Professional Documents
Culture Documents
Amato 2001
Amato 2001
The present study updates the P. R. Amato and B. Keith (1991) meta-analysis of
children and divorce with a new analysis of 67 studies published in the 1990s.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Compared with children with continuously married parents, children with divorced
This document is copyrighted by the American Psychological Association or one of its allied publishers.
Interest in the effects of divorce on children the quality of relationships with mothers and
remains strong among researchers, practitio- fathers. The mean effect sizes tended to be
ners, policy makers, and the general public. modest, however, ranging from -.26 for
This interest is a consequence of the continuing father-child relationships to - . 0 8 for psy-
high divorce rate. Although after 1980 the di- chological adjustment. The results of this
vorce rate in the United States declined slightly, meta-analysis provided a benchmark that sub-
recent projections indicate that between 40% sequent researchers have used to evaluate the
and 50% of first marriages contracted in the magnitude of effect sizes generated in their
1990s will end in divorce (Schoen & Standish, own research. A computer search of the So-
2000). During the last decade, over one million cial Science Citation Index indicates that the
children experienced parental divorce every Amato and Keith meta-analysis has been cited
year (U.S. Bureau of the Census, 1999). over 250 times.
In 1991, Amato and Keith published a meta- The purpose of the present study is to update
analysis of 92 studies that compared children the Amato and Keith (1991) meta-analysis, with
residing with divorced and continuously mar- a focus on studies published in the 1990s. One
ried parents. Their analysis, based on studies of the more intriguing findings reported by
conducted in the 1950s through the 1980s, in- Amato and Keith was a shift in the magnitude of
dicated that children with divorced parents effect sizes over time. In particular, effect sizes
scored significantly lower on a variety of indi- for academic achievement, conduct, self-
cators of well-being, including measures of ac- concept, and mother—child relations were sig-
ademic achievement, conduct, psychological nificantly lower in studies conducted in more
adjustment, self-concept, social relations, and recent decades (the 1970s and 1980s) than in
earlier decades (the 1950s and 1960s). As
Amato and Keith concluded, "These results
suggest that the implications of parental divorce
I thank Alan Booth and Robert Emery for helpful for children's well-being have become less pro-
comments on a draft of this article and Stacy Rogers nounced since the 1950s and 1960s" (p. 34). By
for assistance with the reliability analysis.
calculating effect sizes for studies published in
Correspondence concerning this article should be
addressed to Paul R. Amato, Department of Sociol- the 1990s, it is possible to see if the gap in
ogy, The Pennsylvania State University, 211 Oswald well-being between children with divorced and
Tower, University Park, Pennsylvania 16802-6207. continuously married parents has continued to
Electronic mail may be sent to pxa6@psu.edu. change during the last 10 years. Good reasons
355
356 AMATO
exist for suspecting that such a shift has may be causes of parental divorce as well as
occurred. child outcomes, such as parents' predivorce in-
come (McLanahan & Sandefur, 1994), parents'
Changes in Effect Sizes Associated antisocial personality (Simons, 1996), and par-
With Divorce ents' marital conflict (Forehand, Neighbors, De-
vine, & Armistead, 1994). Other researchers
Effect sizes in the scientific literature on chil- were able to examine children's postdivorce
dren and divorce may be declining because of behavior problems by controlling for children's
the manner in which research is conducted. behavior problems prior to parental separation
Amato and Keith (1991) found that more meth- (Morrison & Cherlin, 1995). It is probable that
odologically sophisticated studies tended to
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
changed (either decreased or increased) since indicated that children with divorced parents were
the 1980s. To accomplish this task, I calculate worse off than other children.
effect sizes for various child outcomes from Effect sizes were calculated for each "independent
studies published in the 1990s and compare sample" in a study. Independent samples existed
these with effect sizes calculated from studies when data from a single study were reported sepa-
published in earlier decades. Second, I consider rately for two nonoverlapping groups, such as boys
how methodological characteristics of studies and girls. If an independent sample contributed two
or more effect sizes within the same outcome cate-
may have affected effect sizes in the 1990s, as
gory (such as academic achievement), then the
well as in earlier decades. And third, because
within-category mean of the effect sizes was calcu-
gender and age differences are of perennial in- lated. This procedure ensured that each independent
terest in this literature, I consider how effect
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
child relations and father-child relations, but these self-concept during the 1980s and 1990s than in
outcomes were not coded in the present study. earlier decades. Researchers appeared to give
Consistent with Amato and Keith (1991), four roughly equal attention to children of each gen-
methodological characteristics of studies were coded: der, although the most common situation, irre-
whether the sample was selected on the basis of a spective of decade, was to report results for
random process as opposed to convenience (1 =
mixed samples of boys and girls. With respect
random, 0 = convenience), whether the study used
multiple-item or single-item measures of outcomes to age, studies in the 1990s gave little attention
(1 = multiple items, 0 = single item), whether the to children of preschool age—even less than in
study used control variables or reported zero-order earlier decades. There appeared to be a corre-
associations (1 = control variables, 0 = no control sponding modest increase in the use of college-
variables), and the sample size. (Clinical samples of age samples in the 1990s. In each time period,
children, although rare in the 1990s, were counted as however, the most common focus was on chil-
convenience samples.) With respect to gender, sam- dren of primary school age.
