Behavior Problems and Mental Health Referrals of International Adoptees:  A Meta-analysis FREE

International adoption is an increasing phenomenon involving more than 40 000 children a year moving between more than 100 countries.^1- 2 By setting uniform norms and standards, the 1993 Hague Convention³ endorsed and facilitated international adoption. International adoption may offer the advantage of a permanent family to a child for whom a family cannot be found in the country of origin. In 2004, most international adoptions in the United States (22 884) were from China, Russia, Guatemala, South Korea, and Kazakhstan,⁴ whereas most international adoptions in Europe (15 847 in 2003) were from China, Russia, Colombia, Ukraine, and Bulgaria.² Since the 1970s, domestic adoptions in North America and Europe drastically decreased, whereas at the same time the number of international adoptions increased.¹

International adoptees often experience inadequate prenatal and perinatal medical care, maternal separation, psychological deprivation, insufficient health services, neglect, abuse, and malnutrition in orphanages or poor families before adoptive placement.^5- 7Animal models have shown that early maternal separation and deprivation can seriously harm infant functioning and later development.^8- 9 Psychological deprivation in orphanages can result in maladjustment in children.^6- 7,10 In addition, after adoptive placement, adoptees have to cope with integrating the loss of their culture and birth family into their lives.¹¹ In contrast to domestic adoptees who are adopted within the same country, international adoptees may face problems regarding their divergent identity,¹² as most international adoptees are raised by parents who do not share their racial and cultural background.

Adoption usually offers improved medical, physical, educational, and psychological opportunities for institutionalized children,^13- 14 and research has documented children’s substantial recovery from deprivation after adoption,^14- 15 which may partly be due to the possibility that some adopted children were selected for adoption because they seemed brighter or had better social skills. Nevertheless, several studies found that adopted children were overrepresented in mental health populations and showed more externalizing disorders.¹⁶ Some studies found more mental health problems in international adoptees compared with nonadopted controls, in particular in male adoptees,^7,17 in adolescence,^7,18- 19 and in children placed beyond infancy.^20- 21 However, the majority of adoptees were functioning well.^7,15,22 In a large national cohort study in Sweden involving more than 11 000 international adoptees, a significantly higher risk of suicide, psychiatric illness, and social maladjustment was found compared with nonadopted controls although most adoptees were doing well.¹⁸ The authors stated that further studies with less severe outcomes are needed as the main differences between adoptees and nonadopted controls were found in only a small number of international adoptees.

We report the first meta-analyses on behavior problems and mental health referrals of international adoptees comparing them to nonadopted controls and domestic adoptees. We hypothesized that international adoptees present more behavior problems and are referred to mental health services more often than nonadopted controls¹⁶ or domestic adoptees.^5,12,18 We hypothesized that those with preadoption adversity,^6,15 older ages at international adoptive placement (>12 months),^20- 21 and males^7,17 would have an increased risk for behavior problems and mental health referrals. International adoptees were also expected to show more behavior problems in adolescence compared with the years before adolescence.^7,18 We studied domestic adoptions in Western countries only because the increasing domestic adoptions in developing countries, eg, India,²³ have not been systematically studied yet.

The guidelines published by Stroup et al²⁴ for the meta-analysis of observational studies were followed. The aims of our meta-analysis were (1) to compare all adoptees with nonadopted controls; (2) to compare international adoptees with nonadopted controls; (3) to compare international adoptees with domestic adoptees; and (4) to examine moderators for the international adoption outcomes. Empirical studies documenting adoptees’ behavior problems and use of mental health services were collected systematically, using 3 search strategies.²⁵ First, MEDLINE (US National Library of Medicine), PsychLit (Psychological Literature), and ERIC (Education Resource Information Center) were searched for case-control studies published between 1950 and January 2005 with the key words adopt*, combined with (behavior) problem(s), disorder(s), (mal)adjustment, (behavioral) development, clinical or psychiatric (referral) , or mental health. Second, the references of the collected journal articles, books, book chapters, and reports were searched for relevant studies. Third, experts in the field were asked for relevant studies. The search was not limited to English-language publications. Our selection criteria were broad in order to include as many studies as possible. Adoptees in all age groups were included, from early childhood through adulthood. In case of a longitudinal study, the first assessment with adequate data was used to ensure that every adoptee was counted only once in the pertinent meta-analyses. Similarly, a study sample described in several articles or chapters was used only once. We included studies using the Child Behavior CheckList²⁶ or related measures to measure problem behavior. Studies involving clinically referred adoptees were included in so far as their rate of mental health referrals could be compared with the rate of adoptees in the general population. We excluded studies that exclusively sampled adopted children exposed to alcohol or drugs in utero,²⁷ physically or mentally handicapped children, and other special needs children, such as hard-to-place children.²⁸

