CHILDREN AND YOUTH IN FOSTER CARE: DISENTANGLING THE RELATIONSHIP BETWEEN PROBLEM BEHAVIORS AND NUMBER OF PLACEMENTS

Transcription

1 Pergamon PII S (00) Child Abuse & Neglect, Vol. 24, No. 10, pp , 2000 Copyright 2000 Elsevier Science Ltd. Printed in the USA. All rights reserved /00/$ see front matter CHILDREN AND YOUTH IN FOSTER CARE: DISENTANGLING THE RELATIONSHIP BETWEEN PROBLEM BEHAVIORS AND NUMBER OF PLACEMENTS RAE R. NEWTON Department of Sociology, California State University and Child and Family Research Group, Child and Adolescent Health Services Research Center, Children s Hospital and Health Center, San Diego, CA, USA ALAN J. LITROWNIK Department of Psychology, San Diego State University and Child and Family Research Group, Child and Adolescent Health Services Research Center, Children s Hospital and Health Center, San Diego, CA, USA JOHN A. LANDSVERK Department of Social Work, San Diego State University and Child and Family Research Group, Child and Adolescent Health Services Research Center, Children s Hospital and Health Center, San Diego, CA, USA ABSTRACT Objective: The purpose of this research was to provide a prospective look at the relationship between change in placement and problem behaviors over a 12-month period among a cohort of foster children. Method: The sample contained 415 youth, and was part of a larger cohort of children who entered foster care in San Diego, California and remained in placement for at least 5 months. The Child Behavior Check List was used to assess behavior problems. Every change of placement during the first 18 months after entry into the foster care system was abstracted from case records. Results: The results suggest that volatile placement histories contribute negatively to both internalizing and externalizing behavior of foster children, and that children who experience numerous changes in placement may be at particularly high risk for these deleterious effects. Initial externalizing behaviors proved to be the strongest predictor of placement changes for the entire sample and for a sub-sample of those who initially evidenced problem behaviors on at least one broad-band CBCL scale. Our findings also suggest that children who initially score within normal ranges on the CBCL may be particularly vulnerable to the detrimental effects of placement breakdowns. Conclusions: On the basis of these findings we argue for an analytical approach that views behavior problems as both a cause and as a consequence of placement disruption. Children who do not evidence behavior problems may in fact constitute a neglected population that responds to multiple disruptions of their primary relationships with increasingly self-defeating behaviors Elsevier Science Ltd. Key Words Placement change, Foster care stability, CBCL, Foster care youth, Foster care placement. Submitted for publication July 15, 1999; final revision received February 28, 2000; accepted March 1, Requests for reprints should be sent to Rae R. Newton, PhD, Department of Sociology, California State University, P.O. Box 6846, Fullerton, CA

2 1364 R. R. Newton, A. J. Litrownik, and J. A. Landsverk INTRODUCTION RESEARCHERS AND CHILD welfare workers agree that placement stability is critical for the success of foster care placement. Once removed from one dangerous or neglectful environment, a child confronting further disruption through numerous placement failures is likely to experience difficulties trusting adults or forming attachments with adults and children. It is well known that such children, entering the foster care system, exhibit behavioral and mental health problems, or are at risk for those problems. McIntyre and Keesler (1986) found that 50% of the foster children they studied evidenced behavioral problems on the Child Behavior Check List (CBCL; Achenbach, 1991). Hochstadt, Jaudes, Zimo, and Schachter (1987) report that 56.9% of children entering foster care in Chicago during a 1 month sampling had significant emotional and/or behavioral problems requiring treatment and Clausen, Landsverk, Ganger, Chadwick, and Litrownik (1998) report that 61% of their sample of foster children in San Diego showed indications of a mental health problem as assessed by the CBCL. It is also well known that children entering the foster care system are likely to experience change in placement, and that some are likely to experience a large number of such placement changes. For example, Pardeck (1984) found that 22% of children had three or more placements during a median length of 2.5 years in foster care. Millham, Bullock, Hosie, and Haak (1986) report that after 2 years in care 56% of children had three or more placements and Kufeldt, Armstrong, and Dorosh (1989) report that 48% of children had three or more placements during a median length of stay of 4 years. According to Palmer (1996) citing the Millham and colleagues (1986) and Kufeldt and colleagues (1989) studies, some children experienced an unconscionable number of moves [emphasis added]. In the Millham study 14% had five or more, in the Kufeldt study 18% had six or more. There is also evidence that placement disruption and behavior problems are associated and that multiple placements in out-of-home care are associated with both immediate and long-term negative outcomes for the child (Cook, 1994; Festinger, 1983; Kufeldt et al., 1989; Millham et al., 1986; Pardeck, 1984; Penzerro & Lein, 1995; Piliavin, Sosin, Westerfelt, & Matsueda, 1992). The conditions responsible for placement disruption are varied. Some scholars consider disordered attachments directly responsible and describe a cohort of children who display exceptionally clear patterns of alienation in relation to transitions from placement to placement (Penzerro & Lein, 1995, p. 351). However, the reasons for failed placements are varied and may include factors independent of the behavior of either the child or social worker (Staff & Fein, 1995). Empirical inquiries examining the correlates of placement disruption confirm the importance of behavioral and emotional problems (Pardeck, 1984) and issues with attachment (Palmer, 1996; Pardeck, 1984; Walsh & Walsh, 1990). However, other studies have found a variety of additional factors related to placement change, including sex and age of children, expectations of foster families, social workers training, and type of foster care (Doelling & Johnson, 1990; Palmer, 1996; Proch & Taber, 1985; Ray & Horner, 1990; Stone & Stone, 1983). Taken as a whole, studies of foster care stability are difficult to compare and suffer from a number of methodological problems. The studies we reviewed are primarily cross-sectional or qualitative, thus precluding any analysis of the child s behavioral trajectories as a result of multiple placements in out-of-home care. Thus, the role of internalizing or externalizing behaviors as a cause or a consequence of multiple placements is difficult, if not impossible, to assess. Second, most studies contain measurement protocols that lack objective, standardized measures with known psychometric properties. In addition, the sample and/or measurement protocols that have been utilized are typically too narrow to permit partialling the relative influence of variables found to be important in the various studies. Finally, the definition of a placement and a placement change is itself problematic. Some studies include short stays in receiving facilities when counting the total number of placements, while others ignore these short term evaluative stays and do not count movement into, or out of, a receiving facility as a placement change. For example, if each foster

3 Placement change and problem behaviors 1365 care or group home placement is preceded by a 2-day evaluative stay in a receiving facility two 1-month stays in foster care could be reported as four placements by one study, but only two by another. The purpose of this research is to provide a prospective look at the relationship between change in placement and problem behaviors over a 12-month period among a cohort of foster children. We remedy some of the weaknesses of previous studies by examining the data within a longitudinal framework with a large sample and a well standardized measurement package. Sample METHODS Participants included in the present study were part of a larger cohort of children between 0 17 who entered foster care in San Diego, California between May 1990 and October 1991, and remained in placement for at least 5 months (N 1,221). Data were obtained for 78% of the total cohort (N 934). (Primary reasons for not obtaining interview data were refusal to participate by the caretaker and inability to contact the caretaker.) Children under 2 years of age (N 232) were excluded because the measure of behavior problems we utilized, the CBCL, is inappropriate for children under 2. Parents, or parent substitutes, were interviewed at two time points, with the first interview beginning at approximately 5 months (Time 1 interview) and the second beginning at approximately 17 months (Time 2 interview) following initial entry into foster care. These interviews provided the standardized measures of child functioning used in this analysis. Because some children showed tremendous variability in the timing of these two interviews we limited the range between interview periods to no less than 6 and no more than 18 months (mean 13.4 months, standard deviation 4.8 months). Of the 465 participants meeting these criteria, those who did not have complete CBCL data for both time periods (N 40), or for whom the number of placement changes could not be calculated (N 10) were subsequently excluded from the analysis. The final sample contained 415 subjects, of whom 46.5% were male, 45.0% were Anglo, 17.1% were Hispanic, and 34.5% were African-American (2.4% were of other ethnic origins). The mean age at entry into foster care for this subgroup was 6.6 years with a standard deviation of 3.9 years. Measure of Behavior Problems The Child Behavior Check List (parallel CBCL versions 2 3 and 4 18, Achenbach, 1991) was used to assess behavior problems. The CBCL is a well-established measure, its validity having been supported in numerous studies (Achenbach, 1991). The 4 18 CBCL includes 113 questions about behavior problems, while the 2 3 CBCL contains 99. Fifty-nine of the 99 items on the 2 3 CBCL directly overlap with the 4 18 CBCL Items. For more detail regarding the differences between these two forms see Achenbach (1991, p. 214). Items are scored on a 3-point Likert type scale representing the parent or substitute parent s perception of the frequency of behavior problems (i.e., not true, somewhat or sometimes true, very often or often true). The behavior problem items are scored to create a total behavior problems scale, eight to nine factor based scores that represent narrow-band syndromes (e.g., withdrawn, somatic complaints, anxious/depressed) and two broad-band syndromes, internalizing and externalizing, that were originally developed through the use of second-order factor analyses of the narrow-band scales. These analyses used only the two broad band scales and total behavior problems (hereafter referred to as the CBCL scales ). Achenbach generally recommends that raw scores be used for research purposes as opposed to T scores; however, since our sample contained considerable age heterogeneity and we wished to combine subjects who had used different versions of the CBCL, the use of standard T scores was recommended.

