Introduction

Contemporary researchers of adolescent substance abuse recognize that a host of risk and protective factors interact in complex ways to promote or reduce adolescent alcohol and other drug use (e.g. Arthur et al. 2002; Cleveland et al. 2008). Rates of drug use (and other risky behaviors) rise during adolescence, and a greater understanding of the risk and protective factors that influence early using behavior are necessary precursors to the development of effective interventions (Hawkins et al. 2002; Griffin and Botvin 2010). Nargiso et al. (2013), in a recent analysis of youth risk factors for alcohol use, conceptualize risk as occurring within an ecological framework where the domains of peers and family represent proximal influences and neighborhood, community and larger cultural domains represent more distal influences. Within each of these domains, both risk and protection elements co-exist (Bronfenbrenner 1989). While the field has made advancements in understanding risk and protection overall, there are important gaps in the research regarding specialized sub-populations of youth and their unique risk and protection profiles, especially as it relates to substance abuse development.

Youth in foster care represent a particularly highly vulnerable subset of the overall youth population, and recent practice guidelines and public policies have been developed to attempt to address the multiple needs and vulnerabililties of foster youth who are transitioning into adulthood (Foster Care Independence Act 1999; Fostering Connections to Success and Increasing Adoptions Act 2008). Courtney et al. (2011) important research on young adults aging out of foster care has established that poorer outcomes exist for this population in terms of employment, housing stability, high risk behaviors, and criminal justice system involvement. Research has also shown that foster youth have higher rates of substance use disorders when compared to their non-foster care counterparts (McDonald et al. 2012; Narendorf and McMillen 2010).

Despite recent welcome advances in understanding risk, protection and poor outcomes associated with vulnerabilities, limitations remain regarding the interplay of risk and protection. The research presented in this work has attempted to shed light on this interplay for the foster youth population through examination of data available through the Communities That Care Youth Survey.

Literature Review

The current picture of the influence of risk and protective factors for adolescent alcohol and other drug use in the general population (i.e., non-foster care) is complex given variations in study methodology, operationalization of risk and protective factors, and measurement of drug use outcomes. Still, the weight of the evidence from the past several decades points to powerful influences in several key directions.

Risk factors, which can occur at multiple levels (e.g. the individual, family, or community) are generally defined as “those characteristics, variables, or hazard, that if present for a given individual, make it more likely that this individual, rather than someone selected from the general population, will develop a disorder” (Mrazek and Haggerty 1994, p. 127). At the individual level, risks include disruptive and mood disorders, perceptions of risk, age of onset, personality characteristics, impulse control, and neurochemical and other constitutional factors that impact reactivity to alcohol and other drugs. Family risk factors include direct modeling of behavior and attitudes toward drugs as well as the influence of parenting styles and family bonding. Similarly at the school and community level, lack of engagement and consistent relationships can lead to academic failure and disorganized, unsafe communities (Griffin and Botvin 2010; OAS SAMHSA 2009).

Protective factors are no longer viewed as simply the absence of risk or factors that mediate risk, but rather are factors with unique dimensions that contribute independently to the prediction of drug use (Newcomb and Felix-Ortiz 1992; Scheier et al. 1994). Beginning with their seminal review article in Psychological Bulletin, Hawkins et al. (1992) categorized these factors as intrapersonal and environmental variables. Using a prevention science approach, this research organized risk and protective factors into community, family, school, peer, and individual domains. They assert that alcohol and other drug abuse are negative health outcomes which can be prevented by reducing risk factors and enhancing protective factors throughout time, both in individuals and their environments (Hawkins et al. 2002).

Within community based prevention paradigms, debate exists regarding the interplay of risk and protection, and the use of risk reduction versus strengths enhancement approaches. Some scholars have called for strengths oriented individual and community interventions that do not actively work to target risk, while others call for risk and contextually oriented interventions that emphasize non-individual characteristics, such as community violence levels (Pollard et al. 1999). Yet another body of research exists that calls for both strengthening and reducing risk in order to achieve the best results. In a 1999 study, Pollard, Hawkins, & Arthur found that while overall level of risk exposure explains most problematic behavioral outcomes, protective factors do influence negative effects of exposure to risk, but at varying levels depending on the amount of risk.

