The Hidden Cost of Foster-Care: New Evidence on the Inter-Generational Transmission of Foster-Care Experiences

Abstract

This study investigates inter-generational transmission of foster-care, to test the extent to which an overrepresentation of children of foster-care alumni in a group of children in care persists after controlling for parents’ additional resources (such as criminal history, crime and labour market attachment). For this purpose, we use administrative data from Statistics Denmark, which we analyse using simple descriptive statistics and probit models. Results show that, while children of foster-care alumni are seven to ten times more likely than other children to experience foster-care, this overrepresentation is halved when we control for parental resources.

Introduction

The extensive literature on foster-care provides solid evidence of the disadvantages of children placed in out-of-home care. Not only is their foster-care experience a direct result of significant individual or family-level problems (e.g. Berrick et al., 1993; Berger, 2006; Barth et al., 2006; Ehrle and Geen, 2002; Yampolskaya et al., 2007; Chin and Phillips, 2004; Franzén and Vinnerljung, 2006; Viner and Taylor, 2005; Sidebotham et al., 2002), but individuals with childhood experiences of out-of-home care are also very likely to fall victims of a range of negative outcomes in adulthood (Cheung and Heath, 1994; Colton and Heath, 1994; Jackson, 1994; Courtney et al., 2001; Barth, 1990; Vinnerljung et al., 2005; Hjern et al., 2004; Barth and Blackwell, 1998; Vinnerljung and Ribe, 2001).

Less is known about the inter-generational transmission of foster-care, although the importance of the topic is characterised by not only significant individual and social costs of placing children in out-of-home care, but also detrimental effects on parents having a child removed from home (Fallesen, 2013). From parallel literature on parent/child correlation in experiences with abuse, we know that as many as 30 per cent of children of abused parents become victims of abuse (Fusco, 2015). However, one of the main purposes of foster-care is to remove children from such disadvantaged environments and to help them build skills to live less disadvantaged lives. In fact, in many countries, foster-care has the stated purpose of strengthening and securing the normal functioning of children in care to the extent where their lives and possibilities are comparable to the lives of children with no foster-care experiences (as described in e.g. §46 in the Danish Act of Social Services). This would also imply that foster-care alumni are as likely (or little likely) as any other parent to have their own children placed in care, and that we should expect the parent/child correlation in foster-care experiences to be significantly lower than the parent/child correlation in abuse.

The overall childhood risk of experiencing foster-care is approximately 5 per cent (found in various contexts; see Fallesen et al., 2014; Wildeman and Emanuel, 2013; Wildeman and Waldfogel, 2014); however, according to the few existing studies on inter-generational transmission of foster-care, as many as 8–22 per cent of all children of foster-care alumni spend part or all of their childhood in out-of-home care (Pecora et al., 2010; Courtney et al., 2010). Thus, compared to the average child, children of foster-care alumni are overrepresented by a factor 1.6–4.4 in the group of children in foster-care.

In learning about this disproportionate representation, two concerns spring to mind. First, the mere range of the estimate presented in previous studies is a strong indication that this topic deserves further attention. The samples used for acquiring these estimates are relatively small and represent select groups of foster-care alumni, possibly biasing the results. Second, the overrepresentation may reflect both parents’ foster-care experiences as well as the additional disadvantages often found in such families. While time spent in foster-care (hopefully) improves the life outcomes of the foster-care alumni, these individuals may still end up with fewer resources than the average person, simply because they have had a more disadvantaged point of departure. At the same time, we know that children from low-resource families have an increased risk of experiencing foster-care. The fewer resources of the foster-care alumni will then almost automatically lead to a higher risk of foster-care among their children—not necessarily because they are foster-care alumni, but because they have fewer resources. Thus, to fully grasp inter-generational transmission of foster-care, it is crucial that we separate the importance of the actual foster-care experience from the lack of resources among foster-care alumni.

To improve our knowledge in this field, our study provides new evidence on inter-generational transmission of foster-care. For this purpose, we use population data from administrative sources which we analyse using descriptive statistics and probit models. It adds to existing knowledge in two ways. First, it presents evidence based on population data that does not suffer from problems pertaining to sample selection bias or attrition bias, and which allows us to fully understand the consequences of the steps we take to select a useful sample. Second, it tests how controlling for important observable disadvantages of the foster-care alumni modifies the correlation between parents’ foster-care experience and children’s risk of experiencing foster-care.