ples were coded as boys, girls, or mixed. With respect
to age, samples were coded into five levels: pre-
The largest changes in Table 1 are reflected in
school, primary school, secondary school, mixed pri- methodology. The use of random samples be-
mary and secondary school, and college. Finally, data came increasingly common over time, as did the
were recorded on the year of publication and the year use of control variables to adjust for predivorce
in which the sample was collected. If the year of data factors. The use of multiple-item instruments
collection was not reported, then it was coded as two was common in the 1990s, although the per-
years prior to the date of publication. centage did not increase since the 1980s. Sam-
In the original Amato and Keith (1991) meta- ples were larger in the 1990s than in earlier
analysis, two coders independently calculated effect decades. (Medians rather than means appear in
sizes from a sample of articles, with a resulting Table 1 because the distributions were posi-
intercoder correlation of .98. Two coders also inde- tively skewed.) Curiously, sample sizes were
pendently classified measures into outcomes catego- somewhat smaller in the 1980s than in earlier or
ries (academic achievement, conduct, etc.), resulting later decades.
in a kappa reliability coefficient of .84. In the present
meta-analysis, two coders independently calculated a
sample of effect sizes, with a resulting intercoder Effect Sizes in the 1990s
correlation of .90. The coders also independently
classified child outcomes, with a kappa reliability of As a preliminary step, the distribution of 177
.95. In general, the coding of data achieved a high unweighted effect sizes from studies published
standard of reliability. in the 1990s was examined without regard to
outcome. Effect sizes ranged from —1.25 to
0.37, with a mean of —0.29 and a median of
Results
-0.24. In other words, the typical result was
Study Characteristics Across Decades one in which children with divorced parents
scored about one-fourth of a standard deviation
Table 1 presents data on characteristics of lower than children with continuously married
studies published between 1950 and 1979, 1980 parents. Of the 177 effect sizes, 88% were neg-
and 1989, and 1990 and 1999. Data for the ative and 42% were negative and significant
1990s were obtained from 67 studies involving (p < .05). Only one effect size was positive and
98 independent samples. These 98 samples, in significant. Therefore, although the majority of
turn, contributed 177 effect sizes to the meta- comparisons were negative, most (58%) did not
360 AMATO
Table 1
Characteristics of Effect Sizes by Decade of Publication
Decade of publication
1950- 1980- 1990-
Effect size characteristic 1979 1989 1999
Number of effect sizes 68 142 177
Number of independent samples 38 76 98
Number of studies 26 53 67
Child outcome (%)
Academic achievement 25 13 22
Conduct 32 24 23
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Psychological adjustment 24 23 23
This document is copyrighted by the American Psychological Association or one of its allied publishers.
Self-concept 4 19 16
Social relations 15 20 16
Gender of sample (%)
Boys only 27 17 27
Girls only 16 16 24
Mixed gender 57 67 49
Age of sample (%)
Preschool 9 10 2
Primary school 37 41 41
Mixed primary/secondary 27 19 10
Secondary school 22 23 31
College 6 7 16
Methodology (%)
Random samples 15 31 53
Control variables 18 25 40
Multiple item measures 56 88 88
Mdn size divorced sample 57.5 42.0 72.0
Mdn size nondivorced sample 188.0 74.5 207.0
Note. Percentages are based on the total number of effect sizes.
lead to a rejection of the null hypothesis. (Note accurate estimates of population parameters.
that all significance tests reported in the present Consequently, the mean weighted effect size
study are two-tailed.) (which weights the individual effect size by its
Table 2 reports the mean unweighted and sample size) provides a better estimate of the
weighted effect sizes from studies published in effect size in the population. In the present
the 1990s. The mean unweighted effect size is a meta-analysis, effect sizes were weighted and
good indicator of the typical study result, be- combined using formulas for fixed-effect mod-
cause each study (or sample) is weighted els, as described by Hedges (1994) and Shadish
equally. Larger samples, however, have smaller and Haddock (1994).
