The role of welfare and work in predicting foster care reunification rates for children of welfare recipients

The role of welfare and work in predicting foster care reunification rates for children of welfare recipients

Children and Youth Services Review 26 (2004) 577–590 The role of welfare and work in predicting foster care reunification rates for children of welfa...

112KB Sizes 2 Downloads 47 Views

Children and Youth Services Review 26 (2004) 577–590

The role of welfare and work in predicting foster care reunification rates for children of welfare recipients Katherine Kortenkamp, Rob Geen*, Matthew Stagner The Urban Institute, 2100 M Street NW, Washington, DC 20037, USA Available Online 23 March 2004

Abstract Most research on children’s experiences in foster care has focused on child and system level factors related to length of stay and returns home. This article reports results from an analysis of family factors—welfare and work activities—that predict foster care reunification. The sample is 133 children of welfare recipients in California who entered foster care after December 1, 1992. In this sample, 42% of children reunified with their parents during the observation period, while 27% remained in foster care. We used a Cox proportional hazards model to predict reunification rates. We focused on the welfare and work activities of the families from which the children had been removed, examining welfare history, welfare and work at the time of placement, loss of welfare after placement and work during placement. Control variables included child’s age and race, mother’s marital status, education and health, county, removal reason, placement type, number of placements and whether there were siblings placed. We found that work at the time a child was placed increased the likelihood of reunification and losing welfare benefits following a placement decreased the likelihood. Findings are discussed in terms of implications for welfare and child welfare policy and practice. 䊚 2004 Elsevier Ltd. All rights reserved. Keywords: Reunification rates; Cox proportional hazards model; Foster care

1. Introduction Of the 250 000 children who exited foster care in the US in 1999, 59% were reunified with their parents or caretakers (Department of Health and Human Services, *Corresponding author. Tel.: q1-202-261-5822. E-mail address: [email protected] (R. Geen). 0190-7409/04/$ - see front matter 䊚 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.childyouth.2004.04.012

578

K. Kortenkamp et al. / Children and Youth Services Review 26 (2004) 577–590

2001). Over half of these children had been in the child welfare system for less than a year, and their median length of stay was 11 months. However, there is great variability when it comes to how long children stay in foster care, while 21% exited in less than a month, another 20% stayed for 3 or more years. Past studies looking at predictors of reunification rates and length of stay in foster care have generally focused on child and child welfare system factors (Goerge, 1990; Benedict & White, 1991; Courtney, 1994; Courtney, Piliavin & Entner Wright, 1997; Vogel, 1999; Glisson, Bailey & Post, 2000; Wells & Guo, in press). All of these studies had large samples of foster care entry cohorts (378–20 000 children), controlled for several covariates in their models, and all but one used event history analysis. Although every study was longitudinal, the length of time of the analyses varied greatly.1 Also, most of the studies looked only at predictors of reunification, but two combined other exits like adoption, independent living, and guardianship. Overall, at the child level these studies show that children who are minorities, have disabilities or developmental, health or behavioral problems and are younger (particularity under age one) have lower rates of reunification or longer stays. Studies that examined factors related to the child’s experience in the child welfare system find that children having more placements tend to have lower rates of reunification. Results are mixed for the effects of placement type. Some studies find that children placed in kinship care have lower reunification rates or longer lengths of stay (Goerge, 1990; Benedict & White, 1991; Courtney, 1994; Vogel, 1999). Other studies find no effect of kinship placements (Courtney et al., 1997; Glisson et al., 2000; Wells & Guo, in press). However, Courtney and colleagues did find an interaction effect such that children in kinship placements who were also AFDC eligible were less likely to reunify. Little research has focused on factors related to the families from which these children were removed. Conceptually, it seems that family factors would be as much if not more related to the likelihood of reunification as child factors, and child factors are often just proxies for parent factors anyway. It is ultimately the parents who must display a desire and ability to adequately care for their children before reunification can occur. Parents must demonstrate that they will be available to care for their child or have access to child care, and they must have adequate financial resources and housing. Younger children and children with special needs or behavior problems may be more difficult to parent and require an even larger commitment from parents, and thus these child factors indirectly influence reunification through their effects on parents. Removal reason is a common family factor included in analyses, and it has often been found that those children removed because of neglect have lower reunification rates compared to those removed for abuse (Courtney, 1994; Courtney et al., 1997; Wells & Guo, in press), but higher rates have been shown for neglect compared to those removed for dependency (Goerge, 1990). A few studies have included other family factors in their analyses. They have shown that children removed from single 1 Studies varied on the length of the entry cohorts, ranging from 6 months to 8 years, and the length of time children were followed, ranging from 1.5 years to 8 years.

