a comparison of foster care entry risk at three

Upload: veronica-tapia

Post on 03-Jun-2018

223 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/12/2019 A Comparison of Foster Care Entry Risk at Three

    1/15

    Substance Use & Misuse, 43:223237

    Copyright 2008 Informa Healthcare USA, Inc.

    ISSN: 1082-6084 (print); 1532-2491 (online)

    DOI: 10.1080/10826080701690631

    A Comparison of Foster Care Entry Risk at ThreeSpatial Scales

    BRIDGETTE LERY

    Chapin Hall Center for Children, University of Chicago, Chicago, Illinois, USA

    This study addresses the problem of operationalizing neighborhood boundaries by inves-tigating foster care entry risk at three spatial scales. Foster care entries from a Californiacounty between 2000 and 2003 (n = 3,311) are geocoded to each of the three scales(N =46 zip codes, 320 census tracts, and 983 block groups). Exploratory spatial dataanalysis is used to compare spatial autocorrelation of entry rates among scales. Resultssuggest that depending on how neighborhoods are defined, the geographic pattern offoster care incidence changes. Implications for accurately targeting services to high-risk

    neighborhoods and future research directions are noted.

    Keywords spatial scale; child welfare; foster care; spatial analysis; modifiable arealunit problem; risk factors; high risk neighborhood

    Introduction

    The push for neighborhood-based child welfare and other social services demands bettertools to investigate the role of location in child and family outcomes. Geographic infor-

    mation systems (GIS) and spatial analysis are now commonly used to map, for instance,

    rates of child maltreatment in neighborhoods and to model its relationship to neighbor-

    hood social conditions such as impoverishment or alcohol outlet density (Coulton, Korbin,

    Su, and Chow, 1995; Freisthler, 2004). The spatial representation of social phenomena

    on GIS maps exposes patterns and relationships that are not easily recognized using con-

    ventional analytical procedures. However, there is a methodological gap between the de-

    scriptive mapping of social data across geographic areas and the statistical inference of

    the relationships of such variables across space. Most studies of neighborhood effects

    on child maltreatment and subsequent policies promoting community-based interventions

    have not considered the unique statistical biases that often arise in maps and models that use

    geographic data.

    The problem stems from the fact that there is no consensus on what is the best unit

    of analysis to call neighborhood. Researchers chose administrative units such as census

    tracts to represent neighborhoods in ecological studies out of convenience in the absence

    of an available systematic method for doing so. This creates two related problems. For one,

    administrative units are usually not very good proxies for neighborhoods. For another, the

    variation of a phenomenon over a geographic area may change if the data are represented

    over a different geographic scale (Coulton, Cook, and Irwin, 2004). These issues are known

    as thezone problemand thescale problem, respectively, and fall within a larger definition

    Address correspondence to Dr. Bridgette Lery, Chapin Hall Center for Children at the Universityof Chicago, 1313 East 60th Street, Chicago, IL 60637. E-mail: [email protected]

    223

    Forpe

    rsonaluseonly.

  • 8/12/2019 A Comparison of Foster Care Entry Risk at Three

    2/15

    224 Lery

    of problems in ecological studies, the modifiable areal unit problem (MAUP; Openshaw

    and Taylor, 1979; Robinson, 1950).

    The zone problem refers to the fact that administrative units are typically poor rep-

    resentations for neighborhoods because the concentration of a phenomenon in one spa-

    tial unit will tend to spill over into adjacent units, making the relationships between the

    units not independent. In other words, attributes of areal units may be correlated across

    space. The strength of the correlation depends largely on where the unit boundaries are

    drawn, and this correlation may bias the standard errors in multivariate ordinary least

    squares regression models of ecological data, which assume independence among the

    units of analysis (Bailey and Gatrell, 1995). Further complicating the matter of choos-

    ing a reliable unit, neighborhood boundaries are permeable. People move in and out of

    neighborhoods, making boundaries dynamic over time (Freisthler, Lery, Gruenewald, and

    Chow, 2006).