4 1366 R. R. Newton, A. J. Litrownik, and J. A. Landsverk Table 1. Five Hundred Fifty Days in Out Of-Home Placement Entry Date Exit Date Type of Facility Length of Stay (Days) Running Total Placement Number OOO Receiving Foster Care OOO Receiving Foster Care Foster Care Foster Care O Group Home Foster Care O Group Home Medical Facility O Group Home OOOOO Group Home Group Home OOO Receiving OOOOO End of Study Group Home Notes. The child represented above is a 10-year-old African-American male removed from the home for physical abuse. Arrows connect identical placements. Number of Placement Changes Every change of placement during the first 18 months after entry into the foster care system was abstracted from case records. These included the initial entry into the foster care receiving facility through which most, but not all, children in San Diego pass prior to placement; thus, every child had at least one placement, even if the child entered the receiving facility and was subsequently reunified with his or her biological parents. Children who changed placement through re-entry into the receiving facility and placement elsewhere would be recorded as having experienced two changes. This recording strategy may in part account for the seemingly large number of placement changes in our sample compared to those reported by other studies. In this sample the total number of placement changes ranged from 1 to 15, with a mean of 4.23 (SD 2.00 ) and a median of 4. Because the authors questioned how a single child could experience 15 placement changes in 18 months, the raw data for this case were re-abstracted. The placement data for this 10 year old African-American male, removed from the home for physical abuse, proved valid and is presented in Table 1 as an exemplary case study of multiple placements. During the 550 days for which we recorded this child s 15 placement episodes his shortest stay was 1 day and his longest was 169 days, both representing stays in foster care. It is interesting to note that only three of the 15 placements represent transition stays in receiving facilities. The child was in five different foster care settings, and six group homes, being placed in one group home three times and another twice, where he remained at the end of the study. RESULTS The general analytic strategy used to guide this analysis is based on the assumption that the initial position of a child vis-à-vis his or her psycho-social functioning forms a critical point of departure for evaluation of the relationship between change of placement and functioning. The sample was initially divided into two groups on the basis of CBCL scores at the time of the first interview (Time 1). The two groups were composed of those children below all borderline clinical cutpoints (T 59) and those above at least one cutpoint (T 60) on the three CBCL broad-band

5 Placement change and problem behaviors 1367 Table 2. Child Behavior Checklist and Placement Change Statistics for Two Time Points and CBCL Groupings Child Behavior Checklist Groupings (CBCL Status at Time 1) Below All Cutpoints (N 173) Above One or More Cutpoints (N 242) Time 1 Time 2 Time 1 Time 2 Group Differences Mean SD Mean SD t Mean SD Mean SD t Time 1 (t) Time 2 (t) CBCL Broad Band Scales Behavior Problems * * 28.1* 8.3* Internalizing * * 21.7* 6.5* Externalizing * * 24.9* 1 7.3* No. of Placement Changes Below All (N 173) Above One (N 242) Mean SD Mean SD t Notes. Below All indicates that the child was below the borderline cutpoint for all CBCL Broad Band Scales. Above One means that the child was above the borderline cutpoint for at least one CBCL Broad Band Scale. The variances were significantly different. The t-test was adjusted. P.004. *p.001. scales that we assessed, representing internalizing behaviors, externalizing behaviors and total behavior problems. Approximately 42% (N 173) of the sample was below all three borderline cutpoints, while the remaining 58% (N 242) was above at least one of these three cutpoints. This division of the sample is subsequently referred to as ACBCL status at Time 1 and forms the basis for the bivariate analyses contained in Tables 2 and 3 which report t-tests and bivariate correlations, respectively. These analyses demonstrate the importance of the initial (Time 1) assessment of child behavior problems for understanding the results. The final two tables (Tables 4& 5) contain hierarchical regression models that utilize CBCL scores at Time 1 as covariates. These tables explore the role of placement changes in predicting Time 2 child behavior problems, while controlling for initial problems. Table 3. Correlations Between Number of Placement Changes and CBCL Time 1 and Time 2 Scores by CBCL Status at Time 1 CBCL Status at Time 1 Below All Borderlines (N 173) Above One or More Borderline (N 242) Total Sample (N 415) Time 1 CBCL Broad Band Scales Total behavior problems ** Internalizing * Externalizing **.189** Time 2 CBCL Broad Band Scales Total behavior problems.244** ** Internalizing.260** ** Externalizing.250** ** *p.01; **p.001.

