Measuring the Developmental Nature of Multiple Drug Use


There have been a number of studies in which attempts have been made to measure or assess multiple drug use. Some of these are from general populations while others are focused on specific subpopulations of users. The studies are grouped more on the basis of the approach taken to assessing multiple drug use than on the patterns uncovered. There are at least four different groupings of studies and some studies fit into more than one grouping.

Developmental Patterns of Onset of Use

One of the most influential attempts to describe patterns of multiple drug use is the “stages of drug use” model developed by Kandel. Kandel posited that persons proceed from licit to illicit drugs and from use of less to more serious drugs. The stages of drug use involvement that she identified were: (1) no use of any drugs; (2) use of beer or wine; (3) use of cigarettes and/or hard liquor; (4) use of marijuana; and (5) use of illicit drugs other than marijuana. Although it is not made explicit by Kandel, there is an implication that the drugs from the earlier stages of development are “carried forward” into the later stages of drug involvement. Thus, a marijuana user is likely to continue his or her use of cigarettes/hard liquor and beer or wine.

Kandel and her colleagues utilized unidimensional Guttman scaling techniques and assumptions to assess the fit of data to the stages of development model. These analyses allowed her to state that use of marijuana is a “necessary” condition for use of other illicit drugs such as heroin and cocaine. This interpretation has been confirmed independently in a number of other studies.” Recently, Donovan and Jessor suggested that problem drinking is a stage that exists subsequent to marijuana use and prior to use of heroin and cocaine. Although their point is to add another sequential marker to the “unidimensional” conceptual defintion of drug use, Donovan and Jessor’s claim underscores the reality of multiple drug use that is implied by the developmental stages model. Yamaguchi and Kandel conducted a follow-up interview of persons 24-25 years old who were initially interviewed 9 years earlier in a study of a representative sample of 10-llth graders in New York state. They found the following:

For men, the pattern of progression is one in which the use of alcohol precedes marijuana; alcohol and marijuana precede other illicit drugs; and alcohol, marijuana and cigarettes precede the use of prescribed psychoactive drugs. Eighty-seven percent of men (82 percent not by chance) are characterized by this pattern. For women, the pattern of progression is one in which either alcohol or cigarettes precedes marijuana; alcohol, cigarettes, and marijuana precede other illicit drugs; alcohol and either cigarettes or marijuana precede prescribed psychoactive drugs. Eighty-six percent of women (77 percent not by chance) share this pattern.

There are at least two things to note about the relatively invariant stages of drug involvement. First, the stages are invariant at this point in history. The order could change at some future point for a number of reasons, especially changes in the sociolegal aspects of drugs or new knowledge about negative effects. Second, it should be remembered that the stages-of-development model “implies” use of multiple substances, not just a sequence in which drugs at later stages are “substituted” for drugs used at prior stages. Drug users carry forward their use of substances appearing at earlier stages, often at higher or more intensive levels of frequency and quantity consumed.

Clusters of Drugs

The development in recent years of more sophisticated statistical techniques for uncovering the “latent” constructs for measured variables (i.e., frequency and extent of use of various drugs) has provided a different window or perspective on multiple drug use. Huba et al. used data from 1634 students in the 7th through the 11th grades in Los Angeles to compare their “common factor” model with Kandel’s “simplex” model. The respondents were asked the number of times they had used 13 different classes of drugs. Using maximum likelihood confirmatory factor analysis techniques, Huba and his colleagues found three latent constructs:

There are positive loadings for beer, wine, liquor, cigarette, and inhalant use on the first latent variable of alcohol use. The second latent variable of cannabis use has positive loadings for marijuana, hashish, and cigarettes, and a significant negative loading for wine. The third latent variable of hard drug use has significant positive loadings for cocaine, tranquilizers, drugstore medication, heroin, hashish, inhalants, hallucin-ogenics, amphetamines, and liquor.

