This article was downloaded by: [McMaster University]On: 21 December 2014, At: 01:23Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Substance AbusePublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/wsub20

Measuring College Student DrinkingJiun-Hau Huang ScD a , William Dejong PhD b , Shari Kessel SchneiderMSPH c & Laura Gomberg Towvim MSPH da Department of Society, Human Development, and Health , HarvardSchool of Public Health , 677 Huntington Avenue, Boston, MA,02115, USAb Department of Social and Behavioral Sciences and Youth AlcoholPrevention Center , Boston University School of Public Health , 715Albany Street, Boston, MA, 02118, USAc Social Norms Marketing Research Project, Education DevelopmentCenter, Inc. , 55 Chapel Street, Newton, MA, 02158, USAd Center for College Health and Safety Education DevelopmentCenter, Inc. , 55 Chapel Street, Newton, MA, 02158, USAPublished online: 23 Sep 2008.

To cite this article: Jiun-Hau Huang ScD , William Dejong PhD , Shari Kessel Schneider MSPH & LauraGomberg Towvim MSPH (2006) Measuring College Student Drinking, Substance Abuse, 27:1-2, 33-45

To link to this article: http://dx.doi.org/10.1300/J465v27n01_05

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the“Content”) contained in the publications on our platform. However, Taylor & Francis,our agents, and our licensors make no representations or warranties whatsoever as tothe accuracy, completeness, or suitability for any purpose of the Content. Any opinionsand views expressed in this publication are the opinions and views of the authors,and are not the views of or endorsed by Taylor & Francis. The accuracy of the Contentshould not be relied upon and should be independently verified with primary sourcesof information. Taylor and Francis shall not be liable for any losses, actions, claims,proceedings, demands, costs, expenses, damages, and other liabilities whatsoever orhowsoever caused arising directly or indirectly in connection with, in relation to or arisingout of the use of the Content.

This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &

Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

Dow

nloa

ded

by [

McM

aste

r U

nive

rsity

] at

01:

23 2

1 D

ecem

ber

2014

Measuring College Student Drinking:Illustrating the Feasibility of a Composite Drinking Scale

Jiun-Hau Huang, ScDWilliam DeJong, PhD

Shari Kessel Schneider, MSPHLaura Gomberg Towvim, MSPH

ABSTRACT. This study explored the feasibility of a Composite Drinking Scale (CDS) designedto capture fully the phenomenon of problem drinking among college students while allowing easypublic understanding. A survey conducted at 32 four-year U.S. colleges included four consump-tion measures: 30-day frequency; average number of drinks per week; number of drinks usuallyconsumed when partying; and greatest number of drinks in one sitting in the past two weeks. Re-sponses were normalized and added to create a continuous distribution, which was then subdividedinto quartiles (CDS/Q1-Q4). The CDS is an easily understood scoring system, but compared to thesimplistic “binge drinking” measure, it captures a broader range of relative risks and more clearlyestablishes the quadratic relationship between consumption and alcohol-related problems. Devel-opment of the CDS will require further exploring the best set of questions to include, establishingU.S. norms for the general population, and then transforming those scores to a simple measure-ment yardstick whose meaning can be easily communicated to the public. [Article copies availablefor a fee from The Haworth Document Delivery Service: 1-800-HAWORTH. E-mail address: <docdelivery@haworthpress.com> Website: <http://www.HaworthPress.com> © 2006 by The Haworth Press, Inc. Allrights reserved.]

KEYWORDS. Alcohol consumption, measurement, psychometric properties, college, student

Jiun-Hau Huang is Postdoctoral Fellow, Department of Society, Human Development, and Health, HarvardSchool of Public Health, 677 Huntington Avenue, Boston, MA 02115.

William DeJong is Professor, Department of Social and Behavioral Sciences and Youth Alcohol PreventionCenter, Boston University School of Public Health, 715 Albany Street, Boston, MA 02118.

Shari Kessel Schneider is Research Director, Social Norms Marketing Research Project, Education Develop-ment Center, Inc., 55 Chapel Street, Newton, MA 02158.

Laura Gomberg Towvim is Deputy Director, Center for College Health and Safety, Education DevelopmentCenter, Inc., 55 Chapel Street, Newton, MA 02158.

Address correspondence to: William DeJong, PhD, Department of Social and Behavioral Sciences, Boston Uni-versity School of Public Health, 715 Albany Street, Boston, MA 02118 (E-mail: wdejong@bu.edu).

The authors are grateful to their many colleagues on the Social Norms Marketing Research Project (SNMRP)who helped conduct the student surveys reported in this article, especially Jennifer Allard, Ingrid Bruns, KarenKaphingst, Amy Stern, and Cameron Ware. They are also grateful to the 32 site-based coordinators, whose hardwork provided the foundation for this study. The authors thank John D. Clapp of San Diego State University for hisinsightful comments on an earlier draft of this article.

This study was supported by grant R01-AA-12471 from the National Institute on Alcohol Abuse and Alcoholism(NIAAA) and the U.S. Department of Education to Education Development Center, Inc. (EDC).

The views expressed in this article are those of the authors and do not necessarily reflect the official position ofthe NIAAA or the U.S. Department of Education.

Substance Abuse, Vol. 27(1/2) 2006Available online at http://suba.haworthpress.com

© 2006 by The Haworth Press, Inc. All rights reserved.doi:10.1300/J465v27n01_05 33

Dow

nloa

ded

by [

McM

aste

r U

nive

rsity

] at

01:

23 2

1 D

ecem

ber

2014

INTRODUCTION

Developing measures of alcohol consump-tion that both capture the phenomenon of prob-lem drinking and allow easy communicationwith the public is an enormous challenge. To bescientifically useful, a measure should be pre-cise, reliable, and valid, of course, but it shouldalso capture the full meaning of problem drink-ing, which is a multifaceted behavior. It shouldalso allow for accurate predictions of alco-hol-related problems. To be useful in publicdiscourse, however, the measure should also besimple and its meaning easily communicated tothe public.

Among the earliest and most commonly usedmeasures of alcohol consumption are quan-tity-frequency methods (QF). In essence, QFmeasures ask respondents to report on howmany days they typically drink within a giventime period and how much they typically drinkon a given drinking day. On the plus side, QFmeasures are relatively easy to administer, donot require much time, and provide a fairly reli-ableestimateofdrinkingfrequencyandtotalal-cohol consumption (1).

