Foster Care Dynamics in Urban and Non-Urban Counties
This paper is available on the Internet at:http://aspe.hhs.gov/hsp/fostercare-issues02/dynamics/index.htm
- Basic Caseload Trends
- Risk of Entry to Care
- Demographic Composition of the Children Admitted to Foster Care
- Placement Duration
- Proportional Hazards Models
- Exit Destinations
- Exit Patterns by Urbanicity
- Discharge Destination by Age at Entry and Urbanicity
- Discharge Destination by Race/Ethnicity and Urbanicity
This issue paper has been prepared by Fred H. Wulczyn and Kristin Brunner Hislop of the Chapin Hall Center for Children at the University of Chicago for the Assistant Secretary for Planning and Evaluation, U.S. Department of Health and Human Services, under contract HHS-100-99-0007, delivery order #4. Additional funding has been provided by the Administration for Children and Families, U.S. Department of Health and Human Services and the Annie E. Casey Foundation. Analyses are based on the Multistate Foster Care Data Archive maintained by the Chapin Hall Center for Children. The opinions expressed in this paper are those of its authors and do not necessarily represent positions of the U.S. Department of Health and Human Services or of the Annie E. Casey Foundation.
In this paper, we describe foster care utilization in the urban and non-urban counties that are part of the Multistate Foster Care Data Archive (the Archive). Our aim is to understand the extent to which placement experiences differ depending on whether children reside in primary urban areas, secondary urban counties, or non-urban (rural) counties. In this analysis, primary urban areas are defined as the counties with the largest child welfare system in a given state; secondary urban counties encompass other urban areas; and non-urban counties include all other counties in the state. We are also particularly interested in how these differences have evolved over time. The indicators we use to further our understanding of foster care utilization are placement rates, child demographic characteristics, and placement history, including the type of placement, placement duration, and exit destination.
There are two primary reasons for studying geographic variation in the utilization of foster care. On the one hand, to the extent that poverty and related measures of social disorganization exhibit a strong spatial component, the analysis of geographic areas provides a way to understand how "need" influences child placement. Foster care utilization (and child maltreatment) is expected to be higher on average in areas where social disorganization is more prevalent. In this context, foster care utilization may be treated as one of several measures that describe the impact of poverty on child and family well being. On the other hand, geographically defined variation can provide important clues about the operation of child welfare programs. That is, in the absence of experimental data, geographic variation can draw attention to performance differences that are rooted in how various system resources are deployed. Small area analyses of this type are strengthened considerably when characteristics of the local residents are used to control (partially) for population differences. In turn, data that describe geographic variation can be used as benchmarks against which future performance may be assessed. Research in this vein adds a temporal component to the underlying geographic comparisons.
Despite its potential value, comparative research that explores geographic variation in the utilization of foster care is relatively scarce because the data simply have not been available. Mech (1983) explored differences in out-of-home placement rates, focusing on between state and within state differences. The findings emphasized both the higher rate of placement and the longer duration of placement observed among African American children. However, the data were derived from a survey of states. The expanded use of administrative data as a basis for foster care research has improved opportunities for comparative research using place as an explanatory variable. However, the primary examples are state-specific studies that treat residence in one or more urban counties within a given state as covariates in multivariate statistical models that are used to evaluate a particular child welfare outcome. For example, Goerge (1990) and Needell (1996) studied placement duration by controlling for whether the child lived in the state's primary urban county. Similarly, Courtney (1995) and Wulczyn (1991) evaluated the likelihood of foster care reentry as a function of the child's characteristics and their county of residence. These studies highlight the fact that children from a state's primary urban area fare poorly when their foster care experiences are compared with similar children from other parts of the same state. National studies, whether compiled using the Voluntary Cooperative Information System or the Adoption and Foster Care Reporting System (AFCARS), have tended to stress state-level analyses, with little attention paid to comparisons based on some other geographic taxonomy.
In this paper, we expand on previous research in the following ways. First, we examine basic caseload trends over the decade from 1990 to 1999 for urban and non-urban areas separately. From a historical perspective, the decade represents a critical period. Nationwide, the estimated number of foster children increased each year, from 400,000 children in 1990 to 568,000 in 1999 (U.S. House of Representatives, 2000), even though child poverty and maltreatment rates have subsided (U.S. Census Bureau, 2001; U.S. Department of Health and Human Services, 2001). The conventional assumption has been that large states with major urban areas, such as California, Illinois, and New York, dominated national trends, yet there is no research that fully explores that assumption. As we show, a more complete grasp of national trends hinges on a balanced understanding of trends in primary urban areas separate and apart from trends in other parts of the states.
Second, in prior Archive reports, we relied on a simple division of counties that distinguished each state's largest county from the "balance of the state." The problem with such a simple coding scheme is that other large urban counties are grouped together with counties that are more rural in character. If urban areas do differ markedly from rural counties, the mixing of smaller urban areas with predominantly rural counties may mask trends that are critical to understanding the underlying dynamics of caseload growth. To remedy this problem, we have adopted a three-tiered coding scheme that distinguishes primary urban counties (with the largest child welfare systems) from other (non-primary) urban counties, and counties that are more rural. There may be other categories that could further minimize within-group heterogeneity, but the scheme applied here serves a useful starting point.
Last, using multivariate analysis applied to stratified populations of children, we are able to establish that state-level differences in length of stay may be tied specifically to the performance of child welfare systems in primary urban areas. That is, when the analyses is limited to just those children from either the secondary urban or the non-urban areas of each state, the results reveal smaller between state differences than are found using statewide models that simply control for urbanicity. Of course, within state and cross-state variation linked to geography may reflect differences in the underlying populations (e.g., poverty rates, family structure). Nevertheless, the data highlight the importance of understanding policy and practice as contributors to overall performance differences. We amplify this conclusion by examining exits from placement in the final section of the paper.