The essential conclusions of this study were that drug use among adolescents is unidimensional even at the first-order latent factor level and that the fit of the data to the common factor model is not significantly better than the fit to the simplex model. Examination of the drugs clustered in each of the three latent variables illustrates the reality of multiple drug use only implied in discussions of sequential conceptualizations of the stages of development in drug use. However, neither the Kandel nor the Huba et ah approaches tell us anything about when the various drugs are/were used or whether use of one drug had anything to do with use of the other drugs. These two models merely allow the researcher to locate respondents on a crude scale of drug involvement; they do not provide a very good basis for substantive conclusions about levels of multiple drug use in society.

Typologies of Drug Users

There is a longstanding tradition in the drug abuse field to attempt to place individuals into descriptive categories or types based on (1) their use of various drugs and (2) their psychosocial characteristics.

Bachman, O’Malley, and Johnston have developed a fairly simple index of multiple “illicit” drug use for the annual survey of high school seniors. The principal assumption underlying the index is that there is a developmental pattern to use of drugs. The index consists of five types: (1) none, (2) marijuana only, (3) few pills, (4) more pills, and (5) any heroin. Types 1, 2, and 5 are quite clear. Types 3 and 4 are defined according to no use of heroin and differences in the level of use of the following drugs: LSD, other psychedelics, cocaine, amphetamines, tranquilizers, methaqualone, barbiturates, and other narcotics. To qualify for inclusion in Type 3 (few pills), the respondent would have used one or more of the above drug classes one or two times and would never have used heroin. Persons in Type 4 would have used one or more of the above drug classes on three or more occasions and no heroin. It is clear that a substantial proportion of persons in the types beyond “marijuana only” had used marijuana as well as alcohol and cigarettes. Therefore, although the labels given the types of drug use do not explicitly mention multiple drug use, the persons in the types are increasingly “multiple” moving from Type 1 to Type 5. The index of illicit drug use developed for the high school senior study is similar to one developed by Miller and her colleagues for data from the 1982 National Survey on Drug Abuse. This index had three categories or types: (1) never used any illicit drug; (2) used marijuana only; (3) used cocaine, hallucinogens, or heroin. This index does not explicitly acknowledge but implies and allows for the possibility that those in Type 3 could also be using marijuana as well as cigarettes and alcohol.

The typological approach to multiple drug use is fairly common among researchers studying clients in drug treatment programs. Simpson and his colleagues identified 28 different patterns of use for a set of eight drug classes (heroin, other opiates, barbiturates, cocaine, amphetamines, hallucinogens, marijuana, and other drugs) in their study of over 11,000 clients in the Drug Abuse Reporting Program (DARP). Frequency of use in the 2 months prior to treatment was recorded along the scale from daily use, weekly use, less than weekly use, to no use. The modal pattern was heroin only (28.3%). Heroin in combination with marijuana (7.8%), with cocaine (8.8%), and with both marijuana and cocaine (7.3%) constituted the other prevalent types of multiple drug use. Alcohol use was not considered in the DARP study of mutiple drug use.

Bray and his colleagues used data from the 3389 clients in the Treatment Outcome Prospective Study (TOPS) to replicate the pattern of multiple drug use found in the DARP study. There were major differences. Heroin only, the modal pattern (28.3%) for the DARP study, constituted only 3.2% of the TOPS clients. TOPS had a much larger percentage in the marijuana-only type (10.3 versus 0.6%), fewer heroin-cocaine users (4.5 versus 8.8%), more heroin-cocaine-marijuana users (15.2 versus 7.3%), more polydrug users defined as heroin and other opiates less than weekly plus any use of three or more nonopiates (9.7 verus 4.9%), and fewer polydrug and opiate users defined as heroin or other opiates more than weekly plus any use of three or more nonopiates (5.5 versus 13.3%). One of the biggest differences between the TOPS and the DARP studies was the percentage of clients who did not fit into the 28 categories identified in the DARP study (1.0 versus 17.9%). One thing is clear from these two studies: the percentage of multiple drug users in the 1979 TOPS sample of drug treatment clients was considerably higher than that found in the DARP sample from the early 1970s.