On the other hand, QF measures do not in-quire about heavy episodic drinking, whichmight not be “typical” for many individuals butnevertheless is strongly associated with alco-hol-relatedproblems.Heavydrinking,which issometimes called “binge drinking,” is com-monlydefinedashavingfiveormoredrinks inarow (2). A “heavy drinker” is typically definedas someonewho drinksat this levelat leastoncein a two-week period. For women, Wechslerand his colleagues defined heavy drinking asfour or more drinks in a row (3), but this stan-dard has not been universally adopted (2).

One study found that 31 percent of heavydrinkers identified through daily diary reportswere misclassified as moderate drinkers whenusing QF measures (4). Other researchers havereported even higher rates of misclassification(5). In a study of high school seniors and drop-outs, Ellickson and her colleagues found thatQF measures failed to identify 54 percent ofthose who engaged in heavy drinking (6).

Some researchers have improved on QFmeasures by folding in a question about heavyepisodic drinking (7). One survey study dem-onstrated that adding questions about atypical

drinkingtoordinaryQFquestions increasedthetotal consumption estimate for a sample ofadultsby 14 percent, illustrating that traditionalQF measures tend to underestimate the “true”level of total alcohol consumption (8).

On theside of simplicity, several researchershave utilized the heavy drinking measure by it-self (2). The principal downside of using thismeasure is its dichotomous nature, which cre-ates a very crude yardstick by which to measurechanges in alcohol consumption. Using thismeasure by itself creates additional problems,including creating inflated public perceptionsof the true extent of dangerous drinking (espe-ciallywhen it is referred to as “binge drinking”)and perpetuating the false notion that drinkingbelow that level is entirely safe (9).

The purpose of the present study was to ex-plore the feasibility of developing a more re-finedmeasureof alcoholconsumption,one thatcombinesseveralquestionsdesignedtocapturedifferent aspects of problem drinking but alsoprovides a simple yardstick that the public atlarge might find easier to understand. A na-tional survey of four-year college and univer-sity students asked four questions, which werebrought together into a Composite DrinkingScale (CDS): (1)numberofoccasionsonwhichalcoholwasusedinthepast30days, (2)averagenumberofdrinksconsumedinaweek, (3)num-ber of drinks usually consumed when partying,and (4) thegreatestnumberof drinks consumedat one sitting during the last two weeks. Re-sponses to these four questions were first usedto assess the abstaining versus drinking statusof each respondent, and then to compute a con-tinuous CDS score representing the overalllevel of alcohol use for each drinker. For pur-poses of simplification, and to facilitate com-parisons with a dichotomous measure of heavyepisodic (“binge”) drinking, the continuousdistribution of CDS scores was then dividedinto quartiles, and each drinker was assigned acorresponding quartile score.

We report here the basic properties, internalconsistency reliability, and construct validityof the CDS. Proponents of the heavy (“binge”)drinking measure have argued that it providesan easy shorthand for identifying problematicdrinkers (3). Accordingly, we compare the util-ity of the CDS in predicting alcohol-relatedproblems in comparison to the heavy drinking

34 SUBSTANCE ABUSE

Dow

nloa

ded

by [

McM

aste

r U

nive

rsity

] at

01:

23 2

1 D

ecem

ber

2014

measure. Finally, we outline the steps requiredto move beyond this illustration to create a newmeasure of alcohol consumption that will meetthe field’s scientific and public communicationsneeds.

METHODS

The Survey of College Alcohol Norms andBehavior (SCANB) is administered annuallyeach spring to students from 32 four-year insti-tutions of higher education that are participat-ing in the Social Norms Marketing ResearchProject (SNMRP). The SNMRP is a five-yearproject designed to assess the effectiveness ofsocial norms marketing campaigns in reducinghigh-risk drinking (10).

The 32 schools come from all four U.S. cen-sus regions (Northeast, North Central, West,and South). They range in size from approxi-mately 2,000 to 31,000 undergraduates andvary in terms of sector (private vs. public), per-centage of residential students, and studentbodydemographics.Theanalysis reportedhereuses SNMRP baseline data.

Survey Content

The SCANB is a self-administered, volun-tary, and anonymous mail survey that asksabout students’ alcohol-related attitudes, per-ceptions, and behaviors. The baseline SCANBadministered to a first cohort of 18 schools(Study 1) in 2000 consisted of 64 questions onstudent characteristics, alcohol use and its con-sequences, reasons for drinking and abstainingfrom alcohol use, perceptions of campus alco-holnorms,campusandcommunityalcoholpol-icies, and perceived social capital.The baselineSCANB administered to a second cohort of 14schools (Study2) in2001consistedof54 items;the number of questions was decreased toencourage higher response rates.

Alcohol Consumption

The SCANB defined a drink as “a bottle ofbeer (12oz),aglassofwine(4oz),awinecooler(12 oz), or a shot of liquor (1 oz) served straightor in a mixed drink.”

Concerning their own alcohol use, the respon-dents answered four key questions: (1) “Duringthe past 30 days, on how many occasions didyou use each of the following substances–alco-hol (beer, wine, liquor)?” The response optionsfor this scale (scored 1-7) were “Never,” “1-2times,” “3-5 times,” “6-9 times,” “10-19 times,”“20-39 times,” and “40 or more times.” For thenext three questions, a numerical fill-in re-sponse box allowed students to mark in 00 to 99drinks, thus creating an interval-level scale. (2)“What is theaveragenumberof drinks you con-sume in a week?” (3) “When you party, howmany drinks do you usually have?” (4) “Thinkback over the last two weeks. What was thegreatest number of drinks you consumed at onesitting? For how many hours did you drink?” Asecond numerical fill-in response box allowedstudents to report the duration of the drinkingepisode.

To be classified as an abstainer, a studenthad to report no alcohol use in response to allfour consumption questions. A student report-ing alcoholuse in response to one or more ques-tions was classified as a drinker. Students whodid not respond to one of the four questionswere still classified, but those who failed to re-spond to two or more questions were excludedfrom the analysis.

