSEA) is an index that does not penalize for model complexity and
will tend to favor models with many parameters. According to Browne and Cudeck
(1993), a RMSEA value of .05 or less indicates a close ﬁt of the model in relation to the
degrees of freedom, while a model with an RMSEA greater than 0.1 should not be
accepted.

In addition to ﬁt indices, it is important to pay attention to the predictiveness of
the model. According to Chin (1998), factor loadings should be at least .60 and ideally
.70 or above to indicate that each measure is accounting for 50 percent or more of the
variance in the underlying latent variable. Standardized paths should be at least .20 and
ideally above .30 in order to be considered meaningful. A path weight of .10 only
represents a 1% explanation of variance (Chin, 1998) which is hardly theoretically
meaningful. As Meehl (1990) points out, paths and loadings must be of substantial
strength to be considered theoretically interesting because “everything correlates to some

extent with everything else...[due to] some complex unknown network of genetic and

87

environmental factors” (p. 209).
Model Building,

Because structural equation modeling works best in a conﬁrmatory mode, it is
important to avoid slipping into an exploratory mode where the ﬁnal results may be
unwittingly inﬂuenced by the idiosyncracies of the data at hand. Chin (1998) warns that
the process of changing and re-estimating the model until it ﬁts the data often results in a
ﬁnal model that is mistakenly believed to be correct. In actuality, use of this exploratory
mode, often resulting in strong goodness of ﬁt, is merely indicative of the researchers
skill at adding and deleting items during the exploratory phase. Ideally, Chin (1998)
recommends that a set of a priori models, based on knowledge of the underlying theories,
be followed in order to arrive at valid and reliable ﬁndings that are unlikely to be the
result of capitalizing on chance. Further, for any given SEM model, there are often
alternative models that are equivalent in terms of overall model ﬁt, but are substantially
different in terms of explaining the data. Breckler (1990) addressed this issue by
surveying extant published SEM studies and found that all but one of the 72 studies he
examined acknowledged the possibility of alternative models. As a result, it is of utmost
importance to remain loyal to one’s original theoretical assumptions when conducting
SEM analyses and not fall victim to exploratory methods with the primary goal of
attaining the greatest goodness of ﬁt indices.

In the current analyses, in order to examine the relationships among risk levels,
stress, and both behavioral and attitudinal parenting outcomes, two separate models were

estimated. Speciﬁcally, because actual physical abuse was not measured, the CAP was

88

used to indicate an attitudinal measure of parenting and the Denver Developmental
Screening Test was used as a behavioral marker of parenting outcome. It was not
considered appropriate to group these measures together as indicators of a latent
“parenting” construct because they are formative, rather than reﬂective indicators.
Because SEM techniques attempt to account for all the covariances among its measures,
all items must be reﬂective to be consistent with the statistical algorithm that assumes
that the correlations among indicators of a particular latent variable are caused by that
latent variable (Chin, 1998). Alternatively, even though estimates from formative
indicators are invalid, they can still result in reasonable goodness of ﬁt, leading to
questionable validity of ﬁndings and incorrect conclusions. Hence, each of these models
will be presented in turn.

Descriptive statistics for all of the variables used in the two models can be seen in
Table 4. The data set was chosen to minimize missing data, with only those subjects
having complete data at all three time periods included in the analyses (N =125). The data
was examined for distribution qualities. Kurtosis and skewness values for the entire
model ranged from -1.52 to +1.85, indicating that excessive kurtosis and skewness were
not present in the current data set. The data reﬂected a relatively normal distribution and

were appropriate for the analyses.

89

Table 4: Descriptive Statistics for Study Variables

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Variable N Mean SD Minimum Maximum
Sum of intra- and inter- 125 1.09 .83 .00 2.00
personal level risk (scored

0, 1, 2+)

T1 PSI (child domain 1) 125 46.29 9.95 15.00 80.00
T1 PSI (child domain 2) 125 42.40 13.87 16.50 98.00
Tl PSI (parent domain 1) 125 52.91 12.99 19.00 83.00
T1 PSI (parent domain 2) 125 75.48 15.11 33.00 111.00
Percentage living below 125 .17 .13 .00 .58
poverty line

Percentage female headed 125 .17 .12 .02 .48
households

Percentage not working 125 .38 .10 .15 .77
Percentage non-white 125 .25 .29 .01 .95
residents

Time 2 CAP Abuse score 125 132.75 100.18 6.00 412.00
T2CAP parental distress 125 79.40 72.82 0 255.1
T2CAP parental rigidity 125 14.77 15.42 0 60.0
T2CAP problems with 125 11.23 13.17 0 38.0
family

T2CAP unhappiness 125 12.60 10.90 0 56.2
T2CAP problems with 125 3.10 5.03 0 24.0
child and self

T2CAP problems with 125 11.40 7.86 0 24.0
others ‘
Time 2 Denver advances 125 1.53 1.81 0 13
Time 2 Denver cautions 125 .70 1.01 0 5
Time 2 Denver delays 125 .34 1.02 0 7

 

 

90

 

Model 1.“ CAP Measuring Outcome

Before interpreting the structural paths, the measurement model was examined for
cross-loading items (based on Modiﬁcation Indices) and items that loaded weakly on their
factor (nonsigniﬁcant parameter estimates, p < .05). Based on these criteria, one item
was removed from the community-level risk factor (neighborhood transiency rate).
Consequently, all indicators loaded signiﬁcantly (p < .05) and substantially (standardized
parameter estimates > .30), and modiﬁcation indices suggested no cross-loading
indicators. This model is presented in Figure 3 and the standardized maximum likelihood
estimates of the parameters are presented in Table 5.

The overall model was partially supported based on some of the practical
goodness of ﬁt indices: x2 (85, N=125) = 255.7, p = .000, CFI = .953, TLI = .934, NFI =
.932. However, the RMSEA value was unacceptable at .127. This model accounted for
10% of the variance in child abuse potential as measured by the CAP. There were no
signiﬁcant relationships between person-level risk and stress or outcome. Community
level risk, however, was found to be signiﬁcantly related to child abuse potential (beta =
.195, p = .05) and parenting stress (beta = .263, p = .01). Parenting stress had a direct
effect on child abuse potential that approached signiﬁcance (beta = .178, p < .10). This
suggests that neighborhood risk is mediated by parenting stress in affecting potentially
abusive parenting practices. This pattern of ﬁndings suggests that person-level risk
factors, as measured in the current study, are unrelated to parenting stress or attitudes.
Conversely, neighborhood level risk at Time 0 (the birth of the baby) can affect stress six

months later as well as predict abusive parenting attitudes at one year, and perhaps

91

beyond.

Table 5: Standardized Parameter Estimates for Model I — CAP Outcome

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Parameter Parameter Estimate p R2
Community Level Risk -'
percent non-white .913 .000 .83
percent below poverty line .855 .000 .73
percent female headed households .987 .000 .98
percent mothers not working .619 .000 .38
Stress -+
parent domain 1 .595 .000 .35
parent domain 2 .546 .000 .30
child domain 1 .798 .000 .64
child domain 2 .853 .000 .73
Child Abuse Potential -r
CAP Distress .852 .000 .73
CAP Problems with family .622 .000 .39
CAP Problems with others .745 .000 .55
CAP Rigidity .443 .000 .20
CAP Unhappiness .455 .000 .21
CAP Problems with self and child .424 .000 .18

 

92

 

Figure 3:
Model of multiple inﬂuences of risk on child abuse potential
(CAP subscales measuring outcome)

 

 

 

 

 

 

 

   
   

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Time 0 Time 1 Time 2
sumrsk t2capdis
05 .1 0 t2capfam
t1pdom1 t2capoth
o o
t1pdom2 . 1 8*
paslifggsng . Child abuse t2caprig
t1chdom2 1 potential
-.02
t1chdom2 _ mam”
'20... t2cappro
0
community-level
risk - -
Chl square - 255.7
df = 85
< .000
(TU = .353
I = . 34
percnonw percbelo percﬂlh notwork NF 1 = .932
. RMSEA = .127
* p < .10
** p < .05

93

Model 11: Denver Developmental Screening Test Measuring Outcome

Similar to the previous model, this measurement model was examined for cross-
loading items (based on Modiﬁcation Indices) and items that loaded weakly on their
factor (nonsigniﬁcant parameter estimates, p < .05). As was the case with the previous
model, these criteria warranted the deletion of neighborhood transiency rate from the
community-level risk factor. Consequently, all indicators loaded signiﬁcantly (p < .05)
and substantially (standardized parameter estimates > .30), and modiﬁcation indices
suggested no cross-loading indicators. This model is presented in Figure 4 and the
standardized maximum likelihood estimates of the parameters are presented in Table 6.

Table 6: Standardized Parameter Estimates for Model 11 — Denver Outcome

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Parameter Parameter Estimate p R2
Community Level Risk -r
percent non-white .913 .000 .83
percent below poverty line .853 .000 .73
percent female headed households .987 .000 .98
percent mothers not working .612 .000 .38
Stress -*
parent domain 1 .565 .000 .32
parent domain 2 .516 .000 .27
child domain 1 .814 .000 .67
child domain 2 .868 .000 .75
Child Development-r
Denver delays .691 .000 .48
Denver cautions .640 .000 .41
Denver Advances -.403 .000 .16

 

 

94

The overall model was partially supported based on some of the practical goodness of ﬁt
indices: x2 (49, N=125) = 173.2, p = .000, CFI = .960, TLI = .936, NFI = .945, although
the RMSEA value was .143, indicating a poor model ﬁt. This model accounted for 8% of
the variance in child developmental outcomes as measured by the Denver.