    Substance Use and Child Welfare in Spatial ContextThe explicit consideration of spatial scale makes a substantive difference when seeking to

    understand how attributes of place may impact the risk of child maltreatment or foster care

    entry and when designing interventions to reduce this risk. For example, there is ample

    evidence that parental substance abuse* is a posited risk factor for child maltreatment and

    that a large portion of children receiving child welfare services are from substance-using

    families (Besinger, Garland, Litrownik and Landsverk, 1999). One recent study suggests

    that integrating substance user and child welfare services might be an effective strategy

    to increase the likelihood of reunification for children placed in out-of-home care (Ryan,

    Marsh, Testa and Louderman, 2006). Coordinating the location of substance user treat-

    ment programs nearby to maltreating parents is one way to do this (DAunno, 1997).However, in order to target the location of treatment services effectively and efficiently, it

    is important to know which areas in a jurisdiction are at the highest risk of using foster

    care.

    From an ecological perspective, it is also important to know whether there is any

    relationship between drug and alcohol availability in neighborhoods and risks to children,

    beyond the known associations at the individual level. Freisthler (2004) found that the

    availability of alcohol in neighborhoods is associated with rates of child maltreatment in

    census tracts, suggesting that a solution could be to reduce the density of bars and other

    alcohol outlets at that spatial scale. However, it is possible that the density of outlets, rates

    of maltreatment, and therefore their relationship to each other changes depending on how

    neighborhoods are defined.

    Defining Neighborhoods

    The child welfare literature offers little guidance on how to measure and define neigh-

    borhoods and, as a consequence, the geographic unit chosen to represent neighborhoods

    varies across studies. A neighborhood can be conceptualized as a geographic place where

    residents share social networks and informal ties. However, such networks are not nec-

    essarily tied to place, and the distinction between neighborhoods and communities can

    become blurred (Newman and Small, 2001). In addition, residents perceptions of their

    *The journals style utilizes the category substance abuseas a diagnostic category. Substancesare used or misused; living organisms are and can be abused. Editors note.

    Forpe

    rsonaluseonly.

  • 8/12/2019 A Comparison of Foster Care Entry Risk at Three

    3/15

    A Comparison of Foster Care Entry Risk at Three Spatial Scales 225

    neighborhoods are unlikely to correspond to administrative boundaries (Coulton, 2005;

    Coulton et al., 2004; Coulton, Korbin, Chan, and Su, 1997). Since social networks and

    residents perceptions are not reliable tools with which to measure homogenous neigh-

    borhoods, researchers typically select either counties (Albert and Barth, 1996; Fryer and

    Miyoshi, 1995; Spearly and Lauderdale, 1983), ZIP codes (Drake and Pandey, 1996), cen-

    sus tracts (Coulton et al., 1995; Deccio, Horner, and Wilson, 1994; Ernst, 2001; Freisthler,

    2004), or block groups (Coulton, Korbin, and Su, 1999; Young and Gately, 1988) as units

    of analysis. Block groups or census tracts are sometimes grouped to model socialization

    spaces that better represent neighborhoods than do individual units (Sampson, Raudenbush

    and Earls, 1997). Census and geographic boundary data are readily available at all of

    these levels of aggregation, allowing for the calculation of rates of social indicators such

    as poverty and, when the data are accessible, rates of child maltreatment and foster care

    entry.

    In addition to the zone problem, administrative units may not represent the scale at

    which the social processes of interest operate (Coulton, 2005). When the level of data

    aggregation is too large, the spatial pattern is obscured (Can, 1993). Maps are merely amodel of reality and can misrepresent data and lead to poor assumptions about relationships

    between variables. People are good at detecting patterns visually, but they also tend to

    identify patterns that are not present (MacEachren, 1995).

    Most neighborhood studies in child welfare limit analysis to one spatial scale without

    formal consideration of the statistical bias that may occur due to spatial autocorrelation

    among units of analysis. Recent evidence suggests that an outcome such as child maltreat-

    ment risk may be affected by neighborhood characteristics differently at different scales

    (Johnston et al., 2004). Chou (1991) found that Morans I (Moran, 1950)the most com-

    monly used measure of spatial autocorrelationincreases systematically with the resolution

    level. If resolution changes the degree to which units are correlated across space, which in

    turn affects the level of bias in the statistical model, then the choice of which spatial scale

    to use in a neighborhood analysis is a nontrivial one.

