A cluster-analytic study of substance problems and mental health among street youths

Based on a cluster analysis of 211 street youths aged 13–24 years interviewed in 1992 in Toronto, Ontario, Canada, we describe the configuration of mental health and substance use outcomes. Eight clusters were suggested: Entrepreneurs (n = 19) were frequently involved in delinquent activity and were highly entrenched in the street lifestyle; Drifters (n = 35) had infrequent social contact, displayed lower than average family dysfunction, and were not highly entrenched in the street lifestyle; Partiers (n = 40) were distinguished by their recreational motivation for alcohol and drug use and their below average entrenchment in the street lifestyle; Retreatists (n = 32) were distinguished by their high coping motivation for substance use; Fringers (n = 48) were involved marginally in the street lifestyle and showed lower than average family dysfunction; Transcenders (n = 21), despite above average physical and sexual abuse, reported below average mental health or substance use problems; Vulnerables (n = 12) were characterized by high family dysfunction (including physical and sexual abuse), elevated mental health outcomes, and use of alcohol and other drugs motivated by coping and escapism; Sex Workers (n = 4) were highly entrenched in the street lifestyle and reported frequent commercial sexual work, above average sexual abuse, and extensive use of crack cocaine. The results showed that distress, self-esteem, psychotic thoughts, attempted suicide, alcohol problems, drug problems, dual substance problems, and dual disorders varied significantly among the eight clusters. Overall, the findings suggest the need for differential programming. The data showed that risk factors, mental health, and substance use outcomes vary among this population. Also, for some the web of mental health and substance use problems is inseparable.

Keywords: Drug problems; Mental health; Street youths

It is exceedingly evident that such vulnerable populations as dislocated adolescents (e.g., the homeless and street youths) experience a constellation of mental health problems, of which substance misuse is but one [1], [2], [3]. High rates of poor health among such youths should not be surprising. Arguably, adolescents experiencing early social dislocation may be more vulnerable to impaired psychological health than are those who become dislocated later in the life course. Indeed, dislocated adolescents would lack the accumulation of human and social capital—the personal and social resources necessary to buffer the negative effects of street life [4]. Unfortunately, because longitudinal studies are rare, research has been unable to disentangle clearly whether preexisting psychological or environmental conditions traject youths to the street, whether street involvement causes psychological impairment, or whether both causal processes are at play.

Correlational studies have documented clearly the high-risk profile of street youths. Between 30% and 40% of street youths report depressed mood [5], [6], [7], a rate 2.8 times that of nonhomeless adolescents [8], [9]. As well, studies of street youths indicate a high prevalence of poor psychological health, including paranoid ideation and conduct disorder [1], [10], [11]. Perhaps no behavioral manifestation is a better indicator of acute psychological distress than is attempted suicide. The percentage of street youths attempting suicide ranges from 26% among U.S. shelter users [12] to 47% among homeless prostitutes in Los Angeles, California [13], with numerous studies finding rates of attempts between 30% and 39% [8], [12], [14], [15] and between 40% and 46% [6], [11], [16], [17].

Another dimension of mental health is substance-related problems. Although elevated rates of drug taking are well documented among street youths [8], [18], [19], [20], [21], [22], [23], few studies have assessed alcohol and drug problems. One exception is a study by Robertson, Koegel, and Ferguson [24], who found that 48% of 93 homeless adolescents in Hollywood, California, met the DSM-III criteria for either lifetime alcohol abuse or dependence. The majority (24%) met the criteria for alcohol abuse only, and 3% met the criteria for dependence only (22% met the criteria for both abuse and dependence). Moreover, 26% of their sample met the criteria for both alcohol and drug abuse.

Despite their elevated risk profile, street youth studies reveal diversity in characteristics and lifestyles [2], [21], [25]. Identification and knowledge of these differences can have important policy implications for prevention and intervention programming since programs can be targeted more effectively to the varying needs of different youths. Moreover, the identification and understanding of group differences may illuminate causal processes. Unfortunately, most street youth typologies are based on limited dimensions and offer a small number of simplistic types, such as "runaways" and "throwaways." Yet, whether such classifications are useful scientifically or programmatically is doubtful given the underlying heterogeneity of street youths. A few classifications, however, have explicitly attempted to classify the complexity of phenomena regarding dislocated adolescents with their adoption of multiple typological groups [25], [26], [27], and more recent work has attempted to investigate classifications among homeless adults [28].

