Volume 1, Issue 2, January 1999 ISSN 1096-4886 http://www.westerncriminology.org/Western_Criminology_Review.htm © 1999, The Western Criminology Review. All Rights Reserved.

A Drug Use Typology for Treatment Intervention

Citation: Yacoubian, George S. 1999. "A Drug Use Typology for Treatment Intervention." Western Criminology Review 1 (2). [Online] http://www.westerncriminology.org/documents/WCR/v01n2/Yacoubian/Yacoubian.html

The ultimate goal of drug treatment is to improve the quality of life of a dependent. Many believe this objective is best achieved by matching dependents to individually tailored treatment programs. Treatment efforts can fail, however, because they are not tailored to the specific needs of the dependent. Dependents are often classified by techniques that are ill-suited for successful rehabilitation. In the present study, I use hierarchical and K-means cluster analyses to identify homogeneous clusters of substance users from a sample of 1,118 San Jose arrestees surveyed through the Arrestee Drug Abuse Monitoring Program. Five clusters are yielded in the present analysis. Treatment and policy implications are discussed.

---

A Drug Use Typology for Treatment Intervention¹

Much scholarly attention in the field of drug addiction focuses on the success, or lack thereof, of differing treatment interventions. However, our understanding of the varied needs of drug-using populations has changed very little since the Subcommittee on Drug Addiction of the New York Academy of Medicine (1955: 604) stated more than four decades ago that "it is a temptation to think of addicts as a homogeneous group whereas all that they have in common is their addiction." Many treatment interventions remain designed for a generic addict, a rudimentary designation that assumes all addicts possess like characteristics, and, as such, can be treated similarly. A review of recent research in the field suggests that this assumption is an erroneous one. Moreover, the current insufficient comprehension of effective strategies for the treatment of addiction stems, in part, from a failure to differentiate characteristics peculiar to sub-populations of drug-addicted people. Following a brief review of previous attempts to identify classifications of drug users, and findings that link treatment outcomes to specific programs tailored to individual client needs, I describe my identification of a drug use typology that could lead to program assignation better fitted to addict's needs, and, therefore, more likely to result in successful treatment outcomes.

Previous Research

During the past three decades, several attempts have been made to identify classifications of drug users (Yacoubian and Kane 1998; Solomon 1968; Cohen 1986; Nurco et al. 1991). Solomon (1968), for example, suggested a three-type classification based on the addict's symptoms, fears, and purported goals of ingesting drugs. The "social addict" uses drugs because of personal inadequacies; the neurotic addict uses drugs to combat anxiety; and the "psychotic addict" uses drugs to escape the fears of annihilation (Solomon 1968). Following the successful identification of these classifications, Solomon (1968) suggested treatment plans. He recommended education for the "social addict," psychological treatment for the "neurotic addict," and intensive medical treatment for the "psychotic addict." Although the classifications were not derived from sophisticated analytical techniques, they nevertheless illustrate an early effort to distinguish characteristics among addicts.

Cohen (1986) proposed a client-based typology based on data from 520 addicts surveyed in ten methadone clinics and one therapeutic community. Cohen's (1986) analysis yielded nine clusters, including the "emotionally sick addict," the "professional criminal addict," and the "inadequately socialized addict." Like Solomon (1968), Cohen (1986) also recommended treatment goals and techniques. The "emotionally sick addict" requires reconstructive psychotherapy to achieve emotional stability; the "professional criminal addict" requires value clarification; and the "inadequately socialized addict" requires therapeutic relations to establish positive object relations (Cohen 1986). While the typology is not ideal, it supports the notion that drug addicts possess varied characteristics.

Nurco et al. (1991) interviewed 250 male narcotic addicts attending methadone maintenance treatment centers in Baltimore and New York City. The researchers obtained data regarding the type and amount of criminal activity perpetrated during a period of narcotics addiction. A factor analysis and hierarchical cluster analysis were conducted to derive classifications of narcotic addicts. Their analysis yielded ten clusters, including "drug-distributors/dealers," who specialized in crimes of drug distribution; "thieves," who engaged in shoplifting; and "violent generalists,"who specialized in violent crimes, such as assault or robbery. The two largest clusters were the "drug distributors/dealers" and "thieves," which comprised nineteen percent (n=47) and eighteen percent (n=46) of the total sample, respectively. Again, the classifications confirm that drug users possess varying characteristics and cannot be generically categorized as "addicts."

