JSLHR

Review Article

A Meta-Analysis of Cross Sectional

Studies Investigating Language
in Maltreated Children
Jarrad A. G. Lum,a Martine Powell,a Lydia Timms,b and Pamela Snowc

Purpose: In this review article, meta-analysis was used showed that maltreated children demonstrated
to summarize research investigating language skills in consistently poorer language skills with respect to
maltreated children. receptive vocabulary (k = 19; standardized mean
Method: A systematic search of published studies was difference [SMD] = .463; 95% confidence interval [CI;
undertaken. Studies were included in the meta-analysis .293, .634]; p < .001), expressive language (k = 4;
if they investigated language skills in groups comprising SMD =.860; 95% CI [.557, 1.163]; p < .001), and receptive
maltreated and nonmaltreated children. Studies were language (k = 9; SMD =.528; 95% CI [.220, .837];
selected if these 2 groups of children were of comparable p < .001).
age and from a similar socioeconomic background. Conclusion: Together, these results indicate a reliable
Results: A total of 26 studies were identified that met association between child maltreatment and poor language
the inclusion criteria. Results from the meta-analysis skills.

C
hild maltreatment, which encompasses both abuse known to disrupt brain development (De Bellis, 2001; Hart
(physical, sexual, and emotional) and neglect & Rubia, 2012; Twardosz & Lutzker, 2010). Research un-
(physical or emotional), is a significant public dertaken with human and nonhuman animals shows that
health and social welfare problem (Gilbert et al., 2009). the chronic production of cortisol in response to sustained
Affected individuals have higher risk of mortality (Overpeck, stress exposure has a neurotoxic effect on the brain (Teicher
Brenner, Trumble, Trifiletti, & Berendes, 1998; Palusci & et al., 2003). This includes the temporal lobes that support
Covington, 2014) as well as medical, behavioral, emotional, learning, memory, and language (Anderson, Teicher,
and mental health problems (Brown, Fang, & Florence, 2011; Polcari, & Renshaw, 2002; De Bellis et al., 2002; De Bellis &
Éthier, Lemelin, & Lacharité, 2004; Jaffee & Maikovich- Kuchibhatla, 2006; De Brito et al., 2013; Teicher, Anderson,
Fong, 2011; Keyes et al., 2012; Rhodes et al., 2012). This & Polcari, 2012; Teicher et al., 1997; Woon & Hedges, 2008).
article reviews research that has investigated language func- Second, language development may be particularly vulnera-
tioning in maltreated children. Using meta-analysis, this ble to the effects of maltreatment because development in
literature was summarized and an overall estimate of the this domain is contingent on adequate stimulation from the
association between maltreatment and language functioning environment. In order for children to learn language, they
was estimated. require direct, child-focused exposure and sustained oppor-
There are a number of reasons for suspecting poor tunities to engage in meaningful communicative interactions
language skills in children exposed to maltreatment. First, with caregivers, other adults, and peers (Kuhl, Williams,
the stress experienced as a consequence of maltreatment is Lacerda, Stevens, & Lindblom, 1992; Tomasello, 2003;
Vihman, 1996). In cases of maltreatment, the necessary en-
gagement and stimulation required for language learning
a
Deakin University, Melbourne, Australia
may not be present. For instance, research suggests maltreat-
b
Curtin University, Perth, Australia ing caregivers interact less, ignore their children more (Allen
c
Monash University, Melbourne, Australia & Wasserman, 1985; Eigsti & Cicchetti, 2004; Kavanagh,
Correspondence to Jarrad A. G. Lum: jarrad.lum@deakin.edu.au Youngblade, Reid, & Fagot, 1988; Wasserman, Green, &
Editor: Rhea Paul Allen, 1983), react infrequently when their children talk
Associate Editor: Elizabeth Kay-Raining Bird (Kavanagh et al., 1988), and use a less diverse range of
Received February 18, 2014 vocabulary and syntactic structures during communication
Revision received August 9, 2014
Accepted March 12, 2015 Disclosure: The authors have declared that no competing interests existed at the time
DOI: 10.1044/2015_JSLHR-L-14-0056 of publication.

Journal of Speech, Language, and Hearing Research • Vol. 58 • 961–976 • June 2015 • Copyright © 2015 American Speech-Language-Hearing Association 961
(Allen & Wasserman, 1985; Eigsti & Cicchetti, 2004; this gap in the literature, the current review used meta-
Kavanagh et al., 1988; Wasserman et al., 1983). analysis to summarize the relationship between maltreat-
To date, language functioning in maltreated children ment with respect to vocabulary and the comprehension
has been the subject of several reviews. A systematic review and production of language.
of the child maltreatment literature by Veltman and Browne
(2001) identified a total of 42 studies (published and unpub-
lished) investigating language skills in maltreated children. Method
Of this total, 36 (or 86%) reported poorer language skills in Study Design
maltreated children. Nonsystematic narrative reviews have
Studies were identified following a systematic search of
also concluded maltreatment negatively impacts language
titles, abstracts, and keywords in four electronic databases.
skills (Hwa-Froelich, 2012; Law & Conway, 1992). To our
The electronic databases searched were Education Resources
knowledge, a quantitative review of studies investigating lan-
Information Center (ERIC), Cumulative Index to Nursing and
guage functioning in maltreated children using meta-analysis
Allied Health Literature (CINAHL), Medline, and PsycINFO.
has yet to be undertaken. In meta-analysis, effect sizes from
Only studies (written in any language) that had been pub-
individual studies are pooled and an average effect size is com-
lished or were in press in a peer-reviewed journal were in-
puted (Hunter, Schmidt, & Jackson, 1982). In the absence of a
cluded. Details of all keywords, fields searched, and search
prior meta-analysis, the magnitude of the association between
syntax for all databases are presented in the online supple-
maltreatment and language functioning is not yet known.
mental file. The search strategy identified studies published
A challenge in quantifying the relationship between
between January 1960 and October 2013. An initial search
language problems and maltreatment is ensuring extrane-
was executed in June 2013 and repeated in October 2013.
ous variables are controlled. To determine the impact of
maltreatment on language skills, it is important that socio-
economic status (SES) is controlled. This is because chil- Study Inclusion Criteria
dren’s language skills have been shown to covary with Studies were included in the meta-analysis if they met
socioeconomic level (Hart & Risley, 1995; Hoff & Tian, the following criteria. First, the study was required to have
2005; Hoff-Ginsberg, 1998). Also, elevated stress levels examined language in a group of children found to have been
have been found in children reared in low SES environ- maltreated. Studies undertaken exclusively with cases of sus-
ments (Bradley & Corwyn, 2002; McLoyd, 1998). Because pected, but unconfirmed, maltreatment were excluded (e.g.,
maltreatment is disproportionately associated with low Heath, Colton, & Aldgate, 1994). Second, the study was
SES (Garbarino & Crouter, 1978; Garbarino & Kostelny, required to have compared either the expressive or receptive
1992; Trickett, Aber, Carlson, & Cicchetti, 1991), it could language or vocabulary skills of the maltreated group to a
be that environmental factors and not maltreatment largely nonmaltreated comparison group. Third, the nonmaltreated/
contribute to the language outcomes for this group. comparison group was required to comprise participants of
Also missing from the literature at present is a sys- similar age to the maltreatment group and be of equivalent
tematic review into the relationship between maltreatment SES as noted in the study’s method section. Fourth, a study
and different aspects of language functioning (for a narrative was only included if the mean age of the participants was
review of this literature, see Law & Conway, 1992). In the between 2;6 (years;months) and 11;0 years of age.
child maltreatment literature, the most widely used method
to assess language has been to examine the comprehension or
understanding of single words, otherwise known as receptive Study Selection
vocabulary (e.g., Cicchetti & Rogosch, 1997; Noll et al., 2010; After the search strategy was executed and duplicates
Toth, Stronach, Rogosch, Caplan, & Cicchetti, 2011; Trickett, were removed, three reviewers (one of the authors and two
1993). However, it can be problematic to use this type of as- research assistants) inspected titles and abstracts of re-
sessment as a proxy for general language skills. This is because cords. Full-text articles were sought for records considered
vocabulary tests measure only one component of linguistic to be potentially relevant. A record was considered to be
knowledge. Vocabulary tests also have low levels of diagnos- potentially relevant if the study investigated language or
tic accuracy with respect to identifying children with lan- any cognitive ability in maltreated children. Also, a study
guage problems (Spaulding, Plante, & Farinella, 2006). In investigating any aspect of maltreatment that compared a
contrast, tests that measure how well children can compre- group of maltreated children to a control group on any
hend and produce sentences have higher diagnostic accuracy outcome measure was referred for further screening. Reli-
levels (i.e., sensitivity & specificity levels > 80%, see Spaulding ability of the reviewers screening was assessed. Ten percent
et al., 2006). It is therefore of interest to determine whether of articles were randomly selected and examined by all re-
the literature indicates maltreated children have difficulties viewers. The agreement between reviewers for this compo-
comprehending and producing language. nent of the screening was high (k = .989). There was 100%
Although prior reviews suggest language skills are agreement for the suitability of full-text articles to be in-
poorer in maltreated children, the magnitude of the rela- cluded in the meta-analysis.
tionship between maltreatment and different domains of Our search strategy led to the identification of several
language functioning has yet to be ascertained. To address studies undertaken at the same laboratory. To ensure these