standard errors and for this reason provide more The mean unweighted and weighted effect
Table 2
Mean Effect Sizes Derived From Studies Published in the 1990s That Compared Children With
Divorced and Continuously Married Parents
Mean Mean
n of unweighted weighted Fail-safe
Child outcome samples effect size effect size Heterogeneity N values
Academic achievement 39 -0.26*** -0.16*** 119.47*** 1,745
Conduct 40 -0.33*** -0.22*** 103.04*** 1,771
Psychological adjustment 41 -0.31*** -0.21*** 104.38*** 1,939
Self-concept 28 -0.24*** -0.12** 58.28** 256
Social relations 29 -0.28*** -0.15*** 114.79*** 544
**p < .01. ***p < .001.
CHILDREN OF DIVORCE 361
sizes were negative and statistically significant duct. For this outcome category, the mean
for all outcomes. Unweighted mean effect sizes weighted effect size was -0.28 among the 15
ranged from —0.24 for self-concept to —0.33 all-boy samples and -0.16 among the 15 all-
for conduct; the corresponding weighted mean girl samples (z = 2.04, p < .05). It is well-
effect sizes were -0.12 and -0.22. The fact known that boys are more prone to behavioral
that the weighted effect sizes were consistently problems than girls, and divorce may exacer-
lower than the unweighted effect sizes indicates bate this difference. In the Amato and Keith
that larger studies revealed narrower group dif- (1991) meta-analysis, effect sizes associated
ferences than smaller studies. In general, al- with marital disruption were significantly
though the mean weighted effect sizes for the higher among all-boy samples than all-girl sam-
1990s were modest in magnitude, they demon- ples with respect to social relations, but not with
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
strate the continuing gap in achievement, ad- respect to conduct. It is not clear why earlier
justment, and well-being between children with studies suggest a gender difference in social
divorced and continuously married parents. relations and later studies suggest a gender dif-
The present meta-analysis almost certainly ference in conduct. One possibility is that the
failed to locate and include some studies pub- apparent discrepancy reflects overlap between
lished in the 1990s, and the sampling strategy the two outcome domains. Researchers may
deliberately excluded unpublished studies. If consider aggression among boys to be a conduct
most of these excluded studies reported null problem, but aggressive boys also experience
results, then the significance tests reported in difficulty forming and maintaining networks of
Table 2 are too liberal. To address this problem, close friends. In general, the present meta-
fail-safe N values were calculated (Rosenthal, analysis, considered with the earlier Amato and
1979). This statistic indicates the number of Keith meta-analysis, provides modest support
additional studies with null results that would for the notion that divorce has stronger effects
need to be added to the meta-analysis to reduce on boys than girls, at least in some domains.
the mean effect sizes in Table 2 to insignifi- Nevertheless, the mean weighted effect sizes
cance. All of the fail-safe N values were large. were significant for girls as well as boys across
For example, there would need to be an addi- all five outcomes. The central finding, therefore,
tional 1,745 studies of academic achievement— is that divorce is associated with a range of
all with null results—to reduce the mean effect poor outcomes among children irrespective of
size of —0.16 to an insignificant level. Be- gender.
cause the fail-safe N values were large, it is
reasonable to conclude that the significance Children's ages. Although children's ages
tests are valid. That is, in the larger population were coded into five categories, the number of
from which these samples were drawn, children samples involving preschool children was too
with divorced parents scored lower than chil- small for analysis, and the mixed category (part
dren with continuously married parents on these primary school students, part high school stu-
outcomes. dents) was difficult to interpret. Consequently, a
series of comparisons were carried out between
Table 2 also contains heterogeneity values
children in primary and secondary school. Of
(Hedges, 1994; Hedges & Olkin, 1985). The
statistical significance of these values indicates these comparisons, two were statistically signif-
that the variability in effect sizes across samples icant. The mean weighted effect size for aca-
is greater than expected by chance. If these demic achievement was stronger for children in
values are significant, then one is justified primary school than in secondary school (—0.20
in exploring between-studies (or between- vs. -0.14; z = 2.48, p < .01). In contrast, the
samples) characteristics that might account for mean weighted effect size for psychological
variability in effect sizes. adjustment was weaker for children in primary
school than in secondary school (—0.15 vs.