[Go To Contents]
The data used for these analyses are from the Multistate Foster Care Data Archive maintained by the Chapin Hall Center for Children at the University of Chicago.(1) The Archive is a database constructed from information drawn directly from the administrative databases that state agencies use to manage and operate their child welfare programs. The Archive currently maintains data from twelve states: Alabama, California, Illinois, Iowa, Maryland, Michigan, Missouri, New Jersey, New York, North Carolina, Ohio and Wisconsin. At last count, the Archive included the placement records for over 1.3 million children placed into foster care.
For each child whose placement is recorded as part of the Archive, the following pieces of information are maintained: type of placement, date of entry, date of exit, and exit destination, as well as the child's gender, age, race/ethnicity and county of residence. Using this information, we process the data to identify child spells that consist of a continuous period of time in care. A single spell may consist of multiple placements and a child may have more than one spell in care.
To enhance the comparability of Archive data across states, we include the following three modifications in the calculation of child spells:
- Spells in care that last fewer than 5 days are excluded from analyses. We have done this because these shorter spells, which are typically court-vacated protective custodies, are reported only in certain states. Including these spells in analyses could distort duration comparisons so we have chosen to exclude them.
- When spells in care were terminated for reasons other than reunification or adoption, and reentry to care occurred within one week, we 'bridged' the gap and treated the two spells as one single spell. This was done to account for local differences in reporting sensitivity.
- Spells in the Archive are "terminated" on the child's twenty-first birthday regardless of whether the state's administrative data indicate an exit from care. We have done this to account for differing state policies regarding the participation of older adolescents in substitute care.
Although the Archive stores the same data elements for each of the states, and all states have provided data on activity through December 31, 1999, state data provided to the Archive differ in two ways. First, the time period covered varies. The Archive includes information from the late 1970's forward for Illinois, from the early to mid 1980's forward for Michigan, Missouri, New Jersey, and New York, from the late 1980's forward for Alabama, California, and Maryland, from 1990 forward for Ohio and Wisconsin and from 1995 forward for Iowa and North Carolina. In addition, seven of the states provide data on children who were already in care at the beginning of the reporting period, while five states provide entry-cohort data that consists of information on children who entered care for the first time during the time periods shown.
Unless indicated otherwise, we selected data from ten Archive states: Alabama, California, Illinois, Michigan, Missouri, New Jersey, New York, Ohio and Wisconsin. We chose these states because their data encompass all children who first entered substitute care between 1990 and 1999. As described above, the Archive data are arranged to describe spells in care. Although a child may have more than one spell in care, for the work presented in this paper, we utilized data that describe the initial spell in care for children entering in or after 1990.
In order to compare placement experiences of children from urban areas to those in other areas, we assigned children to one of three different classes based on their county of residence: primary urban, secondary urban and non-urban counties. To do this, we used a combination of data already stored in the Archive, input from Archive states and data from the U.S. Census Bureau. As a starting point, the Archive maintains data on county of residence for each child placed in care. To identify the primary urban counties in each state, we consulted with state child welfare officials, asking them to identify the largest child welfare system in their state. The emphasis on the largest child welfare system in each state reflects the evolution of the Archive. When the Archive was originally assembled, data from three states - Illinois, Michigan, and New York - were the first to be included. At that time, it was assumed that the child welfare systems in Chicago (Cook County), New York City, and Detroit (Wayne County) faced a set of unique challenges attributable to concentrated urban poverty. That perception was supported by the data insofar as admissions and length of stay in those three jurisdictions exceeded statewide averages. As new states were added to the Archive, we preserved the idea of separating the largest county from other regions of the state. Again, the practice was supported by the data. For example, in the most recent Archive report, the reported length of stay was significantly higher in each of the primary urban counties than it was in the balance of each state, except in Iowa and Missouri (Wulczyn et al 2000).
Our thinking about county-based differences in outcomes is still guided by a perception that child welfare systems in "large" urban counties face challenges defined in part by the social ecology of the principal cities(2). Nevertheless, over time, it has become apparent that a simple dichotomy may be too limited. Specifically, mixing all other urban counties with more rural counties may mask meaningful differences associated with other urban areas. As a partial remedy, we have opted to separate secondary urban counties from non-urban counties. Our goal is to explore the utility of an alternate coding scheme, with the understanding that some other taxonomy could prove to be more theoretically interesting.
|State||Primary Urban Area||Secondary Urban Areas|
|Alabama||Jefferson County (Birmingham)||Madison County, Mobile County, Montgomery County|
|California||Los Angeles County (Los Angeles)||Alameda County, Contra Costa County, Marin County, Orange County, Sacramento County, San Bernardino County, San Diego County, San Francisco County, San Mateo County, Santa Barbara County, Santa Clara County, Santa Cruz County, Solano County, Ventura County|
|Illinois||Cook County (Chicago)||DuPage County, Kane County, Lake County, Macon County, Madison County, Peoria County, Rock Island County, St. Clair County, Tazewell County, Winnebago County|
|Maryland||Baltimore City||Anne Arundel County, Baltimore County, Howard County, Montgomery County, Prince George's County|
|Michigan||Wayne County (Detroit)||Genesee County, Ingham County, Kent County, Macomb County, Oakland County|
|Missouri||Jackson County (Kansas City) and St. Louis||Buchanan County, Clay County, Greene County, St. Charles County|
|New Jersey||Essex County (Newark)||Bergen County, Camden County, Gloucester County, Hudson County, Middlesex County, Monmouth County, Morris County, Passaic County, Somerset County, Union County|
|New York||New York City||Albany County, Erie County, Monroe County, Nassau County, Onondaga County, Rockland County, Schenectady County, Suffolk County, Westchester County|
|Ohio||Cuyahoga County (Cleveland)||Butler County, Franklin County, Hamilton County, Lake County, Lorain County, Lucas County, Mahoning County, Montgomery County, Summit County|
|Wisconsin||Milwaukee County (Milwaukee)||Brown County, La Crosse County, Winnebago County|
To identify counties as secondary urban areas, we used data from the 1990 U.S. Census that indicate the percent of persons in each county who lived within an urbanized area. The U.S. Census Bureau defines an urbanized area as one or more places and the adjacent densely settled surrounding territory that combined have a minimum of 50,000 persons. We used the Census data to define secondary urban areas as counties, other than those identified by the states as their primary urban area, with 75 percent or more of the residents living within an urbanized area. For the purposes of this analysis, we defined non-urban areas as the remaining counties that were not identified as either a primary urban area or a secondary urban area. Table 1 shows the counties designated as primary and secondary urban areas. The remaining counties in each state belong to the third category, non-urban areas.