962 Journal of Speech, Language, and Hearing Research • Vol. 58 • 961–976 • June 2015
studies did not present data from the same participants, Table 1 presents a summary of each study’s partici-
authors were contacted by e-mail and the independence of pants with respect to their age, gender, maltreatment type,
samples was queried. In cases where data from the same par- SES, and residential status. This table shows variability be-
ticipants were reported in multiple studies, only one study tween studies with respect to the mean age of participants.
was used in the meta-analysis. The study selected for inclu- Substantial differences also exist between studies in rela-
sion in the meta-analysis was the one with the largest sample tion to the type of maltreatment experienced by the chil-
size. After applying the above criteria, 26 studies were dren. It should be noted that only two studies (Alessandri,
found that were included in the meta-analysis. The Preferred 1991; Barnett, Vondra, & Shonk, 1996) presented informa-
Reporting Items for Systematic Reviews and Meta-Analyses tion about the onset and/or duration of maltreatment. A
(PRISMA; Moher, Liberati, Tetzlaff, & Altman, 2009) limitation of this meta-analysis was we could not examine
flowchart that summarizes studies removed after application how the onset and/or duration of maltreatment might im-
of the aforementioned criteria is presented in Figure 1. pact children’s language skills. Nearly all studies examined

Figure 1. Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flowchart showing process for selecting articles.
SES = socioeconomic status.

Lum et al.: Language and Child Maltreatment 963

 964
Journal of Speech, Language, and Hearing Research • Vol. 58 • 961–976 • June 2015

Table 1. Summary of studies’ maltreatment and control group characteristics.

% of female Mean age Specific types of maltreatment

Sample size participants (years) reported in cases (%)c,d Exposed to
multiple types Study group Residential
Study Control Study Control Study Control Emotional Physical Sexual of maltreatment from low SES status of
Study group group group group group group Neglect abuse abuse abuse (% of cases) background? study group

Alessandri (1991) 15 15 46.7 46.7 4.3 4.6 53.0 53.0 33.0 33.0 66.7 Yes Family
Allen & Oliver (1982) 51.a 28 33.3 64.0 3.78 4.0 74.5 86.3 60.8 Yes Family
Ayoub et al. (2006) 75 66 Not reported 3.5–6b 3.5–6b 86.7 65.3 36.0 14.7 64.0 Yes Family
Barnett et al. (1996) 50 26 54.0 39.0 8.9 8.5 80.0 44.0 54.0 10.0 70.0 Yes Family
Beers & De Bellis (2002) 14 15 42.9 46.7 11.4 12.2 35.7 14.3 50.0 No Family
Cicchetti et al. (1993) 127 79 39.8 9.6 9.6 44.9 47.2 7.9 Yes Family
Coster et al. (1989) 20 20 45.0 40.0 2.6 75.0 50.0 25.0 5.0 50.0 Yes Family
De Bellis et al. (2009) 61 45 48.7 38.0 7.6 7.8 100.0 0.0 No Mixed: 34% in
foster care.
Eigsti & Cicchetti (2004) 19 14 47.4 50.0 4.8 5.0 94.7 84.2 52.6 47.4 Yes Family
Flores et al. (2005) 76 57 26.3 36.8 8.5 8.9 74.6 83.1 66.2 15.5 77.6 Yes Family
Fox et al. (1988) 30 10 43.3 50.0 5.1 5.4 66.7 33.3 Yes Foster care
Gregory & Beveridge 13 16 46.2 50.0 6.3 6.4 100.0 Yes Family
(1984)
Hoffman-Plotkin & 28.a 14 32.1 35.7 4.2 4.2 50.0 50.0 Yes Family
Twentyman (1984)
Nolin & Ethier (2007) 84.a 53 41.3 49.0 9.1 8.8 100.0 66.7 0.0 66.7 Yes Family (or
temporarily
in foster
care)
Perry et al. (1983) 21 21 42.9 47.6 4.6 5.5 100.0 Yes Family
Prasad et al. (2005) 19 19 47.4 57.9 2.9 2.5 100.0 0.0 Yes Foster
Rieder & Cicchetti (1989) 72 70 50.0 51.4 6.4 6.3 85.0 81.0 50.0 6.0 Yes Foster
Robinson et al. (2012) 70 70 51.7 55.7 7.4 7.6 80.0 9.0 4.0 Mixed: 34% Mixed: 54%
from low in foster
SES care.
Smith & Walden (1999) 15 15 60.0 47.0 4.4 4.6 27.0 19.0 27.0 27.0 46.7 Yes Not stated
Stipanicic et al. (2008) 11 11 54.5 54.5 7.3 7.5 100.0 No Mixed: 27% in
foster care.
Toth & Cicchetti (1996) 29 32 49.0 49.0 9.8 9.4 41.4 27.6 31.0 55.0 Yes Family
Toth et al. (2000) 56 37 29.0 35.0 3.7 3.9 23.2 48.2 28.6 66.0 Yes Family
Toth et al. (2011) 91 43 62.0 67.0 10.1 10.5 86.0 51.0 41.0 14.0 58.0 Yes Family
Trickett et al. (1991, 21 21 33.0 33.0 6.2 6.1 100.0 0.0 Yes Family
NIMH Sample)
Valentino et al. (2008) 96.a 128 35.4 48.4 10.0 10.0 88.0 75.0 52.0 27.0 81.2 Yes Family
Vondra et al. (1990) 12 11 42.0 42.0 5.6 4.9 83.0 50.0 33.0 Yes Family

Note. SES = socioeconomic status; NIMH = National Institute of Mental Health.