-0.32, z = 4.98, p < .001). The explanation for
Effect Sizes and Study Characteristics
this apparent discrepancy is not clear. One pos-
Children's gender. Comparisons of all-girl sibility is diat it is easier to measure psycholog-
and all-boy samples across the five outcomes ical adjustment among adolescents than
revealed few significant differences in effect younger children. However, adolescents with
sizes in the 1990s, with the exception of con- serious academic problems, unlike younger
362 AMATO
children, may drop out of school, leading to Effect Sizes Across Decades
weaker effect sizes for academic achievement
among adolescents. It is difficult to interpret To explore variation in effect sizes across
these results substantively because the data re- decades, in Table 3 I report mean effect sizes by
flect children's ages at the time of data collec- decade of publication. For each outcome, the
tion rather than children's ages at the time of first row (labeled Unadjusted) shows the mean
parental separation. Data on the latter variable weighted effect size for each time period. These
were not available in most studies. One cannot means were obtained from weighted multiple
tell, therefore, if a large effect size for high regression analysis—one analysis for each
outcome—with decade serving as a predictor.
school students reflects the recency of divorce
(See Hedges, 1994, for a description of multi-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
due to experiencing divorce at an early age. In statistic reflects the amount of between-studies
spite of this ambiguity, the data indicate that, for heterogeneity accounted for by decade. If the
most outcomes, mean effect sizes were negative overall Q statistic was significant, then z tests
and significant regardless of children's ages at were used to compare each pair of mean effect
the time of data collection. sizes.
Study methodology. Although relatively With respect to academic achievement, de-
few gender and age differences appeared in the cade of publication was related significantly to
data, methodological characteristics of studies effect sizes (see Table 3, first row), as reflected
appeared to be more consequential. Amato and in the significant Q statistic. Individual z tests
Keith (1991) reported a trend for methodologi- revealed that the mean effect size declined sig-
cal sophistication to be associated with weaker nificantly from —0.29 in the earliest period to
effect sizes, at least for some outcomes. The -0.14 in the 1980s. Similarly, the mean effect
results of studies published in the 1990s were size for the earliest period was significantly
consistent with this earlier observation. Random larger than the mean effect size of —0.16 in the
samples yielded weaker effect sizes than did 1990s. These findings suggest that the effects of
convenience samples in terms of academic divorce on academic achievement were rela-
achievement (-0.20 vs. -0.41, z = 3.67, p < tively strong prior to 1980, then became weaker
.001). Studies that conducted multivariate anal- in the last two decades. The mean effect sizes
yses reported smaller group differences than for the 1980s and 1990s, however, did not
studies that reported zero-order associations differ.
with respect to academic achievement (-0.14 Making comparisons across decade may be
vs. -0.26, z = 1.97, p < .05) and psychological misleading, however, because of the shifts in
adjustment (-0.17 vs. -0.30, z = 3.67, p < study methodology documented in Table 1. To
.001). The use of multiple-item measures, as take changes in research methods into account,
opposed to single-item measures, was associ- I conducted a second weighted multiple regres-
ated with weaker effect sizes for psychologi- sion analysis, with the four methodology vari-
cal adjustment (-0.20 vs. -0.49, z = 3.27, ables (random sample, multivariate analysis,
multiple item indicators, and large sample size)
p < .01) and social relations (-0.13 vs.
included as predictors. The second row for each
-0.27, z = 2.74, p < .01). Finally, studies
outcome (labeled Adjusted) shows the mean
that involved samples of 100 or more children effect sizes controlling for these study charac-
from divorced families revealed weaker effect teristics. In other words, this row shows what
sizes than studies with smaller samples in the mean effect sizes would have been if there
terms of self-esteem (-0.05 vs. -0.20, z = had been no improvements in methodology
2.23, p < .05) and social adjustment (-0.13 across decades. The corresponding Q statistic
vs. -0.24, z = 4.04, p < .001). Overall, these shows the amount of between-studies heteroge-
results are consistent with the prior meta- neity accounted for by decade after the effects
analysis in showing that methodologically so- of methodology were removed from the data.
phisticated studies tend to report smaller dif- (In supplementary analysis, controls also were
ferences between children from divorced and introduced for sample gender and age, but be-
nondivorced families than methodologically cause these study characteristics changed little
simple studies. across decades, these adjustments affected the
CHILDREN OF DIVORCE 363
Table 3
Mean Weighted Effect Sizes by Decade of Study Publication
Decade of publication
1950- 1980- 1990-
Child outcome 1979 1989 1999 Q change
Academic achievement
Unadjusted -0.29 a -0.14b -0.16 b 14.95***
Adjusted -0.25 a -0.09 b -0.17 c 14.27***
N of samples 17 19 39
Conduct
-0.30 a -0.21b -0.22 b 11.76**
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Unadjusted
This document is copyrighted by the American Psychological Association or one of its allied publishers.