The drinkers’ responses to the four con-sumptionquestionswerecombinedintoaCom-posite Drinking Scale (CDS). Due to the vari-ous question formats, the overall distributionfor each of the four questions was normalized,yielding a z-score for each drinker on eachquestion. A CDS score was then calculated asthe sum of a drinker’s four z-scores. If a drinkerdid not respond to one of the four questions, az-score was imputed using the mean z-score forthe other three. Cronbach’s coefficient alphafor the CDS was calculated to assess its internalconsistency reliability (11). Next, the distribu-tion of CDS scores was divided into quartiles,and each drinker was assigned a score (Q1-Q4)on that basis.

The SCANB also asked respondents howmany times they had four or more drinks at onesitting in the last two weeks, and how manytimes they had five or more drinks. Heavy(“binge”) drinking was defined as the con-sumption of five or more drinks for men andfour or more drinks for women (3,12). Non-

Huang et al. 35

Dow

nloa

ded

by [

McM

aste

r U

nive

rsity

] at

01:

23 2

1 D

ecem

ber

2014

heavy drinkers were respondents who con-sumed alcohol but did not drink heavily.Among heavy (“binge”) drinkers, infrequentheavy drinkers were those who drank heavilyoneor two times,while frequentheavydrinkersdrank heavily three or more times.

Self-Reported Drinking Status

The survey asked respondents to describetheir alcohol use, both presently and duringtheir last year of high school. Response catego-ries for both questions included “abstainer,”“abstainer/former problem drinker in recov-ery,” “light drinker,” “moderate drinker,”“heavy drinker,” and “problem drinker.”

Alcohol-Related Problems

The survey asked respondents how manytimes they had experienced or engaged in 26 al-cohol-related problems during the past 30 daysduetotheirownalcoholuse, includingdrivingacar while under the influence of alcohol, ridingwith a drunk driver, and four problems relatedto academic performance (missing a class, get-ting behind in school work, performing poorlyon a test or important project, and turning in anassignment late). The response alternatives(scored 1-5) were “never,” “1-2 times,” “3-5times,” “6-9 times,” and “10 or more times.”Respondents who reported that they did notdrink alcohol skipped this question.

For CDS-defined drinkers, responses foreach of the 26 alcohol-related problems weredichotomized (dummy variable for having theproblem,withnothavingtheproblemastheref-erence group). The total number of indicatedproblems (up to 26) was calculated for each re-spondent who answered 18 or more items (i.e.,who failed to answer eight or fewer items);those who failed to answer nine or more ques-tions were assigned a missing value on this newvariable. Next, the total was dichotomized,with a dummy variable for having experiencedfive or more problems (and with having four orfewer problems as the reference group). Thiscut-off point, which separated the top one-thirdand bottom two-thirds of the distribution, wasused to facilitate comparisons to previousresearch on college students (12).

Finally, the total number of academic prob-lems (up to four) was calculated for each re-spondent who answered three or four of theitems; those who failed to answer two or morequestions were assigned a missing value on thisnew variable. This total was then dichoto-mized, with a dummy variable for having expe-rienced any academic problems (and with nothaving any problems as the reference group).This cut-off point separated the top one-fourthfrom the bottom three-fourths of the distribu-tion.

Control/Background Variables

The SCANB included 16 control/back-ground variables of interest. With recoding,some response categories were combined. Thevariables included gender (dummy variable formale, with female as the reference group); age(continuous variable in years); race/ethnicity(dummy variables for African American/Black, Asian, Hispanic/Latino, and Other, withCaucasian/White as the reference group);personal relationship status (dummy variablefor married/separated/divorced/widowed, withsingle as the reference group); student status(dummy variable for part-time status, withfull-time status as the reference group); studentclassification (dummy variables for sopho-more, junior, and senior, with freshman as thereference group); and grade point average(continuous variable on a four-point scale).

Additional variables included location ofresidence (dummy variable for on-campus,withoff-campusas thereferencegroup); typeofresidence (dummy variables for fraternity/so-rority house, residence hall/dorm, and other,with house/apartment as the reference group);living situation (dummy variables for livingalone and living with family/other, using livingwith one or more roommates as the referencegroup); parental education (dummy variablesfor one parent and both parents being a collegegraduate, with neither parent as the referencegroup);numberofclosestudent friends (dummyvariables for having none and 1-4 close studentfriends,withhaving5+as the referencegroup);yearof surveyadministration (dummyvariablefor2001,with2000as thereferencegroup);andtiming of survey completion (dummy variable

36 SUBSTANCE ABUSE

Dow

nloa

ded

by [

McM

aste

r U

nive

rsity

] at

01:

23 2

1 D

ecem

ber

2014

for during/after spring break, with beforespring break as the reference group).

Finally, the SCANB also asked about in-volvement in a fraternity/sorority (dummyvariable for Greek member/pledge, with non-Greek member/pledge as the reference group)and involvement in intercollegiate athletics(dummy variable for athlete, with non-athleteas thereferencegroup).Studentsweresaid tobea college athlete if they indicated “intercolle-giate athlete” when asked about extracurricularactivities, or if they indicated spending timeeach week on a varsity athletic team.

Sampling Method

The baselineSCANB was mailed to 300 ran-domly selected undergraduate students at eachof the 32 participating schools, for a total of9,600 students. The baseline survey was ad-ministered at 18 schools in the spring of 2000(Study 1) and at the 14 remaining schools in thespring of 2001 (Study 2). Each school’s regis-trar’s office provided a list of all matriculated,degree-seeking undergraduates, including bothfull- and part-time students. This samplingframe excluded students with out-of-state ad-dresses listed as their current or local address.For each school, the random sample of 300 stu-dents was stratified by class year (freshman,sophomore, junior, senior).

Survey Administration

Each school was put on a separate mailingschedule based on its academic calendar, withthe first survey mailing sent out three to fourweeks after the beginning of the spring semes-ter. Prior to the first survey mailing for Study 1,a “teaser” postcard was mailed out to alert stu-dents that the survey would be arriving soon;this was not done for Study 2 one year later.Thesecond survey mailing was scheduled to arriveapproximately two weeks prior to each school’sspring break. The third was sent out 2-3 weeksafter spring break. The fourth mailing, sent outapproximately two weeks after the third mail-ing, was an abbreviated (two-page) version ofthe SCANB.