Turning our attention specifically to substance use and abuse among youth in foster care, we find methodological and practical challenges, also detailed by Thompson and Auslander (2007). First, the population of interest in this research is the current foster care population, and studies in this area have used retrospective approaches that focus on former foster youth (who may have been in care at any age) and who had transitioned out of care at the time of study. Second, comparisons across studies are difficult because the characteristics of the foster care arrangement (kinship versus non-relative, duration of care, number of placements, level of social and educational disruptions) can influence risk in different ways (Courtney and Heuring 2005; Pecora et al. 2003, 2006; Perry 2006). Third, the time period of past use is not uniform across studies—some measure “past 30 day use,” some measure “past year” or “past 6 months,” while others record lifetime use. Further, the drugs of abuse are also not consistently measured. While most measure alcohol and marijuana, other drugs of abuse are not consistent across studies. Finally, the construct of certain terms associated with risk is not uniformly defined. For example, “fighting with peers” can be cited as a risk factor; however, exactly what is meant by this is highly variable—does this mean physical fighting or simple bickering? Another poorly defined construct is that of “association with peers who use substances.” The concept of peer is inconsistently applied—are these children who are the same age and in the same school, or does this refer to a youth’s close friends? The disentangling of many of these constructs may yield greater understanding about the contribution of these influences.

Data released from the 2008 National Survey on Drug Use and Health (Substance Abuse and Mental Health Services Administration [SAMHSA] 2009) indicate that almost 35 % of those ever in foster care used alcohol in the past year compared to 31 % of those never in foster care. Alcohol dependence rates for these same groups were 3.4 and 1.8 %, respectively. Rates of cigarette use and nicotine dependence were also greater among 12–17-year-olds who had ever been in foster care (26.5 %, 14.5 % for use and 4.1 %, 1.9 % for dependence). Further, 29 % of those ever in care reported past year illicit substance use, compared to 19 % of those never in care. Youth who have ever been in foster care also report higher rates of treatment need for substance abuse. Approximately 14 % of youth (ages 12–17) who have ever been in care report a need for treatment for substance abuse compared to 8 % of youth who have never been in care. McDonald et al. (2012) reported that foster youth were more than twice as likely as non-foster youth to have ever used tobacco, marijuana, inhalants, depressants, hallucinogens, and steroids and were more than twice as likely to have used these substances in the past 30 days. In this same study, foster youth were found to be three times more likely to have used methamphetamine and four times more likely to have used heroin in the last 30 days immediately preceding the survey.

Thompson and Auslander (2007) examined risk factors associated with alcohol and marijuana use among 320 15–18 year olds who were currently in foster care. They found that 40 % of the youth reported past 6 month alcohol use, 36 % reported marijuana use, and 25 % reported use of both substances. In their examination of social and individual risk factors for use, 7 of 14 modeled risk factors that were significantly related to risk for alcohol or marijuana use (running away, suspension or expulsion from school, skipping school, fighting with teachers, failing to progress in a grade level, having friends who drink, and having friends who use). Of these factors, skipping school and having friends who use were the most influential.

Recent research by Narendorf and McMillen (2010) found that substance use varied among the older foster youth population (ages 17–19); however, substance use increased during the time of transition out of foster care into independent living. Further, they found that older foster youth had lower rates of substance use overall than the general population, but a higher rate of problematic substance use. Their research finding in the area of increased use during the time of transition into independent living provides important information on the timing of prevention, screening, and intervention efforts and also has implications for the timing of the State’s discharge of these children from foster care. Dworsky and Havlicek (2009) found that, despite federal initiatives which provide financial support and resources for states to extend care, only 14 of 45 states that participated in their study had a provision which would allow youth to remain in care for an extended period of time due to substance abuse or mental health disorders.