results in only minor ways and are not fect size for the 1990s was no longer signifi-
reported.) cantly different from the mean effect sizes in
The second row in Table 3 shows mean effect earlier decades. With respect to psychological
sizes for academic achievement adjusted for adjustment, both the unadjusted and the ad-
study quality. The Q statistic reveals that decade justed means indicate that effect sizes became
of study continued to have a significant associ- stronger in the 1990s than in earlier decades.
ation with effect sizes with study methodology These data suggest that the gap in psychological
held constant. Comparable to the results based and emotional adjustment between children
on unadjusted data, z tests on the adjusted with divorced and continuously married parents
means revealed a significant decline in the es- was wider in the 1990s than at any earlier time.
timated effect of divorce between the earliest The unadjusted results for self-concept suggest
period and the 1980s (i.e., the mean effect size a decline in effect sizes between the earliest
moved closer to zero). But contrary to the re- studies and studies in the 1980s, but the overall
sults based on unadjusted data, the mean for trend (as reflected in the Q statistic) was not
studies in the 1990s was significantly different significant. Adjusting for study characteristics
from the mean for studies published in the made the pattern more curvilinear and fully
1980s. If studies in the 1980s and 1990s had significant. Although not reported in Table 3,
been comparable in methodology, then the ob- the difference between the 1980s and 1990s
served gap in achievement between children approached significance (p < .10). Finally the
with divorced and married parents would have results for social relations, both unadjusted and
increased during the last decade. adjusted, show no time trend.
With respect to conduct, the unadjusted re- Consistent with the findings of Amato and
sults indicate that effect sizes decreased signif- Keith (1991), the data in Table 3 suggest that
icantly between earlier decades and the 1980s, effect sizes for several outcomes were weaker in
then changed little during the next decade. With the 1980s than in earlier decades. The results for
study methodology held constant, the mean ef- the 1990s, however, suggest that effect sizes for
364 AMATO
several outcomes (academic achievement, psy- tween 1960 and the late 1980s, then declined
chological adjustment, and to a lesser extent, slightly (became stronger) during the 1990s.
self-concept) became stronger during the last 10 The curve for conduct reached its peak (was
years. Although Table 3 is suggestive of histor- closest to zero) in the early 1980s, then fell
ical trends, the data are limited by the somewhat during the 1990s. For psychological adjustment,
arbitrary division of time into decades. Shifts the curve rose (became weaker) during the
between one decade and the next may not rep- 1960s, reached its highest (weakest) point in the
resent the most appropriate cutting points for mid 1970s, then fell (became stronger) during
determining trends over time. the next 25 years. The curve for self-concept
To address this issue, additional analyses followed a similar pattern, rising throughout the
involving weighted multiple regression for ef- 1960s and 1970s, reaching its highest point in
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
fect sizes were conducted—one for each the early 1980s, then falling throughout the lat-
This document is copyrighted by the American Psychological Association or one of its allied publishers.
outcome—in which decade was replaced with ter half of the 1980s and the 1990s.
the actual year of publication. Additional pre- One should keep in mind that the year of
dictors included year of publication squared (to publication differs from the year in which data
capture curvilinearity) and the four method- were collected, with data collection sometimes
ological variables described above. These anal- taking place years (or even decades) prior to
yses revealed significant curvilinear trends for publication. To address this issue, a similar set
academic achievement (z = 3.00, p < .01), of regression analyses were carried out based on
conduct (z = 2.41, p < .05), psychological year of data collection rather than year of pub-
adjustment (z = 4.47, p < .001), and self- lication. However, as noted in the Method sec-
concept (z = 3.34, p < .001). The resulting tion, nearly one half of all articles in this anal-
equations were used to predict effect sizes for ysis failed to indicate the year of data collection.
each year of publication between 1960 and To include studies with missing data in the
2000, with methodological characteristics set at analysis, it was assumed that data were col-
the mean of all studies for that particular out- lected 2 years prior to the year of publication.
come. (Too few studies were published prior to The analyses focusing on year of collection,
1960 to result in reasonable estimates, and pro- therefore, were only approximate. Nevertheless,
jections for 2000 represent a slight extrapola- the results of analyses based on year of data
tion from the data.) collection were similar to the results based on
These predicted values appear in Figure 1. year of publication, with significant curvilinear
Note the curvilinear trends for all four out- trends appearing for academic achievement,
comes. For academic achievement, effect sizes conduct, psychological adjustment, and self-
moved closer to zero (i.e., became weaker) be- concept. This similarity is not surprising, given
-.40
"1 1 1 ' 1 i 1 1 1 i ! 1 1 i 1 1 ! 1 i 1
1960 1964 1968 1972 1976 1980 1984 1988 1992 1996 2000
Year of Publication
Academic Conduct
• • Psychological Self-concept
Figure 1. Yearly trend in effect sizes associated with divorce for four child outcomes.