Finally, Yacoubian and Kane (1998) identified homogenous clusters from a population of 1,329 Philadelphia arrestees surveyed through the Arrestee Drug Abuse Monitoring (ADAM) Program. Using hierarchical and K-means cluster analyses, they identify six types, including "recreational users," who use drugs intermittently; "injectors," who indicate high rates of injectable Central Nervous System (CNS) modifying drugs; and "converters," who commit property crimes in order to convert stolen merchandise into cash to support their habits. The results confirm that drug users can be classified into several distinct categories. The Yocoubian and Kane ( 1998) finding strengthens the idea, also noted in other research below, that treatment efforts should not treat the general "addict" but should instead be tailored to the specific needs of the drug user,

Along with efforts to classify drug users, the past three decades have also witnessed concerted efforts to curtail drug use in the United States, culminating in the rise of a wide range of treatment alternatives (Hser 1995; Wellisch, Prendergast, and Anglin 1995; Hepburn 1994; Mieczkowski et al. 1992). As Hser (1995: 209) states, "individuals may choose from a host of therapeutic settings including single-modality approaches or combinations of hospital-based inpatient stay, residential care, outpatient care, and self-help group meetings."

While the development of these alternatives has been beneficial for the drug treatment systems, their proliferation has evolved without the sophisticated techniques necessary to match individuals to the most effective client-specific option. Research suggests that no single treatment approach works for all addicts (Institute of Medicine 1990). It is likely that clients should be matched to alternatives that best work for them. As Hser (1995: 210) affirms, "optimum treatment involves the selection of the most appropriate treatment or treatments that are most likely to facilitate a positive outcome in a particular individual-- or effective matching." Thus, it is critical to indentify the characteristics of those in need of drug. In order for rehabilitation to be successful treatment providers need to be aware of the varying characteristics possessed by their clients. This awareness can then serve as basis for program assignation.

Classification of drug users into more meaningful sets has several advantages over programs that don't distinguish sub-populations of drug users. First, treatment providers can match clients to the alternative best suited to an individual's needs. Second, classifications can provide guidance for the evaluation of existing treatment programs. If clients are accurately matched, evaluation research can be conducted to measure the overall success of specific treatment programs. Third, classifications can assist in the development of programs that are needed. Given that certain addicts possess a wide range of treatment needs, jurisdictions may not have the functional capabilities of providing for those needs. Furthermore, a client with multiple physical, emotional, and psychological needs may require care in more than one program or facility. In such a case, providers will either have to prioritize treatment needs or be prepared to create a new facility more apropos for the divergent drug-using population. For these reasons, a matching system based on professional guidance and scientific analyses could improve the efficacy of jurisdictional drug treatment systems. Next I describe the cluster analysis used to identify unique characteristics possessed by drug users, and to group individuals according to a wide range of characteristics.

METHODOLOGY

To gauge drug use trends in urban areas, in 1987 the National Institute of Justice established the Arrestee Drug Abuse Monitoring (ADAM) Program, formerly the Drug Use Forecasting (DUF) Program. To date, thirty-five cities implement ADAM nationwide.² The program goals of ADAM are to: identify the levels of drug use among arrestees; track changing drug-use patterns; identify specific drugs of abuse in each jurisdiction; alert local officials to trends in drug use and the availability of new drugs; provide data to help understand the drug-crime connection; and serve as a research platform upon which a wide variety of drug-related initiatives can be based.

Researchers ask arrestees who admit to drug use questions about previous use, whether or not they consider themselves drug dependent, at what age they first used drugs, and whether they were under the influence or in need of drugs at the time of arrest. Several questions also focus on treatment-- whether the person has ever received treatment, is currently in a treatment program, or perceives a need for treatment. Participants are also asked several demographic questions, including education level, marital status, employment status, and income level. Demographic information from the interview and from the arrest record, which includes such data as age, sex, and race, can be correlated with information on drug behavior to better target intervention and treatment programs.