a
Combines data from multiple subgroups. bStudy reports age range only. cEmpty cell value denotes data not reported in study. dRow totals exceeding 100% indicate participants
experienced multiple forms of maltreatment.
language skills in children from low SES backgrounds. This maltreatment status of the child. We also coded whether
is to be expected given low SES is a risk factor for maltreat- studies excluded children with developmental problems.
ment (Brown, Cohen, Johnson, & Salzinger, 1998). Table 1 This was important to examine whether there might be
also presents the residential status of the maltreated chil- other reasons a study observed poorer language skills in
dren. That is, whether the maltreated children were residing maltreated children.
with a biological relative (parent, grandparent) or with a Table 3 summarizes the methodological quality of
nonbiological relative after adoption or foster care place- the studies in relation to selection, comparability, outcome,
ment. For most studies, children were residing with a bio- and whether the study excluded children with developmen-
logical relative, with three studies examining language tal disabilities. Cells with black circles indicate a criterion
functioning in maltreated children who were all in foster was met, and empty cells indicate the study did not meet
care. In three studies, the residential status of the maltreated the criterion or did not provide sufficient information to
children was mixed. determine whether a criterion had been met.
Table 2 presents the tests used in each study to mea- Table 3 shows nearly all of the 26 studies used gov-
sure language. The table shows receptive vocabulary re- ernment records to determine the presence of maltreatment.
ceived the most empirical attention having been investigated Also, nearly all studies matched participants with respect
in 19 studies. Only two studies were found that examined to age and gender. However, there is variability in other as-
expressive vocabulary skills. Receptive language skills were pects of study methodological quality. For instance, 12 of
investigated in eight studies. In three of these studies, recep- 26 studies indicated that participants were recruited from
tive language was measured by using a composite score that the same community. Fourteen studies used government re-
summed performance over multiple subtests. Expressive cords to determine maltreatment was not present in the con-
language skills were assessed in four studies. In two studies, trol group. Eleven studies used assessors who were blind to
expressive language was measured by using a composite the maltreatment status of the children. It is interesting to
score. In the remaining two studies (Coster, Gersten, Beeghly, note that eight studies did not meet the criteria for matching
& Cicchetti, 1989; Eigsti & Cicchetti, 2004), expressive on SES according to the Newcastle–Ottawa guidelines. For
language was measured by collecting spontaneous speech seven studies, this was because insufficient SES data were
samples from children playing with an adult. The utterances reported. In one study, the difference between the maltreat-
from the child are later coded for completeness and gram- ment and control groups exceeded a medium effect size,
matical complexity. even though the difference between groups on a SES indica-
tor was not found to be statistically significant.
Table 3 also presents the specific developmental
Study Quality problems that were ruled out in the children. Empty cells
The methodological quality of the studies included indicate this information was not reported. Thirteen of the
in the meta-analysis was assessed by using a modified ver- studies indicated participants were excluded from the re-
sion of the Newcastle–Ottawa Scale (Wells et al., 2000) for search if one or more developmental problem/s was pres-
cohort studies. This scale evaluates the extent studies re- ent. The most common developmental problems excluded
duce bias with respect to participant selection and compa- were neurological and intellectual impairments. Seven
rability of control and study groups. studies reported mean IQ scores for the maltreated and
Following the guidelines of the Newcastle–Ottawa control groups. In five of these studies, the mean IQ dif-
Scale (Wells et al., 2000), bias in participant selection is ference was 10 points or less. In two studies, the difference
reduced if a study indicates participants in the study and was 20 points or more.
control groups were recruited from the same communities,
and government/official legal records were used to deter-
Data Extraction Procedures and
mine the presence/absence of exposure. Comparability
of the study and control groups was also evaluated in rela- Effect Size Calculations
tion to how similar the groups were with respect to SES, From each study included in the meta-analysis, data
age, and gender. A study was considered to have met were extracted to compute an effect size and its variance.
this criterion if the difference between groups was found The effect size computed for this meta-analysis was a stan-
to be less than a medium effect size on one of the above dardized mean difference (SMD). This describes the dif-
variables. At this level, there is 67% or more overlap ference between the maltreated and control group on a
in a distribution of scores or cases (Cohen, 1988). The language measure in standard deviation units. SMD was
Newcastle–Ottawa Scale does not credit a study for dem- calculated so that positive values indicated the control
onstrating groups are comparable when only a statement is group performed better than the maltreated group on the
provided in the publication that groups did not differ on measure. That is, the maltreated group performed poorer
a matching variable. For this criterion to be met, results on the language measure relative to the control group.
from statistical tests and/or summary data must be re- Negative SMD values indicate that the maltreated group
ported. A study was considered to have reduced bias dur- performed better on the language measure relative to the
ing data collection if it was reported that the individual/s control group. SMD values approaching zero indicate a
undertaking the language assessment were blind to the smaller difference between groups. The general formula to

Lum et al.: Language and Child Maltreatment 965

 966
Journal of Speech, Language, and Hearing Research • Vol. 58 • 961–976 • June 2015

Table 2. Summary of language tests and language domain assessed in studies included in the meta-analysis.

Expressive Receptive Expressive Receptive

Study Language test/s used vocabulary vocabulary language language

Alessandri (1991) Expressive One-Word Picture Vocabulary Test (Gardner, 1979); Test of Auditory • •
Comprehension of Language–Revised (Carrow-Woolfold, 1985)
Allen & Oliver (1982) Auditory Verbal Ability and Auditory Comprehension Quotients from the Preschool • •
Language Scale (Zimmerman, Steiner, & Pond, 1979)
Ayoub et al. (2006) Peabody Picture Vocabulary Test–Revised (Dunn, Dunn, & Dunn, 1981) •
Barnett et al. (1996) Peabody Picture Vocabulary Test–Revised (Dunn et al., 1981) •
Cicchetti et al. (1993) Peabody Picture Vocabulary Test–Revised (Dunn et al., 1981) •
Coster et al. (1989) Mean Length of Utterance; Peabody Picture Vocabulary Test–Revised (Dunn et al., 1981) • •
De Bellis et al. (2009) Comprehension of Instructions subtest from the NEPSY (Korkman, Kirk, & Kemp, 1998); • •
Peabody Picture Vocabulary Test–III (Dunn, Dunn, & Williams, 1997)
Eigsti & Cicchetti (2004) Index of Productive Syntax (Scarborough, 1990); Peabody Picture Vocabulary • •
Test–Revised (Dunn et al., 1981)
Flores et al. (2005) Peabody Picture Vocabulary Test–Revised (Dunn et al., 1981) •
Fox et al. (1988) Miller-Yoder Language Comprehension Test–Clinical Edition (Miller & Yoder, 1984); • •
Peabody Picture Vocabulary Test–Revised (Dunn et al., 1981)
Gregory & Beveridge Naming Vocabulary from the British Ability Scale (National Foundation for Educational • •
(1984) Research, 1977); Sentence Comprehension Test (Wheldall, Mittler, & Hobsbaum, 1979)
Hoffman-Plotkin & Peabody Picture Vocabulary Test (edition not reported) •
Twentyman (1984)
Nolin & Ethier (2007) Comprehension of Instructions subtest from the NEPSY (Korkman et al., 1998) •
Perry et al. (1983) Peabody Picture Vocabulary Test (edition not reported) •
Prasad et al. (2005) Expressive and Receptive Language Composite Scores from the Sequenced Inventory of • •
Communication Development (Hendrick, Prather, & Tobin, 1995); Expressive and Receptive
Language Composite Scores from the Clinical Evaluation of Language Fundamentals
(Semel, Wiig, & Secord, 1995)
Rieder & Cicchetti (1989) Peabody Picture Vocabulary Test–Revised (Dunn et al., 1981) •
Robinson et al. (2012) Peabody Picture Vocabulary Test–Revised (Dunn et al., 1981) •
Smith & Walden (1999) Peabody Picture Vocabulary Test–Revised (Dunn et al., 1981) •
Stipanicic et al. (2008) Comprehension of Instructions subtest from the NEPSY (Korkman et al., 1998) •
Toth & Cicchetti (1996) Peabody Picture Vocabulary Test–Revised (Dunn et al., 1981) •
Toth et al. (2000) Peabody Picture Vocabulary Test–Revised (Dunn et al., 1981) •
Toth et al. (2011) Peabody Picture Vocabulary Test–Revised (Dunn et al., 1981) •
Trickett et al. (1991, Peabody Picture Vocabulary Test–Revised (Dunn et al., 1981) •
NIMH Sample)
Valentino et al. (2008) Peabody Picture Vocabulary Test–Revised (Dunn et al., 1981) •
Vondra et al. (1990) Peabody Picture Vocabulary Test–Revised (Dunn et al., 1981) •

Note. NEPSY = A Developmental Neuropsychological Assessment; NIMH = National Institute of Mental Health.
 Table 3. Study quality assessment using a modified version of the Newcastle–Ottawa Scale (Wells et al., 2000).

Selection
Outcome
Both groups Exposure Nonexposure
recruited determined by determined by Comparability Assessor blind to
from same using government using government maltreatment Exclusion criteria for Method used to
Study community records records SES Age Gender status developmental problems measure SES

Alessandri (1991) • • • • Autism; intellectual disability Household income;

(mean FSIQa: maltreated maternal education
group = 84, control group =
88.13); neurological impairments
Allen & Oliver (1982) • • • • • Neurological impairments Household income;
maternal education
Ayoub et al. (2006) • • • • Developmental delay Household Income
Barnett et al. (1996) • • • • • • Hollingshead Four
Factor Index
Beers & De Bellis • • • • • • Developmental delay; head injury; Hollingshead Four
(2002) intellectual impairments Factor Index
(mean FSIQb: maltreated
group = 105.71; control group =
113.20); psychiatric problems
Cicchetti et al. • • • • Household income;
(1993) maternal education
Coster et al. (1989) • • • Neurological impairments Household prestige;
maternal education
De Bellis et al. • • • • • Autism; intellectual impairments Hollingshead Four
(2009) (mean FSIQc: maltreated group = Factor Index
99.73, control group = 101.96);
medical illness; head injury;
low birth weight; neurological
impairment
Eigsti & Cicchetti • • • • • • Household income;
Lum et al.: Language and Child Maltreatment

(2004) maternal education

Flores et al. (2005) • • • • • Hollingshead Two
Factor Index
Fox et al. (1988) • • • Developmental delay; Recipients of
hearing impairments; government
neurological impairments income assistance
Gregory & • • • • Not stated
Beveridge (1984)
Hoffman-Plotkin & • • • • • • Intellectual impairment (mean Household income;
Twentyman (1984) FSIQa: maltreated group = maternal education
81, control group = 102)
Nolin & Ethier (2007) • • • • • • Birth problems; intellectual Household income
impairments (mean nonverbal
IQe: maltreated group =
103.35; control group =
106.35); neurological impairment
Perry et al. (1983) • • • • Hollingshead Two
Factor Index;
maternal education
967

(table continues)
 968
Journal of Speech, Language, and Hearing Research • Vol. 58 • 961–976 • June 2015

Table 3 (Continued).