CHILDREN OF DIVORCE 365
that year of publication and year of data collec- studies conducted in earlier decades. But this
tion correlated at .86 prior to substitution of was not the case. Across most outcomes, studies
missing data and .89 after substitution. The in the 1990s yielded effect sizes comparable to
main difference between the regression lines those in the 1980s. Furthermore, controlling for
based on year of data collection and year of methodological characteristics of studies tended
publication was that the former curves reached to expand the differences in effect sizes between
their highest (and weakest) point 3 to 6 years decades. More detailed analyses, in which effect
earlier, depending on the outcome. sizes were regressed on year of publication (as
well as year of data collection), revealed signif-
Discussion icant curvilinear trends for academic achieve-
ment, conduct, psychological adjustment, and
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
contrast, did not share in the prosperity of the Implications for Practitioners
last decade. To the extent that children benefit
from increases in family income, those with Although marital dissolution has become
married parents (as a group) would have pulled normative in American culture, practitioners
further ahead of those with single parents. Al- should be aware that children with divorced
though this idea seems reasonable, however, no parents, as a group, continue to fare more poorly
direct data are available to test it. than children with continuously married par-
Whatever the reason for the apparent de- ents. In particular, children with divorced par-
ents achieve lower levels of success at school,
cline in the relative well-being of children
are more poorly behaved, exhibit more behav-
with divorced parents during the last decade,
ioral and emotional problems, have lower self-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
this shift occurred in spite of the growth of esteem, and experience more difficulties with
This document is copyrighted by the American Psychological Association or one of its allied publishers.
school-based interventions for children, par- interpersonal relationships. Indeed, the gap in
enting classes for divorcing parents, and di- well-being between the two groups of children
vorce mediation. This conclusion does not appears to have grown larger during the last
mean that the spread of therapeutic and legal decade. Furthermore, this gap is present irre-
interventions has not benefitted children—it spective of children's gender or age. Practitio-
only means that other social forces have op- ners working with divorcing families or chil-
erated even more strongly to disadvantage dren with single parents, therefore, should not
children from divorced families. underestimate the extent to which family dis-
Before concluding, it is necessary to point out ruption continues to represent a risk factor for a
some limitations of the present study. First, this range of undesirable child outcomes.
study did not include unpublished conference At the same time, however, practitioners
papers or dissertations, so the results may have should be aware that the average differences
been influenced by publication bias. The calcu- between children with divorced and continu-
lation of large fail-safe values, however, sug- ously married parents are not large in absolute
gests that this omission did not seriously distort terms. These relatively small differences reflect
the results. Second, the manner in which study the diversity of outcomes among children in
quality was operationalized was crude, as it was both groups. The adjustment of children follow-
based on simple dichotomies such as random ing divorce depends on a variety of factors,
versus convenience sampling. Moreover, many including the level of conflict between parents
important characteristics of studies—using before and after separation, the quality of par-
multiple informants, direct observation of chil- enting from both the custodial and noncustodial
dren, or longitudinal designs—were not consid- parent, changes in the child's standard of living,
ered. Of course, it was necessary to code study and the number of additional stressors to which
quality in this manner to maintain comparability children are exposed, such as moving or chang-
ing schools. Depending on the specific constel-
with the earlier study by Amato and Keith
lation of factors around the time of divorce,
(1991). Finally, pooling the results of multiple
children may exhibit an improvement in func-
studies is not the ideal method of establishing tioning, a modest decline in functioning that
trends in effect sizes over time. The studies improves over time, a substantial long-term de-
included in the present meta-analysis differed in cline in functioning, or little change. Knowl-
many ways that could not be controlled. A edge of group averages, therefore, cannot pre-
better strategy would involve conducting iden- dict how a particular child will adjust to family
tical comparisons, using the same sampling disruption.
strategy, instruments, and analytic techniques, Nevertheless, the persisting—and apparently
on repeated cross-sections of children in differ- increasing—gap between children with di-
ent decades. Unfortunately, such data are diffi- vorced and continuously married parents sug-
cult to obtain. Nevertheless, new research that gests the continuing importance of developing
focuses on cohort differences in children's ad- and evaluating therapeutic and educational pro-
justment to divorce, and the processes that help grams for divorcing families. Classes for di-
to explain them, would represent an important vorcing parents, for example, have been intro-
contribution to our understanding of how mar- duced in many cities and states, but they vary
ital disruption affects children's lives. widely in length and content, and evaluations of
CHILDREN OF DIVORCE 367
these programs are sparse. The same can be said •Brennan, K. A., & Shaver, P. R. (1993). Attachment
of many legal approaches. For example, an in- styles and parental divorce. Journal of Divorce and
creasing number of mothers and fathers are Remarriage, 21, 161-175.
completing parenting plans prior to marital dis- •Brodzinsky, D., Hitt, J. C , & Smith, D. (1993).