K. Kortenkamp et al. / Children and Youth Services Review 26 (2004) 577–590

579

mother or relative homes were less likely to reunify (Courtney, 1994; Courtney et al., 1997; Glisson et al., 2000). Additionally, Courtney found that children from AFDC eligible homes were less likely to be reunified, although Glisson and colleagues found no effect of the socioeconomic status of the child’s home of origin. The present study expands on these family measures. Specifically, our analysis includes welfare and work related factors that may affect whether the family is reunified with the child. A small body of research has shown that welfare and work factors influence whether a family will become involved with the child welfare system. Studies have shown that current welfare receipt increases the likelihood for child welfare involvement (Department of Health and Human Services, 2000; Geen, Kortenkamp & Stagner, 2002; Goerge & Lee, 2000; Slack et al., 2001). Also, total time on welfare, breaks in welfare receipt (Needell, Cuccaro-Alamin, Brookhart & Lee, 1999), and reductions in welfare benefits (Shook, 1999) all increase the odds of involvement. Shook also found that families at greatest risk were those whose benefits were reduced and who were unemployed, and although Geen and colleagues found stable employment reduced the risk of a foster care placement, working while on welfare increased the risk. Not only could these welfare and work factors affect whether families have children removed, but they could also affect their ability to get their children back. Families who lose welfare benefits when their children are taken into state custody may have a more difficult time meeting requirements for reunification because of financial instability. Working while on welfare could have both positive and negative effects on the ability to reunify. It might be difficult to hold down a job while trying to work with the child welfare agency and caseworkers to get a child back. However, parents who are able to find employment already have characteristics, such as motivation, and a basic level of competency that should be helpful to them throughout the reunification process. Little research has examined the link between welfare, work and reunification. Wells and Guo (in press) explored welfare and work as predictors of reunification with a sample of 903 children of single mothers. For 18 months they followed a cohort of children who entered foster care over a 6-month period pre-welfare reform. They found that reunification rates were higher as the mother’s total income (welfare plus wages) increased, but as the amount of wages relative to welfare increased, rates of reunification declined. Mothers who received AFDC and then lost it after the child was removed had lower reunification rates compared to mothers who did not lose welfare benefits. We replicate some of these analyses and examine new factors not previously explored with a sample of welfare recipients who have children placed in foster care in California. 2. Methods 2.1. Sample The sample for this study is 133 children of welfare recipients in California who were on welfare in 1992 and selected to be part of California’s Work Pays