The signiﬁcant relationships elucidated by this model were those between person
level risk at the birth of the baby and developmental outcomes at age one year (beta =
.215, p = .05), and community level risk and parenting stress (beta = .263, p = .01).

Stress was unrelated to developmental outcome as was community level risk.

95

Figure 4:
Model of multiple inﬂuences of risk on child abuse potential
(Denver Developmental Screening Test measuring outcome)

 

 

 

 

 

 

 

 

 

 

 

Time 0 Time 1 Time 2
sumrsk
.21”
.05
t1pdom1
0 o
n pdomz parenting .19 child
stress d I
t1chdom2 1 eve opment
-.02
t1chdom2
.26“ .04
0

community-level
risk

t2denva

 

 

 

t2denvc

 

tZdenvd

 

chi square = 173.2

 

 

 

 

 

 

 

 

 

df = 49

nonw < .000
De": percbeio percﬂ'lh notwork C ' = .960
TLI = .936

NFI = .945

RMSEA = .143

** p < .05

96

Alternative Models

The testing of alternative models is contraindicated in SEM analyses due to the
risk of capitalizing on chance alone. However, post hoc analyses, based on the original
theoretical model, were tested to assess relationships between aspects of risk and
parenting outcome not measured in the above analyses.

Models 111 and IV: The original model was intended to test distinct ecological
factors at the intra-personal, inter-personal, and community levels in relation to parenting
outcome. The measurement qualities of the risk screen used to assess intra- and inter-
personal risk, however, were such that these levels had to be combined into person-level
risk because inter-personal risk was so poorly distributed (0 interpersonal risk factors =
89%, 1 interpersonal risk factor = 10%, 2 or more interpersonal risk factors = 1%).
Combing the intra- and inter- personal levels resulted in normal distribution (see Table 3
in Methods section). Models 111 and IV discard interpersonal level variables (family
crisis, family violence, inadequate social support, 4 or more preschool children at home,
mother-child separations) and look only at intra-personal level risk (maternal age,
maternal physical health, maternal mental illness, maternal history of abuse, maternal
substance abuse, unwanted pregnancy, child physical health, child temperament, and
child accidents) in relation to the CAP and Denver outcomes, respectively. In these
analyses, the CAP abuse scale was used as an observed variable, as opposed to breaking
out the sub-scales as was done in the previous model. This was done to test whether the

aggregate abuse scale accounted for a greater amount of variance.

97

Model 111: The overall model was partially supported based on some of the
practical goodness of ﬁt indices: x2 (31, N=125) = 179.0, p = .000, CFI = .951, TLI =
.913, NFI = .942, although the RMSEA value was .196, indicating a poor model ﬁt. This
model (see Figure 5), using the CAP Abuse Scale, accounted for 10% of the variance in
child abuse potential as did the previous model that employed the Six CAP subscales.

Unlike the previous model, the relationships between infra-personal level risk at
the birth of the baby and child abuse potential at age one year approached signiﬁcance
(beta = .15, p = .08). Similar to the previous model, community level risk was
signiﬁcantly related to parenting stress (beta = .26, p = .01) and abuse potential (beta =
.19, p = .05). Again, parenting stress approached signiﬁcance in its association with
abuse potential (beta = .18, p = .10).

Model IV: The model was partially supported based on some of the practical ﬁt
indices: x2 (49, N=125) = 172.9, p = .000, CFl = .959, TLI = .935, NFI = .945, although
the RMSEA value was .143, indicating a poor model ﬁt. This model (see Figure 6)
accounted for 6% of the variance in child developmental outcome, as opposed to 8%
accounted for by the previous model. Further, the only signiﬁcant relationship elucidated
by this model was between community level risk and parenting stress (beta = .25, p =
.01). Hence, extracting the inter-personal level risk factors from the risk screen nulliﬁed
the relationship between person-level risk and child developmental outcome in this
model. Notably, testing the model with only inter-personal level risk factors, as opposed
to infra-personal level, from the risk screen, revealed similar results: the only signiﬁcant
path was between community level risk and parenting stress, with no association between

interpersonal level risk and outcome.

98

Table 7: Alternative Models in Post Hoc Model-Fitting Process Compared to Original

Hypothesized Models
Models x2 df Ar’ Adf CFI NFI RMSEA
I: Hypothesized Model- 255.7 85 ---- ---- .953 .932 .196
CAP subscales measuring
outcome
11: Hypothesized Model- 173.2 49 ---- ---- .960 .945 .143

Denver measuring outcome

III: Inter-personal level risk 179.0 31 76.7 54 .951 .942 .196
excluded (CAP abuse scale
measuring outcome)

IV: Inter-personal level risk 205.4 49 -32.2 0 .951 .937 .160
excluded (Denver measuring
outcome)

99

Figure 5:
Model 111 - lntrapersonal level risk in relation to model variables
(CAP abuse scale measuring outcome)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Time 0 Time 1 Time 2
sumintra
.06 .15*
t1pdom1
‘ 0
“mom parenting 43*
stress ‘ t2abuse
t1chdom2 1
-.08
t1 chdom2
.26“ .19"
0
community-level
"3k chi square = 179.0
df = 31
p < .000
CFI = .951
TLI = .913
percnonw percbelo percfhh notwork NFI = .942
, RMSEA = .196
* p < .10
** p < .05

100

Figure 6: .
Model IV - lntrapersonal level risk in relation to model variables
(Denver measuring outcome)

 

 

 

 

 

 

 

 

 

 

 

 

Time 0 Time 1 Time 2
sumintra
.16
-.06
t1pdom1
' o 0
"pm“ parenting .20 child
stress development
t1chdom2 1
-.08
t1chdom2
.25“ _'03
00

community-level
nsk

 

1 t2denva

 

 

tZdenvc

 

 

 

tZdenvd

 

 

chi square = 172.9

 

 

 

percnonw percbelo percthh notwork

 

 

 

 

 

 

 

 

 

101

df = 49
< .000
C l= .959
TLI = .935
NFI = .945

RMSEA = .143

“p<.05

DISCUSSION

The data provided partial support for the theoretical model proposed in this study.
In the ﬁrst model, the risk screen that was used to measure person-level risk at the birth of
the baby was unrelated to parenting stress six months later and potentially abusive
parenting attitudes at age one year, and it was unaffected by community-level risk. As
hypothesized, however, community-level risk had a direct and stress-mediated effect on
abusive attitudes. In the second model, person—level risk at birth was signiﬁcantly related
to child developmental outcome at age one year, as was predicted. Parenting stress in this
second model was affected by community-level risk, but it was unrelated to outcome, and
it was unaffected by person-level risk. When taken together, these results indicate that
risk factors at different ecological levels predict differentially to outcome domains. This
supports the assertion that outcome is not a unidimensional construct. In the absence of
actual substantiated cases of abuse, the ability of risk factors to predict outcome depends
on the level of risk being assessed and the component of outcome being tapped. From an
ecological perspective, there are an inﬁnite number of combinations of “risk factors” and
“outcome measures” which casts serious doubt on the practicality of risk assessment. As
has been argued, child maltreatment is very difﬁcult to deﬁne, and the decision where to
draw the line between adequate and inadequate parenting is highly arbitrary. In this
investigation, the line was drawn quite liberally, encompassing attitudes that indicate
potential to abuse and developmental outcomes that may be indicative of inadequate
parenting. These choices resulted in support for some of the hypothesized relationships
in the model, but lacked support for the model as a whole.

Another notable ﬁnding of this study was the ability of risk factors at the birth of

102

the baby to predict outcome more than 12 months later. This represents an improvement
over most risk assessment investigations which measure concurrent risk and outcome(cf.,
Rodwell and Chambers, 1992). We can conclude from the current ﬁndings that
neighborhood strain at the birth of the baby is related to parenting stress six months later,
which in turn affects parental attitudes at one year. This indicates that exosystemic
factors were more temporally stable than person-level risk in relation to parental attitudes.
Likewise, person-level risk at the birth of the baby has the ability to predict
developmental outcome a year later. It would be of interest, however, to conduct a
longitudinal investigation in which early risk is evaluated in relation to outcome over a
much longer span of time than the 12 months evaluated here. This would allow
formulations to be drawn about the stability of risk over time, and evaluation of those
factors that change risk status in both positive and negative directions over time.
Practical Implications of this Study

The most notable ﬁnding of this study was the absence of relationship between
person-level risk factors at the birth of the baby and parenting stress and abusive
parenting attitudes, six and twelve months later, respectively. The risk screen, however,
was the sole instrument employed in the original study to determine intervention
eligibility. The assumption underlying its use was that endorsement of any of the items
indicated an “at risk” status, identifying individuals for intervention services. In light of
the current ﬁndings, it appears that its practical utility is unjustiﬁed. Conversely,
neighborhood characteristics, including ethno-cultural make-up, percentage of residents
living below the poverty line, percentage of female-headed households, and percentage of

families with mothers who work, were predictive of parenting stress and abusive

103

attitudes. These indicators, however, in the practical realm are rarely entered into the
equation to determine individual’s risk status and subsequent program eligibility. This
lends support to Garbarino and Sherman’s (1980) assertion that community variables
create neighborhood strain that fosters dysfunctional parenting practices, and that risk
assessment should be focused broader, at communities rather than individuals and
families.