Our analysis relies on basic measures: population counts, admission rates, placement duration and exit destination. In the first section, we report population counts that reflect the total number of children in foster care December 31st of each year between 1990 and 1999, inclusive. So as to simplify the presentation, no state specific data are presented. Displayed in this way, the data reveal both the geographic distribution of the caseload as well as changes over time in the number of foster children in a manner that is more representative of the nation rather than any one state.
In this section, we also provide the number of first admissions to foster care. Again, the presentation separates geographic areas so that within and between area differences are reflected. Our analysis of admissions is extended to include race and age specific incidence rates. The incidence rates indicate how many children, per 1,000 children in the general population, entered care during a given year. Group specific incidence rates were calculated as follows:
ri = the annual incidence rate for group i, where group i is defined by a unique combination of urban, age and race/ethnicity characteristics (e.g. African-American children under the age of one in primary urban areas),
ni = the number of children in the ith group who entered foster care during a given year, and
Pi= the number of children in the ith group in the population, as estimated by the U.S. Census Bureau.
Placement duration refers to the length of time children remain in foster care until exit. Placement spells end when a child leaves foster care for any one of several reasons. Reunification, one such reason, occurs when a child is returned to the physical custody of his or her biological parents. Children can also be adopted, or discharged to relatives. Two specific measures of placement duration are used. The median duration refers to how much time passes before one-half of an entry cohort exits care. The median duration is estimated using the Kaplan-Meier method (Lancaster, 1990). We also provide the 25th and 75th percentile duration. Respectively, these refer to how much time passes before 25 percent and 75 percent of a given entry cohort leaves care. The second measure of duration is the risk ratio, estimated using the Cox proportional hazard model. The Cox model evaluates the probability of exit per unit time given that an exit has not yet been observed. The hazard model can be used to study independent variables and their effect on discharge rates. The particular advantage of the Cox model and other event history models concerns the treatment of censored data (Lancaster, 1990). Since many of the placement spells that started between 1990 and 1999 had not ended by December 31, 1999 (i.e., there are censored observations), duration estimates are biased downward if the cases not yet discharged are excluded from the analysis. The Cox model uses censored data to estimate exit probabilities.
Finally, we identify three primary exit types: reunification, discharge to relative, and adoption. Reunification refers to those instances when a child is returned to the parent's home, whereas a discharge to a relative means that a child has been released to the custody of an adult, other than the parents, to whom the child is related biologically. The difference between relative foster care and a discharge to a relative has to do with legal custody and payment. Relatives who serve as foster parents receive a stipend from the state, the child typically is in the legal custody of the state, and the placement is supervised by public child welfare officials. When a child is discharged to a relative, the stipend ceases, although states increasingly are relying on subsidized guardianship. The relatives to whom a child is discharged may or may not be the same relative that provided foster care. Adoption refers to those instances when the rights of the biological parents have been severed completely and transferred to a new set of adults who may or may not be relatives.
[Go To Contents]
The foster care caseload in the United States grew dramatically between 1990 and 1999. According to federal estimates, at the beginning of the decade there were 400,000 children placed in foster care; by the end of the decade, the foster care population included 568,000 children (U.S. House of Representatives, 2000). From an historical perspective, the unabated increase in the number of children in foster care is something of an anomaly. The crack cocaine epidemic that started in the late 1980s had largely subsided by 1993; child poverty rates together with child maltreatment reports started to decline in 1993; and the federal government, seeking to stimulate new state investments in preventive service programs, authorized substantial new investments in family preservation, family support, and post-discharge services in 1993 and again in 1997. Still, over the decade, the foster care population grew at a rate of about 4 percent each year.
To better understand these broader changes, we computed the annual census of children in foster care from the placement tracking data provided by seven Archive states: California, Illinois, Michigan, Missouri, New Jersey, New York, and Wisconsin (3). In addition, we aggregated the state specific data for three separate regions - primary urban, secondary urban and non-urban areas - to ascertain whether the broader trends were generally true or specific to certain areas. These data provide useful insights that, together with other data, deepen our understanding of national placement trends.
Presented in Table 2, the foster care census computed from the seven Archive states confirms the underlying growth pattern, at least through 1998. In the seven states, the number of foster children increased from 191,682 in 1990 to 241,066 in 1999, a growth rate of about 2.5 percent annually. The decline of 8,876 children between 1998 and 1999 is one reason why the Archive growth rate is somewhat below the national rate, cited above. Given that the Archive includes a number of the larger states, particularly California, Illinois, and New York, these data suggest that the foster care census nationwide may start to decline, notwithstanding other changes that affect caseload trends.
|Year||Total||Primary Urban||Secondary Urban||Non-Urban||Total||Primary Urban||Secondary Urban||Non-Urban|
1 % change from 1990-1999
As also shown in Table 2, trends in the primary urban areas of the seven states differ markedly from the trends observed in the secondary and non-urban regions. Specifically, between 1997 and 1999, the population of foster children in the primary urban areas declined by nearly 10 percent, from 144,763 to 130,659. During that same two year period, the number of foster children in the secondary and non-urban areas of the states increased by seven and three percent, respectively. As a result, the relative mix of foster children shifted slightly, away from urban areas. In 1996, 59 percent of the foster children in the seven states were in urban areas; by the 1999, the percentage had fallen to 54 percent. Thus, the assumption that the national growth in foster care caseloads was driven by events in primary urban areas was a better description of the first half of the decade than the end of the decade, when caseloads fell in primary urban areas.