Because arrestees have been known to underreport drug use (Harrison 1995), ADAM uses urinalysis to validate self-report data. The Enzyme Multiplied Immunoassay Test (EMIT) screens for ten drugs: opiates (heroin, dilaudid, and morphine), marijuana, cocaine (powder and crack), phencyclidine (PCP), methadone, benzodiazepines (e.g., Valium and Xanax), methaqualone (Quaaludes and ludes), propoxyphene (Darvon), barbiturates (e.g., Phenobarbital and Seconal), and amphetamines ("speed"). All positive results for amphetamines are analyzed by gas chromatography, a method commonly used to confirm other tests because it is highly accurate in detecting the amount (sensitivity) and type (specificity) of a drug. The confirmation test eliminates positives that might be caused by over-the-counter "look-alike" medications. For the other drugs, no confirmation is necessary because they have no licit look-alikes that might cause the user to test positive.

For approximately fourteen consecutive days every quarter, San Jose research personnel obtain voluntary and confidential interviews and urine specimens from a sample of booked arrestees who have been in custody for no more than 48 hours. Approximately 225 adult males and 100 adult females are interviewed each quarter. In addition, ADAM sites operate according to a charge priority system, where non-drug felons, drug felons, non-drug misdemeanants, and drug misdemeanants are prioritized hierarchically. Furthermore, the number of drug offenders (felons and misdemeanants) surveyed during a data collection period cannot exceed twenty percent of the total sample. Consistent with established ADAM protocol, all arrestees are eligible for interviewing except for those whose primary charges involve driving offenses. ADAM researchers exclude these arrestees from the sample a priori.

The varying methods by which arrestees are selected across ADAM sites prohibit both site-specific and inter-jurisdictional generalization. The San Jose site benefits from having a centralized booking unit (the Main Jail) for all adults arrested in Santa Clara County. However, limited cell space at that jail often means that some female arrestees are transferred to the county jail--the Elmwood Correctional Facility (Women's Unit). As such, San Jose personnel occasionally interview female arrestees at two locations. Because all San Jose arrestees eligible to be interviewed are selected according to the established crime charge priority system, no efforts are made to randomly select those arrestees eligible to be interviewed. Given this methodological limitation, the external validity of the present findings is clearly an issue. Generalizing to the jurisdictional level is a monumental task in and of itself. As such, it is unclear whether the current findings are generalizable to other national ADAM and/or non-deviant populations.

DATA ANALYSIS AND FINDINGS

"Cluster analysis" is the generic name for a variety of procedures that can be used to classify objects into meaningful groups. These procedures form "clusters" or groups of highly similar, empirically derived entities around a centroid (cluster center).

Data analysis was accomplished in four successive intervals. First, correlation estimations were used to validate the self-report drug use data. Second, frequency distributions with means, standard deviations, and variances were obtained on all variables in the data set. The selection of variables included in the data reduction and grouping techniques was based on the variance of each variable. Those with non-significant variance were automatically excluded from the subsequent analysis in order to increase the parsimony of subsequent theorizing. Third, a principal components analysis (PCA) was estimated on the remaining variables in order to find common association among variables and to attempt to reduce the data into interpretable constructs. Finally, the principal component scores were saved and included in the succeeding cluster analyses, along with several other variables considered theoretically and/or statistically significant. The results from these analyses are shown below.

Descriptives

The sample is comprised of 1,118 adult San Jose arrestees interviewed during the first quarter of 1996 through the third quarter of 1997. The mean age of the sample is 31.6 years; the range is 19 to 77 years old. Over forty-one percent of the arrestees are Hispanic, thirty-two percent are white, fourteen percent are African-American, ten percent are Asian, and two percent identify with another ethnic group. About seventy-three percent of the arrestees are male. About sixty-two percent of the offenders are either high school graduates or have their GEDs.

Validation

Researchers have consistently questioned the accuracy of offenders' self-reports of drug use due to the invasiveness of the questions being posed. Concordance of self-report drug-use to urinalysis results has been found to be both reliable (Rosenfeld and Decker 1993) and unreliable (Harrison 1990). I validated self-report measures of substance abuse by comparing them with urinalysis results to. Spearman correlations were calculated for all drugs identified by the data collection instrument.³ When a drug arrestees admits to having ever used, they are asked to indicate whether they have used the drug within the past twelve months and within the past seventy-two hours. They were also asked the number of days they have used the drug within the past thirty days. For the purposes of the present analysis, I correlated the seventy-two hour self-report indicator with the urinalysis counterpart. Table 1 is a correlation matrix showing the bivariate relationship between the self-report and urinalysis measures.