Ideally, prevention services should target all aspects of functioning within the
social ecology that place children at risk. Traditionally, however, prevention efforts have
been targeted at parents and children with the goal of altering those behaviors and
attitudes that contribute to an elevated risk for poor parenting or child outcomes (Daro &
McCurdy, 1994). For instance, programs targeting parents seek to enhance knowledge of
child development, strengthen basic parenting skills, and offer ongoing social support to
minimize the negative sequelae of stress. Programs aimed at children attempt to
strengthen their resistence to initial maltreatment, provide coping strategies, and mediate
the negative inﬂuences of environmental deﬁcits (Daro & McCurdy, 1994). The current
ﬁndings underscore the importance of shifting preventive resources to target
neighborhood and community functioning. Given the ﬁnding that neighborhood risk is a
more powerful predictor of parental stress and negative parenting attitudes, attention
needs to be directed at ﬁrst understanding the relationship between community-risk and
outcome (i.e., how community factors exert their effect on attitudes and behaviors) and
then programs need to be tailored toward enhancing community strengths and supports.
Theoretical Implications of this Study

Risk is a very ﬂuid concept that is difﬁcult, if not impossible, to operationalize

104

and quantify. This study attempted to measure risk at the intrapersonal, interpersonal, and
community levels of analysis and relate these risk levels to subsequent outcome. The
focal questions addressed by this study were whether risk at different levels of the
ecological context differentially predicted parent and child outcomes, and whether risk
changed over time or was temporally stable. Moreover, this study asked whether
different ecological risks caused certain negative outcomes, or whether it exerted its
effect by exacerbating parenting stress which in turn affected outcome. In summary, the
ﬁndings indicate that all risk is not equal, and that different ecological contexts of risk
affect outcome domains differently. Stress mediated the relationship between community
level risk and parenting attitudes, but was unrelated to person-level risk and child
developmental outcomes. Moreover, the differential effects of intra- and inter- personal
risk raise important questions. It is unclear why interpersonal risk items were so rarely
endorsed. It is also unclear whether these risk factors are less prevalent, or whether they
were poorly measured by the risk screen. When teased apart and tested separately in the
model, they negated the relationship between person level risk and developmental
outcome. But when tested in relation to parenting attitudes, intra-personal level risk
approached signiﬁcance, whereas the sum of inter- and intra- personal level risk had no
association to this outcome measure. Perhaps this lends credence to the argument that
risk is a very ﬂuid concept that is multidimensional in nature (of, Marsh and Wolfe,
1991), and is affects different outcome domains differently. Overall, the study ﬁndings
were able to provide limited insight into these questions, but not without reservations
because of the imperfect conceptualization and measurement of risk that plagues this ﬁeld

as a whole (cf., Zigler, 1976).

105

Limitations of this Study

Many of the limitations of this study are inherent to the risk assessment literature
as a whole. There are no clear-cut deﬁnitions of child abuse and neglect. Because the
goal of this investigation was to study prevention, the outcome measure was an
approximation of potential to abuse, encompassing attitudinal and behavioral indicators.
Substantiation of abuse is a far from clear-cut practice, and as has been discussed, even
these behavioral deﬁnitions are variable and open to interpretation, as well as miss the
mark of prevention. Future research should focus on the relationship between ecological
risk factors and measures of parental ﬁrnctioning.

This study was limited by the use of the KCHS risk screen to measure intra- and
inter- personal risk. This measure was designed as an aggregate measure to determine
basic risk load. In a post hoc fashion, it was separated into ecological tiers for the
purpose of the current investigation. It is unclear whether the lack of relationship
between person level risk and other variables in the model was due to poor measurement
quality. Overall, participants endorsed very few intra- and inter- personal level risk
factors — almost one-third scored zero on this measure. Such low endorsement of person-
level risk may reﬂect a relatively healthy subject pool, or an invalid measure of this
construct. An area for future research would be to develop a valid and reliable risk screen
that measures risk factors at distinct ecological levels, allowing for a more thorough
evaluation of the relationship between risk and outcome. Another measurement
deﬁciency in this study was the use of the PSI to measure parenting stress. This

instrument lacks face validity due to the broad deﬁnition of stress employed by this

106

measure. Perhaps a more precise deﬁnition and measurement of this construct would
have resulted in more predictive power. Some important factors were omitted from this
model that deserve mention. Speciﬁcally, the current model lacked a cultural risk
measure. The ethno-cultural makeup of each neighborhood was used to account for this
level in the ecology, but the literature indicates that this level of analysis deserves its own
tier within the ecology. Another shortcoming in the current investigation was the absence
of measures of resiliency factors. In a truly ecological, transactional model this is an
important dimension to include because resiliency and compensatory mechanisms have
been shown to play an important role in human functioning in general and risk processes
in particular.

The methodological problems inherent in this study included the small sample
size, and the ﬂawed screening procedure that resulted in unequal intervention and control
groups. The sample size was limited by the availability of KCHS screening data which
was not collected by KCHS research staff, but by program bureaucrats. The screening
procedure was also a result of KCHS program ofﬁcials deviating from research protocol.
Alas, one of the costs associated with community-based ﬁeld research is that the richness
of the data is often hampered by extraneous confounds within the study environment.

The various limitations of this study can identify directions for future research.
Beyond the poor measurement quality of several of the instruments, it is most likely that
the model was only partially supported by the data due to errors of omission. As has been
argued throughout these analyses, risk is a multidimensional construct that is difficult to
deﬁne and quantify. Although it would be impossible to construct a list of all of the

ecological forces exerting an effect on parental functioning, the literature reviewed

107

provides guidelines for a more precise inventory than what was able to be tested here.
For instance, at the individual level, Pianta and colleagues (1989) found that factors such
as low self-esteem, poor impulse control, external locus of control, negative affectivity,
and decreased ability to cope with stress were found to increase the likelihood that an
individual will commit child abuse. Another important factor, that was not explicitly
examined, is attachment. This may affect whether or not abusive practices are passed on
from one generation to the next (Bowlby,l969). Omissions at the interpersonal
ecological level included parenting style, discipline techniques, domestic violence, and
social support, all factors described in the literature as signiﬁcantly tied to child
maltreatment (cf., Herrenkohl et al., 1983; Robataille et al., 1985; Straus et al., 1983). At
the community level, many researchers have added to Garbarino and Sherman’s (1980)
list of neighborhood strains. Some of the risk factors explicated by this literature, but not
tested in the current research, include perception of social support and social integration
(Deccio et al.,l994), number of vacant and single family dwellings in the neighborhood,
degree of urbanization, residential density, neighborhood desirability, and neighborhood
development (Zuravin, 1989). The macrosystemic level of analysis was poorly
represented in the current study. Cultural sanctions of physical discipline must be
accounted for (Parke, 1982), as well as cross-cultural differences in family values,
supervision of children, and provision of medical care (English and Pecora, 1994).
Finally, an ideal risk model must include measures of resiliency and compensatory
mechanisms (cf., Cicchetti and Lynch, 1993; Belsky, 1993). Although it was beyond the
scOpe of the current investigation to test all of the risk factors related to child

maltreatment that are described in the literature(alas, this might be too ambitious an

108

undertaking for any researcher), future studies should attempt to use empirically tested
risk factors to guide their decision-making process. Theoretically driven choices, precise
measurement, and empirically supported relationships might result in a more accurate
model of risk.
Lessons from Other Disciplines

The use of risk assessment is fraught with theoretical, conceptual, and practical
difﬁculties that are in no way unique to the ﬁeld of child abuse prevention. Lessons can
be learned from other disciplines. Gina Kolata (2000) reported in the New York Times
on a new screening test for lung cancer. This test involves the use of a spiral CT scanner
which can detect tumors many times smaller than those found with conventional chest X-
rays. The controversy of this procedure revolves around the fact that even though its use
has become widespread, there is no empirical evidence that early detection reduces the
risk of death from lung cancer. In fact, large clinical trials examining whether such
screenings saved lives found that although more cancers were found in the screened
group, and at earlier stages, the screenings did nothing to cut the death rate from lung
cancer. One physician interviewed summed up these ﬁndings by remarking that “these
conclusions are not based on a lack of evidence of an effect, but are based on good
evidence that there is no effect.” The relevance of this example for the current
investigation is that early screening and prediction may not lead directly to the desired
outcome. Kaplan (2000) argues that the identiﬁcation of a disease state on the pathway
between a risk factor and the outcome is interesting, but not essential. Although investing
resources in the early identiﬁcation and treatment of cancer might increase average life

expectancy — which Kolata (2000) contends it does not — the same beneﬁt might also be

109

achieved by investments to reduce cigarette use or exposure to airborne carcinogens. In
other words, public health can be enhanced by removing the risk factor or altering the
risky behavior, rather than relying on screening procedures to identify cases. It is
important to consider the possibility that within the child abuse paradigm, even if we are
able to predict with accuracy who will maltreat their children, this does not in any way
insure that we will then be able to reduce the number of child maltreatment cases that
Occur. Hence, we must be open to the possibility that prediction (or risk assessment) is
only one avenue to reaching the desired outcome (obliteration of child abuse) and that
alternative avenues must also be explored.
Conclusions

Given the ﬁndings of this study that neighborhood characteristics were more
predictive than person-level risk factors of potential to abuse, it appears that the method
of risk assessment used by the KCHS child abuse prevention program was off-track.
Secondary prevention, by deﬁnition, is individually-focused because it involves
identifying individuals at risk for future dysfunctional behavior. Primary prevention, on
the other hand, circumvents this problem by delivering interventions to a much wider
audience, such as an entire community. Targeting risky neighborhoods for intervention
services rather than individuals could also address sources of parenting stress at the
exosystemic level, which was shown in the current investigation to mediate community
risk factors. Those factors that cause parenting stress could be targeted in community-
wide preventive interventions. Because primary prevention is based on a behavioral
rather than disease model (Kaplan, 2000), it does not require diagnosis. Hence, this

model may be more appropriate to the prevention of child maltreatment because of the

110

arbitrary nature of labeling, quantifying, and diagnosing this phenomenon. Moreover,
within the behavioral model, interventions are behavioral in nature, and may involve
individual behavioral change such as involvement in parenting education groups to public
policy changes such as more accessible and affordable child care.