Non-respondents received a reminder post-card a few days after the first survey mailingand 2-3 weeks after the second survey mailing.

In addition, project staff made reminder tele-phone calls after the second and third surveymailings to answer questions and encouragecompletion of the survey. Both rounds of re-minder calls involved up to three attempts toreach each student in person. On the third at-tempt, the caller left a message requesting thatthe survey be completed.

The cover letter that accompanied each sur-vey mailing served as the informed consentdocument. Students were told that they werenot required to participate and that they couldleaveaquestionblank if theydidnotwant toan-swer it. To preserve students’ anonymity, noidentifying information was put on either thesurvey instrument or its stamped return enve-lope. Rather, with every survey mailing, stu-dents received a separate postcard with aunique code number on it, with instructions tomail the postcard separately from the survey it-self to indicate that they had completed the sur-vey or did not wish to participate and that noadditional follow-up would be necessary.

A series of monetary incentives was used toincrease response rates. A $1 bill was includedwith the first survey mailing as an up-front in-centive (13). Students who filled out the surveybecame eligible for three types of prize draw-ings: one $100 cash prize per school for stu-dentswhoreturned thesurveywithinoneweek;five$50cashprizesper school for studentswhoreturned the survey by the end of the semester;and one $500 grand prize for one student na-tionwide who completed the survey by the endof the semester. With Study 2, the grand prizewas increased to $1,000 to encourage a higherresponse rate.

The human subjects committees at Educa-tion Development Center, Inc., and all 32 par-ticipating colleges and universities approvedthe study procedure.

Response Rate

A total of 330 of the 9,600 surveyed studentswere removed from the sample. Reasons for re-moval included the following: two or morepieces of undeliverable mail were returned;telephone contact revealed that the student wasno longer enrolled, had already graduated, orwas spending the semester abroad; the registrarverified that non-responding students who

Huang et al. 37

Dow

nloa

ded

by [

McM

aste

r U

nive

rsity

] at

01:

23 2

1 D

ecem

ber

2014

could not be reached by telephone were not en-rolled during the survey period; or participantsindicated on their survey that they were not en-rolled or were enrolled in graduate or profes-sional school.

These adjustments resulted in a final samplesize of 9,270 students. The number of com-pleted surveys was 5,210 for a response rate of56.2%. Response rates for individual schoolsrangedfrom45.3%to71.4%.Of thesurveys re-ceived, 4,858 (93.2%) were full-length sur-veys. The remaining 352 surveys (6.8%) wereabbreviated versions sent in the final mailing.

The reported analyses are based on the full-length survey responses from 4,798 students,after excluding 19 students who failed to indi-catewhether they were enrolledas undergradu-ates and 41 students who said they were formerproblem drinkers in recovery, either presentlyor during high school.

If a respondent failed to respond to two ormore CDS questions, then that case was ex-cluded from all CDS-related analyses; therewere 18 such cases (0.4%), leaving 4,780respondents. If a respondent met additional“missing value” criteria, then that case was ex-cluded from analyses that involved the vari-ables in question. As a result of these missingdata, the reported sample sizes for individualanalyses vary slightly.

Statistical Analysis

All statistical analyses were performed us-ingSAS (14). Descriptivestatisticswereexam-ined for all variables.

Frequency data for the alcohol consumptionvariables were examined for extreme or im-plausible values. When asked to specify the av-erage number of drinks per week for a typicalstudent at their school, four respondents gave aresponsegreater than80drinks formales,whilethree respondents did so for females. Giventheir implausibility, these responses were re-duced to 80 drinks. When asked to report howmany drinks students at their school have whenthey party, 15 respondents gave a responsegreater than 30 for males, while nine respon-dents did so for females. Given their implausi-bility, these responses were reduced to 30drinks. Finally, when asked about the greatestnumber of drinks they had consumed at one sit-

ting during the last two weeks, seven respon-dents indicated that the sitting had lasted morethan 24 hours. These data were declared miss-ing.Thealternativeofreducingtheresponses to24 hours was rejected, as that would also affectthe apparent rate of alcohol consumption.

Assessing the construct validity of the CDSinvolved looking at the relationship betweenthe CDS variable and a set of predictor variablespreviously found to be associated with heavy(“binge”) drinking in the College AlcoholStudy (CAS) (15). These six drinking-relatedcharacteristics were as follows: male; under 24years; Caucasian/White (if respondents identi-fied themselves as “Caucasian/White” only,not mixed with others); drinker in high school(if they identified themselves as a light, moder-ate, heavy, or problem drinker during their lastyear of high school); Greek member/resident(if they indicated “fraternity/sorority memberor pledge” when asked about extracurricularactivities, or if they indicated “fraternity/soror-ity house” as their residence); and college ath-lete (if they indicated “intercollegiate athlete”when asked about extracurricular activities, orif they indicated spending time each week on avarsityathletic team).Note that the race/ethnic-ity variable Caucasian/White was defined some-whatdifferently fromthecorrespondingSCANBcontrol/background variable.

With CDS treated as a cate gorical variable,Cochran-Armitage trend tests were used tocompare the proportions of respondents witheach drinking-related characteristic across CDS/Q1-Q4 drinkers. Second, with CDS treated as acontinuous variable, two-sample t-tests wereused to compare the mean CDS scores of col-lege student drinkers with and without each ofthe drinking-related characteristics.

Construct validity analyses also looked atfour measures of alcohol-related problems ex-periencedduringthepast30days.First, thepro-portions of respondents who drove under theinfluence of alcohol were compared acrossnon-heavydrinkers, infrequentheavydrinkers,and frequent heavy drinkers using chi-squaretests and Cochran-Armitage trend tests; thesame analyses were conducted to compareCDS/Q1-Q4 drinkers. Similar sets of analyseswere also conducted on the proportions of re-spondents who rode with a drunk driver, expe-rienced five or more alcohol-related problems

38 SUBSTANCE ABUSE

Dow

nloa

ded

by [

McM

aste

r U

nive

rsity

] at

01:

23 2

1 D

ecem

ber

2014

(out of 26 possible), and experienced one ormoreacademicproblems(outof fourpossible).