In a retrospective study of foster care alumni, White et al. (2008) identified characteristics of past foster care arrangements which were related to current alcohol and drug dependency. Risk factors for alcohol and drug dependence in adulthood included not knowing birth parents, substance abuse by birth parents, and behavior problems as the reason for placement. Protective factors included placement stability, living with family or friends rather than licensed placements, few school changes, educational supports, preparation to leave foster care, and nurturing supports. This research found that characteristics of foster care did impact alcohol and drug dependency rates, and that by making improvements in the foster care experience, youth are more likely to have improved outcomes in the substance abuse domain. Their study (based on a statistical simulation of Northwest Foster Care Alumni) found that the provision of educational services and supports and the presence of a supportive foster family and nurturing supports while in foster care were associated with reduced prevalence of adult substance abuse.

It has not gone unnoticed by these writers that the specific life circumstances that many youth face that precipitate placement into foster care are the same circumstances that predispose individuals in the general population to increased risk for substance abuse: physical, emotional or sexual abuse, parental substance abuse, family chaos and disruption, conduct disorders, and educational and social disruption (Barth 1990; Courtney and Heuring 2005; Elze et al. 2001; Thompson and Auslander 2007; Simms et al. 2000; Young et al. 1998).

Present Study

Despite known vulnerabilities of youth in foster care, there have been few studies of the risk and protective factors related to alcohol and other drug use specific to this population. As was outlined in the literature, the field at large has gained information about risk and protection, constellations of risk that surround the development of substance abuse, and the poorer outcomes experienced by foster youth across a broad spectrum of life domains.

There is a large theoretical and empirical body of work that has been conducted around the Communities That Care program over several decades, and the measurement and structural approach we use in this study will fill an important gap in the research through it’s application with the foster care population. CTC is a coalition-based prevention operating system that uses a public health approach to prevent youth problem behaviors such as violence, delinquency, school dropout, and substance abuse (Hawkins et al. 1992). As such, it is much more than just a student survey. However, the survey is an important tool of the program and its conceptualization and measurement of risk and protective factors is fairly germane to the assessment of community needs and the adoption of program intervention strategies. Differentiation of risks and protective factors is critical to this process and is part of a larger attempt in various helping fields to move away from a medical illness model including strengths-based approaches in social work (Saleebey 2008) and positive psychology (Seligman and Csikszentmihalyi 2000). Our study is not the first to explore the ability of the CTC items and scales to differentiate between risk and protective constructs (Arthur et al. 2002; Feinberg et al. 2007). If the differentiation of risks/protective factors or strengths/deficiencies is to serve fully in assessment and intervention processes, we need to consider the identification of separate constructs, not simply opposite ends of a single continuum (risk and protection).

The main aim of the current study is to evaluate the relationships between individual risk and protective factors and drug use among youth currently living in a foster care arrangement and to explore the role of risk and protection. Specifically, this work seeks to evaluate a range of risk and protective factors, and assess these constructs in order to provide more information about the needs of this specific population. Using a prevention science approach, understanding risk factors, protective factors, and the relationship between the two serves as the foundation for intervention thinking and planning.

Methods

The Communities That Care Youth Survey (CTCYS)

This study utilized the CTCYS as the basis for examination of the relationship between risk and protective factors for youth currently in foster care. The CTCYS is an ongoing cross-sectional survey of perceptions and behaviors of students in grades 6 through 12, and is administered in school settings across the nation. In brief, schools elect to participate on a voluntary basis and the survey administration is completed during one classroom period by the teacher. Participation is voluntary and anonymous and teachers are instructed to remain at the front of the room during survey administration. Parents receive a letter at least 2 weeks prior to survey administration informing them of the survey and offering an opportunity to decline their child’s participation or sign and return an attached release form. All surveys in a given school are completed on the same day and same class period and all surveys in a given state are completed within a two-month period. At the end of the class period students place their survey in an envelope which is sealed by the last student.

A total of 23 states and 837 zip-codes are represented in the CTCYS normative database. The survey is designed to assess levels of risk and protection within the student’s peer group, family, school, and community. The survey includes questions about alcohol, tobacco, drugs (ATODs), and antisocial behaviors such as carrying guns to school or selling illegal drugs. Respondents’ personal information includes age, gender, race-ethnicity, and residential location (urban/non-urban). Validation studies have established the reliability and validity of the survey’s risk and protective factor scales across gender, racial/ethnic, and age groups (Arthur et al. 2002; Glaser et al. 2005), and the utility of the scales in predicting a community’s levels of ATOD prevalence (Hawkins et al. 2004).