580

K. Kortenkamp et al. / Children and Youth Services Review 26 (2004) 577–590

Demonstration Project (WPDP). The children were all placed in foster care at some time between December 1992 and 1999. The data on this sample come from four sources: (1) county welfare administrative records; (2) a telephone survey; (3) county employment data; and (4) county child welfare administrative records. In 1992, California received a federal waiver that allowed the state to test new approaches to encourage welfare recipients to work, increase their earnings, and decrease their time on public assistance. The California Department of Social Services contracted with the University of California, Berkeley Data Archive and Technical Assistance (UC DATA); the University of California, Los Angeles Welfare Policy Research Group; and other researchers to evaluate this demonstration project in four counties: Alameda, Los Angeles, San Joaquin and San Bernardino. The sample pool for the WPDP project was all family group (FG) and unemployed parents (U) cases receiving welfare in the four California counties in October and December 1992 (Ns354 476). From this pool, the WPDP sample of 14 537 was randomly selected to be tracked with administrative data to observe welfare and employment experiences. Welfare administrative records are available for this sample from 1987 to 1999. As part of the WPDP, a telephone interview (WPDP Survey, Wave 1) was conducted with a randomly selected subset of the project’s participants. These interviews collected information on family characteristics, education and health status, access to child care, housing arrangements, employment history and income. The sample pool for the survey included all WPDP participants whose primary language was one of the six used to conduct the interviews: English, Spanish, Vietnamese, Laotian, Cambodian and Armenian. There was an over-sampling of Vietnamese, Laotian, Cambodian and Armenian speakers; cases in Alameda, San Joaquin and San Bernardino counties; and U cases. The sample pool was 5541 cases. Between October 1993 and September 1994, interviews were conducted with 3560 respondents for a response rate of 64%. We obtained employment data for WPDP participants from the California Employment Development Department (EDD).2 UC DATA matched EDD and WPDP records using social security numbers. Data on quarterly earnings and number of employers are available from 1984 to 1999. To obtain a sub-sample of welfare recipients in the WPDP sample who had children taken into child welfare custody, researchers at UC Berkeley conducted probabilistic matching to link the 3560 Wave 1 survey respondents to child welfare administrative records. A total of 165 respondents had children in the child welfare system between July 1986 and October 2000. For the present analysis we include only those child welfare spells that began between December 1, 1992 (the beginning date of the WPDP) and December 31, 1999 (the end of our available EDD and welfare data), for a total of 133 families with children in foster care. 2 The EDD data contain an undercount of people who are actually working, because employees who are working for the federal government, non-profit organizations, railroads, students working at schools, casual laborers, and the self-employed (not electing unemployment or disability insurance) are not included in these files.

K. Kortenkamp et al. / Children and Youth Services Review 26 (2004) 577–590

581

For each of the 133 families we selected one child who was placed in foster care for the analysis. For cases in which multiple children were placed, we used the child with the earliest post December 1992 placement. For cases in which multiple children were placed on the same day, we used the youngest child.3 For each child only the first spell in care after December 1992 was used in this analysis, but that spell may not have been the child’s first or only foster care spell. The spell started when the child entered foster care and ended upon exit or at the end of observation (December 31, 1999), and the spell may have included multiple placements. 2.1.1. Characteristics of the sample Some basic characteristics of the child welfare sample used in this analysis are reported in Table 1. Thirty-eight percent of the children in the sample were white, 35% were black, 15% were Hispanic and 13% were of other racesyethnicities. At the time of their foster care placement, the average age of the child was 8 years. At the time of the survey, the average age of the mother was 32 years. It is not surprising the mothers were older than typical welfare recipients given that this sample is from a cross section of welfare recipients, many of who have long welfare histories. Again, as expected, mothers were more likely to be single, only 23% were married. They had an average of 2.7 children. Overwhelmingly, these mothers were living in poverty, 92% had incomes below 100% of the federal poverty level (FPL). Forty-seven percent of the mothers had no high school or general equivalency degree. 2.1.2. Characteristics of the child welfare experience Table 2 displays some characteristics of the children’s child welfare experiences. The children in the sample were most likely removed from their parents for neglect.4 Short stays in foster care were common, with 33% leaving in less than 3 months and 56% within a year. However, 35% of children stayed longer than a year and a half. Over half of the children only had one placement while in care, but about a quarter had three or more placements. Children were most commonly placed in family settings, but 37% had experienced placements in a group home or shelter. In this sample, 42% of children were reunified with parents, 27% were still in care at the end of the observation period and 31% exited foster care for other reasons. The other reasons for exiting care included entering the home of an adoptive parent or guardian, aging out (turning 18), or other unknown reasons. Half of the children also had a sibling in care during the same time period. 3 We also ran the same analyses selecting a random child in the family, if more than one child was placed on the same day. This did not affect the results. Guo and Wells (in press) discuss concerns with selecting one child from each family for an event history analysis of this sort, and they show data comparing this method to another which uses the ‘WLW’ procedure for dealing with sibling groups. The WLW method found more significant covariates than the randomly-selected child method, most likely due to the increase in sample size and, therefore power. Also, covariates measured at the family level (the covariates of interest in this analysis), as opposed to at the child level, were minimally affected by the two different methods. 4 A large portion (34%) of data on the reason for removal got lost when child welfare computer systems were updated.