Data were entered into a customized meta-analytic database. We used a detailed coding system to extract from every study data on sample characteristics, design, publication outlet, and information on behavioral problems or mental health referrals. Study characteristics and study results were coded independently. The main coder of the study characteristics was blinded to the meta-analytic study results and had no previous familiarity with the adoption field.

The following sample characteristics were extracted: sex, age at adoptive placement, age at assessment, duration of time with the adoptive family, evidence that the participants in the study were international adoptees, and evidence of preadoption adversity. If available, we included findings for males and females (in case this was not reported, the study was placed in the category of mixed) or different age groups separately, considering these groups as subsamples of the same study. We coded whether the adoptees were placed for adoption between 0 and 12 months, 12 and 24 months, or older than 24 months (or NA, not available, if data were not reported or extractable). We also coded the participants’ age at the time of the assessment: between 0 and 4 years, 4 and 12 years, 12 and 18 years, or older than 18 years (or NA). We coded whether the adoptees had been with the adoptive family for 0 to 4 years, more than 4 to 8 years, more than 8 to 12 years, or more than 12 years. We also extracted whether the adoptees were placed internationally or not. Studies were coded as an international adoption study if the report indicated that all participants were adopted internationally. Preadoption adversity was coded if at least 50% of a sample experienced extreme deprivation, such as serious neglect, malnutrition, and/or abuse. As most adoptees experienced at least some deprivation before adoptive placement and because preadoption histories were not known with certainty in most cases, our index of adversity must be considered as a proxy for the most extreme preadoption circumstances.

The following design characteristics were extracted: whether a nonadopted norm group (eg, Child Behavior Checklist norms) or other control group (a general population sample, classmates, or siblings) was used in the study, and the sample sizes of the adoption and control group. Studies that did not include a nonadopted control group were not included in the meta-analysis. Also, country of study was extracted, distinguishing between studies conducted in North America vs other countries. Finally, year of publication was extracted, analyzing studies published before 1959, during 1960-1969, 1970-1979, 1980-1989, and 1990 or later. Quality of study, as outlined for experimental research,^29- 30 was not coded because some crucial criteria, eg, randomization, are not applicable to nonexperimental research. However, sample size was accounted for in the study outcomes and publication outlet was coded as proxy of study quality.¹⁴ Publication outlet was assessed by distinguishing between studies published in refereed scientific journals and in other scientific reports, books, and book chapters. Peer-reviewed journals may set higher standards than nonrefereed outlets. Alternatively, scientific journals may be more hesitant in accepting studies with small sample sizes, nonsignificant outcomes (resulting in a publication bias, see below), or both than books or chapters.

We extracted information on behavior problems, mental health referrals, or both. For behavior problems, we distinguished between total problems, externalizing problems (eg, aggression, delinquency, hyperactivity), and internalizing problems (eg, withdrawn, anxious or depressed).²⁶ In several studies, scores for externalizing problems and internalizing problems were reported but scores for total behavior problems were lacking. In those cases, a weighted average score was constructed for total problems based on the scores for externalizing and internalizing problems because externalizing and internalizing problems are considered as adequately representing total problems.³¹ We also coded for whether the study involved a referred adoption group (eg, referred to a psychiatric clinic), and if so, whether the rate of overrepresentation of mental health referrals could be computed (ie, percentage of adoptees in the clinic population vs percentage of adoptees in the general population). Satisfactory intercoder reliabilities were established (89%; range, 75%-100%; k = 20).