Sample

The present study is an analysis of publically available data on the CTCYS. The data were collected on 310,171 students in grades 6-12 between January 1, 2000 and December 31, 2002. Of the original 310,171 respondents, 2,366 students reported living with at least one foster parent and were given further consideration for inclusion in the analyses. The public use database includes some records that have been flagged during data cleaning and validation as likely to be of poor quality. Data quality criteria include checks for truthfulness (judged by response about use of a fictitious substance or reporting of an implausibly high rate of ATOD use and antisocial behaviors), inconsistent responses (more than one inconsistency in ATOD use items or antisocial behaviors), and missing data (more than 25 percent of the items left blank). For purposes of this research, we used the original validation procedures to eliminate all cases flagged as poor quality. A total of 771 students in this foster care subsample were considered invalid cases according to these validity check procedures. These students were excluded from the analyses, leaving a final sample of 1,595 students currently living in foster care. Sample characteristics are provided in Table 1 which also provides a comparison of the foster youth subsample and non-foster youth in the remaining sample. This comparison speaks to the unique characteristics of the foster youth and the inability to generalize from this study’s findings to the larger population.

Table 1 Foster care sample characteristics
Full size table

Measures

The majority of items in the survey constitute subscales of risk and protective factors. There are four subscales each for risk and protective factors that relate to different youth contexts: peers, community, school, and family. The largest scale is the Peers Risk Factors subscale which is made up of 45 single survey items. The smallest scale is the Peers Protective Factors subscale which is made up of only 9 survey items. The drug factor is measured by three scales. Alcohol Use has 3 subscale indicators that ask the “estimated number of occasions the respondent has used alcohol in the past two weeks, 30 days and lifetime. The Tobacco Use scale is similarly measured by three subscales. Poly Drug use is a scale composed of seven items which ask the estimated number of uses in their lifetime for the seven drugs (inhalants, hallucinogens, methamphetamines, cocaine, heroin, depressants and marijuana or hashish). For a full description of the items and subscales of the CTCYS, see https://www.pmrts.samhsa.gov/pmrts/CommunitiesCares.aspx.

Due to the fact that the survey items are on vastly different metrics, we rescaled the variables using the percent of maximum score (POMS) procedure. In this procedure, each value for a variable is divided by the maximum value on that variable. Thus, all scores are on a continuous proportion metric. Some variables were reverse-coded prior to POMS rescaling based on guidelines set forth in the public use code book guidelines (SAMHSA 2007).

Statistical Methods

Structural equation modeling (SEM) was used to analyze the data (Bollen 1989; Kline 2010). Generally speaking, SEM is an extension of multiple regression (MR) analysis that incorporates the psychometric principles of classical test theory. SEM allows the analyst to examine relationships among study variables (referred to as latent variables or factors) for which measurement error has been removed. MR assumes variables have been measured without error—an assumption that is often violated in practice and can lead to biased results (Brown 2006). Additionally, MR only allows the prediction of a single dependent variable whereas SEM permits several explanatory pathways between variables to be estimated simultaneously. In short, SEM is a powerful technique and an important advancement in social science methodology (Kline 2010).

SEM begins with confirmatory factor analysis (CFA),also known as the measurement model (see Fig. 1). CFA allows researchers to examine the measurement properties of the outcome variables. A model is evaluated according to how well it fits the data and how particular indicators load onto the latent variables. Hypothesized relationships between latent variables are represented as correlations (double-headed curved lines in Fig. 1). Indices of overall fit used in the present study include the Comparative Fit Index (CFI; Bentler 1990), the Tucker-Lewis Index (TLI; Tucker and Lewis 1973), the Root Mean Squared Error of Approximation (RMSEA; Steiger and Lind 1980), and the Standardized Root Mean Squared Residual (SRMR; Bentler 1995). Commonly used guidelines indicating acceptable data-to-model fit for these indices are as follows: (a) CFI and TLI greater than .90 (>.95 = close fit); (b) RMSEA less than .08 (<.05 = close fit, <.08 = fair fit, <.10 = mediocre fit); and (c) SRMR less than .08 (Browne and Cudeck 1993; Hu and Bentler 1999; MacCallum et al. 1996). In addition to these indices, parameter estimates, specifically factor loadings, should be sensible in size (usually absolute value of .32 or larger) (Tabachnick and Fidell 2012) and direction and gross misspecification as indicated by large residuals or modification indices must not be present.