6 1368 R. R. Newton, A. J. Litrownik, and J. A. Landsverk Table 4. Hierarchical Regression Analysis of CBCL Time 2 Broad Band Scales on CBCL Time 1 Broad Band Scales (Step 1), Sex, Age, and Number of Placement Changes (Step 2). Dependent Variable Time 2 CBCL T Scores (N 415) Beta Adj R 2 R 2 Change (%) Total Behavior Problems Step 1 TBP Time Step 2 TDP Time No. changes Internalizing Step 1 Int. Time Step 2 Int. Time No. changes Externalizing Step 1 Ext. Time Step 2 No. changes Notes. At step 2 variables were entered in stepwise fashion (Probability of F to enter was set at.05). All step 2 models included gender, entry age, and number of placement changes. Gender and age were not statistically significant in any model and are not included above. Ethnicity was assessed in a separate analysis of covariance and was not related to CBCL scores. For all models, at step 1 and 2, time 1 CBCL score (the covariate) is significant at less than.001 as is the overall regression equation. For all models, the beta weight for number of placement changes and change in R 2 values attributable to this variable is statistically significant at less than.01. Table 2 contains the means and standard deviations for the CBCL broad-band scales, at both interview periods (Time 1 and Time 2) for the two groups (those Below All or those Above One CBCL broad-band borderline) formed on the basis of their CBCL status at Time 1. Table 2 assesses differences in CBCL scores in two ways. First, the Time 1/Time 2 differences within each group formed on the basis of CBCL status at Time 1 are assessed using pairedcomparison t-tests. Second, the CBCL group (Below All vs. Above One) differences within each time period is assessed using independent groups design t-tests. As would be expected, these later tests, examining CBCL differences formed on the basis of CBCL status at Time 1, produce large test statistics. In addition, Time 1 versus Time 2 differences, within the CBCL status at Time 1 groupings, were also statistically significant. Those initially below all borderline cutpoints increased approximately 3 4 points when assessed at Time 2, while those above at least one borderline decreased between 6 8 points. These differences could be attributable to regression artifacts; however, subsequent analyses will show that CBCL scores at Time 2 are associated in predictable ways with changes in placement. The mean number of placement changes for each CBCL grouping is also included in Table 2. The mean number of placement changes for those children who were initially above at least one CBCL broad-band borderline is significantly higher than the mean of the children below all such borderlines (4.6 vs. 3.9, t[413] 2.9, p.004). The group above one borderline also had a larger, but not significantly different, standard deviation (2.2 vs. 1.9). These differences in distribution characteristics were further examined with boxplots for each group, as shown in Figure 1. The boxplots illustrate one similarity and a number of differences between the two groups. First, the two CBCL groups have the same median (4), but the Above One group has a larger interquartile range (3 6 vs. 3 4). The small interquartile range for the Below All group results in a large number of cases (N 26) being classified as outliers, with 14 of these being classified as extreme outliers (i.e., more than 3 boxwidths above 75th percentile). (The number of cases represented by each symbol is shown in parentheses.) Only 6 cases (2.5%) of the Above One