In conclusion, this investigation contributes to our understanding of risk processes
speciﬁcally, as well as supports the use of more universal prevention guidelines. First, it
was demonstrated that the practical screening of clients for inclusion in preventive
interventions, based on individual and interpersonal characteristics, is unjustiﬁed. In
addition to this empirical ﬁnding, this practice is further ﬂawed by numerous conceptual
and theoretical complications. Second, even if prediction accuracy can be achieved, there
is no evidence that such procedures will lead to decreased incidence rates of child
maltreatment. Medical model evidence shows that secondary prevention does very little
to improve public health and decrease unwanted outcomes (Kaplan, 2000). This
prevention strategy may be limited in its effectiveness even when the prevention target is
a disease state with clear risk factors such as smoking and lung cancer. Primary
prevention, on the other hand, is more cost efﬁcient (Kaplan, 2000), is not based on
diagnosis and targeting techniques, and when universally applied targets those risk factors
at the community level. Notably, these factors were most signiﬁcantly related to outcome

in the current investigation.

111

REFERENCES

Abidin, RR. (1990). Parenting stress index manual (3rd edition). Charlottesville,
VA: Pediatric Psychology Press.

Altemeier, W., O’Connor, 8., Vietze, P., Sandler, H., & Sherrod, K. (1984).
Predicion of child abuse: A prospective study of feasability. Child Abuse and Neglect, 8,
393-400.

American Psychiatric Association. (1994). Diagpostic and statistical manual of

mental disorders (4th ed.). Washington, DC: American Psychiatric Association.

 

Azar, S. (1991). Models of child abuse: A meta-theoretical analysis. Criminal

Justice and Behavior 18, 30-46.

 

Agathonas-Georgopoulou, H., & Brown, K.D. (1997). The prediciton of child
maltreatment in Greek families. Child Abuse and Neglect, 21, 721-735.
Baranowski, M.D., Schilmoeller, G.L., & Higgins, BS. (1990). Parenting

attitudes of adolescent and older mothers. Adolescence 25, 781-790.

 

Baumrind, D. (1995). Child maltreating and optimal caregiving in social

contexts. New York: Garland Publishing, Inc.

Bavolek, SJ. (1989). Assessing and treating high-risk parenting attitudes. Early

Child Development and Care, 42, 99-112.

Belsky, J. (1993). Etiology of child maltreatment: A developmental-ecological

analysis. Psychological Bulletin, 11, 413-434.

Belsky, J. (1980). Child maltreatment: An ecological integration. American
Psychologist, 35, 320-335.

Bentler, P. M. (1990). Comparative ﬁt indexes in structural Models.

112

Psychological Bulletin, 107, 238-246.

Bentler, P. M. (1992). On the ﬁt of models to covariance and methodology to the

bulletin. Psychological Bulletin 112, 400-404.

Bentler, P.M. & Bonett, BC. (1980). Signiﬁcance tests and goodness of ﬁt in the
analysis of covariance structures. Psychological Bulletin, 88, 588-606.

Besharov, DJ. (1981). Toward better research on child abuse and neglect:
Making deﬁnitional issues an explicit methodological concern. Child Abuse and Neglect,
_5_, 383-390.

Bloom, BL. (1965). The ‘medical model’, miasma theory, and community
mental health. Communig Mental Health JournﬂJ, 333-338.

Bogenschneider, K. (1996). An ecological risk/protective theory for building
prevention programs, policies, and community capacity to support youth. My
Relations. 45. 127-138.

Bowlby, J. (1969). _A_ttgchment and Loss. New York: Basic Books.

Bollen, K. & Lennox, R. (1991). Conventional wisdom on measurement: A
structural equation perspective. Psychological Bulletin, 110, 305-314.

Brayden, R.M., Altemeier, W.A., Deitrich, M.S., Tucker, D.D., Christensen, M.J.,
McLaughlin, J ., & Sherrod, KB. (1993). A prospective study of secondary prevention of

child maltreatment. The Journal of Pediatrics 122, 511-516.

 

Brissett-Chapman, S. (1995). Child protection risk assessment and Aﬁican
American children: Cultural ramiﬁcations for families and communities. Child Welfare,
16, 45-61.

Bronfenbrenner, U. (1979). The ecology of human development. Cambridge:

113

Harvard University Press.
Broskowsi, A., & Baker, F. (1974). Professional, organizational, and social
barriers to primary prevention. American Journal of OrthopsyclLiatry. 44. 707-719.
Buchholz, E.S., & Korn-Bursztyn, C. (1993). Children of adolescent mothers:

Are they at risk for abuse? Adolescence 28, 361-3 82.

 

Burke, W.T. (1978). The development of a technique for assessing the stresses

experienced by parents of young children. Unpublished doctoral dissertation, Institute of
Child Psychology, University of Virginia, Charlottesville.

Burrell, 3., Thompson, B., & Sexton, D. (1994). Predicting child abuse potential
across family types. Child Abuse an_d Neglect, 18, 1039-1049.

Byme, B. M. (1994). Structural Equation Modeling with EQS and
EOS/Windows:

Basic Concepts, Applications, and Programming. SAGE: California.

Caldwell, R.A., Bogat, G.A., & Davidson, W.S. (1988). The assessment of child
abuse potential and the prevention of child abuse and neglect: A policy analysis.
American Journal of Communig Psychology, 16, 609-624.

Caliso, J .A., & Milner, J .S. (1992). Childhood history of abuse and child abuse
screening. Child Abuse and Neglect, 16, 647-659.

Cantos, A.L., Neale, J .M., O’Leary, D., & Gaines, RN. (1997). Assessment of
coping strategies of child abusing mothers. Child Abuse and Neglect. 21. 631-636.

Caplan, G. (1964). Principles of preventive psychiatg. New York: Basic Books.

Carty, H. & Ratcliffe, J. (1995). The shaken infant syndrome: parents and others

need to know of its dangers. British Medical Journal, 310, 344-345.

114

Chin, W. W. (1998). Issues and opinion on structural equation modeling. l\_/I_I§
Merly, 22, 7-16.

Cicchetti, D. & Lynch, M. (1993). Toward and ecological/transactional model of
community violence and child maltreatment: Consequences for children’s development.
Psychiat_ry: Integpprsonal & Biological Processes, 56, 96-118.

Cicchinelli, L. & Keller, R. (1990). A comparative analysis of risk assessment
models and systems. Washington, DC: National Center on Child Abuse and Neglect.

Connelly, C.D., & Strauss, MA. (1992). Mother’s age and risk for physical
abuse. Child Abuse and Neglect, 16, 709-718.

Costin, l.B., Karger, H.J., & Stoesz, D. (1996). The politics of child abuse in
Ampric_a. New York: Oxford University Press.

Coulton, C.J., Korbin, J .13., Su, M., & Chow, J. (1995). Community level factors
and child maltreatment. Child Development, 66, 1262-1276.

Cowen, EL. (1996). The ontogenesis of primary prevention: Lengthy strides and
stubbed toes. American Journal of Communin Psychology, 24, 235-249.

Daro, D. & McCurdy, K. (1994). Preventing child abuse and neglect:
Programmatic interventions. Child Welfare, 73, 405-422.

Dean, A., Malik, M., Richards, W., & Stringer, S. (1986). Effects of parental
maltreatment on children’s conceptions of interpersonal relationships. Developmental
Psychology, 22, 617-626.

Deccio, 6., Homer, W.C., & Wilson, D. (1994). High-risk neighborhoods and

high-risk families: Replication research related to the human ecology of child

maltreatment. loumal of Social Service Research. 18, 123-137.

115

DePanﬁlis, D. (1988). Final Report: Determining safeg in child protective

services and child placement decisions. Charlotte, NC: ACTION for Child Protection.

Drake, B. & Pandey, S. (1996). Understanding the relationship between
neighborhood poverty and speciﬁc types of child maltreatment. Child Abuse and
Neglect, 20, 1003-1018.

Dukewich, T.L., Borkowski, J .G., & Whitman, TL. (1996). Adolescent mothers
and child abuse potential: An evaluation of risk factors. Child Abuse and Neglect, 20,
1031-1047. .

Egeland, B., Breitenbucher, M., & Rosenberg, D. (1981). Prospective study of the
risk of life stress in the etiology of child abuse. Annual Progpess in Child Psychiag and
Child Development. 666-682.

Egeland, B., Jacobvitz, D., & Paptola, K. (1987). Intergenerational continuity of
abuse. In R. Gelles & J. Lancaster (Eds), Child abuse Mlect: Biosocial dimensions
(pp.255-276). Chicago, IL: Aldine.

Egeland, B., Jacobvitz, D., & Sroufe, LA. (1988). Breaking the cycle of abuse.
Child Development, 59, 1080-1088.

Egeland, B. & Susman-Stillman, A. (1996). Dissociation as a mediator of child
abuse across generations. Child Abuse and Neglect, 20, 1123-1132.

Emery, RE. (1989). Family violence. American PsycholtgLst. 44. 321-328.

Emery, R.E. & Laumann-Billings, L. (1998). An overview of the nature, causes,
and consequences of abusive family relationships: Toward differentiating maltreatment
and violence. American Psychologist, 53, 121-135.