As Bailey [29] notes, there are several advantages of typological and taxonomic research, including (a) the ability to describe complex patterns in data, (b) the identification of similarities and differences, (c) the ability to reduce the complexity of data, and (d) the potential to study relationships and explanation. Indeed, typologies can provide clinicians with diagnostic indicators, which in turn can be applied to manage interventions and treatment [27] and can help to identify the distinguishing risk and protective characteristics of differing groups.

Although the research literature demonstrates overwhelmingly that street youths experience an array of mental health problems compared to their mainstream counterparts, few studies have attempted to describe the complex variations in these outcomes among street youths. The purpose of our paper is two-fold: First, based on a cluster analysis of 211 street youths, we describe the configuration of various mental health outcomes—specifically distress, self-esteem, psychotic thoughts, attempted suicide, and substance problems; and second, we identify various factors implicated in these mental health configurations.

Our data are based on face-to-face interviews in Toronto, Ontario, Canada, with 217 street youths aged 13 to 24 years. Interviews were conducted between February and May 1992 (152 interviews were derived from social service agencies and 65 from street and word-of-mouth contacts). We chose to capture a broad spectrum of the population by setting the following criteria for participation in the study. First, based on the local youth services age criterion, we interviewed only youths aged 24 years or younger. In addition to the age criterion, we established conditions to ensure a sample of dislocated youths. Youths must have used at least one social service facility directed toward street youths in their lifetime or must have met three of four conditions: (a) left school before completing high school, (b) lived away from their family (or guardian) for at least 2 days during the past year, (c) run away or been thrown out of their home at least once, and (d) been homeless (i.e., without a place to stay) at least once in their lifetime. All participants easily satisfied these criteria (facility usage 98%; left school before grade 12, 90%; away from family 2 or more days, 99%; ran away from home, 93%; and been homeless, 93%).

We employed a two-stage process to select agency-derived youths. With guidance from local service providers, we constructed a sampling frame of 45 agencies serving the downtown core. This first stage of selection was based on information derived from a 1990 telephone survey of social service agencies [30], which allowed us to sample 11 agencies proportional to size based on the estimated number of cases seen per day during the past 3 months [31], [32]. The 11 agencies sampled in 1992 were the same as those selected from the earlier 1990 study. Although the decision to select 11 agencies was purely arbitrary, it did ensure variation in services and youth characteristics and far exceeded those typically sampled in other studies. We defined youth agencyas any facility providing social and other services to youths within the study's geographical area. Services included both residential (e.g., shelters and hostels) and nonresidential (e.g., drop-ins, outreach, education or referral, and counseling) programs. The selected agencies were contacted by the Addiction Research Foundation (ARF) and were asked for permission to interview their clients. No agency refused to cooperate in the study.

Partly due to administrative difficulties and agency concerns, a random selection of youths within selected agencies was rarely feasible. Although the number interviewed per agency ranged from 2 to 30, in 9 of 11 agencies, 9 or more youths were interviewed. All participants volunteered for the study based on advertisements posted at the selected agencies. Interviews were conducted at the agency, at an ARF office, or wherever the youths felt most comfortable. In total, we interviewed 152 agency-derived youths (70% of the total sample) between February and May 1992.

We recruited youths for the street sample through three primary sources: (a) street outreach workers and mobile vans, (b) direct street contacts, and (c) word of mouth. On our request, mobile van and street outreach workers distributed cards describing the study and payment, and a telephone number was provided to arrange an interview; the interviews were conducted in coffee shops, restaurants, at ARF offices, or any place chosen by the youth. To derive a snowball or word-of-mouth sample, we provided additional cards to participants for them to distribute among their friends. We also recruited youths directly by approaching them in known congregating areas. In total, we interviewed 65 youths (30% of the total sample): 7 from mobile vans, 12 from word of mouth and outreach workers, and 46 from direct interviewer contacts. The street-sampling strategy was relatively successful in interviewing those who infrequently used social services. For example, street-derived youths were significantly less likely than agency-derived youths to spend 3 or more of the past 7 days resting in a shelter or hostel (23% vs. 51%; p <. 001) and to report currently living without shelter (20% vs. 10%; p =. 070). However, none of the mental health or substance use outcomes, and the resulting cluster membership, differed significantly between street and agency samples.