Next, a set of multivariate logistic regressionanalyses were used to examine the relationshipbetween drinking status and these four mea-sures of alcohol-related problems experiencedduring the past 30 days. One set of analysescompared frequent heavy drinkers and infre-quent heavy drinkers to non-heavy drinkers.Another set of analyses compared CDS/Q2,CDS/Q3, and CDS/Q4 drinkers to CDS/Q1drinkers. Both models adjusted for the 16SCANB control/background variables.

Multivariate linear regression analyses werealso used to examine the relationship betweendrinking status and the total number of alco-hol-related problems (out of 26 possible) expe-rienced during the past 30 days. One set of anal-yses compared frequent heavy drinkers andinfrequent heavy drinkers to non-heavy drink-ers. Another set of analyses compared CDS/Q2,CDS/Q3,andCDS/Q4drinkers toCDS/Q1drinkers. Again, both models were adjusted forthe 16 SCANB control/background variables.

Further, the mean, minimum, and maximumvalues of the four CDS items were tabulated byCDS quartile to illustrate theprofileof the“typ-ical” drinker at each CDS level. In addition,ANOVA using Scheffe’s Method was per-formed for each CDS item to make pairwisecomparisons of the means across the four CDSquartiles, applying a significance level of 0.05.In addition, CDS/Q1-Q4 drinking status wascross-tabulatedwith self-reporteddrinkingsta-tus and the measure of heavy (“binge”) drink-ing.

Lastly, Spearman correlation coefficientswere computed between the categorical five/four measure of heavy (“binge”) drinking (1 =non-heavy drinker, 2 = infrequent heavydrinker, and 3 = frequent heavy drinker) andboth the continuous CDS and categorical CDS/Q1-Q4 variables (range = 1, 4).

RESULTS

Background Characteristicsof the Student Sample

The background characteristics of the stu-dent sample (N = 4,798) are as follows: The

sample included more women (61.1%) thanmen (38.9%). The mean age of the students was21.8 years (SD = 5.6); the vast majority ofstudents (85.5%) were under 24 years of age.Respondents could indicate membership inmultiple race/ethnicity categories. Just overthree-fourths of the students (76.3%) wereCaucasian/White. Distribution across otherrace/ethnicity categories was as follows: Afri-canAmerican/Black,6.5%;Asian,10.3%;His-panic/Latino, 6.4%; Native American/Ameri-can Indian, 4.7%; and Native Hawaiian/PacificIslander or Other, 4.7%. Fully 91.1% of the stu-dents were single, with 8.9% being either mar-ried, separated, divorced, or widowed. Justover one-third (34.6%) did not have a parentwho was a college graduate, while 27.2% hadone parent and 38.2% had two parents whograduated from college.

Over nine out of ten students (92.4%) werefull-time. There were slightly more juniors(25.8%) and seniors (28.6%) in the sample thanfreshmen (22.9%) and sophomores (22.7%).The mean grade point average was 3.2 (SD =0.6, A = 4.0). Fraternity/sorority members andpledges constituted 13.2% of the student sam-ple. Approximatelyone in ten students (11.6%)was a college athlete. The majority of students(55.3%) lived off-campus. Nearly six out of tenlived in a house or apartment (58.5%), while37.9% lived in a residence hall, 2.2% lived in afraternity or sorority house, and 1.4% livedelsewhere. Nearly six in ten (59.3%) lived withoneormoreroommates;10.7%livedalone,and30.0% lived with family or indicated “Other.”More than half of the students (55.2%) had fiveor more close student friends, while 39.2% had1-4 and 5.6% reported having none.

Just over half of the sample (52.1%) com-pleted the survey in 2000 (Study 1), with theothers completing it in 2001 (Study 2). Finally,most students (84.8%) filled out the survey be-fore spring break, with 15.2% doing so duringor after spring break.

Composite Drinking Scale

The drinkers’ responses to the four con-sumptionquestionswerecombinedintoaCom-posite Drinking Scale (CDS) by normalizingthe responsedistributionsforeachquestionandaddingthez-scores.Formostof theanalyses re-

Huang et al. 39

Dow

nloa

ded

by [

McM

aste

r U

nive

rsity

] at

01:

23 2

1 D

ecem

ber

2014

ported here, the CDS scores were divided intoquartiles(Q1-Q4).Bydefinition,abstainersdidnot report any alcohol consumption.

Internal Consistency Reliability

The Cronbach’s coefficient alpha was 0.89,indicating that the CDS has high internal con-sistency reliability. The item-total correlationsof the four CDS items ranged between 0.65 and0.81.

Construct Validity

As predicted, six characteristics associatedwith heavy (“binge”) drinking were alsostrongly associated with increased CDS scoresamong drinkers: being male, under 24 years ofage, Caucasian/White only (not combined withother race/ethnicity categories), a drinker inhigh school, a Greek member or resident, and acollege athlete. The mean CDS score for col-lege drinkers with each drinking-related char-acteristic was significantly higher than the cor-responding mean for drinkers without eachcharacteristic (all p < .0001, using two-samplet-tests). Moreover, the percentages of collegedrinkers with each drinking-related character-istic increasedfromCDS/Q1toCDS/Q4;allsixCochran-Armitage trend tests were significant(p< .0001). It is noteworthy that thepercentageofCDS-definedabstainerswhohadeachdrink-ing-related characteristic was always less thanthe corresponding percentage of CDS/Q4drinkers, but not consistently less than the cor-responding percentages of CDS/Q1-Q3 drink-ers. For example, 13.6% of abstainers diddrink in high school, compared to 42.0% ofQ1 drinkers, 59.1% of Q2 drinkers, 71.5% ofQ3 drinkers, and 84.8% of Q4 drinkers. In con-trast, the pattern by gender was as follows:42.2% of abstainers were male, as were 27.3%of Q1 drinkers, 29.6% of Q2 drinkers, 35.7% ofQ3 drinkers, and 59.8% of Q4 drinkers.