    If neighborhoods can be defined in a variety of ways, then what criteria should be

    used to examine the geographic distribution of a problem? While most previous studies

    choose one spatial scale to approximate neighborhoods, this study systematically compares

    foster care entry rates across three commonly used operational definitions of neighbor-

    hood to address the question, How can we define neighborhoods operationally so that

    researchers and administrators can choose an appropriate unit of analysis when evaluating

    child welfare outcomes and designing interventions? A related goal is to demonstrate the

    dangers of arbitrarily choosing units of analysis in neighborhood research. Determining

    a practical and reliable unit of analysis to serve as a proxy for neighborhood will im-prove the reliability of research on neighborhoods and the efficiency of place-based service

    allocation.

    Method

    Exploratory spatial data analysis (ESDA; Anselin, 2005) is used to compare spatial auto-

    correlation in foster care entry rates at three geographic scales within Alameda County,

    California. Counties are the administrative jurisdictions for child welfare agencies in Cali-

    fornia. Foster care entry rates vary considerably within the county, making it an illustrative

    choice for study. Table 1 shows foster care entries and child population by year for Alameda

    County.

    Forpe

    rsonaluseonly.

  • 8/12/2019 A Comparison of Foster Care Entry Risk at Three

    4/15

    226 Lery

    Table 1

    Alameda County first entries to foster care, child population, and entry rates per 1,000

    children by year: 20002003

    Total 2000 2001 2002 2003

    First entries to foster care1 3,739 1,005 1,033 860 841

    Child population

  • 8/12/2019 A Comparison of Foster Care Entry Risk at Three

    5/15

    A Comparison of Foster Care Entry Risk at Three Spatial Scales 227

    inflated and are not shown. The purpose of the rates is to establish relative meaning within

    and across geographic scales with respect to variation.

    Independent Variable

    Zip codes, census tracts, and block groups are the geographic units of analysis for compar-

    ative study. Census block groups are nested within tracts, while zip codes are larger than

    tracts but are not part of this hierarchy. Two of 48 zip codes, one of 321 census tracts, and

    12 of 983 block groups are dropped from the analysis because these areas had zero child

    population and therefore no risk of foster care entry. Foster care entries are particularly

    infrequent at the block group scale. Thirty-one percent of block groups had zero entries

    and 41% had between one and five entries over the study period. Figures 1a and 1b show

    quartile maps of the distribution of child population and first entries to foster care across

    the spatial scales. The presence and strength of spatial correlation of the measures across

    neighborhoods at each spatial scale are assessed using ESDA.

    Data Analysis

    First, mapped foster care entry rates are explored at each spatial scale to examine variation

    in risk within and between scales and to look for outliers. Second, the Morans I statistic

    is calculated at each scale based on a neighbor matrix. The matrix assigns spatial weight

    according to neighboring units that share lines and not vertices (referred to as rook cri-

    teria), under the assumption that areas sharing a full boundary line are more alike (more

    spatially dependent) than areas that do not (spatially independent). Griffith (1996) finds that

    misspecification of the spatial weights matrix affects the quality of maximum likelihoodestimators, especially when N is small, and offers five rules of thumb for specifying the

    weights matrix:

    1. It is better to posit some spatial weights matrix than to assume independence among the

    units (extreme underspecification).

    2. It is best to use a partitioning that is in between a square and a hexagonal tessellation.

    3. N should be greater than 30.

    4. Low-order models are preferential to higher-order models.

    5. It is generally better to use a somewhat underspecified than a somewhat overspecified

    spatial weights matrix.

    The choice of rook criteria in this study meets all of these suggestions.

    It is predicted that each spatial scale will have a different Morans coefficient (MC)

    because the distribution of entry rates will vary at different levels of aggregation. Attributes

    of smaller units tend to have greater variances than do larger units because there are fewer

    people to smooth out the mean. For example, the variance of foster care entry risk in a block

    group is likely to exceed the variance in a zip code, which typically has many more people.

    This will affect the value of the MC (Wong, 1996). The presence of significant positive

    spatial autocorrelation at any spatial scale, as measured by the Morans I statistic indicates

    that areas with high rates of foster care entry tend to be clustered together. If the level of

    spatial autocorrelation varies across geographic scales, this will indicate that the pattern of

    risk differs depending on which unit of analysis is chosen to map entry rates.