For both agency and street samples, face-to-face interviews were completed during February through May 1992, with the majority intentionally completed during the cold weather period of February and March. Interviewing during this period should have maximized the size of the service-using population and therefore increased our coverage of this group. Semistructured interviews averaged 75 minutes in length, and on completion participants were paid $15. Prior to formal interviewing, the questionnaire was pretested among youths and evaluated by several youth work professionals. The full questionnaire and a detailed discussion of our experiences in interviewing street youths are available [33].

Characteristics of our sample were diverse. Youths ranged in age between 13 and 24 years (M = 19.9). About three-quarters (74%) were male and four-fifths (82%) reported that their sexual orientation was exclusively heterosexual. Most (92%) left school before completing high school, and 9% left before completing grade 9. Although two-thirds (65%) did not spend a night on the street during the 7 days before the interview, about 19% reported spending 3 or more nights on the street. During the 7 days before the interview, 47% did not use a shelter, while 28% spent the past 7 nights in a shelter. A majority of the sample (61%) were not raised primarily by their biological parents. Although most had lived with one or more of their natural parents at some point (86%), a sizeable percentage also lived in other settings, including foster homes (38%), group homes (38%), detention centers (48%), or with relatives other than their parents (55%). Among the 187 youths who left home, about half departed four or more times. First departures from the family occurred early: 28% left their parents at 10 years of age or less, 48% left between the ages 11 and 15, and 24% left after the age of 16. Not surprisingly, a sizeable percentage had infrequent contact with their families. About 36% had no contact with mothers, and 57% had no contact with fathers during the 3 months before the interview. About two-thirds (67%) reported having been physically abused by a person living with them, and one-fifth (21%) said that they had been sexually abused by someone living with them.

In total, our cluster analysis employed 17 variables, 8 derived from an exploratory factor analysis and an additional 9 variables meant to capture other aspects of street life. For the factor analysis, we selected 31 variables identified by the research literature as being representative of the underlying dimensions of street youth experiences. The first broad dimension identified by the literature relates to the distal influences of family relations [4], [12], [18], [21], [26], variables especially prominent in the youth's initial exit from the family. We included variables that reflected the nature of family displacement and dysfunction (e.g., being forced to leave home, importance of physical or sexual abuse in departure, parental contact, familial substance problems). The second category of variables identified in the literature relates to the proximal characteristics of street lifestyle [4]. We included two forms of street acculturation: entrenchment (e.g., social service usage, street living, street identity, and street evaluation) and delinquent behavior (e.g., engaging in prostitution, theft). In analyzing these 31 indicators, we employed principal axis factoring using maximum likelihood estimation with oblique rotation. An advantage of the maximum likelihood solution is that it provides a chi-square evaluation of the number of factors [34]. The 8-factor solution provided a good fit to the correlations of the 31 variables (χ2 = 246.69; 245 df; p =. 458). Incremental χ2 tests indicated that an 8-factor solution provided a significantly better fit than did a 7-factor solution (Δ χ2 = 295.22; 269 df; p =. 130), whereas a 9-factor model did not significantly improve the fit (Δ χ2 = 215.02; 222 df, p =. 619). Although several of the factors contained only 2 items, the factors made substantive sense, and all items had loadings exceeding. 30, a criterion considered to indicate nontrivial factors [35]. The factors extracted are described in Table 1.

Table 1. Variables Employed in the Cluster Analysis

To generate clusters, we used 17 variables in the cluster analysis, the 8 factors just described and 9 binary variables meant to classify youths further into important typological categories relevant to our analysis. The first 2 relate to experiencing physical and sexual abuse, while the remaining 7 serve to classify youths according to substance use problems (see Table 1).

We employed the resulting cluster taxonomy to examine variation in mental health outcomes. This serves two important goals. First, it allows an examination of substance use and mental health comorbidity. Second, meaningful mental health differences according to type serve to strengthen the validity of the cluster solution. We assessed eight measures of psychological health: distress, self-esteem, psychotic cognitions, suicide attempts, alcohol problems, drug problems, dual substance problems, and dual disorders. Our measure of psychological distress was based on six items from the Center for Epidemiological Studies Depression Scale (CES-D) [36], a frequently used measure of psychological distress among adults and adolescents [37], [38] and homeless populations [39], [40]. The six items indicate how frequently the following were experienced during the 3 months before the survey: (a) felt sad, (b) felt lonely, (c) felt depressed, (d) felt like crying, (e) felt unhappy about the way your life is going, and (f) felt discouraged and worried about your future. Responses were never or rarely, sometimes, often, and always. Summing the six items into an overall distress measure resulted in values from 0 to 18 (M = 7.7; SD = 4.0; α =. 80).