Mechanisms

The literature on inter-generational transmission of disadvantages, such as experiences with childhood abuse and foster-care, presents several reasons why we should expect an overrepresentation of children of foster-care alumni in the group of children in foster-care.

First, according to the social learning perspective, children observe their parents’ positive and negative behaviours and use these observations to form rules regarding their own behaviour. If this behaviour is accompanied by rationalising verbalisations, children are particularly likely to come to view their parents’ behaviour as normative (Dunlap et al., 2002; McWey et al., 2012; Finzi-Dottan and Harel, 2014) and to imitate this behaviour when they become parents themselves. This mechanism will increase the likelihood that foster-care is inter-generationally transmitted. Not only may the children imitate the part of their parents’ behaviour which caused the problems leading to foster-care; they may also be more likely to view foster-care as a solution to such problems, and to appreciate this type of family intervention. Also, case workers might expect foster-care alumni to have limited parenting skills and may therefore be more observant of this group and hence more likely to notice their children’s symptoms of neglect and maltreatment. This will increase foster-care among the children of foster-care alumni even if their problem load is no different from that of other children (Fusco, 2015).

Second, attachment theory also claims that parental behaviour is transmitted to children but suggests that transmission of negative parenting styles is particularly likely when the child lacks or has an incoherent memory of that behaviour (Kaufman and Zigler, 1989): the coherent recollection of negative parenting styles helps the child identify similar negative parenting styles in his or her own behaviour and enables reflection on possible alternatives to this behaviour. In contrast, incoherent recollection limits the child’s understanding of such similarities and reduces the potential for reflection. This then increases the probability that the child imitates the negative parenting style and replicates this style when he or she becomes a parent (Vondra and Belsky, 1993).

Third, the resource perspective suggests that foster-care alumni experience difficulties transitioning properly into adulthood and into parenthood due to lack of personal and social resources. Young people, who have left home, often rely on their parents for emotional, social and financial support. However, for foster-care alumni, such support is often unavailable or limited, either because of poor parent–child relationships or because of the parents’ low resources. In addition, foster-care alumni often have fewer personal resources for coping with the challenges of living alone, because adverse childhood experiences have reduced their overall abilities to cope with stress and transitions. With such lack of network and individual resources, foster-care alumni face a stronger challenge in acquiring resources, such as educational skills, permanent labour market affiliations and stable income, and it increases their risk of engaging in delinquent behaviour. Furthermore, this lack of resources may limit their abilities as parents; they are less mentally prepared to handle the stress of parenting and they have fewer material resources for handling parenthood (Courtney and Hughes-Huering, 2005; Geiger and Schelbe, 2014). From previous studies, we know that children of low-resource parents are more likely to experience foster-care than children of parents with more resources (Andersen and Fallesen, 2010) and the lack of resources among foster-care alumni may therefore increase the risk that their children are placed in out-of-home care.

The three explanations illustrate two main routes through which parents’ foster-care experience increases the risk that their children experience foster-care. One route is through the parenting styles, as suggested by the social learning perspective and attachment theory where foster-care alumni replicate the negative behaviour of their parents, and the other is through the lack of both coping skills and palpable resources such as education and a stable labour market affiliation that often characterise this group of individuals according to the resource perspective. Though the lack of resources reflects their foster-care experience, it is still an independent reason for the possible overrepresentation of children of foster-care alumni in the group of children in foster-care.

Our empirical analysis therefore not only describes the correlation between parents and children’s foster-care experience; it also tests the robustness of this correlation to the inclusion of controls for the resources of the foster-care alumni. Importantly, our data mainly allow for a test of the importance of palpable resources.

Context and data

We studied inter-generational transmission of foster-care using administrative data from Denmark. Denmark is a good case due to the country’s relative stable use of foster-care and the availability of high-quality data.