    Forpe

    rsonaluseonly.

  • 8/12/2019 A Comparison of Foster Care Entry Risk at Three

    6/15

    228 Lery

    Figure 1. Child population at three scales. (a) zip codes. (b) Census tracts. (c) Block groups. Foster

    care entries at three scales. (d) zip codes. (e) Census tracts. (f) Block groups. (Continued)

    Forpe

    rsonaluseonly.

  • 8/12/2019 A Comparison of Foster Care Entry Risk at Three

    7/15

    A Comparison of Foster Care Entry Risk at Three Spatial Scales 229

    Figure 1. (Continued)

    Forpe

    rsonaluseonly.

  • 8/12/2019 A Comparison of Foster Care Entry Risk at Three

    8/15

    230 Lery

    Results

    Figures 1d to 1f show quartile maps of aggregated 2000 through 2003 foster care entry rates

    by zip code, census tract, and census block group for Alameda County. The distribution of

    children across any spatial unit (Figures 1a1c) is not the same as the distribution of foster

    care entry rates across that same unit (Figures 1d1f), indicating that many areas havedisproportionate rates of foster care entry. The relative rarity of foster care entry and low

    population in some areasparticularly at the block group levelcreates unstable rates that

    may be misleading on maps. This issue will be explored using ESDA techniques (Anselin

    and Bao, 1997).

    Exploratory Spatial Data Analysis (ESDA)

    Zip Codes. The quartile map in Figure 1d shows aggregated foster care entry rates between

    2000 and 2003 for Alameda County by zip code and is consistent with other, single-year

    maps of this data. Areas with highest entry rates are in West Oakland, South Berkeley, andHayward. The lowest rates are in the more rural and suburban areas in the southern and

    eastern parts of the county.

    First-order contiguity spatial weights based on rook criteria produced significant global

    spatial autocorrelation (I = .3445, p

  • 8/12/2019 A Comparison of Foster Care Entry Risk at Three

    9/15

    A Comparison of Foster Care Entry Risk at Three Spatial Scales 231

    Figure 2. Univariate Moran scatterplots of foster care entry rates at each spatial scale. (a) Zip codes:

    Morans I =.3445. (b) Census tracts, Morans I =.2446. (c) Block groups, Morans I =.1815.Note:Axis units are in standard deviations.

    Forpe

    rsonaluseonly.

  • 8/12/2019 A Comparison of Foster Care Entry Risk at Three

    10/15

    232 Lery

    weaker at this scale (I = .2446, p < .01), as shown in Table 2 and in the slope of the

    Moran coefficient in Figure 2b. This is to be expected as neighborhoods are disaggre-

    gated to smaller units. The Moran plot in Figure 2b centers the data around zero by sub-

    tracting the mean, so the concentration of points along the y-axis indicates the preva-

    lence of tracts with zero foster care entries. Positive and negative spatial autocorrelation

    are present in the same general geographic areas as when entry rates were mapped in

    zip codes. However, the finer level of granularity among tracts allows for the identifica-

    tion of some different local spatial clusters, as can be seen by comparing Figure 2b to

    Figure 2a.

    Block Groups. The quartile map in Figure 1f shows aggregated foster care entry rates by

    block group. The very fine resolution of this scale breaks up the north-south pattern visible

    in the tract and zip code maps and highlights some new high-risk areas such as Livermore,

    Castro Valley, Newark, and Dublin.

    The Moran plot in Figure 2c shows significant spatial autocorrelation in entry ratesamong block groups. The spatial autocorrelation statistic remains significant at the block

    group level (I =.1815, p < .01), although the difference in the slope coefficient between

    tracts and block groups is smaller than between zip codes and tracts. This is partly because

    in less populated areas tracts and block groups sometimes represent identical geographic

    boundaries. The stacked values along the y-axis show that many of these small areas had

    no foster care entries over the study period.

    The analysis was repeated at each scale with a queen criteria spatial weights matrix

    (units that share a line or a vertex border are considered contiguous spatial neighbors).

    Results were consistent with the rook neighbor specification.