Fig. 1
figure 1

CFA Model. Model fit: \(\chi^{2}\) (36, n = 1595) = 359.496; RMSEA = 0.075(.068–.082); CFI = .954; TLI = .929

Full size image

If the measurement model is deemed acceptable, then a structural model can be estimated. The difference between the measurement and structural models is simply the addition of regression paths between latent variables that represent predictive relationships (indicated in path diagrams, such as Fig. 2, by straight lines with a single arrowhead). Together, these models allow researchers to adequately measure phenomena and estimate relationships among them. Any measurement error in the individual survey questions are removed leaving only shared variance among indicators to represent the latent variables.

Fig. 2
figure 2

Standardized regression estimates for the orthogonalized risk factor. Model fit: \(\chi^{2}\) (37, n = 1595) = 264.391; RMSEA = 0.062(.055–.069); CFI = .944; TLI = .917

Full size image

Parceling was used to reduce the large number of variables in the dataset. Parceling involves the aggregation of observed variables, by taking the average of a group of items and using that average as the parcel score—which was the method used in this study. Parceling allows for a more parsimonious model and results in better model fit, in addition to other psychometric advantages (Little et al. 2002). CTCYS items were parceled into three indicators for the Drug Use latent variable and four indicators each for the Risk Factors and Protective Factors latent variables. Items were chosen for the parcels according to the subscale structure of the CTCYS. For example, all items included in the community risk subscale of the CTCYS were aggregated into a single parceled indicator.

The SEM program Mplus, version 6.11 (Muthén and Muthén 1998–2010) was used to estimate the three-factor CFA and all subsequent SEM models. A maximum likelihood estimator was used for parameter estimation. To set the scale for the unobserved latent variables, the fixed-factor method was used (Brown 2006). Specifically, each of the latent variables’ variances were fixed to 1 and means fixed to 0. Thus, the latent variables were on a standardized or z-score metric. As such, factor loadings for all indicators were freely estimated. Also, several unique variance terms (i.e., variance in observed variables not explained by the latent variables) were allowed to correlate. Specifically, the residual variances of items for the same domain (e.g., community) were allowed to correlate because indicator variance not due to risk or protective factors was assumed to be related to the risk and protective factor domains.

Missing Data

The resulting data set had approximately 15 % missing. Missing data were multiply imputed to allow for unbiased parameter estimation. Multiple Imputation is one of the two state-of-the-art procedures for handling missing data (Enders 2010). The Multiple Imputation procedure in SAS version 9.2 (SAS Institute 2008) was used to impute the data 100 times. Next, a “super matrix approach” was used to calculate sufficient statistics (means, standard deviations, and correlations)on the 100 data sets which were stacked on top of each other (Wu et al. 2009). These sufficient statistics were then read into M plus for model estimation.

Results

A three-factor CFA model was estimated to validate the survey measurement structure (Fig. 1). The three factors were defined as follows. Community, family, school, and peer risk factor subscales loaded onto the latent variable Risk Factors. Likewise, community, family, school, and peer protective factor subscales loaded onto the latent variable Protective Factors. Finally, the alcohol use, tobacco use, and multiple drug use items loaded onto the latent variable Drug Use. Model fit ranged from acceptable to fair fit with an RMSEA of .08 (90 % confidence interval: .07–.08), a SRMR of .05, a TLI of .93, and a CFI of .95. The Chi square test of model fit had a value of 359.50 with 36 degrees of freedom (p < .01). Overall, the measurement model appears to perform well.