The various statistics in the adoption studies were recomputed with Mullen’s advanced basic meta-analysis program²⁵ and transformed into Cohen d.³² For each study we thus calculated an effect size (Cohen d): the standardized difference between the means of the adoptive and the nonadoptive group. According to Cohen’s³² criteria, ds of <0.20 are considered small effects; ds of about 0.50, moderate effects; and ds of about 0.80, large effects. The resulting set of effect sizes were inserted in the Comprehensive Meta-Analysis (CMA, version 1.025) program³³ that computed fixed as well as random-effect model parameters and 95% confidence intervals (CIs) around the point estimate of an effect size. The Q statistics (provided by CMA) were used to test the homogeneity of the specific set of effect sizes and the significance of moderators.^25,33 The set of international adoption studies was homogeneous; therefore, we decided to use the combined effect sizes in the context of the fixed-effect models in the meta-analyses of international adoptees. In the total set of studies (international and domestic adoption), random-effect models were used as several subsets were heterogeneous.³⁴ In the random-effect models, we computed 85% CIs around the point estimate of each set of effect sizes. When testing moderators, inspection of the overlap between these CIs provided a test of the differences between the combined effects of subsets of study effect sizes grouped by moderators. This approach of comparing 85% CIs served as the significance test in the context of a random-effect model for which the Q statistics are not an adequate index of significance of differences.^14,35 Nonoverlapping 85% CIs were considered to indicate a significantly different effect size in subsets of study outcomes.¹⁴ Winsorizing was used to redress outlying sample sizes.³⁶ Also, combined effect sizes and confidence boundaries were recomputed removing 1 study at a time. This method to test the stability of the outcomes is similar to a jackknife procedure that takes an entire sample except for 1 value, and then calculates the test statistic of interest. It repeats the process, each time leaving out a different value, and each time recalculating the test statistic.³³

We used 1 of the methods developed to estimate potential publication bias, namely, the trim-and-fill method (available in CMA³³). Using this method, a funnel plot is constructed of each study’s effect size against its precision (1/SE). These plots should be shaped like a funnel if no publication bias is present. However, since smaller or nonsignificant studies are less likely to be published, studies in the bottom left-hand corner of the plot are often omitted.^37- 38 For the meta-analyses the right-most studies considered to be symmetrically unmatched were trimmed. The trimmed studies were then replaced and their missing counterparts imputed or filled as mirror images of the trimmed outcomes. This then allowed for the computation of an adjusted effect size and CI.^38- 39 Also, a fail-safe number was computed, ie, the number of studies (k) that would be needed to change a significant combined effect size into a nonsignificant outcome.²⁵

We were able to find 101 subsamples (study outcomes or subsamples coded separately, eg, males and females; hereafter, studies) on total behavior problems, including 25 281 cases and 80 260 controls, 64 studies on externalizing problems and 64 studies on internalizing problems (Table 1 and Table 2^{7,13,17- 18,20- 22,40- 96}). We also found 36 studies on mental health referrals (reported in 34 articles; Table 3^97- 129), including 5092 cases and 75 858 controls. The studies were published in English, Spanish, German, Dutch, and Swedish. The studies were conducted in North America (54%): Canada and the United States; Europe (33%): Belgium, Finland, France, Germany, Greece, the Netherlands, Norway, Spain, Sweden, the United Kingdom; Australia and New Zealand (11%); and other countries (2%). The majority of participants in the studies coded as noninternational, domestic adoption studies were placed within the country. A few domestic adoption studies included a minority of international adoptees (eg, 23%⁴³ or 28%⁸⁵).

For total behavior problems, the age at assessment was from 0 to 4 years in 7% of the studies; older than 4 to 12 years in 44.5%; older than 12 to 18 years in 44.5%; and more than 18 years in 4%. Age at adoptive placement was from 0 to 12 months in 45% of the studies; more than 12 to 24 months in 21%; more than 24 months in 24%; and not reported in 10%. Separate data were reported for male and female adoptees in 24% of the studies and in 52% of the studies data for mixed groups were reported. The nonadopted control groups consisted of samples from the general population (50% of the studies), classmates (12%), siblings of the adoptees (6%), and norm groups, eg, Child Behavior Checklist norms (32%).

For mental health referrals, age at assessment was from 4 to 12 years in 53% of the studies; older than 12 to 18 years in 25%; and not reported in 22%. Age at adoptive placement was from 0 to 12 months in 31% of the studies; more than 12 to 24 months in 19%; more than 24 months in 8%; and not reported in 42%. Distinct data for female and male adoptees were not reported. The nonadopted control groups consisted of samples from the general population (8%) and normative data (92%).