To measure self-esteem, we used five items derived from Rosenberg's work [41]. The true-false items were based on the following statements:

• I feel good about myself.
• I feel that I'm a person of worth.
• I am able to do most things as well as other people can.
• All in all, I feel that I am a failure.
• On the whole, I am satisfied with myself.

The summated five-item scale resulted in values from 0 to 5 (M = 3.9; SD = 1.3; α =. 70).

We employed four items derived from the Psychiatric Epidemiology Research Interview (PERI) to evaluate the extent of psychotic perceptions[42]:

• Have you ever heard noises or voices that other people say they can't hear?
• Have you ever had visions or ever seen things that other people say they can't see?
• Have you ever felt that you had special powers that other people don't have?
• Have you ever felt that your mind was taken over by forces you couldn't control?

Responses were never, almost never, sometimes, fairly often, and very often. These four items capture aspects of auditory hallucinations (item 1), visual hallucinations (item 2), and ideas of reference (items 3 and 4) and refer to experiences occurring during the 12 months before the interview, excluding experiences under the influence of alcohol or other drugs. This measure ranged in values from 0 to 15 (M = 2.8; SD = 3.2; α =. 69).

Attempted suicide indicated the percentage of youths reporting at least one attempt during their lifetime (43% of the sample).

To measure alcohol problems we used the following seven items:

• Have you felt you should drink less?
• Have others bothered you by complaining about your drinking?
• Have you felt bad or guilty because of your drinking?
• Have you drunk in the early morning or drank to get rid of a hangover?
• Have you thought you had a problem because of your drinking?
• Have you ever had any medical problem as a result of your drinking?
• Have you been in the hospital because of your drinking?

All questions required yes or no responses, and all referred to the 12 months before the interview. The first four items were derived from the CAGE scale, a self-report screening test designed to identify problem drinkers [43], and the remaining items were derived from the Ontario Student Drug Use Survey [44]. We also computed an overall alcohol problem measure by summing affirmative responses to the seven items (M = 2.26; SD = 2.06; α =. 78), and we also present the percentage reporting two or more of the CAGE items, a cutoff suggestive of problem drinking [43].

To measure drug problems, we used the following eight items:

• Do you ever feel concerned about your drug use?
• Are you always able to stop using drugs when you want ? (Item was reverse coded.)
• Have you been arrested or warned by police because of your drug use?
• Have you ever had "blackouts" or "flashbacks" due to your drug use?
• Have you had any medical problems as a result of your drug use?
• Do you wish you could use fewer drugs than you do now?
• Have you gone to anyone for help for a drug problem?
• Have you ever seen a doctor or been in the hospital because of your drug use?

As with alcohol problems, all items required yes or no responses, and all referred to experiences occurring during the 12 months before the survey. The first five items were derived from the Drug Abuse Screening Test (DAST) [45], and the remainder were derived from the Ontario Student Drug Use Survey. We computed an overall measure of drug problems by summing positive responses to the eight items (M = 2.3; SD = 2.1; α =. 78). Although an arbitrary cutoff, we also computed the percentage reporting two or more problems in order to describe cluster differences in the proportion with drug problems.

To examine the co-occurrence of mental health outcomes, we used two measures. First, dual substance problems indicated the co-occurrence of alcohol and drug problems, that is, the percentage who reported two or more CAGE items and two or more DAST items. Second, dual disorders measured the percentage reporting an alcohol problem or drug problem and poor psychosocial health. Specifically, we assessed the percentage who reported two or more CAGE items or two or more DAST items and reports of both above average distress and psychotic thinking (based on a median split of 5 or more for distress and 2 or more for psychotic thoughts). Our usage of "dual disorders" does not imply the presence of a clinical diagnosis, but indicates positive reports of both mental health and substance problems.