Braucht et al. studied patterns of drug use among 440 clients in four drug and alcohol treatment programs. Using responses concerning 15 drug classes and cluster analysis techniques, they were able to identify six clusters of drugs and empirically identify eight types of drug abusers. The distribution of the sample into the eight types is quite skewed. However, these investigators asked each respondent “how often they had used drugs in combinations to achieve an effect.” The first type, “infrequent experimenters,” constituted 55% of the total sample of 440 clients, and 42% of this group indicated they had “never” used drugs in combinations to achieve an effect. The second type was “alcoholics,” 16% of the sample (65% had never used combinations to achieve an effect). The third type, “barbiturate and minor tranquilizer users,” was only 4.1% of the sample. However, over 30% of these users reported “always” using combinations of drugs to achieve an effect. “Narcotics users” were only 3.8% of the sample and 30% “always” used drugs in combination. The fifth type consisted of persons who are users of “amphetamines, marijuana, and major tranquilizers.” Although they constitute only 2.8% of the sample of clients, 73% reported “always” using drugs in combination, and none of the clients in this category reported “never” using drugs in combination. The sixth type (methaqualone, minor tranquilizers, and barbiturates) and the seventh type (codeine, inhalants, analgesics, and methaqualone) of users each represented less than 2% of the sample. The eighth type (hallucinogenic users) included almost 15% of the sample. This group was primarily involved with marijuana and other hallucinogenic drugs and heavily involved with using various combinations of drugs simultaneously.

The Braucht et al. study demonstrates that among persons who have used several drugs, a large proportion “always” use the drugs in combinations. Although they did not ask the respondents to specify the motivations for multiple use (e.g., to enhance the effects or to counteract the effects), it is likely that those who said they “always” used drugs in combination would list one or both of these reasons for simultaneous multiple drug use.

Composite Indices of Drug Use

The fourth way that researchers have attempted to deal with the “problem of multiple drug use” is to create composite indices or scales of use. This is a very tricky and formidable task. Among the issues that must be dealt with are: (1) the drugs to be included in a composite index; (2) the dimensions of use (e.g., frequency, volume, extent, quantity, and other factors); (3) the procedures to be used in creating the weights for the drugs and the categories of each dimension; and (4) the sample on which the index is to be normed.

Why create a composite index in the first place? The answer to this question is simple. Most researchers would rather conduct statistical analyses once with one score than many times with each drug class treated separately in the analyses. The disadvantage of a composite index is that it makes more difficult the linkage of specific consequences to patterns of use of specific drugs. However, the researcher can always rely on the patterns of use of each drug from which the index is constructed if specific linkages are needed. A more formidable disadvantage is the implication that index scores reflect the relative “seriousness” of use of various drug classes.

How does one assess the relative seriousness of various classes of drugs? One way is to use “expert” judges. Bucky and his colleagues had a panel of judges assign a weight to the following drugs: marijuana, LSD, amphetamines, barbiturates, and heroin. Hoffman and his associates had four experts rate eight drugs with respect to their “hazard for users.” The drugs rated were marijuana, cigarettes, alcohol, stimulants, depressants, methamphetamines, hallucinogens, and narcotics in that order, from least to most serious. The obvious limitation of this approach is the lack of reliability. The weights assigned by the judges reflect arbitrary evaluations and different panels might produce different weights. What is needed are weights for various drug classes that are derived objectively and normed on a representative sample of the population.