7 Placement change and problem behaviors 1369 Table 5. Hierarchical Multiple Regression Analysis of CBCL Time 2 Broad Band Scales on CBCL Time 1 Broad Band Scales (Step 1), Sex, Age, and Number of Placement Changes (Step 2) by Change Group Cutpoints. Dependent Variable Time 2 CBCL T Scores (N 415) Number of Placement Changes (Median Split) Less Than 5 (N 317) 5 or More (N 98) Beta Adj R 2 Beta Adj R 2 R 2 Change (%) Total Behavior Problems Step 1 TBP Time 1.526**.274**.454**.198** Step 2 (stepwise) TBP Time 1.526**.274**.433**.198** No. changes NS NS.270**.264** 6.6 Internalizing Step 1 Int. Time 1.456**.205**.291**.075** Step 2 (Stepwise) TBP Time 1.456**.205**.282**.075** NS NS.325**.173** 9.7 Externalizing Step 1 Ext. Time 1.457**.206**.469**.201** Step 2 (Stepwise) TBP Time 1.457**.206**.423**.201** No. changes NS NS.273**.268** 6.7 Notes. At step 2, variables were entered in stepwise fashion (Probability of F to enter was set at.05) All step two models included gender, entry age, and number of placement changes. Gender and age were not statistially significant in any model and are not included above. Ethnicity was assessed in a separate analysis of covariance and was not related to CBCL scors. group are considered outliers and this group contains no extreme outliers. Thus, as shown in both Table 2 and Figure 1, children who enter foster care exhibiting behavioral problems on at least one CBCL broad band scale appear more likely to experience numerous placement changes. However, among the group of children who do not appear to exhibit behavioral problems as reported by their foster parent, there exists a subgroup that experience a large number of changes relative to other similarly classified children. To further explore the relationships between the manifestation of behavioral problems and multiple placement changes, both early and at later stages of a child s placement history, we correlated the CBCL scores at both Time 1 and Time 2 with the number of placement changes. Table 3 presents the results of these analyses for each CBCL grouping formed on the basis of Time 1 CBCL cutpoints (i.e., Below All, Above One) and the total sample. For the total sample (N 415) correlations were statistically significant (p.01) and ranged from.101 to.189, suggesting a small, but statistically significant relationship between placement instability and behavior problems. When the total sample is divided on the basis of borderline cutpoint status at Time 1, two patterns emerge. First, for the 242 cases classified as above at least one broad-band CBCL borderline at Time 1, the correlations are near zero. The exception appears to be externalizing behavior assessed at Time 1 (r[242].173, p.001). Thus, for those children initially exhibiting some evidence of behavioral problems, externalizing behavior at Time 1 is significantly related to the number of placement changes. Second, for those 173 children not initially exhibiting evidence of behavioral problems, the number of placement changes was significantly related to all three CBCL broad band scale scores as measured at Time 2 (r.244 to.260, p.001), but not at Time 1(r.009 to.053, NS). These findings suggest that for the entire sample, Time 1 CBCL scores, which are recorded early

8 1370 R. R. Newton, A. J. Litrownik, and J. A. Landsverk Figure 1. Boxplots of Placement Change by CBCL Groupings. in the child s placement history, may predict a small portion of the variability in number of placement changes, particularly for the case of externalizing behaviors. In addition, number of placement changes may predict a small portion of the variability in CBCL scores, measured later in the child s placement history. Dividing the sample into two groups on the basis of borderline cutpoints at Time 1 appears to elaborate these relationships. The first assessment of behavior problems indicates that only externalizing behavior among children exhibiting at least one CBCL above borderline score predicts number of placement changes. However, the second assessment of behavioral problems, which occurs after a minimum of 6 months in an out-of-home setting, suggests that number of placement changes may be predictive of CBCL scores and that these relationships are primarily accounted for by those initially below all three borderline cutpoints on the CBCL as assessed at Time 1. Based on the above findings we concluded that a multivariate analysis including Time 1 CBCL status as a covariate concurrently with those variables found by others to be predictors of placement variability should be undertaken. To begin, three, 2-step, hierarchical multiple regression analyses were conducted on the total sample (N 415). One analysis was conducted for each CBCL broad band scale (total behavior problems, internalizing and externalizing) and in all analyses the first step included only the appropriate Time 1 CBCL score, which was used as a covariate. In step two, utilizing a stepwise method, age at entry into foster care, gender and number of placement changes were entered. (Race/ethnicity was assessed in three separate ANCOVAs not reported here. There were no significant race differences in CBCL broad band scale scores at Time 2 when controlling for the respective CBCL scores at Time 1.) All analyses showed that gender and age at entry into foster care play no statistically significant role in the prediction of Time 2 CBCL scores and these variables were excluded from the model