    Discussion

    This study investigates spatial dependence in foster care entry risk at three different spatial

    scales and explores the extent to which risk spills over into adjacent areas depending on

    how neighborhoods are aggregated. Results from ESDA suggest that indeed entry rates

    in each of the three units of analysis used to represent neighborhoods are spatially au-

    tocorrelated. In other words, depending on how neighborhoods are defined in this study,

    the geographic pattern of foster care incidence changes. Zip codes possess the most spa-

    tial dependence in this context, followed by census tracts and census block groups. This

    pattern naturally occurs because, generally, zip codes are the largest geographic units andblock groups are the smallest. Therefore, neighborhood foster care entry risk varies the

    least across zip codes and the most across block groups. This does not imply, however, that

    zip codes best describe discreet, homogeneous neighborhoods as residents would define

    them.

    Zip Codes. Zip codes represent neighborhoods on a crude scale. Exploratory analysis re-

    veals some initial patterns in the way that foster care entry rates are distributed across

    Alameda County. Descriptive maps show clustering, with high rates concentrated in the

    relatively urban areas and low rates grouped in the suburban and rural southern neighbor-

    hoods. This pattern is supported by the significant Morans statistic as a global measure of

    spatial dependence and by the local indicators of spatial association in the Morans plot.

    Forpe

    rsonaluseonly.

  • 8/12/2019 A Comparison of Foster Care Entry Risk at Three

    11/15

    A Comparison of Foster Care Entry Risk at Three Spatial Scales 233

    Census Tracts

    Census tracts are generally smaller geographically than zip codes and when neighborhoods

    are defined in this way, exploratory analysis yields somewhat different spatial patterning

    that is confirmed by a significant Morans statistic. The finer resolution highlights more

    specific high-rate neighborhoods within the broad areas on the zip code maps and onespecific high-rate tract stands out in the east-central part of the county that was hidden

    when neighborhoods were aggregated to zip codes. Variance instability begins to emerge

    at the smaller scales as the base population declines and this can create misleading visual

    patterns of high rates in areas with very few children. In this case, the tract-level maps

    expose at least one important area that is camouflaged in the zip code maps.

    Block Groups

    Census block groups, a subset of census tracts, are more homogenous and have less within-

    unit variability than tracts. When neighborhoods are classified as block groups, the visualpattern of foster care entry across areas is different from the other two scales. Some different

    high-risk neighborhoods newly emerge at this scale, but many neighborhoods have low

    populations and only a few entries, causing the rates to be inflated. This is particularly

    misleading in the geographically large block groups because they are the most striking

    areas on the map. The significant global Morans statistic and the Morans plot support the

    general trend of high entry rates clustering in the urban parts of the county.

    Implications for Practice and Policy

    Zip code representations of neighborhoods are useful when sharing maps with community

    partners because they generally highlight the same highest risk areas as do the smaller

    scale maps. They also are practical for child welfare service planning in rural areas where

    population and resource densities are low and a census tract does not represent enough

    children to produce a stable rate or an area large enough to capture available resources such

    as foster homes or substance abuse treatment programs. Zip code-level data aggregation

    may be the most useful for matching need with service because treatment facilities are likely

    to serve clients from geographic areas larger than census tracts.

    Block groups may be the best proxies for homogeneous neighborhoods but variance

    instability makes them not ideal for estimation. Foster care entry is relatively rare within

    these small areas so that nearly one third of the block groups had zero entries over the 4-year

    study period and most areas had fewer than five entries. Census tracts, being smaller thanzip codes and larger than block groups, possess the strengths and limitations of both of the

    other scales. Therefore, a best or least biased unit of analysis cannot be determined from

    the study; rather, the analysis demonstrates benefits and drawbacks to each spatial scale.

    Furthermore, the process by which spatial relationships and spatial scale are considered

    strengthens the model-building process and offers a set of diagnostic and comparative

    procedures that can be applied broadly to research on neighborhoods and other small areas.

    A few limitations are noted. First, results may not be generalizable to other areas or time

    periods. Other counties may have different demographic distributions and different spatial

    and social configurations of neighborhoods. Second, foster care entry is a relatively rare

    event compared to maltreatment, so the low number of entries in many areas makes it more

    difficult to detect major differences in spatial dependence among the scales. Third, planning

    for neighborhood-based foster care reform should be informed by more developed analyses

    Forpe

    rsonaluseonly.