The correlation between Drug Use and Protect Factors was −.70, a medium to strong correlation. As expected, this suggests drug use increases as the number or quality of protective factors decreases. A strong, positive correlation (.90) was observed between Drug Use and Risk Factors, implying greater drug use among youth with greater risk factors. The correlation between Risk Factors and Protective Factors was −.92. This is a very strong relationship and suggests that Risk Factors and Protective Factors largely define the opposite ends of a uni-dimensional construct. However, there may still be unique contributions from each factor as they relate to Drug Use, a point that will be addressed later in this section.

Parameter estimates for the CFA model are reported in Table 2. Standardized factor loadings ranged between .34 and .81 and all were significantly different from zero. All standardized factor loadings from the CFA model were greater than .5 with the exception of the community and family indicators that load onto Protective Factors (.34 and .44, respectively). It is interesting to note that the peers indicator had the highest standardized loading for both Risk Factors and Protective Factors. Recall that a standardized factor loading is simply a standardized regression coefficient that quantifies the amount of standard deviation change in a manifest variable given a one standard deviation change in the underlying latent variable. Larger loadings imply greater predictability and, thus, the meaning of a latent variable is defined to a greater extent by indicators that are strongly predicted by it. For the present study, it appears that youth in foster care consider characteristics of their peer climate as the strongest representatives of risk and protective factors. Also noteworthy is that the most representative indicator of the Drug Use factor was alcohol, which had a standardized loading of .79.

Table 2 Parameter estimates for the CFA Model
Full size table

The same measurement structure for the latent variables was used in the structural model. In this model, Risk Factors and Protective Factors predicted Drug Use. Furthermore, Risk Factors and Protective Factors remained correlated. The results from this analysis yielded unacceptable parameter estimates. Specifically, the standardized regression coefficient for the regression of Risk Factors on Drug Use was greater than one. Additionally, the strong positive regression coefficient for Protective Factors predicting Drug Use (β = .91) was unexpected given a negative correlation (r = −.70) between Protective Factors and Drug Use. These findings above are symptoms of net suppression (Cohen et al. 2003), which is a form of collinearity. Given that the correlation between Risk Factors and Protective Factors was −.92, the independent contribution of each factor was unclear. Therefore, an orthogonalizing approach (Little et al. 2006) was employed to remedy the situation. This technique uses multiple regression procedures on the indicators in order to remove all collinearity between the indicators of the two factors. The approach requires two steps. First, the indicators on one factor are individually regressed onto all indicators of another factor. The residuals of these regression equations are then used in the model in place of the original indicators. The residuals contain no shared variance among indicators on the other factor, making the information represented by this “residual” factor unique. Consequently, the correlation between the two factors becomes zero.

As such, two models were needed to assess the unique predictive power ofRisk Factors and ProtectiveFactors. In the first model, the original RiskFactor indicators were replaced with the residuals that resulted from the regressions of each RiskFactor indicator onto all Protective Factor indicators, while the Protective Factors indicators remained the same. Given the procedure described above, the correlation between the two latent variables was fixed to zero. All other estimates were specified as in the previous structural model (Fig. 2). The overall fit for this model was acceptable. The model had a Chi square value of 264.40 with 37 degrees of freedom (p < .01). Other fit indices suggested acceptable model fit. The RMSEA was .06 (90 % confidence interval: .06, .07), the SRMR was .04, the CFI was .94, and the TLI was .92. The new standardized beta estimates from both Protective and RiskFactors significantly predicted Drug Use, with estimates of −.68 and .70 respectively.

The second orthogonalized model (Fig. 3) was similar to the first with the exception that the Protective Factors indicators now consisted of the regression residuals from each Protective Factors indicator being predicted by the entire set of Risk Factors indicators. The Risk Factors indicators were those used in the original structural model. The overall fit for this model was also acceptable. The Chi square value was 120.52 with 37 degrees of freedom (p < .01). Other fit indices suggested close fit, with an observed RMSEA of .04 (90 % confidence interval: .03, .05), an observed SRMR of .03, an observed CFI of .98, and an observed TLI of .97. The model and standardized estimates for the structural path are shown in Fig. 3. The new standardized regression estimate for the path between Risk Factors and Drug Use was .90, whereas the new standardized regression estimate for the path between Protective Factors and Drug Use was .11. It is important to note that both betas significantly predicted drug use, as depicted in Fig. 3. In comparing the standardized regression coefficients for the orthogonalized factors from each model—that is, the factors created from the uncorrelated residuals—the results suggest that the Risk Factors variable was a stronger unique predictor of Drug Use (β = .70) than was the Protective Factors variable (β = .11).