Impact of parental separation and divorce on
solution. Yet we know little about the conse-
adopted and nonadopted children. American Jour-
quences of parenting plans for children's post- nal of Orthopsychiatry, 63, 451-461.
divorce relations with parents and emotional *Brubeck, D., & Beer, J. (1992). Depression, self-
well-being. Closer ties between practitioners esteem, suicide ideation, death anxiety, and GPA in
and researchers will be necessary to formulate high school students of divorced and nondivorced
effective strategies for minimizing the disrup- parents. Psychological Reports, 71, 755-763.
tive effects of marital dissolution on children.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
dren's divorce adjustment. Journal of Child and •Hetherington, E. M., & Clingempeel, W. G. (1992).
Family Studies, 2, 23-36. Coping with marital transitions. Monographs of the
Emery, R. E., Kitzmann, K. M., & Waldron, M. Society for Research in Child Development, 57
(1999). Psychological interventions for separated (2-3, Aerial No. 227).
and divorced families. In E. M. Hetherington (Ed.), •Hilton, J. M., & Devall, E. L. (1998). Comparison
Coping with divorce, single parenting, and remar- of parenting and children's behavior in single-
riage: A risk and resiliency perspective (pp. 323- mother, single-father, and intact families. Journal
344). Mahwah, NJ: Erlbaum. of Divorce and Remarriage, 29, 23-54.
*Emery, R. E., Waldron, M., Kitzmann, K. M , & •Jekielek, S. M. (1998). Parental conflict, marital
Aaron, J. (1999). Delinquent behavior, future di- disruption and children's emotional well-being. So-
vorce or nonmarital childbearing, and externalizing cial Forces, 76, 905-935.
behavior among offspring: A 14-year prospective *Jones, D. C. (1992). Parental divorce, family con-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
study. Journal of Family Psychology, 13, 568-579. flict and friendship networks. Journal of Social and
This document is copyrighted by the American Psychological Association or one of its allied publishers.
adolescent academic achievement. Journal of Di- prints of cohabitation: Results from marital status
vorce and Remarriage, 13, 69-88. life tables for the U.S., 1995. Paper presented at the
•Morrison, D. R., & Cherlin, A. J. (1995). The di- Population Research Institute, The Pennsylvania
vorce process and young children's well-being: A State University, University Park, PA.
prospective analysis. Journal of Marriage and the Shadish, W. R., & Haddock, C. K. (1994). Com-
Family, 57, 800-812. bining estimates of effect size. In L. Cooper &
Morrison, D. R., & Coiro, M. J. (1999). Parental L. V. Hedges (Eds.), The handbook of research
conflict and marital disruption: Do children benefit synthesis (pp. 261-284). New York: Russell Sage
when high-conflict marriages are dissolved? Jour- Foundation.
nal of Marriage and the Family, 61, 626-637. •Simons, R. L. (1996). Understanding differences
*Mott, F. L., Kowaleski-Jones, L., & Menaghan, between divorced and intact families. Thousand
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Does a child's gender make a difference? Journal •Smith, J. R., Brooks-Gunn, J., & Klebanov, P. K.
of Marriage and the Family, 59, 103-118. (1997). Consequences of living in poverty for
*Mulholland, D. J., Watt, N. F., Philpott, A., & young children's cognitive and verbal ability and
Sarlin, N. (1991). Academic performance in chil- early school achievement. In G. J. Duncan & J.
dren of divorce: Psychological resilience and vul- Brooks-Gunn (Eds.), Consequences of growing up
nerability. Psychiatry, 54, 268-280. poor (pp. 132-189). New York: Russell Sage Foun-
•Needle, R. J., Su, S. S., & Doherty, W. J. (1990). dation.
Divorce, remarriage, and adolescent substance use: •Smith, T. E. (1990). Parental separation and adoles-
A prospective longitudinal design. Journal of Mar- cents' academic self-concepts: An effort to solve
riage and the Family, 52, 157-169. the puzzle of separation effects. Journal of Mar-
*Pagani, L., Boulerice, B., & Tremblay, R. E. (1997). riage and the Family, 52, 107-118.