We chose cluster analysis for two important reasons. First, unlike factor analytic methods, our goal was to cluster individuals, not variables. Second, unlike procedures such as multiple-group discriminate analysis, the typological structure is unknown and thus cannot be specified beforehand. To cluster analyze the 17 variables, we implemented SPSS's version of hierarchical agglomerative clustering employing Ward's method and squared Euclidean distance. Ward's method, which constructs clusters by minimizing within-cluster sums of squares, has been found to provide reasonable cluster solutions in empirical studies [46], [47]. As well, cluster analysis can employ binary variables that do not display significant skewness [48]. To ensure equal weighting, all variables were z transformed prior to analysis. Determining the number of clusters is particularly difficult since cluster analysis will provide clusters even when applied to unclustered data. We used three strategies to determine the number of clusters: First, we examined the Euclidean distance schedule for large increases in coefficients between agglomeration steps; second, we examined cluster profiles for meaningful cluster differences; and third, for the purposes of increasing the validity of the cluster solution, we compared clusters on variables not employed in the cluster analysis.

Once the cluster solution was determined, we used ordinary least squares (OLS) or logistic regression to examine cluster differences in mental health outcomes. All analyses employed cluster group as an effect-coded categorical independent variable; thus, levels of significance of group coefficients indicate whether cluster values differ significantly from the grand mean. We must also note that the four substance-use outcomes would likely vary by cluster membership since two indicator variables for alcohol and drug problems were included in the cluster solution. However, our objective is to examine cluster differences in level of alcohol and drug problems and the co-occurrence of mental health problems. Because cluster analysis affects the characteristics of variables not employed in the cluster solution, we intentionally do not employ control variables such as gender in order not to mask true cluster differences. However, variables such as gender and sample did not differ significantly by cluster membership. With listwise deletion of missing data, the retained sample was 211.

Our first objective in this paper is to describe the typological characteristics of our sample. Examination of large increases in distance coefficients suggested a four-cluster or eight-cluster solution. On closer examination, the eight-cluster solution provided substantively important distinctions compared with the four-cluster solution. Most important, cluster 3 in the four-cluster solution resulted in two substantively unique groups. Although the original cluster 3 was characterized by high sexual abuse, when furthered partitioned, a resilient cluster that displayed below average alcohol and drug problems, despite their above average sexual and physical abuse, was revealed. As well, cluster 1 in the four-cluster solution revealed four subclusters that displayed substantively unique characteristics, described below.

In Table 2 we present data on the 17 variables employed in the cluster analysis. Significance levels within clusters indicate differences from the average. To understand better the character of the eight clusters and to provide further evidence of meaningful cluster differences, we also present cluster differences for 25 variables not employed in the cluster solution, of which 16 differ significantly (Table 3).

Table 2. Profile of Variables Used in the Cluster Solution

341 Note: p = proportion; z = z score. p-levels within clusters indicate differences from the average. ap <. 001. bp <. 05. cp <. 10. dp <. 01.

Table 3. Variables Not Employed in the Cluster Solution

342 % = percentage; M = mean. ap <. 001. bp <. 01.

The dominant defining characteristics of the eight types of street youths fall into three broad classes. The first is a class of five clusters—Entrepreneurs, Drifters, Partiers, Retreatists, and Fringers—that share below average familial risk factors. Entrepreneurs (n = 19) are frequently involved in an array of delinquent activity and are highly entrenched in the street lifestyle; Drifters (n = 35) are loners who have infrequent contact with family or friends, display lower than average family dysfunction, and are not highly entrenched in the street lifestyle; Partiers (n = 40) are distinguished by their recreational motivation for alcohol and drug use and their below average entrenchment in the street lifestyle; Retreatists (n = 32) are distinguished by their high coping motivation for substance use; and Fringers (n = 48) are marginally involved in the street lifestyle and show lower than average family dysfunction.

The next class is represented by two clusters—Transcenders and Vulnerables—that share above average familial risk factors. Transcenders (n = 21) represent a resilient type; despite above average family dysfunction, including physical and sexual abuse, they report below average mental health or substance use problems. Vulnerables (n = 12) are characterized by high family dysfunction (including physical and sexual abuse) and elevated mental health outcomes; their use of alcohol and other drugs is motivated by coping and escapism. Particularly notable is that, although the Transcenders and Vulnerables both report above average experiences of sexual and physical abuse, the Transcenders are less likely than the Vulnerables to identify abuse as an important reason for their departure from home.