Since the beginning of the twentieth century, approximately 1 per cent of any Danish birth cohort has been placed in care at any given point in time between the ages of zero and eighteen (Bryderup, 2005) and each child has had a cumulative 5–6 per cent risk of experiencing placement during childhood. We see the highest risk of foster-care during the teenage years; while the cumulative risk of experiencing placement in the pre-school years is approximately 1.5 per cent, the cumulative risk of experiencing placement in the teenage years is over 3 per cent (for more information, see Ejrnæs et al., 2011). This is different from a US context, for example, where the placement risk increases more linearly throughout childhood (see e.g. Wildeman and Emanuel, 2013), even though the cumulative risk of foster-care in the USA is also approximately 5 per cent.

In Denmark, all residents have a unique personal number that identifies the resident in many different aspects; when a resident receives welfare benefits, enrols in or graduates from a specific education, or signs a contract with an employer, it is, for instance, recorded by this number. Similarly, the Danish child protection system registers all decisions regarding preventive measures, foster-care and after care by the involved residents’ personal number. Statistics Denmark collects these data annually and makes them available for statistical and research purposes. Using the personal number, we may link registers and connect families. Most of these data are available from 1980 onwards, but data regarding foster-care placements are available from 1977 to 2010. These population data were ideal for studying inter-generational transmission of foster-care.

Our sample

The possibility of linking family members in the registers was particularly useful for our research question. Importantly, though, our sample selection was restricted; first, because the register on foster-care goes back no further than 1977 and ends in 2010 and, second, because we needed information on parents’ foster-care experience through their entire childhood (birth to eighteen years) and to observe their children for a sufficient period of time to know whether they had experienced foster-care.

Given these constraints, we constructed our sample as follows. From the full sample, we selected all women from the 1977 cohort for our first generation—the parent generation—since this was the first cohort that we could observe during their entire period at risk of foster-care (e.g. from when they were born until they turn eighteen years old) in the registers. This included 30,379 women of which 1,569 (5.2 per cent) had foster-care experiences during childhood. We then linked these women to their children (our second generation, the child generation) and to the fathers of these children. Our main unit of analysis was then parents–child units (regardless of whether the parents cohabited or not). Some mothers and fathers—and mother–father constellations—appeared more than once in the data if the individual (or couple) had more children (together).

Ideally, we wanted to observe the second generation for as long as they were at risk of foster-care—from when they were born until they turned eighteen. However, this required that the mothers (born in 1977) had the child at age fifteen (in 1992), which is very uncommon in Denmark. Instead, we observed foster-care experiences of the second generation during their entire pre-school age (until they turn six years old), which implied reducing our sample to only include parents–child units where the child was born prior to 2005. Since the majority of placements among zero-to-eighteen-year-olds in Denmark takes place during the teenage years (Ejrnæs et al., 2011), this necessary sample selection was then a likely cause for underestimating the inter-generational transmission of foster-care.

With these restrictions, our final sample consisted of 15,213 parents–child units containing mothers born in 1977, their children born prior to 2005 and the children’s father. All mothers were twenty-seven years or younger at the time of childbirth. Note that the registers did not have information on the fathers of 1,461 children and that children of mothers with childhood foster-care experiences were overrepresented by a factor of 4 in this group. By assuming that mothers with foster-care experiences were more likely than other mothers to have children with foster-care alumni fathers, this missing data problem was likely to result in an underestimation of the father–child correlation in foster-care experiences.

Generalisability of sample and results?

Since not all mothers in the 1977 cohort, who have one or more children, had their first child before 2005 and since parents who had children before the mother turned twenty-seven were likely to differ from other parents, our results applied to a selected group of individuals. Studies show, for instance, that women with more education are more likely to postpone motherhood (Heck et al., 1997) and, more generally, age at first birth is positively correlated with resources for both men and women, such that individuals with fewer resources tend to give birth to their first child at younger ages (Manlove, 1997; Hoffert and Goldschieder, 2010). Using the fertility behaviour of older cohorts, we got a sense of how biased our sample was.

First, assuming that the fertility behaviour of women from the 1977 cohort resembled that of older cohorts who had completed their fertility (say the 1960 cohort) allowed us to compare the observed proportion of the 1977 cohort who had become mothers by 2005 with the proportion of such older cohorts who were mothers in that year and get an estimate of the expected completed fertility of the 1977 cohort. According to such calculations, our sample consisted of approximately 40 per cent of the women in the 1977 cohort, who eventually became mothers.