English, D.J. & Pecora, P.J. (1994). Risk assessment as a practice method in

116

child protective services. Child Welfare. 73, 451-473.
Ettema, J .C. & Caldwell, R.A. (Submitted). Home visitor prevention programs:
The impact of relationship quality and attachment style on program outcomes.
Frankenburg, W.K., & Dodds, J. (1967). The Denver Developmental Screening

Test. Journal of Pediatrics 71, 181-191.

 

Frankenburg, W.K., Camp, B.W., & Van Natta, PA. (1971). Validity of the
Denver Developmental Screening Test. Child Development, 42, 475-485.

F rankenburg, W.K., Camp, B.W., & Van Natta, PA. (1971). Reliability and
stability of the Denver Developmental Screening Test. Child Development, 42, 1315-
1325.

Gabinet, L. (1983). Child abuse treatment failures reveal need for redeﬁnition of
the problem. Child Abuse and Neglect, 7, 395-402.

Gallup, G.H., Moore, D.W., Schussel, R. (1995). Disciplining children in
America: A Gallup poll report. Princeton, NJ: The Gallup Organization.

Garbarino, J. (1977). The human ecology of child maltreatment: A conceptual
model for research. Jomﬂf Marriage and the Family, 17, 721-735.

Garbarino, J. (1981). An ecological approach to child maltreatment. In L. Pelton
(Ed.) Thp Social Context of Child Abuse and Neglect (pp. 228-267). New York: Human
Science Press.

Garbarino, J ., & Crouter, A. (1978). Deﬁning the community context for parent-
child relations: The correlates of child maltreatment. Child Development, 49, 604-616.

Garbarino, J ., & Kostelny, K. (1992). Child maltreatment as a community

problem. Child Abuse and Neglect. 16, 455-464.

117

Garbarino, J ., & Sherman, D. (1980a). High-risk neighborhoods and high-risk
families: The human ecology of child maltreatment. Child Development, 51, 188-198.
Garbarino, J. & Sherman, D. (1980b). Identifying high risk neighborhoods. In J.

Garbarino & S.H. Stocking (Eds), Protecting children from abuse and neglect:

Developing and maintaining effective support systems for families (pp. 94-108).
Washington, DC: Jossey-Bass Publishers.

Garmezy, N. (1983). Stressors of childhood. In N. Garmezy & R. Rutter (Eds),
Stress, coping and development of children (pp. 43-84). New York: McGraw-Hill.

Garver, M.S. & Mentzer, J .T. (1999). Logistics research methods: Employing
structural equation modeling to test for construct validity. Journal of Business Logistics,
2_0_, 33-57.

Gil, D. (1973). Violence against children. Cambridge, MA: Harvard University
Press.

Gillham, B., Tanner, G., Cheyne, B., Freeman, 1., Rooney, M., Lambie, A.
(1998). Unemployment rates, single parent density, and indices of child poverty: Their
relationship to different categories of child abuse and neglect. Child Abuse and Neglect.
Q, 79-90.

Giovannoni, J. (1989). Deﬁnitional issues in child maltreatment. In D. Cicchetti

& V. Carlson (Eds), Child maltreatment: Theory and research on the causes and

 

consequences of child abuse and neglect (pp. 3-3 7). New York: Cambridge university
Press.
Graham, P., Dingwall, R., & Wolkind, S. (1985). Research issues in child abuse.

Social Science Medicine 21, 1217-1228.

 

118

Guterman, N.B. (1997). Early prevention of physical child abuse and neglect:

Existing evidence and future directions. Child Maltreatment 2, 12-34.

 

Hamilton, BB. (1980). The relationship of maternal patterns of stress, coping,

and support to qualig of infant-mother attachment. Unpublished doctoral dissertation,

Institute of Child Psyhology, university of Virginia, Charlottesville.

Hobfoll, SE. (1998). Ecology, community, and AIDS prevention. American
Journal of Communig Psychology, 26, 133-144.

Holden, B.W., Willis, D.J., & Corcoran, M.M. (1992). Preventing child
maltreatment during the prenatal/perinatal period. In DJ. Willis, E.W. Holden, & M.
Rosenberg (Eds), Prevention of child maltreatment: Developmental and ecological
perspectives, (pp. 17-46). New York: John Wiley & Sons.

Huxley, P. & Warner, R. (1993). Primary prevention of parenting dysfunction in
high risk cases. American Journal of Orthopsychiatgy, 63, 582-588.

Justice, B., & Calvert, A. (1990). Family environment factors associated with
child abuse. Psychological Reports, 66, 458.

Kaufman, J ., & Zigler, E. (1989). The intergenerational transmission of child
abuse. In D. Cicchetti & V. Carlson (Eds), Child maltreatment: Theog and research on
the causes and consequences of child abuse and neglect (pp. 129 - 150). New York:
Cambridge University Press.

Keller, R.A., Ciccinelli, L.F., Gardner, D.M. (1988). A comparative analysis of
risk assessment models. Denver, CO: Applied Research Associates.

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,

119

17-24.

Kolko, D.J., Kazdin, A.E., Thomas, A.M., Day, B. (1993). Heightened child
physical abuse potential. Journal of Intemrsonal Violence. 8. 169-192.

Krishnan, V. & Morrison, K. (1995). An ecological model of child maltreatment
in a Canadian province. Child Abuse and Neglect, 19, 101-1 13.

Layzer, J., Goodson, B., & DeLange, C. (1986). Children in shelters. Requnse,
2, 2-5.

Light, R. & Nagi, S. (1977). Child maltreatment in the United States. New York:
Columbia University Press.

Lloyd, B.H., & Abidin, RR. (1985). Revision of the parenting stress index.
Journ_al of Pediatric Psychology. 10. 169-177.

Maccoby, N., & Altman, D. (1988). Disease prevention in communities: The
Stanford heart disease prevention program. In E. Price (Ed.) 14 Ounces of prevention: A
casebook for practitioners, (pp. 165-174). Washington, DC: American Psychological
Association.

MacMillan, H.L., MacMillan, J .H., Offord, D.R., Grifﬁth, L., & MacMillan, A.
(1993). Primary prevention of physical child abuse and neglect: A critical review. Part 1.

Main, M., Kaplan, N., & Cassidy, J. (1985). Security in infancy, childhood, and
adulthood: A move to the level of representation. Monographs for the Socieg for
Research in Child Development, 50, 66-104.

Marsh, E.J. & Wolfe, DA. (1991). Methodological issues in research on

physical child abuse. Criminal Justice and Behavior 18 8-29.

 

McCloskey, L., Figuerdo, A., & Koss, M. (1995). The effects of systematic

120

family violence on children’s mental health. Child Development. 66. 1239-1261.

McGee, R.A., & Wolfe, DA. (1991). Psychological maltreatment: toward an
operational deﬁnition. Development and Psychopathology, 3, 3-18.

Meehl, PE. (1990). Why summaries of research on psychological theories are
often uninterpretable. Psychological Reports, 66, 195-244.

Milner, J .S., Gold, R.G., Ayoub, C., & Jacewitz, M.M. (1984). Predictive
validity of the Child Potential Inventory. Journ_al of Consulting and Clinical Psychology,
5;, 879-884. Milner, J .S. (1986). The child abusegﬂential inventory manual (2"d
pﬂpp). NC: Psytech Corporation.

Milner, J .S., Robertson, K.R., & Rogers, BL. (1990). Childhood history of
abuse and adult child abuse potential. Journpl ofLamilv Violence. 5. 15-34.

Milner, J .S. (1994). Assessing physical child abuse risk: The Child Abuse
Inventory. Clinical PsychologyI Review. 14. 547-583.

Mitchell, C.M., Davidson, W.S., Chodkowski, J .A., & McVeigh, J. (1985).
Intervention orientation: Quantiﬁcation of “person-blame” versus “situation-blame”
intervention philosophies. American Journpl of Community Psychology. 13. 543-552.

Mollerstrom, W.W., Patchner, M.A., & Milner, J.S. (1992). Family functioning
and child abuse potential. Journal of Clinical Psychology, 48, 445-454.

Morris, NM. (1981). The biological advantages and disadvantages of teenage
pregnancy. American Journal of Public Health. 14. 412-427.

Morton, N. & Browne,K. (1998). Theory and observation of attachment and its
relation to child maltreatment: A review. (1998). Child Abuse and Neglect, 22, 1093-

1104.

121

Murphy, 8., Orkow, B., & Nicola, R. (1985). Prenatal prediction of child abuse
and neglect: A prospective study. Child Abuse and Neglect, 9, 225-235.

Murphy-Berman, V. (1994). A conceptual framework for thinking about risk
assessment and case management in child protective service. Child Abuse and Neglect,
_1_8, 193-201.

Olds, D.L. (1982). The prenatal/early infancy project: An ecological approach to
prevention of developmental disabilities. In J. Belsky (Ed.), In The Beginning. New
York: Columbia University Press.

Oldshaw, L., Walters, G., & Hall, D. (1989). A behavioral approach to the
classiﬁcation of different types of physically abusive mothers. Merrill-Palmer Quarterly,
35, 255-279.

Oliver, M., Ashley-Miller, M, & Wood, D. (1987). General Discussion (pp. 119-
124). In M. Oliver, M. Ashley-Miller, & D. Wood (Eds) Screening for ris_lgof coron_a_ry
heart disease. New York: John Wiley & Sons.

Parke, RD. (1982). Theoretical models of child abuse: Their implications for
prediction, prevention, and modiﬁcation. In R. Starr (Ed.), Child abuse prediction: Policy
implications. Cambridge: Ballinger Publishing Co.