  • 8/12/2019 A Comparison of Foster Care Entry Risk at Three

    12/15

    234 Lery

    that model the relationship between neighborhood characteristics and foster care entry risk.

    Factors such as poverty and alcohol outlet density have been found to be related to child

    maltreatment and foster care outcomes, and taking such factors into account at multiple

    spatial scales would enrich our understanding of how the ecological environment may

    influence risk for children (Coulton et al., 1995; Freisthler, 2004). Such design additions to

    this study could be conducted in cities such as New York and Chicago, where neighborhoods

    tend to be more discreet and homogenous than in California (and therefore where spatial

    autocorrelation may be lower). This information can be used to guide the development

    of geographically based resources and strategies such as foster parent recruitment or the

    development of substance abuse programs for maltreating families in neighborhoods with

    the highest need.

    For instance, Freisthler (2004) found that the number of bars per 1,000 population

    was positively related to child maltreatment rates in census tracts, adjusting for a very

    high level of spatial autocorrelation (I = .5407) and neighborhood social characteristics.

    When maltreatment reports, substantiations, and foster care entries were aggregated to the

    zip code level and measured over time, Freisthler, Gruenewald, Remer, Lery, and Needell(2007) found a relationship between the concentration of off-premises alcohol outlets such

    as convenience and liquor stores and increased risk for all three measured outcomes. The

    persistent findings across scales suggest that an effective strategy for reducing risk to children

    could be to limit the concentration of bars in neighborhoods.

    As neighborhood-based services in child welfare and other social service fields become

    increasingly popular, the accurate identification of high-risk neighborhoods becomes more

    important. Administrators want to target the right areas, not just those that appear to have

    high rates due to the MAUP. Planners using maps should take into account the spatial

    relationships between areas because neighborhoods do not exist in isolation. As the positive

    spatial autocorrelation in this study shows, phenomena in one neighborhood may, in fact,

    spill over into adjacent areas. Agencies can use neighborhood maps at multiple spatial

    scales to capture this artifact of mapping administrative units as proxies for neighborhoods,

    and to eliminate the problems with viewing only one scale, such as over generalized zip

    codes or variance instability across smaller areas. Maps can then be use to determine which

    neighborhoods are most in need of interventions and then, separately, to consider spatial

    scale in planning services at the scale most efficient for reaching the target population.

    RESUME

    Une comparaison des risques de placements sur trois echelles spatiales.

    Letude aborde la question de loperationnalite des perimetres frontaliers / decoupages

    par quartiers par lanalyse des risques dentree en placement sur trois echelles spatiales.

    Entre 2000 et 2003 (n =3 311), les nouveaux placements ont ete geocodes selon trois

    niveaux (N =46 codes postaux, 320 cas recenses et 983 groupes). Un systeme exploratoire

    danalyse de donnees spatiales a ete utilise pour comparer les autocorrelations spatiales des

    taux de placements sur differentes echelles.

    Les resultats de letude demontrent que la facon dont les quartiers sont definis influence

    la facon dont les placements sont effectues. Cette etude conclue en indiquant des pistes pour

    mieux cibler les services a offrir dans les quartiers a haut risques. Des pistes de recherche

    ulterieures sont aussi suggerees.

    Forpe

    rsonaluseonly.

  • 8/12/2019 A Comparison of Foster Care Entry Risk at Three

    13/15

    A Comparison of Foster Care Entry Risk at Three Spatial Scales 235

    RESUMEN

    Una Comparacion de la Entrada al Sistema Acogida Temporal en Tres Escalas

    Espaciales

    Este estudio aborda el problema de la operacionalizacion de los lmites de vecindarios

    a traves de la investigacion de la entrada al sistema de cuidado adoptivo temporal en tres

    escalas espaciales. Entradas al sistema de acogida temporal de un condado de California

    entre los anos 2000 y 2003 (n =3.311) han sido geo-codificadas para cada una de las tres

    escalas (N =46 codigos postales, 320 subdivisiones de condado1 , 983 grupos de calles2 ).