582

K. Kortenkamp et al. / Children and Youth Services Review 26 (2004) 577–590

Table 1 Characteristics of the sample Percent (%)

N

Race of child Black Hispanic White Other

35 15 38 13

46 20 50 17

Child’s age at placement Less than 1 1–5 6–11 12–17

15 29 17 40

20 38 22 53

Mother’s age 17–24 25–29 30–39 40 or more

16 24 47 14

21 32 62 18

Mother’s marital status Married Separatedydivorcedywidowed Never married

23 39 38

31 52 50

Number of children 0–1 2 3 4 or more

18 33 25 24

24 44 33 32

Mother’s education and family’s income No high school degree High school degree or GED Some college Income less than 100% FPL

47 35 19 92

62 46 25 121

2.1.3. Characteristics of welfare and work This analysis is of a particularly long-term welfare sample. All cases had received AFDC or SSI for at least 3 years prior to having a child placed in foster care, and the average amount of time on welfare was just over 8 years (see Table 3). Ninetyone percent of cases were on welfare at the time of placement and about a quarter (26%) lost welfare benefits in the 6 months following the child’s placement. Also, 37% were working at that time. When the child exited the foster care system or was censored, 74% of the cases were on welfare and 47% were working. 2.2. Analysis We used Cox proportional hazards modeling for the analysis, which allows for right hand censoring of the data and time-varying covariates. The event of interest

K. Kortenkamp et al. / Children and Youth Services Review 26 (2004) 577–590

583

Table 2 Characteristics of the children’s child welfare experiences Percent (%)

N

Removal reason Neglect Physical abuse Sexual abuse

38 11 3

50 15 4

Length of first spell (months) -3 3–6 7–12 13–18 19 or more

33 7 16 10 35

44 9 21 13 46

Number of placements 1 2 3 or more

54 20 26

72 27 34

Type of placement Kinship care Other family settings Group home Shelter

32 49 23 14

43 65 30 18

Outcome Reunified Exited for other reasons Still in care

42 31 27

56 41 36

Sibling also in care

50

66

in our analysis is reunification since we expect that families’ welfare and work patterns will affect whether and when they get their children back. The origin of time is December 1, 1992. Children are considered at risk of a reunification event Table 3 Work and welfare characteristics

Welfare and work at time of placement Welfare history Average number of months on On welfare for 3 or more years On welfare currently Lost welfare after placement Working currently Welfare and work at end of placement or at censoring On welfare Working