Selection
Outcome
Both groups Exposure Nonexposure
recruited determined by determined by Comparability Assessor blind to
from same using government using government maltreatment Exclusion criteria for Method used to
Study community records records SES Age Gender status developmental problems measure SES
Prasad et al. (2005) • • • • • Developmental delay; gestational Hollingshead Four
age <32 weeks; neurological Factor Index
impairment
Rieder & Cicchetti • • Neurological impairment; physical Household income;
(1989) impairment maternal education
Robinson et al. • • • • • Intellectual impairments (mean Household income;
(2012) nonverbal IQd: maltreated group = maternal education
96.47, control group = 104.09)
Smith & Walden • • • • • Household income;
(1999) maternal education
Stipanicic et al. • • • • • • • Hearing impairments; intellectual Household income
(2008) impairments (mean FSIQa =
maltreated group = 86.36,
control group = 104.09); visual
impairments
Toth & Cicchetti • • Hollingshead Four
(1996) Factor Index
Toth et al. (2000) • • • • • Hollingshead Four
Factor Index;
maternal education
Toth et al. (2011) • • • • • Hollingshead Four
Factor Index
Trickett et al. (1991, • • Hollingshead Four
NIMH Sample) Factor Index
Valentino et al. • • • • • Hollingshead Four
(2008) Factor Index
Vondra et al. (1990) • • • • • • Hollingshead Four
Factor Index

Note. SES = socioeconomic status; FSIQ = Full Scale Intelligence Quotient; NIMH = National Institute of Mental health.
a
Estimated using the Stanford-Binet Intelligence Scale (Thorndike, Hagen, & Sattler, 1986); bEstimated using the Wechsler Intelligence Scale for Children (Wechsler, 1991); cEstimated
using the Wechsler Abbreviated Scale of Intelligence (Wechsler, 1999); dEstimated using the Kaufman Brief Intelligence Test (Kaufman & Kaufman, 1990); eAverage of Picture Completion
and Block Design subtests from the Wechsler Intelligence Scale for Children (scaled scored transformed to standard score to permit comparison with other test results).
compute SMD and its variance are presented in (1) and (2), reflect sampling error only (Borenstein, Hedges, Higgins,
respectively. & Rothstein, 2011). A random-effects model was chosen
because both systematic influences and sampling error are
likely to lead to differences between study effect sizes. Poten-
X Control
X  Maltreated
SMD ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi (1) tial systematic influences might be differences in study’s
ðnControl
1ÞSDControl þðnMaltreated
1ÞSDMaltreated
2 2
participant characteristics (e.g., Table 1) or methodological
nControlþ nMaltreated
2
quality (e.g., Table 3).
The I2 statistic was also computed for each meta-
analysis. This measures the amount of differences between
nControl þ nMaltreated SMD2 study level effect sizes due to systematic influences or non-
VarðSMDÞ ¼ þ random error. Thus an I2 value of 80% indicates 80% of
nControl  nMaltreated 2ðnControl þ nMaltreated Þ
(2) the heterogeneity (or differences in study level effect sizes)
is attributable to systematic or nonrandom error and 20%
to sampling error. As a guideline, Higgins, Thompson,
Summary data or results from statistical tests were Deeks, and Altman (2003) suggest that values of 25%,
extracted from each study so that separate values for SMD 50%, and 75% correspond to low, moderate, and high levels
and Var(SMD) could be computed. For most studies, these of heterogeneity, respectively. In cases where moderate to
values were computed by using mean and standard devia- high levels of heterogeneity were found, random effects
tions from the language tests that were reported in the re- meta-regression (Thompson & Higgins, 2002) was used to
sults section (e.g., Alessandri, 1991). For three studies, these investigate whether the participant or methodological char-
data were not available (Barnett et al., 1996; Cicchetti & acteristics of a study that were presented in Tables 1 or 3
Rogosch, 1997; Cicchetti, Rogosch, Lynch, & Holt, 1993), predicted effect sizes.
and results from statistical tests comparing the maltreated
and control group on the language measure of interest were
used to compute the effect size and variance. The conver- Results
sion of extracted data to SMD and Var(SMD) was under- Receptive Vocabulary
taken by using Comprehensive Meta-Analysis software
(Borenstein, Hedges, Higgins, & Rothstein, 2005). For five A summary of each study’s effect size examining re-
studies (Allen & Oliver, 1982; De Bellis, Hooper, Spratt, ceptive vocabulary in maltreated children and the weighted
& Woolley, 2009; Hoffman-Plotkin & Twentyman, 1984; average effect size are presented as a forest plot in Figure 2.
Nolin & Ethier, 2007; Valentino, Cicchetti, Rogosch, & Also shown in Figure 2 (and Figures 3 & 4) is the weight
Toth, 2008), separate data were reported for two or more assigned to each study by the random-effects model expressed
maltreated groups. For these studies, data from separate as a percentage. Studies with a higher percentage made a
groups were combined to create a single maltreatment group. larger contribution to the average. The average effect size
This was achieved by computing a weighted average and computed for receptive vocabulary is .463 and significant
standard deviation for the language scores for the groups (p < .001). Calculation of the I2 statistic yielded a value of
(Higgins & Green, 2008). 62.33% indicating moderate to high levels of heterogeneity.
It is noted that the difference in average IQ between mal-
treated and nonmaltreated groups for one study included in
Meta-Analytic Procedures this meta-analysis was greater than 20 points (Hoffman-
A weighted average effect size was computed that Plotkin & Twentyman, 1984). To examine whether group
summarized the differences between maltreated and con- differences in IQ were influencing results, the average effect
trol groups with respect to receptive vocabulary, receptive size was recomputed after removing this study. After recom-
language, and expressive language skills. An average effect puting the average effect size, a significant result was still
size for expressive vocabulary was not computed because observed (SMD = 0.400; 95% confidence interval [0.258,
only two studies were found that investigated this aspect of 0.541]; p < .001). According to Cohen’s (1988) taxonomy,
language. We tested whether the average effect size value this value approximates a medium effect size (i.e., d = .5).
was significantly different from zero. For these and other The positive value of the effect size indicates that, on aver-
significance tests, an a level of 0.05 (two-tails) was used. age, studies are finding poorer performance by maltreated
Comprehensive Meta-Analysis software was used to com- children compared with control groups. To examine poten-
pute average effect sizes (Borenstein et al., 2005). tial publication bias, the fail-safe N (Orwin, 1983) was com-
A random-effects model was used to weight individual puted. This value indicates the number of missing studies
studies to compute each average effect size (Hedges, 1983). required to return p to .05. This analysis indicated that
By using this model, it is assumed that differences between 345 nonsignificant studies would be required to return a
study level effect sizes reflect the contribution of within-study nonsignificant effect size.
error/sampling error and between-study error/systematic in- Random effects meta-regression was used to investi-
fluences. This can be contrasted with a fixed-effects model, gate potential sources of heterogeneity. In these analyses,
which assumes differences between study effect sizes we tested whether a study’s participant characteristics

Lum et al.: Language and Child Maltreatment 969

Figure 2. Study and average effect sizes from studies investigating receptive vocabulary. SMD = standardized mean difference; CI,
confidence interval; NIMH = National Institute of Mental Health.