Pecora, P.J. (1991). Investigating allegations of child maltreatment: The
strengths and limitations of current risk assessment systems. Child and Youth Services.
_15, 7 3-92.

Pianta, R., Egeland, B., & Erickson, M. (1989). The antecedents of
maltreatment: Results of the mother-child interaction research project. In D. Cicchetti &

V. Carlson (Eds), Child maltreatment: Theory and research on the causes and

 

122

cqnsequences of child apuse and nglpct (pp. 203 - 253). New York: Cambridge
University Press.

Pillow, D.R., Sandler, I.N., Braver, S.L., Wolchik, S.A., & Gersten, J .C. (1991).
Theory-based screening for prevention: Focusing on mediating processes in children of
divorce. American Journal of Community Psychology, 19, 809-836.

Polansky, N.A., Gaudin, J .M., Ammons, P.W., & Davis, KB. (1985). The
psychological ecology of the neglectful mother. Child Abuse and Neglect, 9, 265-275.

Ricks, M.H. (1985). The social transmission of parental behavior: Attachment
across generations. Monogpaphs of the Socieg for Research in Child Development, 50,
211-227.

Robitaille, J ., Jones, E., Gold, R.G., Robertson, K.R., & Milner, J .S. (1985).
Child abuse potential and authoritarianism. Journal of Clinical Psychology, 41, 839-844.

Rodwell, M.K., & Chambers, DE. (1992). Primary prevention of child abuse: Is
it really possible? Journal of Sociology and Social Welfag:_,_12, 159-176.

Sameroff, A.J. & Seifer, R. (1990). Early contributors to developmental risk (pp.
52-66). In J. Rolf, A.S. Masten, D. Cicchetti, K.H. Nuechterlein, & S. Weintraub (Eds),
Risk and protective factors in the development of psychopathology. Cambridge:
Cambridge University Press.

Sedlak, A.J. & Broadhurst, DD. (1996). Third National Incidence Study on
Child Abuse and Neglect. Washington, DC: US. Department of Health and Human
Services.

Showers, J. (1992). “Don’t shake the baby:” The effectiveness of a prevention

program. Child Abuse and Neglect, 16, 11-18.

123

Sloan, M.P., & Meier, J .H. (1983). Typology for parents of abused children.

Child Abuse and Neglect, 7, 443-450.

Stratton, P., & Swaffer, R. (1988). Maternal causal beliefs for abused and

 

handicapped children. Journal of Reproductive and Infant Psychology, 5, 201-216.

Straus, M.A., Gelles, R.J., & Steinmetz, SK. (1980). Behind closed doors:
Violence in American families. Garden City, NY: Anchor.

Stringer, S.A., & LaGreca, AM. (1985). Correlates of child abuse potential.
Journal of Abnormal Child Psychology, 13, 217-226.

Tietjen, AM. (1980). Integrating formal and informal support systems: The
Swedish experience. In J. Garbarino & S.H. Stocking (Eds), Protecting children from
abuse and neglect: Developing and maintaining effective support systems for families
(pp. 15-36). Washington, DC: Jossey-Bass Publishers.

Vasta, R. (19820. Physical child abuse: A dual-component analysis.
Developmental Review. 1 2, 125-149.

Widom, CS. (1989). Does violence beget violence? A critical examination of
the literature. Psychological Bulletin, 106, 3-28.

Willis, D.J., Holden, B.W., & Rosenberg, M. (1992). Child malreatment
prevention: Introduction and historical overview. In DJ. Willis, E.W. Holden, & M.
Rosenberg (Eds), Prevention of child maltreatment: Devplopmental and ecological
perspectives. New York: John Wiley & Sons.

Wolf, DA. (1993). Prevention of child neglect: Emerging issues. C_rin_rjp_al

Justice and Behavior 20, 90-111.

 

Zigler, E. (1976). Controlling child abuse in America: An effort doomed to

124

failure. In R. Bourne & E. Newberger (Eds), Critical Persppctives in Child Abuse (pp.
171-213). Lexington, MA: Lexington Books.
Zirpoli, T.J. (1986). Child abuse and children with handicaps. Remedial and

Smcial Education, 7, 39-48.

Zuravin, SJ. (1989). The ecology of child abuse and neglect: Review of the

 

literature and presentation of data. Violence and Victims 4, 101-120.

125

APPENDIX A

Summary of Intervention Effects from Time 1 to Time 2

Change within the Intervention Group from Time 1 to Time 2

Variable

Time 1
mean (sd)

Time 2
mean (sd)

,, F value

 

1. Child Abuse Potential

I 192.64 (87.41) I 186.31 (83.40) .59(n.s.)

 

2. Parenting Stress Inventory

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

child easily distracted 27.17 (4.93) 28.08 (7.60) .60 (n.s.)

mother does not feel reinforced 11.17 (4.07) 10.59 (3.80) .31 (n.s.)

child negative mood 11.15 (3.55) 11.12 (3.36) 1.19 (n.s.

mother's unacceptance of child 13.41 (5.09) 13.73 (5.10) .06 (n.s.)

child unable to adapt to changes 29.38 (6.07) 30.01 (6.46) .12 (n.s.)
mother feels incompetent as a parent 32.44 (6.07) 31.42 (7.29) .03 (n.s.)
lack of attachment to child 13.88 (4.35) 13.42 (4.27) .03 (n.s.)
mother feels restricted by parenting role 21.51 (5.39) 20.85 (4.58) .31 (n.s.)
maternal depression 23.72 (5.94) 22.51 (5.82) 1.91 (n.s.)

poor spousal relationship due to
parenting 20.52 (5.63) 20.06 (4.93) 1.49 (n.s.)
maternal isolation 16.43 (4.61) 16.08 (6.27) .01 (n.s.)
maternal health problems 14.59 (3.57) 13.45 (3.55) 7.79**
demandingness of child 20.58 (5.61) 21.45 (5.53) .95 (n.s.)
overall negative perceptions of child 108.95 (25.87) 114.98 (24.13) 2.55 (n.s.)
overall negative perceptions of parenting 139.59 (24.96) 135.80 (28.63) .08 (n.s.)
overall parenting stress indicators 249.60 (42.26) 250.77 (46.29) 1.87 (n.s.)
3. Child Development-

Denver AdvancesI 1.43 (1.79) I 1.26 (1.70) I0.87 (n.s.)

Denver CautionsI .66 (1.16) I .79 (1.20) I .29 (n.s.)
Denver DelaysI .31 (.87) I .43 (1.22) I1.80 (n.s.)

 

*p<os **g<bn***p<om

126

 

APPENDIX B

Narrative Description of Levels of Risk Given to Raters and List of Risk Factors from
Healthy Start Screening Form

Etiological models of child maltreatment are largely multidimensional (for
example, Belsky’s ecological model of the determinants of parenting), however, risk
assessment instruments tend to be unidimensional, requiring that a total number of risk
factors be added together to arrive at a composite risk factor “score.” Inherent in the
unidimensional, or checklist, approach, is the assumption that all risk factors are equal,
and it is the composite score or total number of risk factors that is critical in predicting
outcome. Another way to look at the assessment of risk factors, that is more reﬂective of
the dimensionality of etiological models, is to develop dimensional risk assessment
instruments. It is predicted that a dimensional conceptualization of risk can lead to more
accurate predictions about later parent functioning because parenting is a multiply-
deterrnined, or multidimensional construct. Some of the ecological dimensions that have
been described in the literature as contributing to parent functioning include: 1) historical
and psychological characteristics of the parents (in_trapersonal variables), 2) attributes of
the child, 3) familial and social network supports and stressors (,i_n_t_e_rpersonal variables),
and 4) community/sociocultural sources of stress and support (community variables).

The current investigation is an attempt to develop a multidimensional model of
risk assessment, and to test the hypothesis that such a model does a better job of
predicting parenting outcomes than traditional checklist approaches to risk assessment.
As a ﬁrst step, it is necessary to sort through the various risk factors used by KCHS, and

try to categorize each into one of the dimensions described above. Below, I have

127

described in more detail each of the dimensions I am interested in looking at. After
reading each description, please use your best judgement to try to ﬁt each risk factor into
one of the categories. If a particular risk factor does not seem to ﬁt any of the categories,
it can be sorted into an “other” category.

1. hampersonal Level:

Several enduring characteristics of the individual have been shown to
inﬂuence parenting. Developmental history (the type and quality of parenting
received as a child, etc.) and personality style (attachment style, temperament,
resilience, etc.) have been shown to affect parenting ability. Linkages between
parental psychological well-being and parental functioning can also be traced.
Psychological well-being may include such factors as self-esteem, depression, and
psychological resources. Age and developmental level are indicators of maturity,
which have also been linked to more positive parenting practices. Lastly,
maternal health can affect parenting ability.

Child characteristics have been shoWn to inﬂuence parental care-giving.
These characteristics include the age of the child, physical health and
development, temperament, responsiveness to parents, and behavioral problems.

2. Inirpersonal Level:

The marital relationship is an inﬂuential support system for parents which
in turn affects parental functioning. General network support can also provide
both support and stress, depending upon the “goodness-of-ﬁt” between the level
of support desired and that received. Other interpersonal relationships such as

those with friends, extended family, co-workers, and helping professionals can

128

also affect parental functioning. In general, all relationship variables would ﬁt in
this category, including the parent-child interaction.
3. Community/sociocultural Level:

This level of analysis includes the “community climate” (number single-
parent households in the neighborhood, criminal activity, transiency, employment
rates, social cohesiveness of neighborhood, involvement of neighborhood in
community activism, availability of community resources, social alienation due to
cultural differences/language barriers, etc.), SES, and social/community resources
and strains.