Analyzing all adoption studies, we computed effect sizes (d)³² for the overall differences between adoptees (both within and between countries) and nonadopted controls (Table 4). Compared with nonadopted controls, adoptees showed more total behavior problems (d, 0.18; 95% CI, 0.13-0.24), more externalizing behavior problems (d, 0.24; 95% CI, 0.16-0.31), and more internalizing behavior problems (d, 0.16; 95% CI, 0.07-0.26), all in heterogeneous sets of studies, but all effect sizes were small.³² Also, adoptees were overrepresented in mental health referrals (d, 0.72; 95% CI, 0.57-0.86) in a heterogeneous set of studies, and this effect size was large. No publication bias was found in these 4 meta-analyses (Lo, 0 in all cases). The fail-safe number was k = 5251 for total problems, k = 3128 for externalizing problems, k = 2758 for internalizing problems, and k = 7282 for mental health referrals. Combined effect sizes and CIs computed with the jackknife procedure remained the same for all 4 meta-analyses.

There were 47 studies involving international adoptees reporting on total behavior problems, 29 studies on externalizing problems, 30 studies on internalizing problems (Table 5), and 7 studies reporting on mental health referrals (Table 4). The adopted children came from Romania or Russia,^{20- 21,59,71,75,82} Korea (and other countries),^{40,44,49,76,87,89} India (and other countries),^22,45,65 Colombia (and other countries),⁵⁷ Thailand,⁶⁷ Indonesia,⁶⁸ China,⁸⁶ Sri Lanka (and other countries),¹⁷ Greece,⁹² and several countries in Asia and South America.^7,18,41,74

Compared with nonadopted controls, international adoptees showed more total behavior problems (d, 0.11; Figure 1). Combined effect size and CIs computed with the jackknife procedure remained the same. With the trim-and-fill procedure, a publication bias was found in this meta-analysis (Lo, 13), resulting in an adjusted effect of d , 0.06 (95% CI, 0.04-0.09). The fail-safe number was k = 577. Compared with nonadopted controls, international adoptees presented more externalizing problems (d, 0.10; 95% CI, 0.07-0.13) in a homogeneous set of studies. The jackknife procedure yielded a similar point estimate and the same CIs. With the trim-and-fill procedure, 3 studies were trimmed and replaced (Lo, 3), resulting in an adjusted effect of 0.09 (95% CI, 0.05-0.12). The fail-safe number was k = 162. International adoptees presented more internalizing problems (d, 0.07; 95% CI, 0.04-0.11) in a homogeneous set of studies. The jackknife procedure produced the same combined effect size and CIs. No publication bias was found (Lo, 0), and the fail-safe number was k = 84. For behavior problems, all effect sizes were small.³²

Finally, international adoptees were overrepresented in mental health referrals (d, 0.37; Figure 2) and this effect size was medium. The jackknife procedure produced the same combined effect size and CIs. No publication bias was found (Lo, 0) and the fail-safe number was k = 195. The 7 studies in this meta-analysis reported on serious problems (Table 3): 4 studies found that international adoptees were more often receiving psychiatric treatment¹⁰⁷ in a clinic^99- 100,127 than nonadopted children, and 3 studies found that international adoptees were placed out of the home into a residential setting^108,128 or mental health facility⁷⁴ more often than nonadopted controls.

We examined whether international adoptees are at higher risk for behavior problems and clinical referrals than domestic adoptees (Table 4). As several subsets of studies in this meta-analysis were heterogeneous, we present the 85% CIs to test the significance of moderators. No publication bias was present in the meta-analyses of international adoptees vs domestic adoptees (Lo = 0 for total behavior problems [fail-safe, k = 2268], for externalizing [fail-safe, k = 1745], and for internalizing problems [fail-safe, k = 1768]). Preliminary analyses showed no differences for sex and age at adoptive placement; therefore, all analyses were conducted without these covariates. However, the sets of international and domestic adoption studies differed in the number of studies that reported evidence of extreme adversity before adoptive placement. Preadoption adversity was described more often in international adoption studies (18 of 21 studies of total behavior problems, 7 of 8 studies of externalizing and internalizing problems, and a single study of mental health referrals).