9 Placement change and problem behaviors 1371 at step 2. (Considerable research shows that length of time in foster care is significantly related to number of placement changes. We were unable to include this variable in our models because a high proportion of the sample remained in foster care at the end of the study period [18 months].) The final results, shown in Table 4, include only Time 1 CBCL score for the relevant broad band scale, entered as a covariate at step one, and number of placement changes, entered at step 2. Thus, other than the covariate, number of placement changes is the only variable to play a statistically significant role in the prediction of behavioral problems. As would be expected, in all analyses, the Time 1 CBCL score was a significant predictor of Time 2 CBCL score. Of greater importance is the role of number of placement changes in predicting Time 2 CBCL scores after controlling for CBCL scores at Time 1. For the total sample, number of changes in placement is a significant predictor of Time 2 CBCL scores, even when the influence of Time 1 CBCL scores is first removed from the analyses. The contribution to the adjusted R 2, and the relative size of the beta weights, indicates that most of the explained variance is accounted for by the covariate. Further consideration of the above findings and the bivariate correlations shown in Table 3 led to the suggestion that those with more volatile placement histories might be particularly at risk for the effects of multiple placements on internalizing and externalizing behaviors. To examine this hypothesis the sample was divided into two groups on the basis of number of placement changes, using a median split (median 4). This resulted in two groups, one representing less than five changes (N 317) and one with five or more changes (N 98). The hierarchical regression analyses for each group was then repeated. These results are presented in Table 5. For all analyses with that portion of the sample (N 317) experiencing less than five placement changes, the influence of placement changes on CBCL scores, after removing the covariate (CBCL scores at Time 1) is not statistically significant. For those with volatile placement histories, the contribution of number of placement changes to adjusted R 2 ranges from to 6.6 to 9.7%. We discuss the implications of these findings in the next section; however, a few points, both statistical and substantive, help elucidate these findings, and summarize our analyses. First, while there is clearly more variability to explain within that group of children who experience more placements, there is also considerably less power (N 98 vs. N 317). Thus, the fact that the more powerful analysis failed to produce a statistically significant finding for number of placement changes, independent of effect size, is important, and suggests that the statistically significant effects for the total sample, shown in Table 4, are largely due to the influence of high change children. Second, the amount of variability accounted for by number of placement changes among the high change children is substantively important, considering the control for initial status. Not only does this finding demonstrate that high change children get worse, but also that getting worse is directly attributable to volatile placement histories, independent of initial starting position, race/ethnicity, age, or gender. DISCUSSION The results of this study suggest that volatile placement histories contribute negatively to both internalizing and externalizing behavior of foster children, and that children who experience numerous changes in placement may be at particularly high risk for these deleterious effects. The results also suggest that while many children enter foster care with clinically elevated behavior problem profiles, disordered attachment may only be a sufficient explanation for placement breakdown in some cases. In this study initial externalizing behaviors proved to be the strongest predictor of placement changes for the entire sample and for the sub-sample of those who initially evidenced problem behaviors on at least one broad-band scale. This makes sense and agrees with the findings of others. Children who are disruptive, aggressive and/or dangerous to others are likely