Second, we tested for an overrepresentation of foster-care alumni in our sample by comparing the age distribution for first-time mothers in the 1977 cohort with and without foster-care experience. For this purpose, we looked at any first-time motherhood in the 1977 cohort which occurred before 2015 and found that, while 72 per cent of foster-care alumni mothers had their first child before age twenty-seven, this only applied to 45 per cent of non-foster-care alumni mothers. These percentages were clear indications of the different fertility behaviour of the foster-care alumni compared to non-foster-care alumni mothers.

Our sample then consisted of a select group of mothers and their partners (both with and without personal foster-care experience) which must be kept in mind when interpreting our results. Importantly, though, our sample selection procedure was likely to have homogenised our analytical sample and cut away high-resource mothers and their partners. This increased the likelihood that we underestimated the overrepresentation of children of foster-care alumni in the group of children in foster-care (because the comparison group was also low-resource with a higher risk of foster-care than the average child). Still, we claim that our results are representative of much broader groups of foster-care alumni than what is seen in previous studies, and that the transparency of our sample selection procedure and its consequences improve the usefulness of our result.

An additional problem was the significant proportion of fathers born before 1977 in our sample. Calculations showed that the fathers in our sample were approximately 3.7 years older than the mothers. For these fathers, we did not have full information on their foster-care history. By comparing the recorded foster-care experience of fathers in our sample with the foster-care experience of fathers from the 1977 cohort (i.e. the first cohort that we were able to observe throughout their whole period of being at risk of foster-care), we got a sense of the extent of the problem. The comparison showed that 8 per cent of the fathers in our parents–child units experienced foster-care compared to 10 per cent of fathers from the 1977 cohort. As a result, there is a group of fathers from whom we did not observe their foster-care in the data—something which may have caused us to underestimate the father–child correlation in foster-care experiences.

Ethical use of administrative data for research purposes

Our use of administrative data in this study followed the ethical guidelines specified in Stiles and Boothroyd (in press). First, the data were securely stored at Statistics Denmark’s servers and individual access to the data only happened upon Statistics Denmark’s official approval and according to their strict guidelines and technical system requirements. Second, all data were anonymised so that any information which may be used for identifying an individual was removed from the data or encrypted. Third, we applied for and obtained written permission from Statistics Denmark to use the data for the specific purpose of this study, just as Statistics Denmark holds general permission from the Danish Data Protection Agency to allow for this use of their data. Fourth, and last, both authors of this paper had extensive experience handling and analysing Danish administrative data and in-depth knowledge of the Danish foster-care system, which made us capable of appropriately handling and using the data.

Dependent variable: foster-care placement of child during pre-school (birth to six years)

With our focus on inter-generational transmission of foster-care and given our sample selection criteria specified above, our variable of interest was a binary indicator—or dummy variable—taking the value 1 if the child experienced foster-care during age zero to six, and 0 otherwise. Two hundred and twenty-one children experienced foster-care during the pre-school years—116 girls and 105 boys.

The explanatory variables

Our main independent variable measured parental foster-care experience between age zero and eighteen; 2,250 of all children (1,096 girls and 1,154 boys) had at least one parent with foster-care experience. Or, vice versa, 7.9 per cent of mothers and 7.8 per cent of fathers had childhood experiences of foster-care. Most parents experienced foster-care during the teenage years. Since Danish children have an average risk of experiencing foster-care of 5 per cent, foster-care alumni were overrepresented in our sample. This overrepresentation reflected our sample selection procedure described above. For a small proportion of these children (N = 216), both parents had childhood foster-care experiences. These children probably had a particularly high risk of experiencing foster-care, because none of their parents had had a ‘normal’ background as a benchmark for parenting styles or as a source of useful resources. We tested this assumption by including a dummy variable which took the value 1 if both parents had childhood foster-care experiences, and 0 otherwise.