Table 1 shows the percentages of non-heavydrinkers, infrequent heavy drinkers, and fre-quent heavy drinkers who drove under the in-fluence of alcohol, rode with a drunk driver,experienced five or more alcohol-related prob-lems (out of 26 possible), and experienced oneor more academic problems (out of four possi-ble) during the past 30 days. As drinking levels

increased, so did the percentages of respon-dents experiencing alcohol-related problems.For example, the percentage of drinkers whoexperienced five or more alcohol-related prob-lems increased from7.7%for non-heavydrink-ers to 43.0% for infrequent heavy drinkers and75.9% for frequent heavy drinkers. All chi-square comparisons and Cochran-Armitagetrend tests were significant (p < .0001). Multi-variate logistic regression analyses were usedto estimate adjusted odds ratios, comparing in-frequent and frequent heavy drinkers to non-heavydrinkersandadjustingfor the16SCANBcontrol/background variables. All of the oddsratios were significant (p < .0001). The rela-tionship between frequency of heavy drinkingand experiencing alcohol-related problemswas clear. For example, adjusting for the 16SCANB variables, infrequent heavy drinkerswere 7.4 times as likely and frequent heavydrinkers were nearly 29.8 times as likely asnon-heavy drinkers to experience five or morealcohol-related problems.

Table1alsoshows theresultsofsimilaranal-ysesusingtheCDS/Q1-Q4measures.Again,asdrinking level increased, so did the percentagesof respondents experiencing alcohol-relatedproblems. For example, the percentage ofdrinkers who experienced five or more alco-hol-related problems increased from 4.4% forCDS/Q1 drinkers to 14.3% for Q2 drinkers,46.8%forQ3drinkers,and77.7%forQ4drink-ers. All chi-square comparisons and Cochran-Armitage trend tests were significant (p <.0001). Multivariate logistic regression analy-ses were used to estimate adjusted odds ratios,comparing CDS/Q2-Q4 drinkers to CDS/Q1drinkers and adjusting for the 16 SCANB con-trol/background variables. All of the odds ra-tios were significant (p < .0001). For example,adjusting for the 16 SCANB variables, CDS/Q2 drinkers were 3.6 times as likely to experi-ence five or more alcohol-related problemscompared to Q1 drinkers, while Q3 drinkerswere 18.3 times as likely and Q4 drinkers werenearly 82.8 times as likely.

Multivariate linear regression analyses wereused to estimate the total number of variousalcohol-related problems experienced by dif-ferent types of drinkers, first comparing infre-quent and frequent heavy drinkers with non-heavydrinkersandthencomparingCDS/Q2-Q4

40 SUBSTANCE ABUSE

Dow

nloa

ded

by [

McM

aste

r U

nive

rsity

] at

01:

23 2

1 D

ecem

ber

2014

with CDS/Q1 drinkers. Both sets of estimateswere adjusted for the 16 SCANB control/back-ground variables. All estimated regression co-efficients were significant (p < .0001). For ex-ample, adjusting for the 16 SCANB variables,infrequentandfrequentheavydrinkersonaver-age experienced 2.9 and 6.1 more alcohol-re-lated problems, respectively, than did non-heavydrinkers (N = 3,240; R2 = .41). Similarly, com-

pared with CDS/Q1 drinkers, Q2-Q4 drinkerson average experienced 1.3, 3.6, and 7.1 morealcohol-related problems, respectively (N =3,260; R2 = .47).

Drinking Levels Reported by CDS/Q1-Q4Drinkers

Table 2 presents the mean, minimum, andmaximum values of each of the component

Huang et al. 41

TABLE 1. Risk of Alcohol-Related Problems Among College Student Drinkers*

A. Comparing Infrequent and Frequent Heavy (“Binge”) Drinkers with Non-Heavy Drinkers†

Alcohol-Related Problem**Non-Heavy Drinkers

(n = 1,573)Infrequent Heavy Drinkers

(n = 1,328)Frequent Heavy Drinkers

(n = 869)

% % Adj. OR (95% CI) % Adj. OR (95% CI)

Drove under the influence ofalcohol

6.5 21.4 4.88 (3.68, 6.46) 31.1 8.00 (5.89, 10.86)

Rode with a drunk driver 4.1 20.1 5.96 (4.31, 8.25) 37.0 14.27 (10.18, 20.02)

Experienced 5 or more differentalcohol-related problems§

7.7 43.0 7.44 (5.84, 9.47) 75.9 29.77 (22.63, 39.16)

Experienced 1 or more differentacademic problems§

9.8 27.2 2.96 (2.34, 3.74) 49.5 7.66 (5.96, 9.84)

B. Comparing CDS/Q2-Q4 Drinkers with CDS/Q1 Drinkers‡

Alcohol-Related Problem**

CDS/Q1Drinkers

CDS/Q2 Drinkers CDS/Q3 Drinkers CDS/Q4 Drinkers

%(n = 956)

%(n = 944)

Adj. OR(95% CI)

%(n = 959)

Adj. OR(95% CI)

%(n = 942)

Adj. OR(95% CI)

Drove under the influence ofalcohol

2.9 9.7 4.43(2.67, 7.35)

23.5 14.33(8.80, 23.34)

33.8 26.22(15.92, 43.19)

Rode with a drunk driver 2.2 7.6 4.30(2.41, 7.70)

21.0 13.61(7.83, 23.66)

38.7 36.58(20.90, 64.00)

Experienced 5 or more differentalcohol-related problems§

4.4 14.3 3.64(2.40, 5.52)

46.8 18.31(12.32, 27.23)

77.7 82.76(54.29, 126.16)

Experienced 1 or more differentacademic problems§

6.0 14.2 2.62(1.81, 3.79)

29.8 6.32(4.44, 8.98)

50.1 16.30(11.36, 23.40)

* Sample sizes vary slightly across type of alcohol-related problem due to missing values. Adj. OR indicates adjusted odds ratio; CI, confidenceinterval. All ORs are adjusted for the 16 SCANB control/background variables; see Methods (Survey Content, Control/Background Variables)for listing of dummy variables and reference groups.

** In total, the SCANB survey asked about 26 alcohol-related problems, including driving under the influence of alcohol, riding with a drunkdriver, and four problems related to academic performance (see below). A problem was counted if it occurred one or more times vs. not at allduring the 30 days preceding the survey.

† Heavy (“binge”) drinking is defined for men as the consumption of 5 or more drinks at one sitting at least once during the two weeks precedingthe survey, and as 4 or more drinks for women. Non-heavy drinkers consume alcohol but do not drink heavily. Infrequent heavy drinkers drinkheavily one or two times in a two-week period, while frequent heavy drinkers drink heavily three or more times. Comparisons of non-heavydrinkers, infrequent heavy drinkers, and frequent heavy drinkers for each alcohol-related problem, using chi-square tests and Cochran-Armitage trend tests, are all significant at p < .0001. All ORs are adjusted for the 16 SCANB control variables, with non-heavy drinkers as thereference group; all are significant at p < .0001.