PercentyMean

N

97 100% 91% 26% 37%

133 133 121 34 49

74% 47%

99 62

584

K. Kortenkamp et al. / Children and Youth Services Review 26 (2004) 577–590

from the day they enter into the foster care system until the day they exit or until the censoring date, which is December 31, 1999. All children still in foster care at the censoring date are censored and all children who exit care for reasons other than reunification (e.g. adoption, aging out) are censored on the day they exit.5 Included in the model are several control variables, some of which have been found to significantly impact reunification rates in past research and some have not been explored previously. Child characteristics include age and race (Black vs. White, Hispanic, Asian, other). Mother characteristics include whether she was married or cohabiting with a partner, whether she had a high school or general equivalency degree, and whether she was in poor health or had a physical, mental or emotional condition, which limits the amount or kind of work she can do. The mother’s characteristics were all taken from the Wave 1 survey. Characteristics of the child welfare experience include the county responsible for the case (Los Angeles vs. Alameda, San Bernardino, San Juaquin6), whether the child was removed for neglect vs. physical abuse, sexual abuse, or other reasons, the number of placements the child had during the spell, whether any of those placements were in kinship care, and whether the child had a sibling placed in foster care between December 1992 and 1999. Several measures of welfare and work were added to the control variables in the model. These variables were all measured at the case level, as opposed to at the mother level, so they are based on whether anyone in the welfare case was on welfare or working. The following variables are time-varying in the model unless noted otherwise. A welfare history variable measures the percent of months from January 1987 to the current month that the case was receiving AFDCyTANF or SSI. Welfare history was included to test whether reunification rates are related to the amount of time spent on welfare. Two time-independent variables measure welfare receipt at the time the child was placed. One indicates that the case was off AFDCyTANF and SSI at placement and another indicates that the case was on welfare at placement but lost benefits in the six months following placement. These variables are similar to those included in Wells and Guo’s (in press) analysis and show how the likelihood of reunification is affected by losing welfare benefits when a child is removed. Two work variables were included in the model. One measures whether anyone in the case was working during the quarter the child was placed in foster care and the other measures work during the last quarter of that placement (or at censoring). In this way, we capture work activity at the time of placement and also post5 This analysis has informative, random censoring due to the inclusion of other exits from the system as censored. This means that knowing how these cases exited the system informs us as to how long they would have stayed in had they not exited through a route besides reunification. The Cox proportional hazards model assumes non-informative random censoring, and a violation of this assumption could cause bias in our estimates. We conducted a sensitivity analysis suggested by Allison (1995) to see how this might have affected results. This analysis showed that treating these cases as non-informative has very little impact on results. 6 We compared Los Angeles to the rest of the counties because we found LA to have the longest length of stay on average compared with the other counties (Needell et al., 2002). Fifteen percent of cases were from other counties in California.

K. Kortenkamp et al. / Children and Youth Services Review 26 (2004) 577–590

585

Table 4 Cox proportional hazards model for reunification Testing the global null hypothesis Without covariates

With covariates

Model chi-square

d.f.

P-value

y2 Log L

361.965

266.022

95.944

15

0.000

Covariates

Parameter estimate

Standard error

P-value

Hazard ratio

Child characteristics Age Black (vs. any other race)

y0.141 y1.746

0.042 0.457

0.001 0.000

0.868 0.175

Mother characteristics Married or cohabiting (vs. single) No high school degree or GED Poor health or limiting condition

0.447 y0.437 1.475

0.333 0.384 0.372

0.180 0.255 0.000

1.563 0.646 4.372

y0.204 y1.428

0.427 0.491

0.633 0.004

0.816 0.240

y1.024 y0.604 y0.360

0.451 0.202 0.439

0.023 0.003 0.413

0.359 0.547 0.698

y0.016 y1.130

0.011 0.435

0.167 0.010

0.984 0.323

y0.303

0.543

0.577

0.739

1.073 y0.246

0.388 0.384

0.006 0.523

2.924 0.782

Child welfare experience Sibling in foster care Removed for neglect (vs. any other reason) Had a kinship care placement Number of placements Los Angeles county Welfare and work Welfare history On welfare at time of placement, then lost welfare after placement (vs. on welfare and stayed on after) Off welfare at time of placement (vs. on welfare and stayed on after) Working at time of placement Working at time of exit

placement. These could have differential effects on reunification since postplacement work could interfere with the process of getting a child back but work at the time of placement should not. 3. Results Results of the proportional hazards model are displayed in Table 4. The hazard ratio gives an indication of whether the reunification rate for a specified group is higher or lower than average. A hazard ratio greater than 1 implies a higher rate of reunification, a hazard ratio less than one implies a lower rate. For a categorical variable, the ratio corresponds to the multiplicative change in the hazard of reunification associated with a particular value of the variable relative to the comparison category for that variable. For a continuous variable, the hazard ratio