Two other attempts to create composite indices of multiple drug use rely on the distributions of use in a sample. Douglass and Khavari obtained information on use of 19 drugs or drug classes. For each, a score of 0 to 7 was assigned to the responses of having never used it, using none now, using it less than monthly, and so on to using it several times a day. The responses were treated as if they constituted an interval level scale of measurement. The mean and standard deviation were computed for each drug and a normal score was computed for each level of use. The difference between levels for each drug is a constant, and it is used as a weight for that drug. For each drug a person had used, the level was multiplied by the weight, and the sum of these products formed the drug use index. There are several problems with this approach. The first and most obvious is that the scores for levels of use are treated as if the intervals between them are equal. This assumption, which is central to the Douglass and Khavari index, is faulty. For example, if one were to estimate the number of times a drug had been used in a month on the basis of response categories such as about once a month, about once a week, several times a week, daily, and several times a day, the result would probably be 1, 4, 10, 30, and 75 or more. The numbers in the Douglass and Khavari index were 1 through 7. With a wider range and more realistic numbers, the means and the standard deviations would have been quite different. The weights for the drugs would have been substantially different. However, the most serious limitation of this index is that it is useful only with the sample on which it is constructed. The weights are not transferable from study to study. They must be constructed anew for each study.

Lu attempted to construct a composite index of multiple drug use by assigning weights to the categories of extent of use of each drug. The weights are determined by the proportion of cases in the total sample that are found in the various categories. The weights reflect the frequency or rarity of a given level of drug use in a sample. Herein lies a major limitation of the Lu index. For example, O’Donnell and his colleagues found in their nationwide sample of young men the following weights for marijuana: no use (.225); experimental use (.534); light use (.664); moderate use (.755); and heavy use (.900). There is a nice and even progression with this more frequently used drug. However, they found the following weights for heroin: no use (.471); under ten times (.955); under 100 times (.978); under 1000 times (.989); and 1000 times or more (.996). As one can see, nonuse of heroin gets almost as high a score as experimental use of marijuana. Furthermore, there is little difference in the scores between use of heroin under ten times and use of heroin 1000 times or more. The Lu indexing approach to multiple drug would therefore be useless for samples drawn from drug treatment centers or heavy-using street populations because everyone would be getting a high score. The Lu index score is obtained by adding the scores for each drug class. Pandina et al. used the Lu index computational scheme by creating scores for each of three dimensions (extent, frequency, recency) for each of ten drugs. O’Donnell et al. found that the correlation between the Lu index score for multiple drug use and a simple count of the number of separate drugs ever used was .971. In simple terms, this approach represents an improvement in the attempt to measure multiple drug use but fails in that attempt because of problems with the ways the weights are obtained and the fact that the scores are based entirely on the distribution of use in each sample.

Clayton and Voss were responsible for developing the most comprehensive composite index of multiple drug use to date. They used data from: (1) the 1977 National Household Survey on Drug Abuse; (2) the 1977 Client Oriented Data Acquisition Process (CODAP) admission file; (3) emergency room mentions from the Drug Abuse Warning Network (DAWN); (4) drug mentions from the DAWN Medical Examiner data; (5) the Drug Enforcement Administration data on illicit prices; (6) data from the Supported Work study of ex-addicts, ex-offenders, unemployed minority youth, and AFDC mothers; and (7) data from the nationwide study of young men. After extensive preliminary analyses, the following weights were derived: marijuana (1); psychedelics (2); opiates other than heroin (3); stimulants (5); sedatives (6); cocaine (7); and heroin (24). Using these weights for the drugs, a researcher can then develop reasonable weights for various dimensions of use such as frequency or extent and multiply for a composite index. For example, in testing this multiple drug use indexing scheme, Clayton and Voss applied the drug use weights listed above to a sample of 20- to 30-year-old men (125 blacks, 98 white, and 78 men from other race/ethnic groups) drawn randomly from high drug use areas in Manhattan. The figures below indicate how the response categories were used to derive multipliers for the drug classes.

Interview response categories Assumed mean no of times for response Weight
Never usedLess than 10 times

Under 100 times

Under 1000 times

1000 times or more









The range of possible scores on this index is from zero or no use of any drug to the sum of the products obtained by multiplying the drug class weight by 120. Thus, the total possible score is marijuana (120) + psychedelics (240) + opiates other than heroin (360) + stimulants (600) + sedatives (720) + cocaine (840) + heroin (2880), or 5760. The higher the score the higher multiple drug use.