Contrary to our expectations, we found that international adoptees showed significantly fewer total behavior problems compared with domestic adoptees, for the 85% CIs of the subsets were not overlapping (d, 0.11; 85% CI, 0.09-0.13 vs d, 0.20; 85% CI, 0.14-0.27, respectively; Table 4). Also, international adoptees showed significantly fewer externalizing problems than domestic adoptees (d, 0.10; 85% CI, 0.08-0.13 vs d, 0.34; 85% CI, 0.26-0.42, respectively) and also significantly fewer internalizing problems (d, 0.07; 85% CI, 0.05-0.10 vs d, 0.23; 85% CI, 0.13-0.32, respectively). Because all international adoption studies were conducted after 1990, we repeated the same analyses including only the domestic adoption studies conducted after 1990. Again, international adoptees showed significantly fewer total behavior problems than domestic adoptees (d, 0.11; 85% CI, 0.09-0.13; k, 47 vs d, 0.22; 85% CI, 0.14-0.29; k, 40, respectively), fewer externalizing problems (d, 0.10; 85% CI, 0.08-0.13; k, 29 vs d, 0.30; 85% CI, 0.20-0.39; k, 25, respectively), and fewer internalizing problems (d, 0.07; 85% CI, 0.05-0.10; k, 30 vs d, 0.27; 85% CI, 0.16-0.37; k, 24, respectively).

International adoptees were significantly less often referred to mental health services compared with domestic adoptees (d, 0.37; 85% CI, 0.22-0.52 vs d, 0.81; 85% CI, 0.67- 0.94; respectively; Table 4). However, the set of pertinent studies involving international adoptees in mental health referrals was small (k = 7). Because all international adoption studies were conducted after 1980, we repeated the same analysis including only the domestic adoption studies conducted after 1980. Again, international adoptees were significantly less often referred to mental health services than domestic adoptees (d, 0.37; 85% CI, 0.22-0.52; k, 7 vs d, 0.78; 85% CI, 0.57-1.00; k = 14; respectively).

Moderator Analyses. For behavior problems, we present fixed models with 95% CIs for the homogeneous set of studies involving international adoptees (Table 5); the Q statistic was used to test contrasts. The set of international adoption studies for mental health referrals was too small (k = 7) to permit moderator analyses.

Sample Characteristics. The following sample characteristics were tested: preadoption adversity, sex, age at adoptive placement, age at assessment, and length of time in the family (Table 5). In 6 out of 9 articles reporting preadoption adversity, children had been adopted from Romanian or Russian orphanages.^{20- 21,59,71,75,82} International adoptees with preadoption adversity showed more total behavior problems than international adoptees without such backgrounds (d, 0.18 vs d, 0.09, respectively; Table 5; contrast: Q₁ = 6.46; P = .01) and more externalizing problems (d, 0.17 vs d, 0.08; respectively; Q₁ = 4.58; P = .03). There was no difference in internalizing problem behavior between international adoptees with and without preadoption adversity (Q₁ = 0.30; P = .58).

We found no significant differences between male and female international adoptees for total behavior problems (Q₁ = 1.30; P = .25), externalizing problems (Q₁ = 1.20; P = .27), or internalizing problems (Q₁ = 0.66; P=.41).

For children adopted as infants (0-12 months) compared with children adopted after their first birthday, there were no differences for total behavior problems (Q₁ = 2.27; P = .13), externalizing problems (Q₁ = 3.44; P = .06), or internalizing problems (Q₁ = 0.23; P = .63; Table 5). Examining children adopted before or after 24 months resulted in similar, nonsignificant outcomes.

As the category of adulthood (>18 years) consisted of only 1 to 2 studies (Table 5), we restricted the analyses of age at assessment to adolescence (12-18 years) vs early and middle childhood (0-12 years). Contrary to our expectations, we found that international adoptees presented fewer total behavior problems in adolescence compared with international adoptees in early and middle childhood (d, 0.09 vs d, 0.23; respectively; Q₁ = 13.89; P <.001). Externalizing problems did not differ for adolescence vs early and middle childhood (d, 0.09 vs d, 0.17; respectively; Q₁ = 2.76; P = .10), nor did internalizing problems (d, 0.06 vs d, 0.14; respectively; Q₁ = 2.04; P = .15).

Children who had been with their adoptive family for more than 12 years showed fewer total behavior problems than children who had been with the family for less than 12 years (d, 0.05 vs d, 0.21; respectively; Q₁ = 24.07; P <.001) and fewer externalizing problems (d, 0.07 vs d, 0.18; respectively; Q₁ = 8.52; P = .003). For internalizing problems, the contrast was not significant (d, 0.06 vs d, 0.12; respectively; Q₁ = 2.82; P =.09).