** p<.01

The correlation coefficients for barbiturates, methadone, and Quaaludes could not be computed because of the small number of arrestees who admitted use. Table 1 indicates that almost all the correlations are strong. The only exception is the correlation between the self-report measure of marijuana and its corresponding urinalysis results (r=.38). While the correlation is somewhat weak, the results are nevertheless significant (p<.01). The remaining correlations range from .43 to .74, all of which are significant at the .01 level or below. Considered collectively, these results indicate a fairly high level of agreement between the self-report and urinalysis measures of drug use. As such, the self-report data appear to be valid and are used as the basis for generating clusters.

Variable Groupings

A principal components analysis was estimated on the reduced data set of forty-five variables. Varimax rotation was used since it maximizes and minimizes variable loadings on the components, which facilitates more interpretable results (Dunteman 1989). The utility of the components was based on a combination of Eigenvalues and the percent of variance explained. This method yields the identification of three principal components. Table 2 shows variable groupings and loadings on each of these principal components.

The first component produced an Eigenvalue of 5.83 with 13.9 percent explained variance; the second an Eigenvalue of 2.48 with 5.9 percent variance; and the third an Eigenvalue of 1.95 with 4.6 percent explained variance. While the final component produced less than 5 percent explained variance, which is often considered the standard cutoff (Munro and Page 1993), the fact that it represents the weight of three variables was enough to include it. A description of these factors follows.

Experimental use of all illicit drugs. This component is characterized by the use of all illicit drugs on at least one occasion during the arrestee's history. As the variable loadings in Table 1 suggest, drugs on this principal component represent psycho-stimulants (such as cocaine and amphetamines, which stimulate the central nervous system); depressants (such as barbiturates and Valium, which lessen the activity of the central nervous system); opiates (such as heroin, which have both analgesic and sedative properties); and psychedelics (such as PCP and LSD, which produce sensory illusions in the central nervous system). This component is distinct in that no one category of drug loads particularly high. Furthermore, no variables indicating recent use load on this factor. As such, this component is hereafter referred to as the "experimental use" variable.

Need/under the influence at time of crime. Variables associated with this component are those that measure a need for drugs and/or alcohol at the time of crime commission, as well as whether the offender was under the influence of alcohol and powder cocaine during the perpetration of the crime(s). This component is hereafter referred to as the "need/influence" variable.

Use of easily obtainable CNS drugs. Variables associated with this component are those that measure recent alcohol use (within the last seventy-two hours), and well as under the influence of alcohol at the time the crime was committed. Because alcohol is an easily obtainable CNS modifying substance, this PCA variable is hereafter referred to as the "easily obtainable CNS" variable.

Although different extraction methods and rotations may have yielded different results, the groupings of variables on the components appear to be both statistically and theoretically sound. Indeed, the three components are mutually exclusive. As such, the principal components were saved and used as basis variables in the subsequent cluster analyses.

Case Groupings

In addition to the variables created by the PCA, four other variables judged substantively and/or statistically significant were included as basis variables in the cluster analysis. These additional four variables include type of charge (Nurco et al. 1991), age, gender, and race. Since cluster analyses are greatly affected by the metrics used to measure variables, variables are recoded to similar metrics. The original age is recoded into a four point scale based on quartiles. Moreover, the "type of charge" variable is recoded into a scale to reflect four categories of offenses: violent, drug/alcohol, property, and miscellaneous.

Two different cluster analysis techniques were used. First, a hierarchical cluster analysis was estimated on a random sample (n=250) of cases. There are two advantages of starting the classification process with the hierarchical method. First, the number of clusters does not need to be initially specified. That is, an examination of the agglomeration schedule produced by the analysis indicates the optimal number of clusters for the sample. Second, the means of all the clusters identified by the hierarchical method can be saved and used as the starting centroids in a more efficient cluster analysis technique such as K-means (Aldenderfer and Blashfield 1984). This hierarchical cluster analysis led to the identification of five clusters. The means of the variables in these clusters were saved and used as initial seeds in the subsequent K-means cluster analysis, which utilized the entire sample of 1,150 arrestees.