586

K. Kortenkamp et al. / Children and Youth Services Review 26 (2004) 577–590

Fig. 1. The relationship between welfare receipt at the time of foster care placement, loss of welfare following placement, and the proportion of children in foster care over time.

shows the multiplicative change in the hazard of reunification related to a one-unit increase in the value of that variable. In general, the control variables had impacts in the model that were expected based on past research. Older children, Black children, children removed because of neglect, children placed in kinship care and children with more placements all had lower rates of reunification. Children whose mothers were in poor health or had a limiting condition were more than four times more likely to reunify. One welfare variable was a significant predictor of reunification. Cases that were on welfare when the child was placed but subsequently lost welfare had lower rates of reunification compared to cases that were on welfare at placement and stayed on after placement. Fig. 1 displays the percent of children remaining in foster care over the seven year observation period for three different groups—those off welfare at placement, those on welfare who stayed on after placement, and those on welfare who lost benefits after placement. This graph shows that those who lost welfare benefits had children who stayed in foster care longer and were less likely to reunify with their children compared with those who stayed on welfare after placement. Lengths of stay and reunification rates for those who were off welfare at the time of placement were not significantly different from the other two groups. The total time the case had spent on welfare was not a significant predictor. Work also had a significant impact in the model. Cases with someone working when the child was placed in care were almost three times more likely to reunify.7 However, whether anyone was working post-placement did not have a significant effect in the model. 7 We also tested variables in our model that measured whether cases ended or started work after the child was placed in foster care but found no effect other than the main effect for working at the time of placement.

K. Kortenkamp et al. / Children and Youth Services Review 26 (2004) 577–590

587

4. Discussion This study has a number of limitations that restricted the analysis and make it difficult to generalize the findings. Firstly, the sample size for this analysis is small, only 133 children in foster care, thus limiting the types of analyses we could perform, the number of independent variables we could include in our model, and the ability to generalize results. Secondly, as noted above, the uniqueness of our sample and the study sites makes it difficult to generalize our findings to other welfare populations in other counties or states. This is a preliminary exploratory analysis, and results should be interpreted with caution. Thirdly, because this study begins with a welfare sample, we needed to use a long entry cohort (7 years) in order to capture enough children entering foster care. This period of time encompasses different policy conditions that could possibly affect reunification rates. However, the distribution of entries over the 7 years was fairly even, so no one period of time dominated the sample. Fourth, the administrative data we use in this study also have limitations. Incorrectly matched data, multiple records for the same person, and missing data all compromise the validity of the administrative data.8 Finally, we were not able to use much of the survey data. Since the survey was not designed to test child welfare research questions it did not have many important variables about the child or family, and the point in time estimate of variables like income are not particularly useful for this type of analysis. Since our sample was ultimately based on the survey sample, there are concerns about non-response bias. We used logistical regression to predict survey nonresponse and to determine how the sample might thus be biased. Using administrative records we were able to examine race, language, aid code (family group or unemployed parent), county, and age. Controlling for these factors, we found that respondents compared with non-respondents were less likely to live in Los Angeles, more likely to speak Vietnamese, Laotian and Armenian, and less likely to speak Spanish. In addition, from survival analyses, we found respondents had longer welfare assistance histories than the general population of recipients. These differences are unlikely to have much impact on reunification rates in this sample, given that 86% of the sample are English speakers and that county and welfare history were not significant predictors in our model. Despite these limitations, if we consider this analysis in relation to current research in this area, we find that most of our results corroborate other studies, adding confidence to the validity of our sample and findings. In support of past research, we found that Black children, children placed in foster care because of neglect, children in kinship care placements, and children with more placements had lower rates of reunification. Unlike past research, we found that as the age of the child increased the likelihood of reunification decreased. This lack of replication 8 Attempts were made by UC DATA to correctly match records between administrative and survey files based on social security numbers, names, and dates of birth (California Work Pays Demonstration Project, 1998, 2000). We dealt with multiple-person records by collapsing all records across cases so that our welfare and employment data are on a case level. We know that 9% of survey respondents are in cases with missing welfare and employment records.