Design. Studies that made use of a norm group as a comparison group for the international adoptees did not differ from studies that used a general population sample, classmates, or siblings (Table 5). For total behavior problems, one of the subsets was heterogeneous (studies not using a norm group) so the 85% CIs were inspected to test for significance. Because the CIs were overlapping (norm group, d, 0.18; 85% CI, 0.14-0.23 vs no norm group, d, 0.13; 85% CI, 0.08-0.18), there was no difference between the 2 subsets. For externalizing and internalizing problems, the 2 subsets were homogeneous and these contrasts were tested with the Q statistic. There was no significant difference between studies using norm groups or other control groups for externalizing problems (Q₁ = 2.28; P = .13) or internalizing problems (Q₁ = 1.30; P = .25).

Country of study was a significant moderator (Table 5). Studies conducted in North America reported more total behavior problems for international adoptees than studies outside North America (d, 0.23 vs d, 0.10; respectively; Q₁ = 4.69; P = .03). Studies in and outside North America did not differ with respect to externalizing problems (Q₁ = 0.04; P = .85) or internalizing problems (Q₁ = 0.50; P = .48).

Publication Outlet. We examined the 85% CIs as one of the subsets for total problems was heterogeneous. Confidence intervals of both subsets were overlapping: journal articles (d, 0.16; 85% CI, 0.11-0.20) did not differ from other outlets (d, 0.11; 85% CI; 0.06-0.16). The contrasts for externalizing problems (Q₁ = 0.04; P = .84) and internalizing problems (Q₁ = 0.31; P = .58) also showed that journal articles did not differ from other outlets.

As expected from their less optimal start in life, international adoptees presented with more total, externalizing, and internalizing behavior problems than their nonadopted peers and are overrepresented in mental health services. However, the rate of behavior problems is modest, indicating that most international adoptees are well-adjusted. These findings converge with those of a large Swedish cohort study¹⁸ that found that the majority of international adoptees are well-adjusted and with those of other studies of international adoptees in the socioemotional¹⁷ and cognitive domains.^14- 15These positive outcomes may be partly explained by the characteristics of the families adopting children from abroad. These adoptive parents are highly motivated to raise children and they usually have ample opportunities to invest in their children’s development because of their relatively high socioeconomic status.^7,15,17- 18 International adoptees, however, experience substantially more mental health referrals, pointing to a relatively large minority of international adoptees seeking clinical treatment. The threshold to seek professional help, however, might be lower for adoptive parents than for birth parents⁹⁵ because of the adoptive parents’ higher socioeconomic status or their expectations of the adopted child.⁶⁷ Having adopted, they are familiar with mental health resources and how to get services. Furthermore, schools may be more aware of the child’s adoptive status and more likely to recommend referral or to report behavior problems. Also, normative crises in adopted children, eg, coming to terms with the loss of their birth family,¹¹ may be misperceived as behavior problems. Finally, a positive explanation of our outcomes may be that the higher referral rate did in fact prevent higher rates of behavior problems, resulting in the small effect sizes for problem behavior.

In contrast to popular beliefs and hypotheses expressed in empirical studies,¹⁸ international adoptions show better behavioral and mental health outcomes than domestic adoptions. Our findings indicate that this is not explained by lower rates of preadoption adversity experienced by the international adoptees compared with domestic adoptees, as evidence of pre-adoption malnutrition, neglect, or abuse was reported more often in the international adoption studies. It is possible that in many transracial international adoptions, physical differences between parents and children are so obvious that the fact of the adoption was never a secret, resulting in more communication and trust in the family. Families choosing international (transracial) adoption may have different parenting qualities compared with parents in more traditional adoptions. No systematic information about parenting abilities is available in our data set. However, in most countries parents undergo a screening procedure to assess their potential fitness for parenting and receive (some) preparation. Finally, genetic risks may differ between international and domestic adoption. Whereas children in international adoption are often adopted because of lack of resources and poverty,¹ relinquishment in domestic adoption may (also) involve mental health problems in the birth parent,⁷⁴ such as substance abuse or psychiatric disorders. Although reasons for relinquishment may overlap, genetic risks predisposing for mental health problems may be less prevalent in international adoptees.