The last class consists of Sex Workers (n = 4), who are highly entrenched in the street lifestyle and who report frequent commercial sexual work, above average sexual abuse, and extensive use of crack cocaine. It is important to note that this cluster of Sex Workers represents only 4 of 21 youths who were involved in the sex trade. Compared to others involved in the sex trade, this cluster of 4 youths was more involved in prostitution and used cocaine and crack cocaine more frequently. Evidently, several of the clusters are small in size; thus, we must be cautious in not overinterpreting these data. Still, despite the small sample sizes, we find a number of sizable and meaningful intercluster differences. Moreover, a close examination of characteristics of the four Sex Workers led us to conclude that this cluster represented a small, but meaningful, group rather than an aggregation of statistical outliers.

Our second objective is to examine variation in mental health outcomes among clusters. As we noted above, the validity of cluster solutions is strengthened if groups differ in meaningful ways for factors not used in the cluster solution. In Table 4, we see that all four mental health outcomes not used in the cluster analysis—distress, self-esteem, psychotic thinking, and suicide attempts—differ significantly among the eight clusters.

Table 4. Psychological Health by Cluster (n = 211)

343 Significantly different from the average: ap <. 05; bp <. 01; cp <. 001; dp <. 10.

Beginning with distress, we see that four clusters differ significantly from the average: Vulnerables and Retreatists experience higher than average distress whereas Fringers and Entrepreneurs experience lower than average distress. Self-esteem differs from the average for three clusters: Fringers and Transcenders report significantly higher than average self-esteem, whereas Vulnerables report the lowest self-esteem of all eight clusters. Sex Workers also report above average self-esteem, but because of the small sample size, this level does not differ significantly from the average. Less variation for psychotic thoughts is seen among clusters than for distress and self-esteem, with only two groups differing substantially from the average: Vulnerables report significantly more psychotic thoughts than any other group, while Fringers report significantly fewer psychotic thoughts. Variation in reports of lifetime suicide attempts is particularly striking. A significantly lower proportion, compared with 92% of Vulnerables, of the Fringers (23%), Drifters (37%), and Partiers (33%) attempted suicide than the average of 44%.

Regarding substance abuse, we see significant cluster variation in all six outcomes: The average number of alcohol problems ranges from 0 to 3.4 problems; the percentage reporting two or more CAGE items ranges from 0% to 78%; the number of drug problems ranges from 0.8 to 3.5; the percentage reporting two or more DAST items ranges from 13% to 75%; the percentage reporting both alcohol and drug problems varies from 0% to 53%; and the percentage reporting both substance and mental health problems ranges from 0% to 67%.

Five distinct patterns emerge from these data. The most salient pattern is for Retreatists, who display one of the highest rates of substance problems for all six outcomes (significantly for three of the six). This group averaged 3.4 alcohol problems and 3.5 drug problems during the past 12 months. In addition, 78% reported two or more CAGE items, 63% reported two or more DAST items, 53% reported dual substance problems, and 41% reported substance and mental health problems. The second pattern occurs for Fringers and, to a lesser extent, Transcenders. For three of six substance outcomes, Fringers reported lower than average values. The third pattern occurs among Sex Workers. Although not statistically significant from the average, 75% of Sex Workers reported two or more DAST items and, on average, reported 3.3 drug problems. Detailed analyses (not reported) showed that elevated substance problems were due to the use of cocaine and crack cocaine. The fourth pattern occurs among the Vulnerables, who display elevated rates of alcohol problems (3.2) and above average, although not statistically significant, dual disorders (67%). The fifth pattern, which occurs among the Drifters and Entrepreneurs, shows elevated alcohol problems. Partiers, the remaining cluster, display average rates of substance problems.

In summary, two clusters stand in stark contrast: Fringers and Vulnerables. At the one extreme, the Fringers fare better in the domain of psychological health than do other clusters; they experience less distress, fewer psychotic thoughts, and greater self-esteem and are the least likely to attempt suicide and to report alcohol and drug problems. In contrast, Vulnerables report above average distress, psychotic thoughts, and alcohol problems and low self-esteem, and nearly all have attempted suicide.