588

K. Kortenkamp et al. / Children and Youth Services Review 26 (2004) 577–590

may have occurred for a number of reasons. For one, we used age as a continuous variable whereas the other studies previously reported categorized ages and used reference groups. Looking at the distribution of reunification rates by age, we saw no pattern requiring categorization. Additionally, our sample of children was much older than in any of the other studies reporting a different result. Forty percent of our sample was 12 or older, in other studies it was only 15–25%, and one (Courtney et al., 1997) did not include children over 12. Given this large group of older children who are more likely to age out of the system, it is not surprising to find this result. The other unexpected result, which has never been explored before, was that mothers in poor health or with limiting conditions had much higher rates of reunification with their children. We hypothesize this may occur because mothers with health conditions may have their children removed from them for short periods of time while they are incapacitated due to their health, but upon recovery can get their children back more quickly than those who abused or showed long-term patterns of neglect. The outcomes of welfare and work in the model show some support for the findings of Wells and Guo (in press). Like them, we found that reunification was not affected by the total time a family had been on welfare. We did find that cases that lost welfare after a child was placed in foster care had lower rates of reunification compared to cases that remained on welfare. It makes sense that losing a substantial amount of income would make it more difficult for a family to meet all the requirements necessary to get a child back. Wells and Guo also found that the more total income (from welfare and work combined) a mother had post-placement, the more likely she was to reunify. Prior to the 1996 federal welfare reforms, welfare agencies were required to terminate assistance payments to parents who no longer had children in their home including parents whose children were removed and placed in foster care. Under the TANF programs, welfare agencies can continue benefits to such parents for up to 160 days if the state elects this option in their state plan. While this study indicates that continuing welfare payments may help parents reunify more quickly with children removed from them, few states extend benefits to such parents beyond 45 days. Many child welfare agencies appear unaware of the state option to extend welfare payments to parents with children in foster care and may want to discuss this option with TANF officials. We examined two periods of work. Firstly, working at the time the child was removed had a positive effect on reunification in our model. Generally this represents working while on welfare, since over 90% of our sample was on welfare at this time. Work which supplements welfare is not likely to substantially increase income, since as wages increase welfare benefits will decrease. Secondly, post-placement work did not affect reunification rates. These results do not support those of Wells and Guo who found that mothers who had more income from wages (relative to welfare) post-placement and mothers who lost AFDC but gained wage income after placement were less likely to reunify with their children. Perhaps if we had been able to measure amount of or changes in wages we would have found a similar result. Instead we found that just having some work experience before having a

K. Kortenkamp et al. / Children and Youth Services Review 26 (2004) 577–590

589

child removed does increase reunification rates. Finding employment requires motivation, organization, and competence, all traits that would help in maneuvering through the child welfare system to meet reunification requirements. It is interesting to note that Geen, Kortenkamp and Stagner (2002) found that welfare recipients who were working were more likely to have a child taken away by the state, while this study finds that those who were working were also more likely to get their children back. Perhaps for some welfare families, struggling with the responsibilities of work and parenting can lead to negative outcomes for children, but these same families are more likely to be capable of getting their lives back in order and bringing their children back home. It would appear that parents who are on welfare and working may be a good target group for intensive pre-placement services. Moreover, child welfare workers may need to collaborate with TANF workers for such dual system families to ensure that parents are not overwhelmed by the requirements of both systems. In conclusion, this study is an early exploration of welfare and work impacts on child welfare outcomes that can contribute to much needed further research on the topic. Any of the studies mentioned earlier that examined predictors of child welfare involvement in a welfare population using administrative data should also be able to look at welfare and work predictors of reunification and length of stay in foster care. Considering the policy relevance of this topic, there is tremendous need for more research with larger samples and better survey instruments. Acknowledgments This study was funded by the Stuart Foundation and The David and Lucile Packard Foundation and conducted as part of the Urban Institute’s Assessing the New Federalism project, a multiyear effort to monitor and assess the devolution of social programs from the federal to the state and local levels. This research uses data collected by the University of California Data Archive & Technical Assistance and the Survey Research Center, Berkeley, CA, as well as foster care administrative data from the Children Services Archive at the University of California Berkeley Child Welfare Research Center. The authors wish to acknowledge Barbara Needell for her assistance in linking the foster care and Work Pays Data. We thank Kathleen Wells for providing valuable comments on earlier drafts of this paper. We also thank Fritz Scheuren and Avner Ahituv for their statistical advice and Stacey Phillips for her research assistance. References Allison, P. D. (1995). Survival analysis using the SAS system: a practical guide. Cary, NC: SAS Institute Inc. Benedict, M. I., & White, R. B. (1991). Factors associated with foster care length of stay. Child Welfare, 70, 45 –58. California Work Pays Demonstration Project. (1998). County Welfare Administrative Data (Codebook, Public Use Version 3). Berkeley: University of California, UC Data Archive & Technical Assistance.