As described, we aimed to isolate the correlation between parents’ foster-care experiences and children’s placement risk, from the correlation between children’s foster-care risk and other parental characteristics, which could have been more or less prevalent in the foster-care alumni. This was to improve our understanding of how low parental resources contributed to the parent–child correlation in foster-care experience, to distinguish this element from the actual ‘effect’ of parents’ foster-care experience. We therefore prepared a set of explanatory variables to control for socio-economic factors that affect the probability of ending up in foster-care, according to previous studies. These indicators included parents’ income, educational background, employment status, early retirement, criminal behaviour, marital status and their labour force status. We measured all indicators the year before childbirth, except for criminal behaviour, which took the value 1 if the parent had been sentenced at least once one to five years prior to childbirth (and 0 otherwise). We also included a dummy which took the value 1 if information on the father was missing. When this indicator was 1, all characteristics of the father were set to 0, allowing us to include these observations in our sample as well. Importantly, the included indicators did not reflect all of the important types of resources, since we did not have information on, for instance, parents’ coping resources. Still, it represented important palpable resources on which the foster-care alumni may differ from other parents, and which we expected affect their children’s placement risk.

Results

Table 1 shows the proportion of children with foster-care experience by parents’ foster-care history. The results demonstrated a substantial overrepresentation of children of foster-care alumni among children who experienced foster-care. More than 7 per cent of the children with a mother from the foster-care alumni experienced foster-care, compared to less than 1 per cent of children of mothers with no foster-care experience. This was an overrepresentation of six percentage points. We found a similar overrepresentation among children of fathers from the foster-care alumni, of approximately four percentage points. Children of foster-care alumni were then between approximately seven and ten times more likely to experience foster-care than other children. More specifically, children of fathers who experienced foster-care during childhood had a risk ratio of experiencing foster-care during their pre-school years of 7.56 (0.0537/0.0071) and children of mothers with childhood foster-care experiences had a risk ratio of experiencing foster-care during their pre-school years of 9.74 (0.0779/0.008). The risk ratio was even higher—18.63—when both parents had foster-care experiences. Here, more than 15 per cent of the children experienced foster-care.

As stated earlier, the increased risk for children of foster-care alumni of experiencing foster-care may have two explanations. According to the resource perspective, the lower resources of the foster-care alumni explain a large part of the variation across the two groups of children (with/without foster-care alumni parents). However, if the differences between the two groups persist after controlling for important characteristics, it may reflect other factors described by the social learning perspective and the attachment theory, such as the parenting behaviour of the foster-care alumni.

We tested the resource perspective by first presenting descriptive evidence of the differences in resources between parents with and without childhood foster-care experience. Table 2 shows the statistics, by parents’ gender. The left column shows statistics for parents belonging to the foster-care alumni and the right column parents with no foster-care experience.

As expected, the foster-care alumni had, on average, lower income, lower employment, lower level of education and they were younger when having their (first) child. Given the strong correlations between parents’ low socio-economic resources and children’s increased risk of experiencing foster-care, these findings may explain the higher foster-care risk of children of foster-care alumni.

Results from probit model

With the evidence presented in Table 2 that the foster-care alumni differed from other parents on a range of parameters, a next step was to clarify the extent to which these differences explained the overrepresentation of children of foster-care alumni in the group of children in care. For this purpose, we further analysed the parent–child association in foster-care experience using a regression model. This allowed for a straightforward test of how adjusting for these differences affected the association. Since our dependent variable took the value 1 if the child experienced foster-care during pre-school and 0 otherwise, we used a probit model (note that our results were robust across model specifications, such that probit, logit and linear regression models led to similar conclusions). We present model findings as marginal effects, which express the association between the outcome variable and the explanatory variables in percentages. This is to facilitate a more straightforward interpretation of the magnitude of the association than regression coefficients. The associated probit coefficients are available from the authors on request.

Our main independent variables were two dummies which took the value 1 if either the mother or father experienced foster-care during ages zero to eighteen, and 0 otherwise. We also included a dummy which took the value 1 if both parents had childhood experiences of foster-care.

We ran two models: one without and one with controls. With this strategy, we tested whether the inclusion of control variables modified the coefficients of our three dummy variables of interest, and thus if socio-economic differences between foster-care alumni and other parents explained differences in their children’s risk of experiencing foster-care.

Table 3 shows our results. Column 1 shows that mother’s foster-care experience—at any point during childhood—increases the child’s placement risk by 3.12 per cent and that father’s foster-care experience increases the child’s placement risk by 1.88 per cent. The coefficient for both parents’ foster-care experience was insignificant. In contrast to what we expected, there is then no extra disadvantage of this parents–child constellation and vice versa; a non-foster-care alumni parent does not seem to buffer the negative consequences of the other parent’s foster-care experience.