    El analisis exploratorio de datos espaciales es usado para comparar la autocorrelacion

    espacial de las tasas de entrada entre las distintas escalas. Los resultados sugieren que

    dependiendo de como los vecindarios son definidos, el patron geografico de la incidencia del

    cuidado adoptivo temporal cambia. Se destacan implicaciones para concentrar los servicios

    en vecindarios de alto riesgo y direcciones para futuras investigaciones.

    THE AUTHOR

    Bridgette Leryis a Researcher at the Chapin Hall Cen-

    ter for Children at the University of Chicago. Much of

    her work focuses on how location and neighborhood so-

    cial structure impacts childrens involvement with the

    child welfare system. She is particularly interested in how

    neighborhoods can be better defined and targeted for so-cial service planning. Dr. Lery has an M.S. in Social Work

    from Columbia University and a Ph.D. in Social Welfare

    from the University of California, Berkeley.

    Glossary

    Neighborhood: can be conceptualized as a geographic place where residents share social

    networks and informal ties.

    References

    Albert, V. N., Barth, R. P. (1996). Predicting growth in child abuse and neglect reports in urban,

    suburban, and rural counties. Social Service Review70:5882.

    Anselin, L. (1995). Local indicators of spatial association LISA.Geographical Analysis27:93115.

    1Census tracts en ingles, de acuerdo al US Census Bureau son pequenas subdivisiones estadcas

    relativamente permanentes de un condado.2Block Group en ingles, de acuerdo al US Census Bureau son la unidad geografica constituyentede losCensus Tractsy corresponden a calles o agrupaciones de estas.

    Forpe

    rsonaluseonly.

  • 8/12/2019 A Comparison of Foster Care Entry Risk at Three

    14/15

    236 Lery

    Anselin, L. (1996). The Moran scatterplot as an ESDA tool to assess local instability in spatial

    association. In Spatial analytical perspectives on GIS in environmental and socio-economic

    sciences(pp. 111125). London: Taylor & Francis.

    Anselin, L. (2005).Exploring spatial data with GeoDa: A workbook. Urbana, IL: Center for Spatially

    Integrated Social Science.

    Anselin, L., Bao, S. (1997). Exploratory spatial data analysis linking SpaceStat and ArcView. InM. Fischer, A. Getis (Eds.) (pp. 3559), Recent developments in spatial analysis. Berlin:

    Springer-Verlag.

    Bailey, T. C., Gatrell, A. C. (1995). Interactive spatial data analysis. Essex: Addison Wesley.

    Besinger, B. A., Garland, A. F., Litrownik, A. J., Lankdsverk, J. A. (1999). Caregiver substance abuse

    among maltreated children placed in out-of-home care. Child Welfare78(2):221239.

    Can, A. (1992). Residential quality assessment: Alternative approaches using GIS. The Annals of

    Regional Science 26:97110.

    Chou, Y. H. (1991). Map resolution and spatial autocorrelation. Geographical Analysis 23(3):228

    246.

    Claritas, Inc. (2003).The Claritas demographic update methodology. San Diego: Author.

    Coulton, C. (2005). The place of community in social work practice research: Conceptual and method-ological developments.Social Work Research29(2):7386.

    Coulton, C. J., Cook, T. D., Irwin, M. (2004). Aggregation issues in neighborhood research: A com-

    parison of several levels of census geography and resident defined neighborhoods. Association

    for Public Policy and Management Fall Research Conference, Atlanta, Georgia.

    Coulton, C. J., Korbin, J., Chan, T., Su, M. (2001). Mapping resident perceptions of neighborhood

    boundaries: A Methodological note.Journal of Community Psychology 29:371383.

    Coulton, C. J., Korbin, J. E., Su, M. (1999). Neighborhoods and child maltreatment: A multi-level

    study.Child Abuse and Neglect23(11):10191040.

    Coulton, C. J., Korbin, J. E., Su, M., Chow, J. (1995). Community level factors and child maltreatment

    rates.Child Development66:12621276.