10 1372 R. R. Newton, A. J. Litrownik, and J. A. Landsverk to be moved to a different setting. However, these findings also suggest that children who initially score within normal ranges on the CBCL may be particularly vulnerable to the detrimental effects of placement breakdowns. For the 173 children in the sample who initially were rated as not evidencing behavioral problems, number of placements was a consistent predictor of increased internalizing, externalizing and total behavioral problems following a second assessment at 18 months. Similarly, when the entire sample (N 415) was assessed via a multivariate regression model that covaries initial behavioral problems, the influence of placement history on internalizing and externalizing problems at a later point proved weak, but reliable. For those who experienced over four (the median) placement changes, number of changes explained 9.7% and 6.7% of the variance in internalizing and externalizing behaviors, respectively. This data, containing a varied mix of ethnic groups (45.0% Anglo, 17.1% Hispanic, 34.5% African-American, and 2.4% of other ethnic origins) a reasonably balanced mix of males and females, and a wide age range, also demonstrates that these effects are independent of racial/ethnic origins, as well as age and gender. This approach, that considers number of placement changes an independent variable, after controlling for initial behavior problems, represents a significant change in the approach taken by previous authors, who typically examine these relationships with the explicit assumption that change in placement constitutes a dependent variable, driven in part by the child s disruptive behavior (c.f., Fanshel & Shinn, 1978; Olsen, 1982). In a recent example, Palmer (1996) utilized number of placements in an 18 month follow-up as the dependent variable and difficulty of child s behavior as an independent variable. Her regression models identify child s behavioral difficulties as the most powerful predictor of placement disruption, and she concludes that, most placement workers are already making every effort to help children modify their behavior, to prevent placement breakdown (p. 599, emphasis added). This analytical approach reinforces the view, well supported by both the quantitative and qualitative literature, that children with externalizing behavioral symptomatology are likely to experience multiple placement disruptions. Clearly these children should be identified and managed via treatment foster care or other therapeutic settings, and, as Palmer states, placement workers are making concerted efforts on behalf of these children. However, in the case of children with seemingly typical profiles, it may be much more difficult to identify those likely to exhibit highly negative reactions to multiple placement breakdowns beyond their control or understanding. In this context, it is interesting to highlight two findings from our analyses. First, the strongest bivariate correlations in our sample are represented by the relationship between CBCL broad band scale scores at Time 2 and number of placement changes, for those children who initially evidenced no symptomatic elevation in CBCL scores at Time 1. Second, in our multivariate analysis of children experiencing 5 or more placement changes, number of placement changes was a stronger predictor of internalizing behavior at Time 2 than the children s internalizing behavior at Time 1. These children may in fact constitute a neglected population that responds to multiple disruptions of their primary relationships by becoming increasingly withdrawn and isolated, as opposed to the more evident population that becomes more hostile or offensive. These results argue for initial screening of all children at entry into foster case, including those that may appear asymptomatic. Our analyses failed to find any influence as a function of major demographic categories, including age, race, or gender. While this is in disagreement with some previous studies, these results have been inconsistent across studies in general, and no previous studies have been able to examine these relationships in a longitudinal context. We suspect that our ability to control for initial levels of problem behaviors, the fact that we used t-scores that adjust for age differences, and the reliability of our instruments, has eliminated much of the non-random error that previous studies may have attributed to demographic differences. While there may be some real differences in children s reactions as a function of age, race and gender, we believe that, in general, a child s response to his or her treatment by the foster care system cannot be predicted on the basis of these categories.