(presented in Table 1) or methodological quality (presented Studies with more than 50% of participants reported to be
in Table 3) predicted individual effect sizes (presented in residing with a biological relative were coded as 1 (1 = par-
Figure 2). There were an insufficient number of studies to ticipants residing with family). An additional analysis ex-
test all predictor variables in a single model (Borenstein amined whether the percentage of maltreated children who
et al., 2011). The influence of each predictor variable on ef- had been neglected predicted effect sizes. It was not possi-
fect sizes was tested in separate models. The predictor vari- ble to examine whether other types of maltreatment pre-
ables tested were mean age of participants, percentage of dicted effect sizes. This was because maltreatment data
female participants, whether participants were from a low were not reported in a sufficient number of studies that ex-
SES background (dummy coded where 1 = participants amined receptive vocabulary.
from low SES, 0 = participants from nonlow SES), and the The influence of study quality (presented in Table 3)
residential status of the maltreatment group (dummy coded on effect sizes was also examined in the meta-regression
where 1 = participants residing with family, 0 = foster care). analyses. The predictor variables tested were whether

Figure 3. Study and average effect sizes from studies investigating expressive language. SMD = standardized mean difference; CI,
confidence interval.

970 Journal of Speech, Language, and Hearing Research • Vol. 58 • 961–976 • June 2015
Figure 4. Study and average effect sizes from studies investigating receptive language. SMD = standardized mean difference; CI, confidence
interval.

groups were recruited from the same community, matched 9 (p = .061), and 10 (p = .080) indicate that a larger difference
on age, gender, or SES, and if government records were was observed when participants were recruited from the same
used to determine the presence and absence of maltreat- community and groups were matched on age and gender.
ment. Also tested was whether the use of blinding proce-
dures and excluding participants with developmental
disabilities predicted effect sizes. Each of the aforemen- Expressive Vocabulary
tioned variables was dummy coded for the analyses. A The systematic search of the literature only found
value of 1 was used if the study met the criterion and two studies (Alessandri, 1991; Gregory & Beveridge, 1984)
0 if the criterion was not met. The results of the meta- that examined expressive vocabulary in maltreated chil-
regression analyses are presented in Table 4. dren. Both studies reported poorer expressive vocabulary
Overall, no significant predictors of effect sizes were skills in maltreated children. However, the difference was
found. However, we note that four models approached not significant, and the observed effect sizes were small to
statistical significance (ps < .100). First, Model 4 indicates medium (Alessandri [1991] SMD = .327, p = .373; Gregory
studies observed a larger difference between maltreated & Beveridge [1984] SMD = .278, p = .458). Because there
children and controls when participants were not from were only two studies examining expressive vocabulary, it
low SES backgrounds (p = .091). Models 6 (p = .085), was not feasible to submit their results to a meta-analysis.

Table 4. Results from meta-regression examining predictors of receptive vocabulary effect sizes.

Model No. Predictor Variable k df R2 QModel QResidual β p

1 Mean age of participants 19 1,17 0.079 1.8153 21.1703 −0.281 .178

2 Percentage of female participants 18.a 1,16 0.0237 0.5093 20.9578 −0.154 .476
3 Percentage of participants exposed to neglect 17.b 1,15 0.0199 0.4098 20.1933 0.141 .522
4 Participants from low SES background 19 1,17 0.1245 2.8624 20.1233 −0.3529 .091
5 Residential status of study group 18.c 1,16 0.0125 0.277 21.798 −0.112 .599
6 Maltreated and control groups recruited from same community 19 1,17 0.1288 2.9599 20.0257 0.3588 .085
7 Maltreatment and nonmaltreatment determined by government records 19 1,17 0.0028 0.0634 22.9222 0.0525 .801
8 Maltreatment and control groups matched on maternal education 19 1,17 0.0727 1.6702 21.3154 −0.2696 .196
9 Groups matched on age 19 1,17 0.1525 3.5051 19.4805 0.3905 .061
10 Groups matched on gender 19 1,17 0.1337 3.0735 19.9122 0.3657 .080
11 Groups matched on SES 19 1,17 0.0561 1.2898 21.6959 0.2369 .256
12 Blinding procedure used during data collection 19 1,17 0.0088 0.2017 22.7839 0.0937 .653
13 Study excludes children with developmental disabilities 19 1,17 0.0317 0.7294 22.2562 0.1781 .393

Note. SES = socioeconomic status.

a
One study did not report the number of female participants and was included in this analysis. bTwo studies did not report the percentage of
children who had been neglected and were excluded from this analysis. cOne study was not included in this analysis because residential
status of the children was not reported.

Lum et al.: Language and Child Maltreatment 971

Expressive Language (p < .100). This was the model testing the influence of match-
ing groups by using SES (Model 8; p = .079). The direc-
A forest plot showing effect sizes for studies that investi-
tion of the beta value (which was positive) indicates studies
gated expressive language in maltreated children is presented
that matched groups on SES observed larger effect sizes.
in Figure 3. The average effect size was .860 and significant
(p < .001). That is, on average, studies are reporting that mal-
treated children are performing around .9 SD lower on ex- Discussion
pressive language tests compared with nonmaltreated children
This review article used meta-analysis to summarize
from similar SES backgrounds. Calculation of the fail-safe
the literature investigating language skills in maltreated
N indicated that 27 unpublished studies with nonsignificant
children. The analyses consistently showed language skills
results would be required to return a to .05. The magnitude
are poorer in maltreated children when compared with
of the effect size is large according to Cohen’s (1988) guide-
nonmaltreated children from similar SES backgrounds. In
lines. There was little variability in study level effect sizes.
particular, statistically significant average effect sizes, indi-
The I 2 statistic for this set of effect sizes was found to be
cating poorer language skills in maltreated children, were
less than 1%. Thus all of the heterogeneity between effect
observed when combining studies that measured expressive
sizes for these studies can be attributable to random error.
language, receptive language, and receptive vocabulary
skills. An insufficient number of studies were identified to
permit meaningful meta-analysis of expressive vocabulary.
Receptive Language The overall findings from this meta-analysis showing
Effect sizes for studies examining receptive language poorer language skills in maltreated children is in line with
in maltreated children are presented in a forest plot in conclusions from a previous systematic review (Veltman
Figure 4. The average weighted effect size for studies ex- & Browne, 2001) and narrative reviews of the literature
amining receptive language in maltreated children and con- (Hwa-Froelich, 2012; Law & Conway, 1992). This review
trols was found to be .528 (p < .001). The mean Full Scale article builds on past work by providing an estimate of
Intelligence Quotient difference between maltreated and the effect size that measures the association between mal-
nonmaltreated children for one of the studies in this meta- treatment and poor language skills. Stated in standardized
analysis differed more than 20 points (Stipanicic, Nolin, scores (i.e., a distribution that has a mean of 100 and
Fortin, & Gobeil, 2008). When this study was removed and standard deviation of 15), the average finding is that for
the average effect size recalculated, a significant positive av- receptive vocabulary, maltreated children are scoring
erage effect size that approached a medium effect size was seven points lower than controls of comparable SES. For
still observed (SMD = .467; 95% confidence interval [.161, expressive language skills, the average finding is that mal-
.774]; p = .003). Calculation of the fail-safe N indicated that treated children are 13 points lower and eight points lower
24 unpublished studies with nonsignificant results would for receptive language.
return a to .05. Figure 4 does indicate variability in study Another finding to emerge from this meta-analysis
level effect sizes. Two observed very large effect sizes be- concerns the heterogeneity or differences between individual
tween maltreated and control groups (SMD > 1) and two study effect sizes. Around 60% of the differences between
studies found that the control group performed poorer effect sizes derived from studies investigating receptive vo-
on the comprehension test compared with the maltreated cabulary and receptive language were not attributable to
group. Calculation of the I 2 indicated that 62.5% of the het- sampling or random error. This suggests there are one or
erogeneity between studies could be attributed to system- more systematic influences impacting study findings. Using
atic influences or nonrandom error. meta-regression, we tested whether the characteristics of a
Meta-regression analyses examined whether participant study’s participants or methodology might be influencing
or methodological factors might account for the heterogene- results. Although statistically significant predictors of effect
ity in receptive language effect sizes. Most of the predictor sizes were not found, five meta-regression models were close
variables used to examine influences on receptive vocabulary to statistical significance (ps < .100) and reveal potentially
effect sizes were also used in these analyses (see Table 4). informative trends in the literature.
However, it was not possible to test the influence of recruiting First, one model (Table 4, Model 4, p = .091) poten-
children from the same community and whether participants tially indicates the difference between maltreated and non-
were matched on gender. This was because there was no vari- maltreated groups on a measure of receptive vocabulary is
ability in these predictor variables; all studies examining re- larger when children were not from a low SES background.
ceptive language recruited children from the same community In explaining this trend, one possibility is that maltreatment
and matched on the basis of gender. Also, it was not possible has an equally detrimental impact on children’s language
to test the influence of specific types of maltreatment, includ- skills, irrespective of socioeconomic background. However,
ing neglect, because this information was reported in less than in nonmaltreated samples, lower language skills tend to be
half of the studies examining receptive language. observed when children are only from disadvantaged socio-
Results from the meta-regression analyses are presented economic backgrounds (Hart & Risley, 1995; Hoff & Tian,
in Table 5. None of the models were found to be significant. 2005; Hoff-Ginsberg, 1998). Thus, in studies where children
However, one model approached statistical significance were from nondisadvantaged SES backgrounds, it appears

972 Journal of Speech, Language, and Hearing Research • Vol. 58 • 961–976 • June 2015
Table 5. Results from meta-regression examining predictors of receptive language effect sizes.