‡ See Methods (Survey Content, Alcohol Consumption) for details on the classification of Q1-Q4 drinkers based on the Composite DrinkingScale (CDS); CDS/Q4 signifies the heaviest level of alcohol consumption. Comparisons of CDS/Q1-Q4 drinkers for each alcohol-related prob-lem, using chi-square tests and Cochran-Armitage trend tests, are all significant at p < .0001. All ORs are adjusted for the 16 SCANB controlvariables, with CDS/Q1 drinkers as the reference group; all are significant at p < .0001.

§ The SCANB survey asked about four problems related to academic performance: missing a class, getting behind in school work, performingpoorly on a test or important project, and turning in an assignment late.

Dow

nloa

ded

by [

McM

aste

r U

nive

rsity

] at

01:

23 2

1 D

ecem

ber

2014

CDS items across the CDS/Q1-Q4 groups.ANOVA using Scheffe’s Method was per-formed for each CDS item to make pairwisecomparisons of the means across the four CDSquartiles. All of these comparisons were signif-icant (p < .05).

Self-Reported Drinking by CDS Status

Cross-tabulation of self-reported drinkingwith CDS status revealed that approximatelyoneoutof five respondents (19.1%)whoclassi-fied themselves as an “abstainer” reporteddrinking at least some alcohol when answeringthe four CDS questions: Q1 (16.5%), Q2(2.0%), Q3 (0.6%), and Q4 (0.0%). Most of theself-reported “light” drinkers were classifiedby theCDSasQ1(36.1%)orQ2(37.4%)drink-ers,whileaboutoneinfive(19.2%)wereclassi-fied as Q3 drinkers and just less than five per-cent (4.9%) were classified as Q4 drinkers; asmall number of the so-called “light” drinkers(2.3%) were classified by the CDS as abstainers.Among self-reported “moderate” drinkers,nearly half (47.7%) fell into the CDS/Q4 cate-

gory, putting them in the highest quartile; theothers were classified by the CDS as Q1 (1.6%),Q2(10.4%),orQ3(40.8%)drinkers.Almostallof the self-reported “heavy” drinkers (96.6%)were classified as Q4 drinkers.

Comparison of CDS and Heavy (“Binge”)Drinking Measures

The respondents’ heavy (“binge”) drinkingstatus was cross-tabulated with CDS status, asshown inTable3.More thanhalf (54.3%)of thenon-heavy drinkers were CDS/Q1 drinkers,and more than one-third (35.6%) were Q2drinkers. Interestingly, almost one-tenth (9.5%)were Q3 drinkers, putting them in the topone-half of drinkers according to the CDS.Most heavy (“binge”) drinkers were either Q3(36.5%) or Q4 drinkers (42.0%), while almostone in five (17.1%) was a Q2 drinker, puttingthem in the bottom one-half of drinkers accord-ing to the CDS.

Infrequent and frequent heavy drinkers werealso examined separately. Nearly half (47.5%)of the infrequentheavydrinkerswereclassified

42 SUBSTANCE ABUSE

TABLE 2. Distributions of Individual Composite Drinking Scale (CDS) Items: Profiles of CDS/Q1-Q4 Col-lege Student Drinkers*

CDS Item Description CDS Quartile Mean (SD)† Min Max

Frequency of alcohol use in the past 30 days‡ Q1 1.8 (0.5) 1 4Q2 2.7 (0.7) 1 5Q3 3.8 (0.9) 1 7Q4 4.8 (1.1) 1 7

Average number of drinks consumed in a week Q1 0.6 (0.8) 0 5Q2 2.4 (1.6) 0 12Q3 6.2 (3.0) 0 22Q4 17.9 (11.3) 0 80

Greatest number of drinks consumed at 1 sitting in last 2 weeks Q1 0.7 (1.1) 0 6Q2 3.0 (1.9) 0 10Q3 5.8 (2.3) 0 17Q4 11.3 (4.8) 0 30

Average number of drinks consumed when partying Q1 1.7 (1.3) 0 6Q2 3.5 (1.6) 0 10Q3 4.9 (1.7) 0 17Q4 8.8 (3.3) 3 24

* The Composite Drinking Scale (CDS) consists of the four items listed in the table. See Methods Section (Survey Content, Alcohol Consump-tion) for details on the construction of the CDS and the classification of CDS/Q1-Q4 drinkers; CDS/Q4 signifies the heaviest level of alcoholconsumption.

† For each item, ANOVA pairwise comparisons of the means across the four CDS quartiles were executed using Scheffe’s Method. All pairwisecomparisons are significant at the 0.05 level.

‡ Question: “During the past 30 days, on how many occasions did you use each of the following substances–alcohol (beer, wine, liquor)? Donot include drugs used as prescribed by a medical doctor.” The response options (scored 1-7) were: never, 1-2 times, 3-5 times, 6-9 times,10-19 times, 20-39 times, and 40 or more times.

Dow

nloa

ded

by [

McM

aste

r U

nive

rsity

] at

01:

23 2

1 D

ecem

ber

2014

as CDS/Q3 drinkers, while almost one-fourth(24.4%) were Q2 drinkers and more thanone-fifth (22.4%) were Q4 drinkers. Frequentheavy drinkers were predominantly Q4 drink-ers (72.0%), although one in five (19.6%) was aQ3 drinker. Interestingly, just less than 6 per-cent(5.9%)wereclassifiedasCDS/Q2drinkers.

Spearman correlation coefficients showedthat the heavy (“binge”) drinking measure cor-related0.77 with the continuous CDS score and0.74withthecategoricalCDSquartilemeasure.