However, estimation of the initial K-means cluster solution revealed that race was not statistically significant, which suggests that its inclusion in the model as a basis variable was inappropriate. ⁴This necessitated the re-estimation of the hierarchical cluster analysis on another random sample of cases (n=250) without the race variable. As with the initial hierarchical solution, the agglomeration schedule suggested that five is the optimal number of clusters. Based on this analysis, a five-cluster solution was specified using the initialseeds in a K-means cluster analysis. All variables were statistically significant (p<.01) alpha level. Analysis revealed outliers (n=32). A five-cluster K-means solution was estimated on the remaining sample using the initial seeds from the previous hierarchical cluster analysis. Analysis reveals that all variables are statistically significant, and there are no outliers. Table 3 shows the changes in the results before and after outliers are removed.

Table 3
ANOVA Tables and Cluster Memberships Based on K-Means
Clustering Solutions Before and After Removal of Outliers

All ETA² values are significant at p<.01

 The classification of cases is illustrated in Table 4, and narrative descriptions follow below.

Boozers. Subjects in this cluster (n=195) have the highest rate of use of easily obtainable CNS drugs. They are primarily male (79 percent) and between thirty-one and thirty-seven years of age. Although members of this cluster have the lowest rates of experimental drug use, they have a moderately high need for drugs or alcohol and/or rate of being under the influence of drugs or alcohol at the time of their offense. Their charges are generally for "drug/alcohol" offenses. Individuals in this cluster have excessively high rates of alcohol abuse, as well as excessive needs and/or rates of committing their crimes while under the influence of drugs or alcohol.

Solicitors. Subjects in this cluster (n=395) have a moderately high rate of experimental drug use, although they have the lowest rate of needing or being under the influence of drugs or alcohol at the time of their offense. They also have the lowest rate of easily obtainable CNS drug use. They are between the ages of thirty-one and thirty-seven and are fairly equal with respect to gender (60 percent male; 40 percent female). Individuals in this cluster commit primarily miscellaneous crimes, suggesting that they may be arrested for prostitution or soliciting commercial sex. Curiously, however, members display no need to engage in "miscellaneous" crimes, like prostitution, to support a drug and/or alcohol habit.

Converters. Subjects in this cluster (n=156) have a high rate of experimental use and the highest rates of needing or being under the influence of drugs or alcohol at the time of arrest. They are nineteen to twenty-three years old, predominantly male, and commit primarily property offenses. The correlation between need and property variables suggests that members of this cluster generally abuse easily obtainable CNS drugs and commit property crimes, like burglary, in order to "convert" stolen merchandise to cash in order to support their habits.

Violent Alcoholics. Subjects in this cluster (n=88) indicate the highest rate of experimental use and a high abuse rate of easily obtainable CNS drugs. This group is the oldest (over thirty-seven years old), mostly male, and primarily commit violent offenses. They demonstrate a clear drug experimenting history, culminating with what appears to be an addiction to alcohol. Given the correlation between age and offending, their age, and their high rate of being under the influence of drugs and/or alcohol at the time of crime commission, I inferred that individuals in this cluster explode after a drunken binge and are subsequently arrested for violent offenses, such as assault or domestic violence.

Enablers. Subjects in this group (n=284) indicate moderately low rates of both experimental use and of needing or being under the influence of drugs/alcohol while committing their crimes. Members of this cluster are primarily male, between twenty-four and thirty years of age, and commit primarily property crimes. Considering the low rates of drug use that are associated with this cluster, Enablers are likely to be under the influence of drugs/alcohol when committing their offenses rather than engaging in criminal behavior out of a need for drugs. Theoretical support for this cluster may be found in the geography of crime literature. In their study of suburban burglary, for example, Rengert and Wasilchick (1985) find that some burglary offenders smoke marijuana before committing their crimes because it lowers their inhibitions.