    DAunno, T. (1997). Linking substance-abuse treatment and primary health care. In J. A. Egertson,

    D. M. Fox, A. L. Leshner (Eds.),Treating drug abusers effectively(pp. 311331). Vlalden, MA:Blackwell.

    Deccio, G., Horner, W. C., Wilson, D. (1994). High-risk neighborhoods and high-risk families: Repli-

    cation research related to the human ecology of child maltreatment. Journal of Social Service

    Research18(3/4):123137.

    Drake, B., Pandey, S. (1996). Understanding the relationship between neighborhood poverty and

    specific types of child maltreatment. Child Abuse and Neglect20(11):10031018.

    Environmental Systems Research Institute. (2003). ArcGIS 8.3 (Version 8.3). Redlands, CA: Author.

    Ernst, J. S. (2001). Community-level factors and child maltreatment in a suburban county. Social

    Work Research25(3):133142.

    Freisthler, B. (2004). A spatial analysis of social disorganization, alcohol access, and rates of child

    maltreatment in neighborhoods.Children and Youth Services Review 26(9):803819.Freisthler, B., Gruenewald, P. J., Remer, L. R., Lery, B., Needell, B. (2007) Examining child abuse

    and neglect over space and time: Exploring the spatial dynamics of alcohol outlets and CPS

    referrals, substantiations, and foster care entries. Child Maltreatment12(2):114124.

    Freisthler, B., Lery, B., Gruenewald, P. J., Chow, J. (2006). Methods and challenges of analyzing

    spatial data for social work problems: The case of examining child maltreatment geographically.

    Social Work Research, 30(4):198210.

    Fryer, G. E. J., Miyoshi, T. J. (1995). A clusteranalysis of detected and substantiatedchild maltreatment

    incidents in rural Colorado.Child Abuse and Neglect19(3):363369.

    Griffith, D. A. (1996). Some guidelines for specifying the geographic weights matrix contained in

    spatial statistical models. In S. L. Arlinghaus, D. A. Griffith, W. C. Arlinghaus, W. D. Drake, J.

    D. Nystuen (Eds.),Practical handbook of spatial statistics(pp. 6582). New York: CRC Press.

    Johnston, R., Jones, K., Burgess, S., Propper, C. I., Sarker, R., Bolster, A. (2004). Scale, factor

    analyses, and neighborhood effects.Geographic Analysis36(4):350368.

    Forpe

    rsonaluseonly.

  • 8/12/2019 A Comparison of Foster Care Entry Risk at Three

    15/15

    A Comparison of Foster Care Entry Risk at Three Spatial Scales 237

    MacEachren, A. M. (1995). How maps work: Representation, visualization and design. New York:

    Guilford.

    Moran, P. A. P. (1950). Notes on continuous stochastic phenomena.Biometrika37:1723.

    Newman, K., Small, M. L. (2001). Urban poverty after the truly disadvantaged: The rediscovery of

    the family, the neighborhood, and culture.Annual Review of Sociology 27:2345.

    Openshaw, S. and Taylor, P. J. (1979). A million or so correlation coefficients: Three experimentson the modifiable areal unit problem. In N. Wrigley (Ed.), Statistical methods in the spatial

    sciences. London: Pion.

    Robinson, W. S. (1950). Ecological correlation and the behavior of individuals.American Sociological

    Review15:351357.

    Ryan, J. P., Marsh, J. C., Testa, M. F., Louderman, R. (2006). Integrating substance abuse treatment

    and child welfare services: Findings from the Illinois alcohol and other drug abuse waiver

    demonstration.Social Work Research30:95107.

    Sampson, R. J., Raudenbush, S. W., Earls, F. (1997). Neighborhoods and violent crime: A multilevel

    study of collective efficacy. Science277(5328):918924.

    Spearly, J. L., Lauderdale, M. (1983). Community characteristics and ethnicity in the prediciton of

    child maltreatment rates.Child Abuse and Neglect7:91105.Wong, D. (1996). Aggregation effects in geo-referenced data. In S. L. Arlinghaus (Ed.), Practical

    handbook of spatial statistics. New York: CRC Press.

    Young, G., Gately, T. (1988). Neighborhood impoverishment and child maltreatment: An analysis

    from the ecological perspective.Journal of Family Issues 9:240254.

    Forpe

    rsonaluseonly.