Model No. Predictor Variable k df R2 QModel QResidual β p

1 Mean age of participants 9 1,7 0.0639 0.5946 8.7069 −0.2528 .441

2 Proportion of female participants 9 1,7 0.0153 0.1428 9.1587 0.1239 .706
3 Participants from low SES background 9 1,7 0.0987 0.9182 8.3833 −0.3142 .338
4 Residential status of study group 9 1,7 0.2413 2.2449 7.0566 −0.4913 .134
5 Maltreatment and nonmaltreatment determined by government records 9 1,7 0.0164 0.1527 9.1488 −0.1281 .696
6 Maltreatment and control groups matched on maternal education 9 1,7 0.2324 2.1619 7.1396 0.4821 .142
7 Groups matched on age 9 1,7 0.2703 2.5142 6.7873 0.5199 .113
8 Groups matched on SES 9 1,7 0.3316 3.0844 6.2171 0.5759 .079
9 Blinding procedure used during data collection 9 1,7 0.1269 1.1801 8.1214 −0.3562 .277
10 Study excludes children with developmental disabilities 9 1,7 0.2429 2.2593 7.0422 0.4928 .133

Note. SES = socioeconomic status.

that only maltreated children obtained lower language The results of this meta-analysis have implications
scores leading to a large effect size (or difference between for applied settings. This review article provides strong evi-
groups). It is interesting to note that the corresponding dence that children exposed to maltreatment may have dif-
model examining this same variable for receptive language ficulties understanding and producing language. Thus, for
(Table 5, Model 3), whilst not statistically significant, was professionals working with maltreated children, this review
comparable with respect to the R2 value and direction, as highlights communicative difficulties that will most likely
well as the magnitude of the beta-value. From this perspec- be present in this group. It is not clear from this meta-
tive, low SES might be masking the impact of maltreatment analysis whether language skills are disproportionally af-
on children’s language skills. This proposal will need to be fected compared with other cognitive abilities in maltreated
examined in future research. children. In this review article, we provide some evidence
Second, models testing whether maltreated and non- that the language problems in maltreated children may not
maltreated groups were matched on age (Table 4, Model 9), be related to other developmental problems or explained
gender (Table 4, Model 10), SES background (Table 5, by IQ. The meta-regression analyses indicated that whether
Model 8), and recruited from the same community (Table 4, a study excluded children with a developmental problem
Model 6) approached statistical significance. Each of did not predict effect sizes (Table 4, Model 13; Table 5,
these models indicated studies matching on these variables Model 10). Also, removing studies from the meta-analysis
observed a larger effect size or difference between mal- that had a large discrepancy in IQ between maltreated and
treated and control groups. This might arise because stud- control groups had a negligible impact on the average effect
ies that controlled for nuisance variables might be able to size. However, the influence of co-occurring developmental
measure the association between maltreatment and lan- disabilities on maltreated children’s language skills will
guage with smaller levels of interference. Thus the meta- need to be clarified in future research. A limitation with the
regression analyses presented in this review article reiterate analyses undertaken in this review article was that informa-
the need for future studies to control for confounding par- tion about children’s IQ was only reported in seven of the
ticipant characteristics where possible. 26 studies. It is also possible that maltreated children pre-
Failure to find significant meta-regression models sented with cognitive deficits outside of IQ. Research has
also indicates that reliable predictor/s of language out- found that maltreated children perform poorer on neuro-
comes in maltreated samples have yet to be identified. As psychological tests that assess attention, learning, and mem-
noted earlier, chronic stress experienced as a consequence ory (De Bellis et al., 2009). Thus it could be that poor
of maltreatment is known to disrupt brain development language skills might be only one of a number of abilities
(De Bellis et al., 2002; De Bellis & Kuchibhatla, 2006). vulnerable to the effects of maltreatment.
Also, there is evidence to suggest that maltreated children A strength of meta-analysis and meta-regression is
do not receive sufficient linguistic stimulation, or experi- data from multiple studies are pooled producing more reli-
ence sufficient meaningful, child-focused interactions with able and precise effect estimates compared with results
caregivers (Allen & Wasserman, 1985; Eigsti & Cicchetti, from an individual study (Borenstein et al., 2011; Hunter,
2004; Kavanagh et al., 1988; Wasserman et al., 1983). Schmidt, & Jackson, 1982). However, it is important to
Given this, maltreatment type and history may be key pre- note limitations with this approach as well. First, this
dictors of language outcomes. A limitation of this review meta-analysis is unable to determine how maltreatment
article was these variables could not be quantitatively ex- negatively impacts language development. For instance, all
amined because insufficient data were reported in primary studies analyzed were cross-sectional in design. The devel-
studies. Thus one avenue for future research would be to opmental trajectory of expressive and receptive language
examine in detail how maltreatment history and type im- skills following maltreatment will need to be investigated in
pacts children’s language skills. longitudinal research. Second, caution is required when