It is important to note that while cluster analyses cannot assign a case to more than one group, there is a degree of overlap with the age and gender variables. As the findings in Table 3 indicate, Boozers and Solicitors share the same age category (thirty-one to thirty-seven years old). Moreover, there appears to be a significant amount of overlap among clusters in terms of gender. While no two clusters share the exact gender proportions, they are all somewhat similar. This distribution is suggested by the ETA-squared obtained from the ANOVA estimation (refer to Table 3). While the between-cluster variances are significant for the gender variable (F=5.84), the ETA-squared is only .02, suggesting an almost equal distribution of sex across clusters. Thus, while it is interesting to include age and gender in the cluster descriptions, these two variables add very little to the discrimination between groups.

In sum, the analysis discerns several distinct subpopulations of drug users who use drugs for differing reasons and, as a consequence, are likely to require treatment modalities that take these differences into account. Next, I discuss some implications of these findings for policy and future research.

DISCUSSION

Matching drug-addicted clients and treatment programs continues to be a critical concern for clinicians and researchers in the field of substance. There is little dispute that some level of specificity is essential to efficaceous rehabilitation policy. Clients with distinct addiction histories and varying motivation for treatment will respond to rehabilitation differently. As Hser (1995: 220) states, "improved matching of clients to treatment programs has importance ranging from policy decisions on resource allocation to clinical decisions on developing treatment plans." Evaluation research has established that no single treatment approach is effective for all clients with drug-related problems, but rather a range of alternatives, tailored to varied individual needs, is required to provide effective drug treatment (Hser 1995; Institute of Medicine 1990, 1989). Curiously, however, treatment programs have been developed without the benefit of effective means of matching clients to the most appropriate treatment program.

The research taking place at ADAM, part of which is described in this paper, offers an important opportunity for identifying, differentiating, and targeting drug-using populations from within the criminal justice system. The present analysis has yielded five drug-using classifications for San Jose arrestees, clearly accomplishing the objective of targeting illicit drug users. The second step toward maximizing effective drug treatment policy is matching drug users to client-specific treatment programs. Clients will respond to rehabilitation differently depending on their addiction histories and motivation for treatment. This study provides an important starting point for accomplishing this objective.

Nonetheless, some methodological cautions are in order, which future research may address. Although the present analysis identifies a drug use typology of San Jose arrestees, the external validity of the study is an issue and the current findings by themselves are not generalizable to other populations. Given the sampling techniques utilized, the findings are, at best, generalizable to the San Jose arrestee population. While the current classifications may resemble other jurisdictional typologies, the issue of external validity is an empirical question that can only be addressed with future research. This work must first confirm the current findings for San Jose arrestees through a more rigorous methodological approach; second, compare the findings to other ADAM populations; and third, compare the results with non-arrestee populations. As jurisdictional treatment systems do not provide for arrestee clients alone, any classification system should also be based on a sample of non-arrestee respondents. If program developers are to use drug-using classifications as a foundation for program development, the typology needs to be widely applicable.

In addition, the current analysis is limited to the variables collected by the ADAM protocol. For example, the ADAM instrument does not measure mental illness, a critical component of successful substance abuse treatment (Cohen 1986; Solomon 1968). The ability of programs to treat clients, or the willingness for clients to be treated, can certainly be influenced by psychological deficiencies. Future research might include an addendum to the standard ADAM instrument that would measure mental illness.

The ultimate goal of drug treatment is to diminish or completely obviate a dependent's need for drugs and improve the quality of life of addicted people and those with whom they live and work. There is no doubt that certain treatment interventions within the criminal justice process ameliorate drug dependence problems (Anglin and Hser 1990). This study adds to our understanding of the need for client-specific program assignation. Despite these moderate successes, little progress has been made in identifying what kinds of dependents will benefit from which type of program. In sum, future research should consider identifying the specific needs of dependents in particular jurisdictions and evaluate the efficacy of those programs that claim to provide effective drug treatment.

Aldenderfer, Mark S. and Roger K. Blashfield. 1984. Cluster Analysis. London: Sage Publications.

Anglin, M. Douglas and Yih-Ing Hser. 1990. "Treatment of Drug Abuse." Pp. 393-460 in Drugs and Crime, edited by Michael Tonry and James Q. Wilson. Chicago: University of Chicago Press.

Cohen, Arie. 1986. "A Psychosocial Typology of Drug Addicts and Implications for Treatment." International Journal of the Addictions 21:147-154.