One reason primary urban areas reported a declining foster care population during the later part of the decade has to do with admission patterns. Presented in Table 3, total first admissions to foster care declined by 3.4 percent between 1990 and 1999. Most of the decline came after 1997 when first admissions peaked at 76,205. However, the drop in admissions overall was largely restricted to primary urban areas. In 1999, first admissions to urban foster care systems were nearly 20 percent below the level recorded in 1990. Between 1996 and 1999, first admissions in primary urban areas declined by 26 percent. In contrast, the secondary urban areas and non-urban areas reported admissions increases of 13.5 and 9.3 percent, respectively. Consequently, by 1999 only 39 percent of all first admissions were in primary urban areas, compared to 48 percent in 1996.
|Year||Total||Primary Urban||Secondary Urban||Non-Urban||Total||Primary Urban||Secondary Urban||Non-Urban|
|(1) % change from 1990-1999|
Changes in admission patterns such as those shown in Table 3 may be affected by changes in the underlying populations. To adjust for the number of children in the population of each area, we computed annual incidence rates by area, age, and race/ethnicity. The result provides an estimate of the risk or likelihood that a child will enter care (4).
Figure 1 shows the annual incidence rate per 1,000 children for primary, secondary, and non-urban areas. The graph reveals two fundamental patterns. First, the rate of entry into foster care in primary urban areas is considerably above the rate of entry in other areas. The higher rate of placement is consistent with the relatively higher levels of social disorganization found in the primary urban areas. The rate of entry found in secondary urban areas was slightly below the rate recorded in non-urban areas. Second, the incidence rates for secondary urban areas and non-urban areas were relatively steady over the decade while the incidence rate for the primary urban area decreased from 5.0 in 1996 to 3.8 in 1999, a drop in the admission rate that coincides with the overall drop in admissions.
Figure 1 Rate of First Admission to Foster Care by Year and Urbanicity: Children Ages 0 - 17 (AL, CA, IL, MD, MI, MO, NJ, NY, OH, WI)
Table 4 shows the admission rate by age at entry and urbanicity. For each age group, incidence was highest in the primary urban areas. Age specific incidence rates in the secondary urban areas are slightly lower than non-urban rates for each of the age groups except for infants. Children under the age of one had the highest risk of entering care; infants in the primary urban areas were the most likely to enter foster care. During the ten-year period we examined, the incidence rates for children under the age of one in primary urban areas were twice the rates for the same age group in other areas and more than four times the rates for one to five year olds.
|Age at Admission and Urbanicity||1990||1991||1992||1993||1994||1995||1996||1997||1998||1999|
|Less than 1 year|
1 to 5 years
6 to 12 years
13 to 17 years
(Rate per 1,000)
Over time, there have been important changes in the rate of placement. Specifically, the rate of placement for infants in primary urban areas declined, from 19.0 in 1990 to 14.9 in 1999. The downward trend is largely the result of changes in New York City and Chicago. However, infant placement rates in the secondary and non-urban areas increased over the decade. Among older children, placement rates generally declined in the primary urban areas and increased in other areas. The lone exception involved 13 to 17 year olds. In all of the designated areas, the rate of placement for the oldest children declined.
|Race/Ethnicity and Urbanicity||1990||1991||1992||1993||1994||1995||1996||1997||1998||1999|
(Rate per 1,000)
To further understand placement trends, we calculated incidence rates by urbanicity and race/ethnicity for children entering care under the age of fifteen (5). The data, presented in Table 5, reveal the following patterns. First, placement rates for African American children exceeded the rate reported for white and Hispanic children, regardless of the geographic area. Second, the overall rate of placement among African American children declined from 9.0 per 1,000 children in 1990 to 7.3 in 1999, the largest change in the rate of placement. Third, placement rates for African American and Hispanic children were highest in the primary urban areas at the beginning of the decade. As placement rates in the primary urban areas declined, geographic differences diminished. Among Hispanic children, placement rates from one area to another were nearly identical by 1999, largely because of a downward trend in the rate of placement in the primary urban areas between 1996 and 1999. Among African American children, rates of placement were highest in the primary urban areas, but less so at the end of the decade than at the beginning. Fourth, among white children, the rate of placement was highest in the non-urban areas. Moreover, the observed rate of placement for white children was essentially unchanged over the decade.
The pattern of placement risk, together with differences in the "local" population, influences the demographic composition of the caseload in a way that is tied directly to geography. The clearest illustration of this point is found in Figure 2. The data presented show the distribution of first admissions to foster care by race/ethnicity and urbanicity. About 55 percent of the children admitted to foster care in primary urban areas between 1990 and 1999 were African American children. Hispanic children accounted for another 18 percent. White children made up the smallest racial/ethnic group in the primary urban areas, the result of both a low placement rate and the smaller number of resident white children.
Figure 2 Percentage Distribution by First Admissions to Foster Care, Race/Ethnicity, and Urbanicity: 1990 - 1999 (AL, CA, IL, MD, MI, MO, NJ, NY, OH, WI)
The racial and ethnic makeup of the children admitted to foster care in non-urban areas was sharply different. Despite the considerably lower placement rates shown in Table 5, the majority of children admitted to foster care in non-urban areas was white. Clearly, this is because the more rural areas have higher concentrations of whites in the general population. In the secondary urban areas, the mix of white and African American children is nearly even, with slightly more white children among those children admitted to foster care. Again, the composition reflects an interaction between risk and the characteristics of the general population. Specifically, there are more African Americans among the children entering foster care than indicated by the general population because the rate of placement among African American children is so much higher. Presumably, the higher rate of placement is connected to poverty rates. However, the higher incidence could also reflect deficiencies in the service delivery systems serving African American communities, or other forms of bias.
The age distribution of children entering foster care also has a distinct structure. According to previous studies carried out using the Archive, the fundamental and enduring structure is bi-modal. The largest group of children entering foster care is made up of children who start their initial placement prior to their first birthday. This basic pattern has been true since at least the mid-1980s, with some minor changes during the crack cocaine epidemic that resulted in more babies entering care. The second distinct subgroup is adolescents or children between the ages of 12 and 15. These features are highlighted in Figure 3. Children between the ages of 1 and 11 make up yet a third population, inasmuch as the children in this group are a consistent proportion of all children admitted to placement, regardless of geography.