Lum et al.: Language and Child Maltreatment 973

interpreting meta-regression results. This is because other Bradley, R. H., & Corwyn, R. F. (2002). Socioeconomic status and
study characteristics may be highly correlated with one or child development. Annual Review of Psychology, 53, 371–399.
more predictor variables (Thompson & Higgins, 2002). As Brown, D. S., Fang, X., & Florence, C. S. (2011). Medical costs at-
tributable to child maltreatment: A systematic review of short-
a consequence, associations between a predictor variable
and long-term effects. American Journal of Preventive Medicine,
and effect size might reflect the influence of another vari- 41, 627–635.
able. Given this, it is important that results from meta- Brown, J., Cohen, P., Johnson, J. G., & Salzinger, S. (1998). A
regression are investigated by additional primary research. longitudinal analysis of risk factors for child maltreatment:
Findings of a 17-year prospective study of officially recorded
and self-reported child abuse and neglect. Child Abuse &
Conclusions Neglect, 22, 1065–1078.
Carrow-Woolfolk, E. (1985). Test for Auditory Comprehension of
This meta-analysis summarized results from research Language–Revised. Allen, TX: DLM Teaching Resources.
investigating language skills in maltreated children. Based Cicchetti, D., & Rogosch, F. A. (1997). The role of self-organization
on the findings from 26 studies that represent data from in the promotion of resilience in maltreated children. Develop-
1,176 maltreated children and 936 controls, the analyses ment and Psychopathology, 9, 797–815.
showed poorer language skills in maltreated children, com- *Cicchetti, D., Rogosch, F. A., Lynch, M., & Holt, K. D. (1993).
pared with matched SES controls. Additional research is Resilience in maltreated children: Processes leading to adaptive
warranted to better understand how maltreatment disrupts outcome. Development and Psychopathology, 5, 629–647.
language development and to determine the best methods Cohen, J. (1988). Statistical power analysis for the behavioral
sciences. London, England: Routledge.
for improving language skills and developmental outcomes *Coster, W. J., Gersten, M. S., Beeghly, M., & Cicchetti, D. (1989).
in this group. Communicative functioning in maltreated toddlers. Develop-
mental Psychology, 25, 1020–1029.
De Bellis, M. D. (2001). Developmental traumatology: The psycho-
Acknowledgments biological development of maltreated children and its impli-
This research was supported by the Australian Research cations for research, treatment, and policy. Development and
Council (Grant DP1095509). We thank Gillian Clark and Bronwen Psychopathology, 13, 539–564.
Manger for their assistance with collecting and reviewing articles *De Bellis, M. D., Hooper, S. R., Spratt, E. G., & Woolley, D. P.
for this article. (2009). Neuropsychological findings in childhood neglect and
their relationships to pediatric PTSD. Journal of the Interna-
tional Neuropsychological Society, 15, 868–878.
De Bellis, M. D., Keshavan, M. S., Shifflett, H., Iyengar, S., Beers,
References S. R., Hall, J., & Moritz, G. (2002). Brain structures in pediatric
*Denotes used in meta-analysis maltreatment-related posttraumatic stress disorder: A socio-
*Alessandri, S. M. (1991). Play and social behavior in maltreated demographically matched study. Biological Psychiatry, 52,
preschoolers. Development and Psychopathology, 3, 191–205. 1066–1078.
doi:10.1017/S0954579400000079 De Bellis, M. D., & Kuchibhatla, M. (2006). Cerebellar volumes in
Allen, R., & Wasserman, G. A. (1985). Origins of language delay pediatric maltreatment-related posttraumatic stress disorder.
in abused infants. Child Abuse & Neglect, 9, 335–340. Biological Psychiatry, 60, 697–703.
*Allen, R. E., & Oliver, J. M. (1982). The effects of child mal- De Brito, S. A., Viding, E., Sebastian, C. L., Kelly, P. A.,
treatment on language development. Child Abuse & Neglect, 6, Mechelli, A., Maris, H., & McCrory, E. J. (2013). Reduced
299–305. orbitofrontal and temporal grey matter in a community sam-
Anderson, C. M., Teicher, M. H., Polcari, A., & Renshaw, P. F. ple of maltreated children. The Journal of Child Psychology
(2002). Abnormal T2 relaxation time in the cerebellar vermis and Psychiatry, 54, 105–112.
of adults sexually abused in childhood: Potential role of the Dunn, L. M., Dunn, L. L., & Dunn, D. M. (1981). Peabody Picture
vermis in stress-enhanced risk for drug abuse. Psychoneuro- Vocabulary Test–Revised. Circle Pines, MN: AGS.
endocrinology, 27, 231–244. Dunn, L. M., Dunn, L. M., & Williams, K. T. (1997). Peabody
*Ayoub, C. C., O’Connor, E., Rappolt-Schlichtmann, G., Fischer, Picture Vocabulary Test–III. Circle Pines, MN: AGS.
K. W., Rogosch, F. A., Toth, S. L., & Cicchetti, D. (2006). Cog- *Eigsti, I.-M., & Cicchetti, D. (2004). The impact of child mal-
nitive and emotional differences in young maltreated children: treatment on expressive syntax at 60 months. Developmental
A translational application of dynamic skill theory. Develop- Science, 7, 88–102.
ment and Psychopathology, 18, 679–706. Éthier, L. S., Lemelin, J.-P., & Lacharité, C. (2004). A longitudi-
*Barnett, D., Vondra, J. I., & Shonk, S. M. (1996). Self-perceptions, nal study of the effects of chronic maltreatment on children’s
motivation, and school functioning of low-income maltreated behavioral and emotional problems. Child Abuse & Neglect,
and comparison children. Child Abuse & Neglect, 20, 397–410. 28, 1265–1278.
*Beers, S. R., & De Bellis, M. D. (2002). Neuropsychological *Flores, E., Cicchetti, D., & Rogosch, F. A. (2005). Predictors of
function in children with maltreatment-related posttraumatic resilience in maltreated and nonmaltreated Latino children.
stress disorder. The American Journal of Psychiatry, 159, Developmental Psychology, 41, 338–351.
483–486. *Fox, L., Long, S. H., & Langlois, A. (1988). Patterns of language
Borenstein, M., Hedges, L., Higgins, J., & Rothstein, H. (2005). comprehension deficit in abused and neglected children. Journal
Comprehensive meta-analysis: Version 2. Englewood, NJ: Biostat. of Speech and Hearing Disorders, 53, 239–244.
Borenstein, M., Hedges, L. V., Higgins, J. P., & Rothstein, H. R. Garbarino, J., & Crouter, A. (1978). Defining the community
(2011). Introduction to meta-analysis. West Sussex, England: context for parent-child relations: The correlates of child mal-
Wiley. treatment. Child Development, 49, 604–616.

974 Journal of Speech, Language, and Hearing Research • Vol. 58 • 961–976 • June 2015
Garbarino, J., & Kostelny, K. (1992). Child maltreatment as a Law, J., & Conway, J. (1992). Effect of abuse and neglect on the
community problem. Child Abuse & Neglect, 16, 455–464. development of children’s speech and language. Developmental
Gardner, M. (1979). Expressive One-Word Picture Vocabulary Medicine & Child Neurology, 34, 943–948.
Test. Novato, CA: Academic Therapy Publications. McLoyd, V. C. (1998). Socioeconomic disadvantage and child
Gilbert, R., Widom, C. S., Browne, K., Fergusson, D., Webb, E., development. American Psychologist, 53, 185–204.
& Janson, S. (2009). Burden and consequences of child mal- Miller, J. F., & Yoder, D. E. (1984). Miller-Yoder Language Com-
treatment in high-income countries. The Lancet, 373, 68–81. prehension Test–Clinical Edition. Baltimore, MD: University
*Gregory, H. M., & Beveridge, M. (1984). The social and educa- Park Press.
tional adjustment of abused children. Child Abuse & Neglect, 8, Moher, D., Liberati, A., Tetzlaff, J., & Altman, D. G. (2009).
525–531. Preferred reporting items for systematic reviews and meta-
Hart, B., & Risley, T. R. (1995). Meaningful differences in the analyses: The PRISMA statement. PLOS Medicine, 6,
everyday experience of young American children. Baltimore, e1000097. doi:10.1371/journal.pmed.1000097
MD: Brookes. National Foundation for Educational Research. (1977). British
Hart, H., & Rubia, K. (2012). Neuroimaging of child abuse: A Ability Scales. Slough, England: National Foundation for
critical review. Frontiers in Human Neuroscience, 6, 1–24. Educational Research.
doi:10.3389/fnhum.2012.00052 *Nolin, P., & Ethier, L. (2007). Using neuropsychological profiles
Heath, A. F., Colton, M. J., & Aldgate, J. (1994). Failure to es- to classify neglected children with or without physical abuse.
cape: A longitudinal study of foster children’s educational Child Abuse & Neglect, 31, 631–643.
attainment. British Journal of Social Work, 24, 241–260. Noll, J. G., Shenk, C. E., Yeh, M. T., Ji, J., Putnam, F. W., &
Hedges, L. V. (1983). A random effects model for effect sizes. Trickett, P. K. (2010). Receptive language and educational
Psychological Bulletin, 93, 388–395. attainment for sexually abused females. Pediatrics, 126,
Hendrick, D. L., Prather, E. M., & Tobin, A. R. (1995). Sequenced e615–e622. doi:10.1542/peds.2010-0496
Inventory of Communication Development–Revised. Seattle: Orwin, R. G. (1983). A fail-safe N for effect size in meta-analysis.
University of Washington Press. Journal of Educational Statistics, 8, 157–159.
Higgins, J. P., & Green, S. (2008). Cochrane handbook for system- Overpeck, M. D., Brenner, R. A., Trumble, A. C., Trifiletti, L. B.,
atic reviews of interventions (Vol. 5). Chichester, England: & Berendes, H. W. (1998). Risk factors for infant homicide in the
Wiley Online Library. United States. New England Journal of Medicine, 339, 1211–1216.
Higgins, J. P., Thompson, S. G., Deeks, J. J., & Altman, D. G. Palusci, V. J., & Covington, T. M. (2014). Child maltreatment
(2003). Measuring inconsistency in meta-analyses. British deaths in the U.S. national child death review case reporting
Medical Journal, 327, 557–560. system. Child Abuse & Neglect, 38, 25–36. doi:10.1016/j.
Hoff, E., & Tian, C. (2005). Socioeconomic status and cultural chiabu.2013.08.014
influences on language. Journal of Communication Disorders, *Perry, M. A., Doran, L. D., & Wells, E. A. (1983). Developmen-
38, 271–278. tal and behavioral characteristics of the physically abused
Hoff-Ginsberg, E. (1998). The relation of birth order and socio- child. Journal of Clinical Child Psychology, 12, 320–324.
economic status to children’s language experience and language *Prasad, M. R., Kramer, L. A., & Ewing-Cobbs, L. (2005). Cogni-
development. Applied Psycholinguistics, 19, 603–630. tive and neuroimaging findings in physically abused pre-
*Hoffman-Plotkin, D., & Twentyman, C. T. (1984). A multimodal schoolers. Archives of Disease in Childhood, 90, 82–85.
assessment of behavioral and cognitive deficits in abused and Rhodes, A. E., Boyle, M. H., Bethell, J., Wekerle, C., Tonmyr, L.,
neglected preschoolers. Child Development, 55, 794–802. Goodman, D., . . . Manion, I. (2012). Child maltreatment and
Hunter, J. E., Schmidt, F. L., & Jackson, G. B. (1982). Meta- repeat presentations to the emergency department for suicide-
analysis. Beverly Hills, CA: Sage. related behaviors. Child Abuse & Neglect, 37, 139–149.
Hwa-Froelich, D. (2012). Childhood maltreatment and commu- *Rieder, C., & Cicchetti, D. (1989). Organizational perspective on
nication development. Perspectives on School-Based Issues, 13, cognitive control functioning and cognitive-affective balance in
43–53. maltreated children. Developmental Psychology, 25, 382–393.
Jaffee, S. R., & Maikovich-Fong, A. K. (2011). Effects of chronic *Robinson, L. R., Boris, N. W., Heller, S. S., Rice, J., Zeanah,
maltreatment and maltreatment timing on children’s behavior C. H., Clark, C., & Hawkins, S. (2012). The good enough
and cognitive abilities. The Journal of Child Psychology and home? Home environment and outcomes of young maltreated
Psychiatry, 52, 184–194. children. Child & Youth Care Forum, 41, 73–88.
Kavanagh, K. A., Youngblade, L., Reid, J. B., & Fagot, B. I. Scarborough, H. S. (1990). Index of productive syntax. Applied
(1988). Interactions between children and abusive versus con- Psycholinguistics, 11, 1–22.
trol parents. Journal of Clinical Child Psychology, 17, 137–142. Semel, E., Wiig, E. H., & Secord, W. A. (1995). Clinical Evalua-
doi:10.1207/s15374424jccp1702_5 tion of Language Fundamentals–Third Edition. San Antonio,
Kaufman, A. S., & Kaufman, N. L. (1990). Kaufman Brief Intelli- TX: The Psychological Corporation.
gence Test. Circle Pines, MN: AGS. *Smith, M., & Walden, T. (1999). Understanding feelings and
Keyes, K. M., Eaton, N. R., Krueger, R. F., McLaughlin, K. A., coping with emotional situations: A comparison of maltreated
Wall, M. M., Grant, B. F., & Hasin, D. S. (2012). Childhood and nonmaltreated preschoolers. Social Development, 8, 93–116.
maltreatment and the structure of common psychiatric dis- doi:10.1111/1467-9507.00082
orders. The British Journal of Psychiatry, 200, 107–115. Spaulding, T. J., Plante, E., & Farinella, K. A. (2006). Eligibility
Korkman, M., Kirk, U., & Kemp, S. (1998). NEPSY: A Develop- criteria for language impairment: Is the low end of normal al-
mental Neuropsychological Assessment. San Antonio, TX: ways appropriate? Language, Speech, and Hearing Services in
The Psychological Corporation. Schools, 37, 61–72.
Kuhl, P. K., Williams, K. A., Lacerda, F., Stevens, K. N., & *Stipanicic, A., Nolin, P., Fortin, G., & Gobeil, M. F. (2008). Com-
Lindblom, B. (1992). Linguistic experience alters phonetic per- parative study of the cognitive sequelae of school-aged victims
ception in infants by 6 months of age. Science, 255, 606–608. of shaken baby syndrome. Child Abuse & Neglect, 32, 415–428.