Fig. 3
figure 3

Standardized regression estimates for the orthogonalized protective factor. Model fit: \(\chi^{2}\) (37, n = 1,595) = 120.524; RMSEA = 0.038(.030–.045); CFI = .982; TLI = .973

Full size image

Discussion

The high correlation observed between the risk and protective constructs suggests a single dimension for this sample and for the scales as currently constructed; however, the orthogonalizing approach used here also provides support for considering them separately. Our results suggest that risk factors play a greater role in predicting drug use than do protective factors among foster youth. Our findings indicate that both risk and protective factors are important, but risk has increased importance with this population. This is an important distinction in the understanding of the interactive role that risk and protection may play in the development of foster youth substance abuse. These findings are consistent with those in the general population presented by Pollard et al. (1999). Although either decreasing risk factors or increasing protective factors will mitigate drug use in foster care youth, our findings suggest that the development and implementation of targeted risk reduction strategies for this population should take on increased importance and may result in a larger payoff not only in the reduction of substance abuse but also to reduce the higher rates of school failure, unemployment, legal system involvement, and homelessness experienced by this group of adolescents (McDonald et al. 2012).

This single construct finding may be partially created and understood by looking at the contributions of the four domains for each. Our analyses indicate that foster care youth define risk and protective factors largely through peer relationship characteristics. In this study, youth in foster care considered characteristics of their peer climate as the strongest agents of risk and protection. This finding is noteworthy but is not surprising given the age group under study, and the fact that the population of study likely has family relationships that have been disrupted due to abuse or neglect and foster care placement, thus the role of peers may taken on greater than expected importance for these youth compared to non-foster youth. Other research lends credence to this finding. Shook et al. (2009) found that different levels of peer affiliations are associated with externalizing (negative) behaviors. In Samuels and Pryce’s (2008) study of foster care affiliated youth, they note that relational skills are often challenging for these youth who have experienced a developmental trajectory that is characterized by losses and disconnection. They present research that suggests that while a risk and protective factors framework is helpful, it is also important to consider practices aimed at the creation of a holistic life course framework that allows the youth to form inter dependence in a broader social context that is less reliant on any one factor (risk or protective in nature).

Interestingly, a significant portion of the early understanding of the Social Development Model and the risk and protective factors for positive youth development promote parent training and family management as a primary vehicle for impacting undesirable outcomes for youth. With this particular population, family may represent increased risk instead of vehicles for protection. In this study, risk and protection were highly correlated, and our findings suggest that these concepts may represent a spectrum (or continuum) of factors rather than discrete constructs.

In terms of policy/practice implications of our study, these findings support the promotion of multi-dimensional prevention and intervention planning that incorporates protection, but also vigorously promotes risk reduction with increased priority. Practitioners are also alerted to our findings regarding the importance given to the role of peers for youth in foster care. Samuels and Pryce’s (2008) work also provided information about the need for strong peer and supportive alliances among foster youth, and cited it’s importance in helping them transition successfully to adulthood.

Though we believe this study extends current knowledge about risk and protection by empirically exploring the relationship between aspects of each, more research is needed to understand the influence of risk and it’s differential impact on behavior, and our findings should be taken in context of limitations. The broad conceptualization of peer influence, as is used here, should be noted as a limitation particularly because the data are cross sectional in nature and youth may transition between beliefs about the role of family, friends, and the importance of parent/child relationship. The risk and protective factors analyzed here are also limited to those measures in the CTC survey and replication with a broader range of factors would provide a more robust understanding, and thus is an area for future research. Finally, our study did not take into account youth gender, which may be influential in the understanding of risk and protection (Nargiso et al. 2013).