Figure 3 Percentage Distribution by Age at First Admission to Foster Care and Urbanicity: 1990 - 1999 (AL, CA, IL, MD, MI, MO, NJ, NY, OH, WI)
Figure 3 also reveals that the age distribution exhibits a subtle, but important variation from the general pattern when the distributions for each geographic region are separated. Specifically, the age distribution in the primary urban areas tended to be younger (i.e., a higher proportion of babies), whereas in the non-urban areas, the population was older. The age distribution in secondary areas falls in the middle: somewhat older in that there are fewer babies than found in the primary urban areas, but fewer adolescents than found in the non-urban areas. According to the data in Figure 3, 25 percent of the admissions in primary urban areas involved babies, compared with 15 percent in non-urban areas. At the other end of the age spectrum, adolescents (children between the ages of 12 and 15 at the time of admission) accounted for 26 percent of the first admissions in non-urban areas, 20 percent of the admissions in secondary urban areas, and 17 percent in the primary urban areas. Together, the demographic patterns suggest a pattern dominated by placement of very young (i.e., infant) African-American children in urban areas, and white adolescent children in non-urban areas.
[Go To Contents]
In this section, we examine differences in placement duration that are related to child characteristics, placement history, and urbanicity using the Cox proportional hazards model. For the most part, the results from these analyses replicate findings published elsewhere. However, the findings presented here include results based on stratified samples that separate children based on geographic area. The use of stratified samples isolates time trends that are otherwise more difficult to detect in a general sample.
Our analysis of placement duration starts with simple descriptive statistics. Specifically, we calculated duration quartiles for each of the three geographic areas we have used so far to organize our analysis. The first quartile indicates how much time elapsed before 25 percent of the children admitted between 1990 and 1999 were discharged from their first spell in foster care. Similarly, the 50th and 75th quartiles indicate how much time passed before 50 and 75 percent of the children were discharged from care. Table 6 provides these data by urbanicity for first placement spells that started between 1990 and 1999.
In non-urban and secondary urban areas, 25 percent of the children completed their first spell in slightly more than 2 months time. In contrast, 25 percent of the children in primary urban areas completed their first spell in 5.3 months, more than twice the duration for all other children. Following the rapid initial rate of discharge, children from the secondary urban areas slowed down, at least when compared with children from non-urban areas. For the second quartile, 50 percent of the children admitted to foster care in the secondary urban areas had completed their initial spells in slightly more than one year. The comparable figure for children from non-urban areas was under 10 months. For children admitted in primary urban areas, it took almost two years for 50 percent to complete their first placement spell. The third quartile shows a similar pattern. In the non-urban areas, the elapsed time between when the 50th percentile was reached and when the 75th percentile was reached was 16 months. The comparable figure in the secondary urban areas was 20 months; in the primary urban areas, the elapsed time was 34 months. These results point to the fact that placements in primary urban areas were completed much slower initially, and that the disparity persisted with the passage of time.
|Quartiles||Statewide||Primary Urban||Secondary Urban||Non-Urban|
The duration quartiles described above indicate that the length of time in care varies by the degree of urbanicity. We also know from prior work with Archive data that duration is affected by such factors as the child's age at entry and race/ethnicity. Since the population of children entering placement in each of the three areas differs with respect to these attributes, it is possible that the results described in Table 6 are a function of the population differences. To evaluate this possibility, we studied placement duration using the Cox proportional hazards model. The proportional hazard model measures the probability a child will exit foster care per unit time. The hazard model also considers the effect of a given independent variable on duration, controlling for other variables in the model. The independent variables included are the year the spell began, urbanicity, race/ethnicity, age at entry, primary care type and state. The effects are measured as risk ratios that indicate the relative risk a child will exit per unit time, other things being equal. For each independent variable in the model, one category is designated as the standard against which the risk ratios for other categories can be compared. The "standard" category is assigned a risk ratio of 1.00. Risk ratios above one imply shorter spells and risk ratios less than one imply longer spells. For example, for Admission Year, we designated 1990 as the standard. The placement experiences (duration) of children admitted in other years were compared with the experience of children admitted in 1990. In Table 7, the risk ratio for children entering in 1999 is 0.93. That is, controlling for all the other variables in the model, children entering care in 1999 had a lower risk of exiting care compared to the children entering in 1990. In other words, the rate of exit appears to have slowed (7). Table 7 shows the results from four separate hazard models. The statewide model considers the experience of all children admitted for the first time between 1990 and 1999 in the Archive states. The three supplementary models rely on stratified samples that include only children admitted to foster care from the counties that comprise the three geographic areas.
The statewide model indicates that overall, exit rates declined between 1997 and 1999, as compared to 1990. Prior to 1997, there had been slight increases in the exit rates. These results confirm that lower admissions in the later part of the decade were offset by an increase in placement duration, thereby slowing the effect of falling admissions on the overall size of the population.The statewide results also indicate a lower risk of exit and therefore longer length of stay in the primary urban regions. The risk ratio reported for secondary urban areas was below the ratio for non-urban areas, but above the ratio for primary urban areas. The findings verify that observed differences reported in Table 6 are not due to population differences and that the use of more refined geographic categories is warranted. The statewide models also show that African American children leave foster care at rates that are slower than those observed for either white or Hispanic children. Some limited insights into why African American children stay longer in foster care are provided in the stratified samples.
With respect to age at admission and placement type, the results found in Table 7 replicate findings reported previously. Children admitted as infants (less than 1 year) and children placed in kinship homes remained in foster care considerably longer than all other children did. Finally, there are important state differences. When compared with Illinois, children from other states spent less time in foster care. Again, this finding has been reported in prior Archive studies.
The models based on the stratified samples amplify these findings in the following ways. Generally, effects associated with time, race/ethnicity, age, and placement type were most pronounced in the primary urban areas. For example, exit rates slowed more rapidly in the primary urban areas. Early in the decade, however, movement through the system was actually increasing in the primary urban areas. In the secondary urban areas, exit rates relative to 1990 were generally higher each year, whereas exit rates were slower in non-urban areas. Thus, trends in underlying placement duration differ in important ways, depending on the historical period (early vs. later in the decade) and whether one is studying primary urban, secondary urban, or non-urban areas.