Table Analysis of Variance: Lifetime Drug Use Index by Race and Educational Attainment for the Manhattan Young Men Sample

Independent variables N Lifetime drug use index
Mean S.D.
Total sample 294 469.90 1036.77
Black 125 737.21 1188.43
Less than high school 40 1176.80 1304.44
High school graduate 40 806.58 1325.09
Some college 34 352.03 823.32
College graduate 11 77.00 127.97
White 98 223.20 789.46
Less than high school 9 1282.44 2342.60
High school graduate 12 249.00 480.58
Some college 31 151.16 231.57
College graduate 46 57.78 139.65
Other 71 339.80 950.61
Less than high school 18 418.11 908.82
High school graduate 25 234.40 964.34
Some college 22 477.09 1102.75
College graduate 6 40.67 62.84

F for race 3.477 p .032

F for education 7.548 p .001

F for race-education interaction 1.593 p .149

The utility as well as the disadvantages of measuring multiple drug use in this manner can be seen in the data in Table Analysis of Variance: Lifetime Drug Use Index by Race and Educational Attainment for the Manhattan Young Men Sample. The primary advantage is that the “dependent variable,” drug use, is captured in a single, interval level score. This allows the researcher to use more robust parametric level statistical techniques such as analysis of variance. Expected relationships such as the one between educational attainment and drug use can be readily examined within categories of race/ethnicity. In addition, if one has longitudinal measures on the same sample or repeated cross-sectional data on similar samples, it is possible to use a single score to measure change. However, a limitation of this approach to assessing multiple drug use is that one cannot know readily which drugs go into making up the score. Further, the Clayton and Voss index is most useful with samples from the “normal” population. It is not very useful for samples of treatment clients where use of illicit drugs is extensive and intensive.

Measurement of Multiple Drug Use: Preliminary Conclusions

Clinicians and survey researchers agree that multiple drug use is a major problem, both in terms of measurement and treatment. Ask them how to measure this phenomenon and they will throw up their hands in despair. Drugs differ in their pharmacological composition, the biochemical effects they have on the human organism, in their abuse liability, and in the way they are used and abused in this society. Drugs are often used in combinations to enhance effects, to counteract effects, to substitute for drugs that are not available, or because combined use is normative. However, the problem of how to measure and treat multiple drug use is as elusive as the Loch Ness monster.

From the foregoing discussion the following can be assumed. First, persons go through relatively invariant stages of involvement with drugs. Second, drugs begun at earlier stages are continued into subsequent stages, often with higher frequency and quantity of ingestion. Third, because of their abuse liability and possible physical and medical consequences, some drugs are considered more serious to use than others. Fourth, some drugs are often used in combination with each other for various purposes. Fifth, the long-term health and other consequences associated with use of licit drugs such as alcohol and tobacco may be substantially higher, given the extent to which they are used, than those associated with the illicit drugs. Sixth, persons who use drugs at latter stages of development usually smoke cigarettes and drink alcoholic beverages, often in conjunction with their use of illicit drugs.

At this time, emphasis on developing adequate measures of multiple drug use is relatively new. The attempts that have been made along this line are at best crude and limited. The goal of such research should be to create a composite index of multiple drug use that is comprehensive, sensitive to differences between drug classes yet sensitive as well to patterns of use within classes, objectively and empirically sound, easy to construct and use, and able to discriminate between groups known to differ on this phenomenon. The ultimate goal of epidemiology and etiologic research on multiple drug use should be the development of a diagnostic tool that would allow clinicians to assess the problem of multiple drug use and use that information to link clients to appropriate treatment.


Selections from the book: “Recent Developments in Alcoholism. Volume 4: Combined Alcohol and Drug Abuse. Typologies of Alcoholics. The Withdrawal Syndrome. Renal and Electrolyte Consequences.” Edited by Marc Galanter. An Official Publication of the American Medical Society on Alcoholism, the Research Society on Alcoholism, and the National Council on Alcoholism. 1986.