4. “Other”

List of risk factors from Healthy Start Screening Form:

jaundice

asphyxia

meconium aspiration

failure to thrive

congenital abnormalities

feeding problems

premie/low birth weight/less than 51bs. 8oz.
excessive irritability

positive drug screen (baby)

IUGR

small gestational age (SGA)

lead poisoning

atypical/recurrent accidents

chronic otitis media (ear infections)
bronchopulmonary dysplasia
immunization concerns
unplanned/unwanted pregnancy
less than 18 years old, and no support system
adolescent mother

primary caregiver/child separation
mental illness/low functioning

129

pre-existing medical condition

Single parent

4 or more preschool children

history of reproductive problems

no prenatal care/late to receive prenatal care
severe prenatal complications

severe perinatal complications

poor nutrition (mom)
overweight/underweight

smoking

language barrier

acute family crisis

no medical insurance

inadequate health care

ﬁnancial need

unemployed

history of medical/genetic problems
family violence

substance abuse

severe chronic physical illness
homeless/other dangerous living condition
history of abuse/neglect (morn)

parenting concerns

other concerns that may result in infant death, injury, impairment, or illness
isolation/lack of support

130

APPENDIX C
Parenting Stress Index (Abidin, 1990)

Directions: In answering the following questions, please think about the child in the
KCHS program.

The questions on the following pages ask you to mark an answer which best describes
your feelings. While you may not ﬁnd an answer which exactly states your feelings,
please mark the answer which comes closest to describing how you feel. YOUR FIRST
REACTION TO EACH QUESTION SHOULD BE YOUR ANSWER.

Please mark the degree to which you agree or disagree with the following statements by
ﬁlling in the number which best matches how you feel. If you are not sure, please ﬁll in

#3.
l 2 3 4 5

Strongly Agree Not Sure Disagree Strongly
Agree Disagree

1. When my child wants something, my child usually keeps trying to get it.

2. My child is so active that it exhausts me.

3. My child appears disorganized and is easily distracted.

4. Compared to most, my child has more difﬁculty concentrating and paying attention.

5. My child will often stay occupied with a toy for more than 10 minutes.

6. My child wanders away much more than I expected.

7. My child is much more active than I expected.

8. My child squirms and kicks a great deal when being dressed or bathed. 9. My child

can be easily distracted from wanting something.
10. My child rarely does things for me that make me feel good.
11. Most times I feel that my child likes me and wants to be close to me.
12. Sometimes I feel my child doesn't like me and doesn't want to be close to me.
13. My child smiles at me much less than I expected.
14. When I do things for my child I get the feeling that my efforts are not appreciated
very much.
15. Which statement best describes your child?
1. almost always likes to play with me,
2. sometimes likes to play with me,

4. usually doesn't like to play with me,

131

16.

l7.
18.
19.
20.
21.
22.

5. almost never likes to play with me.
My child cries and fusses:
1. much less than I had expected,
2. less than I expected,
3. about as much as I expected,
4. much more than I expected,
5. it seems almost constant,
My child seems to cry or fuss more often than most children.
When playing, my child doesn't often giggle or laugh.
My child generally wakes up in a bad mood.
I feel that my child is very moody and easily upset.
My child looks a little different than I expected and it bothers me at times.

In some areas my child seems to have forgotten past learnings and has gone back to

doing things characteristic of younger children.

23.
24.
25.
26.
27.
28.

29.
30.
31.

32.
33.
34.
35.
36.
37.
38.
39.
40.

My child doesn’t seem to learn as quickly as most children.
My child doesn’t seem to smile as much as most children.
My child does a few things which bother me a great deal.
My child is not able to do as much as I expected.
My child does not like to be cuddled or touched very much.
When my child came home from the hospital, I had doubtful feelings about
my abilities to handle being a parent.
Being a parent is harder than I thought it would be.
I feel capable and on top of things when I am caring for my child.
Compared to the average child, my child has a great deal of difﬁculty in
getting used to changes in schedules or changes around the house.
My child reacts very strongly when something happens that my child doesn’t like.
Leaving my child with a baby-sitter is usually a problem.
My child gets upset easily over the smallest thing.
My child easily notices and overreacts to loud sounds and bright lights.
My child’s sleeping or eating schedule was much harder to establish than I expected.
My child usually avoids a new toy for a while before beginning to play with it.
It takes a long time and is very hard for my child to get used to new things.
My child doesn’t seem comfortable when meeting strangers.

When upset, my child is:

132

easy to calm down,

harder to calm down than I expected,

:51"?

very difﬁcult to calm down,
5. nothing I do helps to calm my child.
41. l have found that getting my child to do something or stop doing something is;
1. much harder than I expected,
2. somewhat harder than I expected,
3. about as hard as I expected,
4. somewhat easier than I expected,
5. much easier than I expected.
42. Think carefully and count the number of things which your child does that bothers
you. For example: dawdles, refuses to listen, overactive, cries, interrupts, ﬁghts, whines,
etc. Please ﬁll in the number which includes the number of things you counted.
1. 1-3
2. 4-5
3. 6-7
4. 8-9
5. 10+
43. When my child cries it usually lasts:
1. less than 2 minutes
2. 2-5 minutes
3. 5-10 minutes
4. 10-15 minutes
5. more than 15 minutes
44. There are some things my child does that really bother me a lot.
45. My child has had more health problems than I expected.
46. As my child has grown older and become more independent, I ﬁnd myself more
worried that
my child will get hurt or into trouble.
47. My child turned out to be more of a problem than I had expected.
48. My child seems to be much harder to care for than most.
49. My child is always hanging on me.
50. My child makes more demands on me than most children.

51. I can't make decisions without help.

133

52. I have had many more problems raising children than I expected
53. I enjoy being a parent.
54. I feel that I am successful most of the time when I try to get my child to do or not do
something.
55. Since I brought my last child home from the hospital, I ﬁnd that I am not able to take
care of

this child as well as I thought I could. I need help.
56. I often have the feeling that I cannot handle things very well. 57. When I think about
myself as a parent I believe:

1. I can handle anything that happens

2. I can handle most things pretty well

3. Sometimes I have doubts, but ﬁnd that I handle most things without any
problems.

4. I have some doubts about being able to handle things

5. I don't think I handle things very well at all.
58. I feel that I am:

1. a very good parent

2. a better than average parent

3. an average parent

4. a person who has some trouble being a parent

5. not very good at being a parent
59. What were the highest levels in school or college you and the child's father/mother
completed?

Mother:

. 1-8th grade
. 9-12th grade
. Vocational or some college

. College graduate

MAWNt-s

. Graduate or professional school
60. Father:

. 1-8th grade

. 9-12th grade

. Vocational or some college

#0.)th

. College graduate

134

61.

62.
63.

me.
64.

65

66.
67.

68
69

5. Graduate or professional school
How easy is it for you to understand what your child wants or needs?
1. very easy,
2. easy,
3. somewhat difﬁcult,
4. it is very hard,
5. I usually can't ﬁgure out what the problem is.
It takes a long time for parents to develop close, warm feelings for their children.

I expected to have closer and warmer feelings for my child than I do and this bothers

Sometimes my child does things that bother me just to be mean.

When I was young, I never felt comfortable holding or taking care of children.
My child knows I am his or her parent and wants me more than other people.
The number of children that I have now is too many.

Most of my life is spent doing things for my child.

I ﬁnd myself giving up more of my life to meet my children's needs than I ever

expected.

70.
71.
72.
73.
74.

75.
76.
77.

78.
79.
80.
81.

I feel trapped by me responsibilities.

I often feel than my child's needs control my life.

Since having this child I have been unable to do new and different things.

Since having a child I feel that I am almost never able to do things that I like to do.

It is hard to ﬁnd a place in our home where I can go to be by myself.
When I think about the kind of parent I am, I often feel guilty or bad about myself.
I am unhappy with the last purchase of clothing I made for myself.
When my child misbehaves or fusses too much I feel responsible, as if I didn't
do something right.
I feel every time my child does something wrong it is really my fault. .
I often feel guilty about the way I feel towards my child.
There are quite a few things that bother me about my life.
I felt sadder and more depressed than I expected after leaving the hospital with my

baby.

82.

I wind up feeling guilty when I get angry at my child and this bothers me.

135

83. After my child had been home from the hospital for about a month, I noticed that I
was feeling more sad and depressed than I expected84. Since having my child, my
spouse (male/female ﬁiend) has not given me as much help and support as I expected85.
Having a child has caused more problems than I expected in my relationship with my
spouse

(male/female friend).
86. Since having a child my spouse (or male/female friend) and I don’t do as many
things together.
87. Since having my child, my spouse (or male/female friend) and I don’t spend as much
time together as a family as I expected
88. Since having my last child, I have had less interest in sex.
89. Having a child seems to have increased the number of problems we have with in-
laws

and relatives.
90. Having children has been much more expensive than I had expected.
91. I feel alone and without friends. ,
92. When I go to a party I usually expect not to enjoy myself.
93. I am not as interested in people as I used to be.
94. I often have the feeling that other people my own age don't particularly like my
company. ‘
95. When I run into a problem taking care of my children I have a lot of people to whom
I can talk to get help or advice
96. Since having children I have a lot fewer chances to see my friends and to make new
friends.
97. During the past six months I have been sicker than usual or have had more aches and
pains

than I normally do.
98. Physically, I feel good most of the time.
99. Having a child has caused changes in the way I sleep.
100. I don't enjoy things as I used to.
101. Since I've had my child:

136

l. I have been sick a great deal,
2. I haven't felt as good,
3. I haven't' noticed any change in my health,

4. I have been healthier.

137

APPENDIX D
Child Abuse Potential Inventory (Milner, 1986)

The following questionnaire includes a series of statements which may be applied to
yourself. Read each of the statements and determine if you AGREE or DISAGREE
with the statement.