590

K. Kortenkamp et al. / Children and Youth Services Review 26 (2004) 577–590

California Work Pays Demonstration Project. (2000). Wages: Four County Research Sample, Quarter1, 1984–Quarter 4, 1998 (Codebook, Version 5). Berkeley: University of California, UC Data Archive & Technical Assistance. Courtney, M. E. (1994). Factors associated with the reunification of foster children with their families. Social Services Review, 68, 81 –108. Courtney, M. E., Piliavin, I., & Entner Wright, B. R. (1997). Transitions from and returns to out-ofhome care. Social Service Review, 71, 652 –667. Geen, R., Kortenkamp, K., & Stagner, M. (2002). Foster care experiences of long-term welfare recipients in California. Social Services Review, 76, 552 –574. Glisson, C., Bailey, J. W., & Post, J. A. (2000). Predicting the time children spend in state custody. Social Services Review, 22, 253 –280. Goerge, R. M. (1990). The reunification process in substitute care. Social Service Review, 12, 422 – 457. Goerge, R. M., & Lee, B. J. (2000). Changes in child social program participation in the 1990s: initial findings from Illinois. Chicago, IL: Chapin Hall Center for Children. Guo, S., & Wells, K. (In press). Research on timing of foster-care outcomes: one methodological problem and approaches to its solution. Social Services Review. Needell, B., Cuccaro-Alamin, S., Brookhart, A., & Lee, S. (1999). Transitions from AFDC to child welfare in California. Children and Youth Services Review, 21, 815 –841. Needell, B., Webster, D., Cuccaro-Alamin, S., Armijo, M., Lee, S., Brookhart, A., Lery, B., Shaw, T., & Kim, H. (2002, May 7). Child Welfare Services Reports for California wOn-linex. Available: University of California at Berkeley Center for Social Services Research website, http:yy cssr.berkeley.eduyCWSCMSreports. Shook, K. (1999). Does the loss of welfare income increase the risk of involvement with the child welfare system?. Children and Youth Services Review, 21, 781 –814. Slack, K. S., Holl, J., Lee, B. J., McDaniel, M., Altenbernd, L., & Stevens, A. B. (2001, November). Child protection intervention in the context of welfare reform. Paper presented at the Annual Research Conference of the Association for Public Policy Analysis and Management, Washington, DC. US Department of Health and Human Services. (2000). Dynamics of children’s movement among the AFDC, Medicaid, and foster care programs prior to welfare reform: 1995–1996. Washington, DC: Author. US Department of Health and Human Services (2001, June). The AFCARS report: interim FY 1999 estimates (Report No. 6) wOn-linex. Available: http:yywww.acf.dhhs.govyprogramsycbydisyafcarsy cwstats.html. Vogel, C. A. (1999). Using administrative databases to examine factors affecting length of stay in substitute care. Children and Youth Services Review, 21, 677 –690. Wells, K., & Guo, S. (In press). Mothers’ welfare and work income and reunification with children in foster care. Children and Youth Services Review.