Dunteman, George H. 1989. Principal Components Analysis. London: Sage Publications.

Harrison, Lana. 1995. "The Validity of Self-Reported Data on Drug Use." Journal of Drug Issues 25:91-111.

Hepburn, John R. 1994. "Classifying Drug Offenders for Treatment." Pp. 172-187 in Drugs and Crime: Evaluating Public Policy Initiatives, edited by Doris MacKenzie and C. Uchida. California: Sage Publications.

Hser, Yih-Ing. 1995. "A Referral System that Matches Drug Users to Treatment Programs: Existing Research and Relevant Issues." Journal of Drug Issues 25:209-224.

Institute of Medicine. 1989. Prevention and Treatment of Alcohol Problems: Research Opportunities. Washington, DC: National Academy Press.

Institute of Medicine, Committee for the Study of Treatment and Rehabilitation for Alcoholism and Alcohol Abuse. 1990. "Matching." Pp. 279-312 in Broadening the Base of Treatment for Alcohol Problems. Washington, DC: National Academy Press.

Mieczkowski, Tom, M. Douglas Anglin, Shirley Coletti, Bruce D. Johnson, Ethan A. Nadelmann, and Eric D. Wish. 1992. "Responding to America's Drug Problems." Journal of Urban Affairs 14:337-357.

Nurco, David, Thomas E. Hanlon, Mitchell B. Balter, Timothy W. Kinlock, and Evelyn Slaght. 1991. "A Classification of Narcotic Addicts Based on Type, Amount and Severity of Crime." Journal of Drug Issues 21:429-448.

Rengert, George and John Wasilchick. 1985. Suburban Burglary: A Time and A Place for Everything. Illinois: Charles C. Thomas.

Rosenfeld, Richard and Scott Decker. 1993. "Discrepant Values, Correlated Measures: Cross-City and Longitudinal Comparisons of Self-Reports and Urine Tests of Cocaine Use Among Arrestees." Journal of Criminal Justice 21: 223-230.

Shoukri, Mohamed M. 1996. Statistical Methods for Health Care Sciences. Boca Raton: CRC Press.

Solomon, Philip. 1968. "Medical Management of Drug Dependence." Journal of the American Medical Association 206:1521-1526.

Subcommittee on Drug Addiction. 1955. Bulletin of New York Academy of Medicine 31: 592-607. New York: New York Academy of Medicine.

Wellisch, Jean, Michael L. Prendergast, M. Douglas Anglin. 1995. "Toward a Drug Abuse Treatment System." Journal of Drug Issues 25:759-782.

Yacoubian Jr., George S. and Robert J. Kane. 1998. "Identifying a Drug Use Typology of Philadelphia Arrestees: A Cluster Analysis." Journal of Drug Issues 28:559-574.

Endnotes

¹The author would like to thank Jack R. Greene, Director of the Center for Public Policy at Temple University in Philadelphia, and several anonymous reviewers, for their contributions to this paper.

²Albuquerque, Anchorage, Atlanta, Birmingham, Chicago, Cleveland, Dallas, Denver, Des Moines, Detroit, Fort Lauderdale, Houston, Indianapolis, Laredo, Las Vegas, Los Angeles, Miami, Minneapolis, New Orleans, New York, Oklahoma City, Omaha, Philadelphia, Phoenix, Portland, Sacramento, Salt Lake City, San Antonio, San Diego, San Jose, Seattle, Spokane, St. Louis, Tucson, and Washington, DC.

³ Spearman correlation is a non-parametric version of the Pearson correlation, and is typically used when data do not meet the assumption of normality. The Spearman correlation is based on the ranks of the data rather than on the actual values. The absolute value of the correlation coefficient represents the strength of the association between the two variables.

⁴ The measure of association used was the F-test of ETA. The F-test of the ETA produces an ANOVA table testing the null hypothesis that the mean of the dependent variable is equal in all the groups (e.g., clusters) defined by the first layer independent variable. The ETA-squared reveals the amount of explained variance in the cluster by the dependent variable. Rejecting the null hypothesis of no difference indicates that the dependent variable (e.g., age) varies significantly across clusters, thus suggesting its utility as a basis variable in a cluster analysis.

Top of Page