Lum et al.: Language and Child Maltreatment 975

Teicher, M. H., Andersen, S. L., Polcari, A., Anderson, C. M., developmental sequelae of physical child abuse. Developmental
Navalta, C. P., & Kim, D. M. (2003). The neurobiological Psychology, 27, 148–158.
consequences of early stress and childhood maltreatment. Twardosz, S., & Lutzker, J. R. (2010). Child maltreatment and
Neuroscience & Biobehavioral Reviews, 27, 33–44. the developing brain: A review of neuroscience perspectives.
Teicher, M. H., Anderson, C. M., & Polcari, A. (2012). Child- Aggression and Violent Behavior, 15, 59–68.
hood maltreatment is associated with reduced volume in *Valentino, K., Cicchetti, D., Rogosch, F. A., & Toth, S. L.
the hippocampal subfields CA3, dentate gyrus, and subiculum. (2008). Memory, maternal representations, and internalizing
Proceedings of the National Academy of Sciences, 109, symptomatology among abused, neglected, and nonmaltreated
E563–E572. children. Child Development, 79, 705–719.
Teicher, M. H., Ito, Y., Glod, C. A., Andersen, S. L., Dumont, N., Veltman, M. W. M., & Browne, K. D. (2001). Three decades
& Ackerman, E. (1997). Preliminary evidence for abnormal of child maltreatment research: Implications for the school
cortical development in physically and sexually abused chil- years. Trauma, Violence, & Abuse, 2, 215–239. doi:10.1177/
dren using EEG Coherence and MRI. Annals of the New York 1524838001002003002
Academy of Sciences, 821, 160–175. Vihman, M. M. (1996). Phonological development: The origins of
Thompson, S. G., & Higgins, J. (2002). How should meta- language in the child. Malden, MA: Blackwell.
regression analyses be undertaken and interpreted? Statistics *Vondra, J. I., Barnett, D., & Cicchetti, D. (1990). Self-concept,
in Medicine, 21, 1559–1573. motivation, and competence among preschoolers from maltreat-
Thorndike, R. L., Hagen, E. P., & Sattler, J. M. (1986). Stanford- ing and comparison families. Child Abuse & Neglect, 14, 525–540.
Binet Intelligence Scale–Fourth Edition. Chicago, IL: Riverside. Wasserman, G. A., Green, A., & Allen, R. (1983). Going beyond
Tomasello, M. (2003). Constructing a language: A usage-based abuse: Maladaptive patterns of interaction in abusing mother-
theory of language acquisition. Cambridge, MA: Harvard infant pairs. Journal of the American Academy of Child Psychi-
University Press. atry, 22, 245–252.
*Toth, S. L., & Cicchetti, D. (1996). The impact of relatedness Wechsler, D. (1999). Wechsler Abbreviated Scale of Intelligence.
with mother on school functioning in maltreated children. San Antonio, TX: The Psychological Corporation.
Journal of School Psychology, 34, 247–266. Wechsler, D. (1991). Wechsler Intelligence Scale For Children–
*Toth, S. L., Cicchetti, D., Macfie, J., Maughan, A., & VanMeenen, Third Edition. San Antonio, TX: Psychological Corporation.
K. (2000). Narrative representations of caregivers and self in Wells, G., Shea, B., O’Connell, D., Peterson, J., Welch, V., Losos,
maltreated pre-schoolers. Attachment & Human Development, M., & Tugwell, P. (2000). The Newcastle–Ottawa Scale (NOS)
2, 271–305. for assessing the quality of nonrandomised studies in meta-
*Toth, S. L., Stronach, E. P., Rogosch, F. A., Caplan, R., & analyses. Ottawa, Canada: Ottawa Health Research Institute.
Cicchetti, D. (2011). Illogical thinking and thought disorder in Wheldall, K., Mittler, P., & Hobsbaum, A. (1979). Sentence Com-
maltreated children. Journal of the American Academy of Child prehension Test. Slogh, England: National Foundation for
& Adolescent Psychiatry, 50, 659–668. Educational Research.
Trickett, P. K. (1993). Maladaptive development of school-aged, Woon, F. L., & Hedges, D. W. (2008). Hippocampal and amygdala
physically abused children: Relationships with the child-rearing volumes in children and adults with childhood maltreatment-
context. Journal of Family Psychology, 7, 134–147. doi:10.1037/ related posttraumatic stress disorder: A meta-analysis. Hippo-
0893-3200.7.1.134 campus, 18, 729–736. doi:10.1002/hipo.20437
*Trickett, P. K., Aber, J. L., Carlson, V., & Cicchetti, D. (1991). Zimmerman, I. L., Steiner, V. G., & Pond, R. E. (1979). Preschool
Relationship of socioeconomic status to the etiology and Language Scale. Columbus, OH: Charles E. Merrill.

976 Journal of Speech, Language, and Hearing Research • Vol. 58 • 961–976 • June 2015
Copyright of Journal of Speech, Language & Hearing Research is the property of American
Speech-Language-Hearing Association and its content may not be copied or emailed to
multiple sites or posted to a listserv without the copyright holder's express written permission.
However, users may print, download, or email articles for individual use.