I never feel sorry for others.

I enjoy having pets.

I have always been strong and healthy.

I like most people.

I am a confused person.6. I do not trust most people.

People expect too much from me.

Children should never be bad.

. I am often mixed up.

10. Spanking that only bruises a child is okay.

11. I always try to check on my child when it’s crying.

12. I sometimes act without thinking.

13. You cannot depend on others.

14. I am a happy person.

15. I like to do things with my family.

16. Teenage girls need to be protected

17. I am often angry inside.

18. Sometimes I feel all alone in the world.

19. Everything in a home should always be in its place.

20. I sometimes worry that I cannot meet the needs of a child.

21. Knives are dangerous for children

22. I often feel rejected.

23. I am often lonely inside.

24. Little boys should never learn sissy games.

25. I often feel very ﬁ'ustrated.26. Children should never disobey.
27. I love all children.

28. Sometimes I fear that I will lose control of myself.

29. I sometimes wish that my father would have loved me more.
30. I have a child who is clumsy.3 1. I know what is the right and wrong way to act.
32. My telephone number is unlisted.

33. The birth of a child will usually cause problems in a marriage.
34. I am always a good person.

35. I never worry about my health

36. I sometimes worry that I will not have enough to eat.

37. I have never wanted to hurt someone else.

38. I am an unlucky person.

39. I am usually a quiet person.

40. Children are pests.

41. Things have usually gone against me in life.

42. Picking up a baby whenever he cries spoils him

owswewwr

138

43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
63.
64.
65.
66.
67.
68.
69.
70.
71.
72.
73.
74.
75.
76.
77.
78.
79.
80.
81.
82.
83.
84.
85.
86.
87.

I sometimes am very quiet.

I sometimes lose my temper.

I have a child who is bad.

I sometimes think of myself ﬁrst.

I sometimes feel worthless.

My parents did not really care about me.

I am sometimes very sad.

Children are really little adults.

I have a child who breaks things.

I often feel worried.

It is okay to let a child stay in dirty diapers for a while.
A child should never talk back.

Sometimes my behavior is childish.

I am often easily upset.

Sometimes I have bad thoughts.

Everyone must think of himself ﬁrst.

A crying child will never be happy.

I have never hated another person.

Children should not learn how to swim.

I always do what is right.

I am often worried inside.

I have a child who is sick a lot.

Sometimes I do not like the way I act.

I sometimes fail to keep all of my promises.
People have caused me a lot of pain.

Children should stay clean.

I have a child who gets into trouble a lot

I never get mad at others.

I always get along with others.

I often think about what I have to do.

I ﬁnd it hard to relax.

These days a person doesn’t really know on whom one can count
My life is happy

I have a physical handicap.

Children should have play clothes and good clothes.
Other people do not understand how I feel .

A ﬁve year old who wets his bed is bad.
Children should be quiet and listen.

I have several close ﬁiends in my neighborhood.
The school is primarily responsible for educating the child.
My family ﬁghts a lot.

I have headaches.

As a child I was abused.

Spanking is the best punishment.

1 do not like to be touched by others.

139

88.
89.
90.
91.
92.
93.
94.
95.
96.
97.
98.
99.

100.
101.
102.
103.
104.
105.
106.
107.
108.
109.
110.
111.
112.
113.
114.
115.
116.
117.
118.
119.
120.
121.
122.
123.
124.
125.
126.
127.
128.
129.
130.
131.
132.

People who ask for help are weak.

Children should be washed before bed.

1 do not laugh very much.

I have several close friends.

People should take care of their own needs.

I have fears no one knows about

My family has problems getting along

Life often seems useless to me.

A child should be potty trained by the time he’s one year old.
A child in a mud puddle is a happy sight.

People do not understand me.

I often feel worthless.

Other people have made my life unhappy.

I am always a kind person.

Sometimes I do not know why I act as I do.

I have many personal problems.

I have a child who often hurts himself.

I often feel very upset.

People sometimes take advantage of me.

My life is good

A home should be spotless.

I am easily upset by my problems.

I never listen to gossip.

My parents did not understand me.

Many things in life make me angry.

My child has special problems.

I do not like most children.

Children should be seen and not heard.

Most children are alike.

It is important for children to read.

I am often depressed.

Children should occasionally be thoughtful of their parents.
I am often upset.

People don’t get along with me.

A good child keeps his toys and clothes neat and orderly.
Children should always make their parents happy.

It is natural for a child to sometimes talk back.

I am never unfair to others.

Occasionally, I enjoy not having to take care of my child.
Children should always be neat.

I have a child who is slow.

A parent must use punishment if he wants to control a child’s behavior.
Children should never cause trouble.

I usually punish my child when it is crying.

A child needs very strict rules.

140

133.
134.
135.
136.
137.
138.
139.
140.
141.
142.
143.
144.
145.
146.
147.
148.
149.
150.
151.
152.
153.
154.
155.
156.
157.
158.
159.
160.

Children should never go against their parents’ orders.
I often feel better than others.

Children sometimes get on my nerves.

As a child I was often afraid.

Children should always be quiet and polite.

I am often upset and do not know why.

My daily work upsets me.

I sometimes fear that my children will not love me.
I have a good sex life.

I have read articles and books on child rearing.

I often feel very alone.

People should not show anger.

I often feel alone.

I sometimes say bad words.

Right now, I am deeply in love.

My family has many problems.

I never do anything that is bad for my health.

I am always happy with what I have.

Other people have made my life hard.

I laugh some almost every day.

I sometimes worry that my needs will not be met.
I often feel afraid.

I sometimes act silly.

A person should keep his business to himself.

I never raise my voice in anger

As a child I was knocked around by my parents.

I sometimes think of myself before others.

I always tell the truth.

141

APPENDIX E

Summary of Intervention - Control Group Comparisons

Comparison of Intervention and Control Groups on Time 1 Demographic

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

variables
Intervention Control
ALariable mean (sd) mean (sd) F valnp
1. mother's education 11.01 (2.49) 11.59 (2.30) 2.41 (n.s.)
2. father's education 11.43 (2.65) 12.38 (2.26) 5.05*
3. mother's age 23.36 (6.00) 23.83 (5.10) .31 (n.s.)
4. father's age 27.02 (7.28) 26.77 (5.81) .05 (n.s.)
percentage percentage chi square
5. marital status 1.34 (n.s.)
married 30% 36%
single 68% 63%
divorced 1% 1%
separated 1% 0%
6. mother employed ,7 2.20 (n.s.)
yesI 32% I 40%
ml 68% I 60%
7. father employed ,, .32 (n.s.)
yesI 78% I 73%
ml 22% I 27%
8. mother's citizenship 4.00*
US. citizenI 87% I 96%
non-U.S. citizenI 13% I 4%
9. father's citenship , , 5.53"I
us. citizenI 86% J 97%
non-U.S. citizenI 14% J 3%

 

 

 

142

 

Comparison of Intervention and Control Groups on Time 1 Parenting Variables,

Child Development, and Community Resource Participation

Variable

Intervention

mean (sd)

Carnal

mean (sd)

F value

 

 

1. Child Abuse Potential

191.45 (87.44)

128.02 (75.45) , 27.44 ***

 

2. Parenting Stress Inventory

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

child easily distracted 27.27 (4.86) 26.14 (4.28) 2.80 (n.s.)
mother does not feel reinforced 11.35 (4.17) 9.78 (3.53) 7.65 **
child negative mood 11.16 (3.45) 9.68 (2.80) 9.60 **
mother's unacceptance of child 13.43 (5.13) 11.29 (3.76) 9.91 **
child unable to adapt to changes 29.43 (6.35) 26.33 (5.70) 12.01 **
mother feels incompetent as a
parent 32.57 (6.03) 27.32 (5.54) 35.52 ***
lack of attachment to child 13.93 (4.32) 11.81 (2.93) 14.12 ***
mother feels restricted by 21.36 (5.45) 18.14 (5.32) 17.01 ***
maternal depression 23.62 (5.97) 18.67 (5.25) 36.01***
poor spousal relationship 20.24 (5.61) 17.72 (5.94) 9.20"
maternal isolation 16.20 (4.22) 13.86 (4.15) 14.86***
maternal health problems 14.55 (3.59) 13.01 (3.90) 8.36"
demandingness of child 20.68 (5.68) 16.58 (4.47) 2848*"

 

143

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

negative perceptions of child 109.31 (26.23) 98.85 (18.81) 9.22 **
overall negative perceptions 139.28 (24.28) 120.52 (24.51) 28.66“”
overall parenting stress indicators 249.57 (41.57) 2193713804) 2699*”
3. Child Development
Denver II Advances 1.43 L179) .90 (1.24) 4.79*
Denver II Cautions .66 (1.16) .32 (.65) 503*
Denver 11 Delays .31 (.87) .04 (.21) 6.00*
4. Community Resources
number community resources 6.27 (4.00) 6.92 (25H 2.43 (n.s.)
number community resources 5.57 (4.33) 5.28 L244) 1.14 (n.s.)
number community resources 3.47 (2.85) 2.63 (1.63) 3.66*

 

 

 

 

*Q<-05, H: p<.01, *IHI Q< .001

144

 

 

llilIIIMIMIIIIIIIIIIIIIIII