African American children stay longer in foster care than white and Hispanic children, no matter where they live. However, the observed difference in time spent in foster care by African American children was greatest in the primary urban areas.
The age effects were also more pronounced in the primary urban areas. Compared to infants, children placed into foster care between their first and fifth birthday were about 18 percent more likely to exit foster care per unit time than infants were. The comparable figure for one to five year olds in secondary and non-urban areas was 6 percent and 9 percent, respectively. Older children generally had higher exit rates, with the noteworthy exception of children 6 to 12 years from secondary urban areas. According to the results in Table 7, children in this age group can be expected to stay in foster care longer than even infants do.
|Characteristic||Statewide||Primary Urban Area||Secondary Urban Area||Non-Urban|
Year of Admission
Balance of State
Primary Urban Region
Age at Admission
Less than 1 year
1 to 5 years
6 to 12 years
13 to 17 years
Primary Care Type
* p < .05
Within placement type variation was also greater in the urban areas. When compared with foster family placements (foster boarding homes), kinship placement lasted much longer (risk of ratio of .73). In non-urban areas, exit rates from relative homes, although slower, were not as sharply different (risk ratio of .90). Congregate care placements tended to be shorter than foster care placements in both primary and secondary urban areas; the difference in non-urban areas was less pronounced.
Finally, patterns of state variation are least pronounced in secondary urban areas, particularly when compared with urban areas. For example, in urban areas, when states are compared with Illinois, the range of risk ratios runs from 1.37 in Maryland (MD) to 2.26 in Missouri (MO). In the secondary urban areas, the risk ratio comparisons range from 1.02 in Missouri to 1.66 in Ohio. These findings suggest the movement of children through the foster care system in St. Louis and Kansas City (the primary urban counties in Missouri) is very different than the movement of children in Cook County (the primary urban county in Illinois). At the same time, the data suggest that secondary urban areas in Missouri and Illinois do not differ significantly. Indeed, the overall pattern of state variation in secondary urban areas is less remarkable than is the case in either urban or non-urban counties. Whether the "similarity" is attributable to performance differences in the systems that provide service in secondary urban areas or are the result of still unmeasured characteristics of the children is a fundamental question for future research.
[Go To Contents]
In this last section, we examine the type of discharge or exit destination. Children leave foster care in one of several ways, however, the preferred path is to have the children returned home if possible, or adopted when reunification is not possible. Increasingly, children leave foster care to live with family members other than their parents. As we discussed previously, discharge to a relative differs from kinship or relative foster care because legal custody typically shifts from the child welfare agency to the relative. Some states now provide subsidies to relatives caring for family members, but these arrangements tend to exist outside the traditional foster care system. Although data in the Archive track children who are discharged to relatives, we have not yet adapted coding schemes that differentiate subsidized guardianship arrangements from other relative discharges. In addition, for some older children, reunification is not possible, and adoption is less likely. If children in this situation reach the age of majority while still in foster care, they are said to have "aged out."
We focused our analysis on first placements spells for children who entered care between 1988 and 1995 (8). The time period was shifted "backward" to allow for the greatest number of exits. In the case of adoption, extending the observation period is particularly important, given that adoptions take so long to complete. Our presentation follows the structure of previous sections. First, we consider geographic differences; then we evaluate age and race/ethnicity effects.
Table 8 shows the exit distribution by destination and urbanicity. Primary urban areas had the lowest rate of exit to reunification and the highest rates of exit to adoption. Conversely, non-urban areas had sharply higher rates of family reunification and much lower rates of adoption. If other relative exits are added to family reunification, to form a category of "family exits," then 54 percent of first placement spells in urban areas ended because the child returned to his or her family. The comparable figure in secondary urban areas was 63 percent. Two-thirds of all children placed in non-urban areas exited to family.
About 3 percent of all the placements ended because the child reached the age of majority. Runaways were slightly more common in the primary urban counties, an unusual finding given that primary urban areas tended to have fewer adolescents, the children most likely to runaway. Finally, other exit types, which include a mix of exits tracked idiosyncratically by states, totaled to about 15 percent of all exits.
|Destination as Percent of All Discharges|
|Urbanicity||Number of Children||Percent Discharged Prior to 12/31/1999||All Discharges||Family Reunif.||Other Relative||Adoption||Reach Age of Majority||Runaway||Other Exit|
Age at admission is an important determinant of exit destination. Young children, the children most likely to enter placement in the primary urban areas, are more likely to be adopted than older children, the children who are most likely to be admitted to foster care in non-urban areas. In other words, the differences observed in Table 8 could be attributable to the age mix of children served in the various regions. To refine further our understanding of exit patterns, we adjusted the results provided in Table 8 using age at admission. These results can be found in Table 9.
|Destination as Percent of All Discharges|
Age at Admission and Urbanicity
|Number of Children||Percent Discharged Prior to 12/31/1999||All||Family Reunif.||Other Relative||Adoption||Reach Age of Majority||Runaway||Other Exit|
Ages 0 - 5
Ages 6 - 12
Ages 13 - 17
The data confirm two basic patterns. First, age is strongly correlated with exit type. Across all geographic areas, 45 percent of the children admitted between the ages of 0 and 5 were reunified; 29 percent were adopted. Among 6 to 12 year olds, family reunification accounted for 55 percent of the completed placements. Adoptions accounted for just 9 percent of the exits. Second, although the age is an important factor, urbanicity is also a critical factor. For example, among the youngest children, 51 percent placed from non-urban counties went home to their parents, compared to 41 percent in primary urban areas. In contrast, one third of the 0 to 5 year olds admitted in primary urban areas were adopted. The rate for children of the same age from other areas was about 25 percent. Given that a large proportion (13 percent) of the 0 to 5 years from primary urban areas were still in care as of December 31, 1999, we can expect the proportion adopted to rise substantially.
Table 9 also shows that the highest rates of runaway were for teenagers and especially those in primary urban areas where nearly one quarter of teenagers exited placement by running away. That rate was two and a half times the rate reported for youth (13 to 17 year olds) from non-urban areas. Six to twelve year olds in primary urban areas were also twice as likely to leave by running away as children the same age from other regions.