11 Placement change and problem behaviors 1373 There were a number of limitations of the current study that could be remedied in future research. First, there was an initial lag between a child s first entry into the foster care system and our ability to conduct an interview with that child s caretaker. There were a number of reasons for this, including the lag time between system entry and inclusion in the study, and the sheer size of the cohort of children to be studied (N 1221). We believe that our first assessment, at an average of 5 months after initial placement, produced conservative results in that the amount of variability explained by change in placement would have been larger if initial assessment had been earlier. Second, nearly all of the children in this study remained in foster care at the end of the study; thus, it was not possible to assess the relationship between length of stay and number of placements. An ideal model would not only assess the child early in his or her placement career, but would follow that child at multiple time points over a longer period of time. Such a model would permit our understanding of the reciprocal causal mechanisms involved in the placement change/problem behavior system. Third, we used only one measure of child behavior, the CBCL. While the CBCL is a well respected measure with demonstrated reliability and validity, it has also been criticized, particularly when utilizing multiple informants (c.f., Achenbach, McConaughy, & Howell, 1987; Clausen, Landsverk, Newton, & Ganger, 1995). Measures taken directly from the target child, such as the CBCL Youth Self-Report, would be particularly valuable in this context. In sum, it seems clear that we need to study how parents, foster parents, and social workers, prepare for and cope with the consequences of placement change for children. We also need to closely examine how non-symptomatic children manage change interpersonally, with a view toward developing strategies that minimize the distinctly negative consequences of multiple placements for many of these children. Measures to prevent placement disruption, including care in initial placement screening, placement management over time and early intervention to prevent placement disruption are all recommended. Early assessment of all children, and careful monitoring of those children who experience frequent placement disruption, even those who appear resilient, seems warranted. REFERENCES Achenbach, T. M. (1991). Manual for the Child Behavior Checklist/4 18 and 1991 profile. Burlington, VT: University of Vermont Department of Psychiatry. Achenbach, T. M., McConaughy, S. H., & Howell, C. T. (1987). Child/adolescent behavioral and emotional problems: Implications of cross-informant correlations for situational specificity. Psychological Bulletin, 101, Cook, R. J. (1994). Are we helping foster care youth prepare for their future? Children and Youth Services Review, 16, Clausen, J., Landsverk J., Newton, R., & Ganger, W. (1995, August). Mental health assessment of children in foster care: Cross-informant differences. Poster session presented at the annual meeting of the American Psychological Society, New York. Clausen, J. M., Landsverk J., Ganger, W., Chadwick, D.,& Litrownik, A. (1998). Mental health problems of children in foster care. Journal of Child and Family Studies, 7, Doelling, J. L., & Johnson, J. H. (1990). Predicting success in foster placement: The contribution of parent-child temperament characteristics. American Journal of Orthopsychiatry, 60, Fanshel, D., & Shinn, E. B. (1978). Children in foster care: A longitudinal investigation. New York: Columbia University Press. Festinger, T. (1983). No one ever asked us...a postscript to foster care. New York: Columbia University Press. Hochstadt, N. J., Jaudes, P. K., Zimo, D. A., & Schachter, J. (1987). The medical and psychosocial needs of children entering foster care. Child Abuse & Neglect, 11, Kufeldt, K., Armstrong, J., & Dorosh, M. (1989). In care, in contact? In J. U. Hudson & B. Galaway (Eds.), The state as parent (pp ). Dordrecht: Kluwer Academic Publishers. McIntyre, A., & Keesler, T. Y. (1986). Psychological disorders among foster children. Journal of Clinical Child Psychology, 15, Millham, S., Bullock, R., Hosie, K., & Haak, M. (1986). Lost in care. London: Gower. Olsen, L. J. (1982). Predicting the permanency status of children in foster care. Social Work Research and Abstracts, 18, Pardeck, J. T. (1984). Multiple placement of children in foster family care: An empirical analysis. Social Work, 29,

12 1374 R. R. Newton, A. J. Litrownik, and J. A. Landsverk Palmer, S. E. (1996). Placement stability and inclusive practice in foster care: An empirical study. Children and Youth Services Review, 18, Penzerro, R. M., & Lein, L. (1995). Burning their bridges: Disordered attachment and foster care discharge. Child Welfare, LXXIV, Piliavin, I., Sosin, M., Westerfelt, A. H., & Matsueda, M. (1992). Toward a longitudinal analysis of homelessness. Journal of Social Issues, 46, Proch, K., & Taber, M. (1985). Placement disruption: A review of research. Children and Youth Services Review, 7, Ray, J., & Horner, W. C. (1990). Correlates of effective therapeutic foster parenting. Residential Treatment for Children and Youth, 7, Staff, I., & Fein, E. (1995). Stability and change: Initial findings in a study of treatment foster care placements. Children and Youth Services Review, 17, Stone, N. M., & Stone, S. F. (1983). The prediction of successful foster placement. Social Casework, 64, Walsh, J. A., & Walsh, R. A. (1990). Studies of the maintenance of subsidized foster placements in the Casey family program. Child Welfare, 69, This article is being published without benefit of the authors review of the corrected proof as this was not available at press time.