Having described the mental health configurations of the eight clusters, can we identify from the cluster profile what characteristics of the Vulnerables appear to be distinguishing factors in their poor psychological profile? If we integrate the data from Tables 2, 3, 4, we find that Vulnerables are the only cluster to display the poorest mental health and to attach significant importance to family dysfunction. Note that the Vulnerables also report above average physical and sexual abuse; however, these factors do not engender poor psychological health unless coupled with negative appraisals of familial dysfunction. Indeed, the Transcenders, who experience similar rates of sexual and physical abuse, are not as susceptible to poor psychological health as are the Vulnerables. The only dominant characteristic that distinguishes these two groups is the youth's appraisal of family dysfunction, which is lower among the Transcenders. Regarding the Fringers, we find that their above average health seems attributable to familial factors: They are the only cluster to report the lowest rate of physical and sexual abuse, in addition to placing the least importance on familial factors as motivators for their departure from home.

Our results echo comments that "Homeless people are not homogeneous; even homeless alcoholics and drug addicts are not homogeneous" ([49], p. 629). Indeed, the results of our cluster analysis of adolescent street youths replicate this finding. Although our sample displayed sizable variation in mental health characteristics of street youths, we must not lose sight of their population risk: 44% attempted suicide, 67% were physically abused, 70% reported substance problems, and 46% sought treatment for substance problems. Indeed, street youth experience an array of risk factors compared to mainstream adolescents [7], [13]. Still, an examination of street youth characteristics show striking differences, many of which may provide implications for intervention and programming.

Before discussing the potential implications of our analysis, we should outline the limitations of this study. First, we must recognize that, because we did not employ a full-probability sample design, we cannot conclude that our sample is fully representative of all street youths in Toronto. However, we did employ probability sampling of 11 agencies, and we made careful attempts to ensure no systematic biases in our selection process. The second limitation centers on the cluster solution. Our moderate sample size did not allow a thorough examination of cluster differences given the small sample size of several clusters, especially the Sex Workers. As well, although our clusters provided meaningful differences, because of the exploratory nature of cluster analysis, we cannot know to what extent our taxonomy is stable and replicable. Third, our mental health measures were not based on diagnostic outcomes; however, they were sufficient to measure variation and discriminate clusters. Fourth, our data are cross-sectional; consequently, we can only establish associations among mental health outcomes and various influences. Finally, our data are based on self-reports. However, like others who have commented on interviewing similar populations [39], [50], [51], we believe that responses provided by youths were honest and accurate. Although participants were clearly cognizant of the $15 payment, they also realized their responses would not affect their personal needs, unlike interviews with authorities that might determine their qualification for various social services.

In our view, there are several important findings in our data. One finding that seems particularly striking is the extent to which characteristics, risk factors, and mental health outcomes vary among street youths. Indeed, our findings dismantle several stereotypes held of street youths: that all are sexually and physically abused, that all are products of gross family dysfunction, that all reject conventional goals of society, and that all are without the resources or desire to become productive members of society.

Another finding from our study is the obvious web of mental health issues. Indeed, for many, separating substance abuse problems from other mental health difficulties may be futile at best. Indeed, those who report above average alcohol and drug problems also report other mental health difficulties. Attempting to intervene in only one of these areas may be far from efficacious. Also interesting is the seemingly critical role played by appraisals of family dysfunction. As symbolic-interactionists have long argued, the subjective interpretation of family dysfunction played a greater role in poor mental health than did the actual reports of physical and sexual abuse. Another finding that requires future research is the processes involved in resiliency. Indeed, a critical study of resilient youths would greatly inform our understanding of the interplay between risk and protective factors.

Assuming the findings from our analysis are fairly representative, there might be some suggestion that, compared to other countries such as Brazil [52], a sizable proportion of street youths in Toronto are less entrenched in the underside of the street culture. Indeed, the two clusters that were marginally connected to the street world—Fringers and Drifters—together account for 40% of youths in our sample. This would suggest that interventions would have reasonable potential in reaching this group. As well, the clusters suggest the need for differential programming. Youths representative of Fringers and Drifters would benefit most from programs designed to increase opportunities for job skills, employment, and education. On the other hand, youths representative of Vulnerables and Retreatists require multifaceted programs to manage their mental health issues in addition to job skill and employment needs. Informed social policy must be based on a thorough understanding of the characteristics of the population. And, it is evident that simplistic typologies such as "throwaways" and "curb kids" cannot describe adequately the heterogeneity of this population.

We gratefully acknowledge the contributions of Reginald Smart, Gordon Walsh, and Frank Ivis.