Table 10 extends the analysis by using race/ethnicity to refine our understanding of exit patterns. Adding race and ethnicity to the analysis does not alter the findings substantially: within race-specific categories, children from primary urban areas are less likely to be reunified and more likely to be adopted. This is true for white children, African American children, and Hispanic children. Adoption was most common among African American children from the primary urban areas. African American children were generally more likely to be discharged to a family member other than the parent(s). Overall, the highest rates of reunification were for Hispanic children in secondary urban areas (63 percent) and non-urban areas (61 percent) while the lowest rate of reunification was for African American children in primary urban areas (40 percent). In addition, Hispanic children were somewhat more likely to run away from placement. White children were more likely to reach the age of majority, but this may be because white children tend to be much older at the time of admission.
[Go To Contents]
National foster care trends have been a source of persistent concern for some time. The unabated growth in the number of foster children throughout the 90s stands in sharp contrast to changes in the incidence of childhood poverty and maltreatment. To the extent that poverty and maltreatment are linked to the need for foster care, the growth in the number of foster children defies conventional explanations.
In part, the reason why trends in poverty and maltreatment do not align with changes in the number of children in foster care has to do with how the underlying dynamics are untangled. Our analysis of foster care trends in the Archive states disaggregates the underlying admission and discharge dynamics and reveals patterns that account for a rising foster care population in the midst of improving social conditions. Moreover, overall trends are more readily understood after counties are grouped by the degree of urbanicity. In primary urban counties, the number of admissions and corresponding placement rates did in fact fall in the late 1990s, a pattern that is more consistent with trends in maltreatment and poverty. However, the change in admissions was limited to the primary urban counties. In the secondary urban and non-urban areas, new admissions were actually higher in 1999 than they were in 1990.
|Destination as Percent of All Discharges|
|Race/Ethnicity and Urbanicity||Number of Children||Percent Discharged Prior to 12/31/1999||All||Family Reunif.||Other Relative||Adoption||Reach Age of Majority||Runaway||Other Exit|
The data also indicate that changes in admissions were mitigated at least somewhat by discharge patterns. During the period of falling admissions recorded in the primary urban areas, an increase in the length of stay sustained the population at a level that was higher than it would have otherwise been given the number of admissions. In the secondary urban counties, where the number of admissions increased over the decade, the opposite was true-accelerated movement through the foster care system somewhat reduced the impact of higher admissions. Taken together, these data show the importance of looking at the interaction of admissions and discharges. In particular, causal models that link changes in social conditions to changes in the foster care population have to consider both regional variation and whether underlying social conditions affect admission dynamics differently than discharge dynamics. These data suggest that such distinctions are indeed important.
The findings are important also because they clarify the context that interacts with public policy. Federal child welfare law and policy creates a framework that shapes the development of child welfare systems around the country. This includes both procedural and systemic expectations as well as the definition of child welfare outcomes that have been established federally, in consultation with state and local officials and other stakeholders. Nevertheless, the operational path to achieving those ends inevitably depends on a starting point that is defined by geography. In some instances, the geography will be defined by political boundaries, such as states. In other instances, it will prove productive to consider geographical boundaries that are defined in social and economic terms. Our use of county groupings based on urbanicity is one such example, albeit a limited one. Research that explores taxonomies that are more theoretically meaningful should help sort out the underlying issues.
The importance of geographic variation is underscored when the characteristics of the children served in foster care are added to the analysis. Our principal finding has to do with the exceedingly high rate of placement among children who were under the age of one at the time of placement. Although the infant placement rate fell between 1990 and 1999, infants in urban areas were 4 to 5 times more likely to be placed in foster care than children of other ages were. The differences were nearly as profound in the other regions. Moreover, in the secondary and non-urban areas of the Archive states, the risk of placement for infants actually increased. Thus, placement rates were declining in general over the decade, but trends involving the very youngest children were moving in the opposite direction in some parts of the states. In addition, when the basic risk of placement for infants is compared with the placement rate for adolescent children, there is little doubt that infants make up the single most important sub-population served by the nations foster care system.(9) There is little doubt that initiatives on the part of federal, state and local government aimed specifically at the needs of at-risk families with pregnant women would have a significant impact on the utilization of foster care nationwide, provided programs that work are deployed.
Adding race and ethnicity provides yet another layer of insight to our understanding of placement trends. First, the risk of placement among African American children is higher than the risk of placement for either white or Hispanic children. Overall, the likelihood of placement among African Americans has been 3 to 4 times greater, although the differences grew smaller over the course of the decade. Second, the risk of placement for African American children was greatest in urban areas. Published elsewhere, the admission data show that among African Americans, infants account for about 35 percent of all new admissions in the Archive states. Given the racial and economic segregation that characterizes major cities in primary urban counties, these data demonstrate a clear need to build service capacity in low-income neighborhoods that targets interventions to families expecting a newborn.
Geography, together with the age and race of the children served in the foster care system, deepens our understanding of national trends and brings a focus to how interventions might be developed. Yet, questions remain unanswered. Of special importance, the persistent differences that define the experiences of African American children in the foster care system stand out. In this regard, the analysis of geographic variation provides some clarification. Our findings suggest that admission rates and placement duration are greatest for African American children who live in primary urban areas. Since placement rates are greatest for all children in urban areas, the data suggest that racial differences have to be considered in light of underlying patterns of need. In other words, we expect that placement rates will be higher in urban areas because social disorganization is more common. Since African American children are more likely to live in areas of concentrated urban poverty, the elevated risk of placement is to some extent expected. The fundamental question to be answered is whether need in its many forms accounts fully for the observed variation in foster care utilization. Barth and his colleagues (2000) have recently completed a study that attempts to consider this basic set of issues.