Column 2 shows how the results change when we include control variables. The coefficient for mother’s foster-care experience is now 1.24 per cent and the coefficient for father’s experience is now 0.75. Thus, approximately 60 per cent of both the mother–child and the father–child variance in foster-care experiences reflect parents’ low resources.

The remaining explanatory variables showed children’s reduced risk of foster-care when their mother was older, is in school (regardless of level) and had a job. Father’s school attendance and higher household income were also correlated with a reduced risk of children’s foster-care experience. In contrast, mother’s lack of education above elementary school, mother’s reliance on social security, mother’s early retirement and mother’s criminal activities increased the risk that her child experiences foster-care. Likewise, father’s early retirement pension increased his children’s risk of foster-care. While most variables concerning mothers’ characteristics were significant, most variables concerning the fathers were not. This may reflect that mother’s resources matter the most, or it may be an indication of a positive correlation between parents’ resources. Importantly, multicollinearity diagnostics test shows no evidence of problems pertaining to multicollinearity between our indicators. Given the insignificant effect of child gender, there did not seem to be a gender difference in placement risk.

Conclusion

Even if one of the explicit purposes of placing a child in foster-care is to improve his or her later outcomes, mounting evidence suggests that foster-care alumni still underperform compared to the average person. They have worse labour market outcomes, lower educational attainments, and experience more social and personal problems. In addition, empirical evidence suggests that their children are overrepresented in the group of children in care, even if the quality of this evidence is limited by small sample sizes and very select samples. The ambition of our study has been to test not only the extent of the inter-generational correlation in foster-care experience using population data, but also to test the degree to which differences in palpable resources between foster-care alumni and other parents explain this overrepresentation.

Our results show that children of foster-care alumni are almost between seven and ten times more likely to experience foster-care during their pre-school years compared to other children. Since most placements take place during the teenage years, our results most likely represent a lower bound estimate of this overrepresentation. The overrepresentation is significantly larger than found in previous studies (mentioned in the introduction) and may reflect both unique features of the Danish child protection system as well as the higher quality of the data used in this study compared to previous studies.

The overrepresentation reduces substantially when we control for resources such as parents’ education, labour market affiliation, criminal activities, etc., but the significant effects of parents’ foster-care experience persist. Thus, while part of the increased risk of foster-care experienced by children of foster-care alumni reflects parents’ reduced resources, another part is less tangible. Theories in this area—such as the social learning perspective and attachment theory—may explain this remaining part by referring to the foster-care alumni’s parenting skills, as these individuals are likely to either consciously or unconsciously replicate the parenting behaviour of their own parents. However, the resource perspective also implies that besides from the palpable resources used in our analysis, other resources, such as skills for coping with stress and transitions, may matter. Unfortunately, our data do not provide meaningful information on such softer indicators.

The findings of our study comply well with findings from similar studies on inter-generational transmission of disadvantages. From the extensive literature on child poverty, we know that, while there is a direct transmission of poverty from parents to children, other factors related to household dynamics and individual resources may serve to mediate at least part of the direct link (for a review, see Bird, 2007). The same is true for inter-generational transmission of criminal and other types of antisocial behaviour (Farrington et al., 2009; Thornberry et al., 2003). In lieu of these parallel literatures, the findings presented in the current study are unsurprising, but help piece together our overall understanding of inter-generational transmission of disadvantages.

The remaining question is then how to use this knowledge; should we pay extra attention to foster-care alumni and help and support them in their roles as parents to prevent their children from ending up in foster-care? Knowledge in this area is still limited and a first step would be to investigate further the reasons for the documented overrepresentation of children of foster-care alumni in care. Is it parents’ personal coping skills or their parenting behaviour which causes the persistent correlation, or do other palpable resources, not included in our study, explain more of the variation? In case of the former, it could be viable to consider the independent living programmes offered to foster-care alumni. While many countries offer such programmes, they are often poorly executed and have limited coverage. An improvement in that area may not only help the foster-care alumni and their children, but may also turn out to be important social investments in terms of reducing future cost of needing to sent these children to foster-care.

Acknowledgements

The research reported in this paper was funded by the Rockwool Foundation.

References