The influence of poverty on children's classroom •Smith, T. E. (1997). Differences between Black and
placement and behavior problems. In G. J. Duncan White students in the effects of parental separation
& J. Brooks-Gunn (Eds.), Consequences of grow- on school grades. Journal of Divorce and Remar-
ing up poor (pp. 311-339). New York: Russell riage, 27, 25-42.
Sage Foundation.
•Spreecher, R., Cate, R., & Levin, L. (1998). Paren-
•Parish, T. S. (1991). Ratings of self and parents by tal divorce and young people's beliefs about love.
youth: Are they affected by family status, gender, Journal of Divorce and Remarriage, 28, 107-
and birth order? Adolescence, 26, 105-112.
120.
Pedro-Carroll, J., & Alpert-Gillis, L. (1997). Preven-
Stolberg, A. L., & Mahler, J. (1994). Enhancing
tative interventions for children of divorce: A de-
treatment gains in school-based intervention for
velopmental model for 5 and 6 year-old children.
children of divorce through skill training, parental
Journal of Primary Prevention, 18, 5-23.
involvement, and transfer procedures. Journal of
•Peres, Y., & Pasternack, R. (1991). To what extent
Consulting and Clinical Psychology, 62, 147-156.
can the school reduce the gaps between children
raised by divorced and intact families? Journal of •Summers, M., Summers, C. R., & Ascione, F. R.
Divorce and Remarriage, 15, 143-158. (1993). A comparison of sibling interaction in intact
•Pett, M. A., Wampold, B. E., Turner, C. W., & and single-parent families. Journal of Divorce and
Vaughan-Cole, B. (1999). Paths of influence of Remarriage, 20, 215-227.
divorce on preschool children's psychosocial ad- •Teachman, J. D., Paasch, K., & Carver, K. (1996).
justment. Journal of Family Psychology, 13, 145— Social capital and dropping out of school early.
164. Journal of Marriage and the Family, 58, 773-783.
•Rodgers, B. (1994). Pathways between parental di- •Teha, S., & Stolberg, A. L. (1993). Peer support,
vorce and adult depression. Journal of Child Psy- divorce, and children's adjustment. Journal of Di-
chology and Psychiatry, 35, 1289-1294. vorce and Remarriage, 20, 45-64.
•Roizblat, A., Rivera, S., Fuchs, T., Toso, P., Ossan- Thornton, A. (1985). Changing attitudes toward sep-
dran, E., & Guelford, M. (1997). Children of di- aration and divorce: Causes and consequences.
vorce: Academic outcomes. Journal of Divorce and American Journal of Sociology, 90, 856-872.
Remarriage, 26, 51-56. U.S. Bureau of the Census. (1999). Statistical ab-
Rosenthal, R. (1979). The "file drawer problem" and stract of the United States 1999. 119* edition.
tolerance for null results. Psychological Bulletin, Washington, DC: U.S. Government Printing Office.
86, 638-641. •Vandewater, E. A., & Lansford, J. E. (1998). Influ-
Rosenthal, R. (1994). Parametric measures of effect ences of family structure and parental conflict on
size. In H. Cooper and L. V. Hedges (Eds.), The children's well-being. Family Relations, 47, 323-
handbook of research synthesis (pp. 231-244). 330.
New York: Russell Sage. •Weiner, J., Harlow, L., Adams, J., & Grebstein, L.
Schoen, R., & Standish, N. (2000, April). The foot- (1995). Psychological adjustment of college stu-
370 AMATO
dents from families of divorce. Journal of Divorce stances and family outcomes: A review of the
and Remarriage, 23, 75-95. 1990s. Journal of Marriage and the Family, 62,
*Wenk, D., Hardesty, C. L., Morgan, C. S., & Blair, 1035-1051.
S. L. (1994). The influence of parental involvement
on the well-being of sons and daughters. Journal of
Marriage and the Family, 56, 229-234. Received August 23, 2000
White, L., & Rogers, S. J. (2000). Economic circum- Accepted February 12, 2001 •
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Manuscript submission patterns make the precise date of completion of the 2002
volumes uncertain. Current editors, Kevin R. Murphy, PhD, Michael Pressley, PhD,
Philip C. Kendall, PhD, Chester A. Insko, PhD, and Nancy Eisenberg, PhD, respec-
tively, will receive and consider manuscripts through December 31,2001. Should 2002
volumes be completed before that date, manuscripts will be redirected to the new editors
for consideration in 2003 volumes.