The analysis presented here also suggests that the underlying administrative processes may be influenced by attributes of place other than the social conditions typically associated with urban areas. Children of all races are more likely to be adopted in urban areas, whereas children in secondary and non-primary urban areas are more likely to be reunified. Unmeasured child level differences linked to geography may account for these outcomes, but it would be well worth the effort needed to identify the organizational characteristics that contribute to outcomes. In other words, it is rather unlikely that child, family, and community level differences account for all the reasons why some children spend more time in foster care than others do. The possibility that place-specific administrative processes are an important factor is reinforced by the fact that geographic variation in performance is more compressed in the secondary urban areas than it is in the primary urban areas. That is, the states are less different when placement duration is studied in the secondary urban areas than when primary urban areas are compared. Since administrative practices and procedures are more amenable to policy interventions, the need to better understand these differences is an inevitable direction for future research. Among the organizational differences, funding levels in relation to the economic base, staffing patterns (including turnover), the use and availability of data, and other administrative procedures represent an excellent starting point.
The nations foster care system consists of fifty state systems. Understanding national trends depends largely on the events that transpire at the state level. Still, states are not socially and economically homogeneous. National trends are also influenced by trends shaping subdivisions within states. Our analysis demonstrates the fact that grouping political subdivisions (counties) together by their urban character preserves a pattern of variation in the underlying state data that is in fact more revealing than the state data alone. As federal, state, and local child welfare administrators press on with outcome monitoring, it is difficult to overstate the importance of the variation linked to place and time.
[Go To Contents]
Barth, R., Miller, J., Green, R., & Baumgartner, J. (2000). Children of Color in the Child Welfare System: Toward Explaining Their Disproportionate Involvement in Comparison to Their Numbers in the General Population. Washington, DC: U.S. Department of Health and Human Services, Administration for Children, Youth, and Families.
Courtney, M. (1995). Reentry to foster care of children returned to their families. Social Service Review 69(2), 226-241.
Goerge, R. (1990). The reunification process in substitute care. Social Service Review 64(3), 422-457.
Lancaster, T. (1990). The Econometric Analysis of Transition Data. Cambridge, England: Cambridge University Press.
Mech E.V. (1983). Out-of-home placement rates. Social Service Review 57(4) 659-667.
Needell, B. (1996). Placement stability and permanence for children entering foster care as infants. Digital Dissertations. (UMI Proquest. No. AAT 9723126).
U.S. Census Bureau. (2001). " Historical Poverty Tables, (Table) 3. Poverty Status of People, by Age, Race, and Hispanic Origin: 1959 to 2000. [on-line]. Available http://www.census.gov/hhes/poverty/histpov/hstpov3.html.
U.S. Department of Heath and Human Services, Administration on Children, Youth and Families. (2001). Child Maltreatment 1999. [on-line]. Available: http://www.acf.dhhs.gov/programs/cb/publications/cm99/index.htm.
U.S. House of Representatives, Committee on Ways and Means. (2000). 2000 Green Book: Background material and data on programs within the jurisdiction of the Committee on Ways and Means, Oct. 6, 2000. 106th Cong. 2nd sess.
Wulczyn, F. (1991). Caseload dynamics and foster care reentry. Social Service Review 65(1), 133-156.
Wulczyn, F., Brunner, K. (2000). Infants and Toddlers in Foster Care. Protecting Children, 16 (1), 4-11.
Wulczyn, F., Hislop, K, & Harden, B. J. (2001). The Placement of Infants in Foster Care. Infant Mental Health Journal, (in press).
Wulczyn, F., Hislop, K. B. & Goerge, R. M. (2001). An Update from the Multistate Foster Care Data Archive Foster Care Dynamics 1983-1998. Chicago: Chapin Hall Center for Children.
[Go To Contents]
(1) For a more complete description of the Multistate Foster Care Data Archive, see Wulczyn, F.H, Kristen Brunner Hislop, and Robert M. Goerge. Foster Care Dynamics, 1988-1998: An Update from the Multistate Foster Care Data Archive. Chicago: Chapin Hall Center for Children, 2000.
(2) The principal cities in the primary urban counties included in this study are: Birmingham, Los Angeles, Baltimore, Detroit, St. Louis, Kansas City, Newark, New York City, Cleveland, and Milwaukee. Of these cities, only Birmingham and Newark are not included in the City Kids Count, published by the Annie E. Casey Foundation. The City Kids Count volume tracks child well-being in the 50 largest cities in the United States. According to the City Kids Count, the cities located in the primary urban areas of the Archive states had more children living in distressed neighborhoods and higher child poverty rates overall than did other cities.
(3) We used this subset of states because only these states provide a full census for each of the years. The other states provide the number of new admissions since 1990, but the data from these states are missing the children in care on December 31, 1990.
(4) Note that the analyses presented from this point forward include three additional states: Alabama, Maryland, and Ohio. Later, in the section that describes exit patterns, California is dropped from the analysis because after the conversion to the SACWIS data system, the quality of exit data received from California declined.
(5) A truncated age range was used because county-level population projections for race and age produced by the U.S. Census Bureau are based on age groups rather than single years.
(6)Wulczyn, F.H, Kristen Brunner Hislop, and Robert M. Goerge. Foster Care Dynamics, 1988-1998: An Update from the Multistate Foster Care Data Archive. Chicago: Chapin Hall Center for Children, 2000. Wulczyn, Fred, Kristen Brunner Hislop and Brenda Jones Harden. Infant Placements into Foster Care. Infant Mental Health Journal, forthcoming.
(7) Of course, unmeasured attributes of the children might account for the observed trends.
(8) Note, also, that California has been dropped from the analysis. This is because exit data are not reported on the most recent data dumps. Specifically, the exit date is available, but the destination is not reported. Thus California can be used in the analysis of duration, but not in analyses where specific destinations are being studied. Efforts are underway to improve the quality of exit data from California.
(9) It also is important to look at subpopulations within infants. In another paper, we analyzed the population of infants placed into foster care and found that 48 percent of all infants were placed before 30 days. Outcomes for children placed so soon after birth were also different. When compared with children placed between 4 and 12 months of age, children placed within 3 months of being born were much more likely to adopted. Wulczyn, Fred, Kristen Brunner Hislop, and Brenda Jones Harden. The Placement of Infants in Foster Care. Journal of Infant Mental Health, (in press).