The relatively positive outcomes of international adoption do not imply that international adoption should be preferred to domestic adoption in the sending countries. In our meta-analyses, domestic adoptions in developing countries could not be included due to the lack of empirical studies.

Our meta-analytic outcomes confirm the hypothesized greater risk for internationally adopted children with backgrounds of extreme deprivation, neglect, malnutrition, or abuse. Clinicians and mental health professionals should be aware of this risk and support adoptive parents with preventive or therapeutic help.

In contrast to some evidence,^7,17 internationally adopted males do not present more behavior problems than internationally adopted females. Furthermore, we did not find convincing evidence that age at adoptive placement is a decisive factor for international adoptees’ behavior problems. Contrary to previous research,^7,130 we found that international adoptees showed fewer total behavior problems in adolescence compared with international adoptees in early and middle childhood. Although it might be true in general that adoptees are questioning their identity more intensively in adolescence,^11,130 international adoptees may begin struggling with identity issues much earlier because racial and cultural differences between adoptive parents and adoptees are more obvious than in domestic adoption. Some behavior problems in adoptees may occur on a different time schedule than in nonadopted children. For example, identity issues may surface earlier in adoptees than in their nonadopted peers. Therefore, mental health professionals should be aware of increased rates of behavior problems in families with international adoptees during the years before adolescence. We also found that children who had been with the adoptive family for more than 12 years showed fewer total and externalizing behavior problems than children who had been in the family for less than 12 years. This may indicate that a longer stay in the adoptive family offers children opportunities to recover from their problem behavior. Finally, we found more total behavior problems in studies conducted in North America. On the basis of our data base, we are unable to suggest explanations for this finding. Future research should examine this issue.

Limitations of our series of meta-analyses are, first, the small number of studies of international adoptees with mental health referrals. More studies are needed to consolidate these findings. However, the meta-analytic findings on the behavioral outcomes of international adoptees converge with the mental health referral findings. The small number of studies on mental health referrals of international adoptees also precluded moderator analyses. A second limitation is that our definitions of international and domestic adoption and preadoption adversity may have introduced bias, as in some domestic adoption studies a minority of international adoptees were included and in samples without adversity some adoptees may have been neglected or abused. However, if such bias had been present, it would have resulted in an underestimation of our effects. Based on the positive outcomes for international adoptees and the negative outcomes for preadoption adversity, even larger differences in favor of international adoptees without preadoption adversity may be expected in totally unbiased samples. A third limitation is that we used only the first assessment of longitudinal adoption studies, possibly resulting in a bias toward fewer behavior problems. However, we know of only 1 study^7,130 for which this would apply, restricting the possibility of such a bias to a minimum. A fourth limitation is that our findings may not generalize to the large group of Chinese children adopted in the United States, Canada, and Europe in recent years because their development has not been studied well yet (with one exception⁸⁶). A fifth limitation is that we were unable to compare the international adoptees and nonadopted controls on demographic background variables although in most studies it was reported that adoptive parents were somewhat older and more highly educated than the parents of the controls.^{7,17- 18,60,71,74- 75} It is unknown how the demographics would affect the outcomes of our meta-analysis. A final limitation is that we only included studies with nonadopted control groups, thus excluding articles comparing international adoptees with other comparison groups, such as children in foster care or children remaining in institutions. In a meta-analysis of adopted children’s cognitive development, we found that adopted children outperformed their peers and siblings who remained in the children’s home or birth family.¹⁴ In the current meta-analysis, such a comparison was not possible because there were no studies available addressing this issue. For future studies, it is important to compare internationally adopted children not only with nonadopted controls but also with these other relevant groups.

In sum, our series of meta-analyses showed that the majority of international adoptees are well-adjusted although more adoptees are referred to mental health services compared with nonadopted controls. Contrary to common opinion, international adoptees present fewer behavior problems than domestic adoptees, and they have lower rates of mental health referral. Unexpectedly, age at adoption does not appear to be important for the development of behavioral problems. International adoptees with backgrounds of extreme adversity are at risk for more behavior problems, in particular externalizing problems, compared with international adoptees without preadoption adversity. Clinicians should be aware of higher risks for problem behaviors in domestic adoptees and in international adoptees who experienced neglect or maltreatment in the preadoptive period.