DISCUSSION

The Composite Drinking Scale (CDS) com-bines fouralcoholconsumptionmeasures,eachof which can capture a different aspect of prob-lem drinking: (1) number of occasions onwhich alcohol was used in the past 30 days, (2)average number of drinks consumed in a week,(3) number of drinks consumed when partying,and (4) thegreatestnumberof drinks consumedat one sitting during the last two weeks. Formost of the analyses reported here, the continu-ous distribution of CDS scores was then di-vided into quartiles to create a simplified four-pointscoringsystem(CDS/Q1-Q4).Basedonalarge sample of students attending four-yearcolleges and universities in the U.S., the CDSwas found to have high internal consistency re-liability and good construct validity, as indi-cated by the strong association between theCDS/Q1-Q4 and characteristics known to beassociated with heavy (“binge”) drinking andreported alcohol-related problems.

The data reported here illustrate several ad-vantages to the CDS over other measures ofconsumption. Clearly, self-reported drinkingstatus is an inadequate measure. About one infour self-reported “light” drinkers had a CDSscore thatput theminthe tophalfof thedistribu-tion (Q3 or Q4), while nearly one-half of self-reported “moderate” drinkers fell in the highestquartile (Q4). Even declarations of being an“abstainer” cannot be fully trusted, as approxi-mately one out of five respondents who so clas-sified themselves reported drinking at leastsome alcohol when answering the four CDSquestions.

Measures of heavy (“binge”) drinking alsoappear to be inadequate for classifying studentsas problem drinkers. Heavy drinking was de-fined for men as having five or more drinks atone sitting in the last two weeks and for womenas four or more drinks (3). A key problem ispossiblemisclassification.About oneoutof tennon-heavydrinkerswasclassifiedasaCDS/Q3or Q4 drinker, which placed them in the top halfof the distribution of all college student drink-ers. At the same time, about one in five heavydrinkers was classified as a CDS/Q1 or Q2drinker,whichplacedtheminthebottomhalfofthedistributionofallcollegestudentdrinkers.

At the root of these misclassifications is thefact that the heavy drinking measure is a singledichotomous measure, whereas the CDS iscontinuous measure composed of four stronglycorrelated items. The correlation between theheavy drinking measure and the continuousCDS score was 0.77. This means that the heavydrinking measure could not explain approxi-mately41percentofthevarianceinCDSscores.

Huang et al. 43

TABLE 3. Comparisons of Heavy (“Binge”) Drinking Status and Behavior-Based CDS Drinking StatusAmong College Student Drinkers*

Heavy (“Binge”) Drinking StatusBehavior-Based CDS Drinking Status–Number (Row %)

Q1 Drinker Q2 Drinker Q3 Drinker Q4 Drinker

Non-Heavy Drinker 854 (54.3) 560 (35.6) 150 (9.5) 9 (0.6)

Heavy Drinker 98 (4.5) 375 (17.1) 801 (36.5) 923 (42.0)

Infrequent Heavy Drinker 76 (5.7) 324 (24.4) 631 (47.5) 297 (22.4)

Frequent Heavy Drinker 22 (2.5) 51 (5.9) 170 (19.6) 626 (72.0)

* See Table 1 for definitions of heavy (“binge”) drinking, non-heavy drinkers, infrequent heavy drinkers, and frequent heavy drinkers. Heavydrinkers include all drinkers who drink heavily, whether infrequently or frequently. See Methods Section (Survey Content, Alcohol Consump-tion) for details on the construction of the CDS and the classification of CDS/Q1-Q4 drinkers; CDS/Q4 signifies the heaviest level of alcoholconsumption.

Dow

nloa

ded

by [

McM

aste

r U

nive

rsity

] at

01:

23 2

1 D

ecem

ber

2014

This same problem would arise, of course, ifanyoneof thecomponentCDS itemswereusedas an individual measure. Note that the item-to-tal correlations of the four CDS items rangedbetween 0.65 and 0.81. The problem is also re-vealed when looking at the range of scores foreach of the component CDS items, brokendown by CDS quartile (see Table 2). In each ofthe CDS/Q1-Q4 groups, for example, therewere some respondents who reported consum-ing zero drinks in an average week. Relying onthis measure alone, some number of actualdrinkerswouldbemisclassifiedasnon-drinkers.

As noted, proponents of the heavy (“binge”)drinking measure have argued that it providesan easy shorthand for identifying problemdrinkers (3). It is therefore instructive to com-pare the predictive utilities of the heavy drink-ing measure and the simplified CDS/Q1-Q4scores. Consider the consequence of experi-encing five or more alcohol-related problems.Comparedwithnon-heavydrinkers, infrequentand frequent heavy drinkers were 7.4 and 29.8times as likely, respectively, to report this num-ber of problems, adjusting for the 16 SCANBcontrol/backgroundvariables.Thecomparableodds ratios for CDS/Q2, Q3, and Q4 drinkers,compared to CDS/Q1 drinkers, were 3.6, 18.3,and 82.8, respectively. Clearly, the CDS is ableto capture a broader range of relative risks thanthe heavy drinking measure, while also moreclearly establishing the quadratic relationshipbetween alcohol consumption and experiencingalcohol-related problems.

This advantageof theCDS is alsoevident forother measures of alcohol-related problems.For example, compared with non-heavy drink-ers, infrequent and frequent heavy drinkerswere 3.0 and7.7 timesas likely, respectively, toreport experiencing one or more academicproblems, adjusting for the 16 SCANB control/background variables. The comparable oddsratios for CDS/Q2, Q3, and Q4 drinkers, com-pared to CDS/Q1 drinkers, were 2.6, 6.3, and16.3, respectively.

Likewise, compared with non-heavy drink-ers, infrequent and frequent heavy drinkers ex-perienced 2.9 and 6.1 more alcohol-relatedproblems, respectively, again adjusting for the16 SCANB control/background variables.Compared with CDS/Q1 drinkers, Q2-Q4drinkers experienced 1.3, 3.6, and 7.1 more al-

cohol-related problems, respectively. Resultsof multivariate linear regression analysesshowed that use of the CDS, as opposed to the“binge”drinkingmeasure, increased theamountof variance accounted for from 41% to 47% (a14.6% relative increase).

Regarding limitations, the SCANB survey,because itwasdesignedtoassess the impactofacampus-based prevention program, did not in-clude the full range of possible consumptionmeasures that might be included in a compositedrinking scale. While the four questions askedhere did create a useful scale, far more work isneeded to identify the optimal question set toapply.

In principle, this type of composite measureshould also be superior to quantity-frequency(QF) measures, which gauge general drinkinglevels but do not include inquiries about heavydrinking episodes specifically. As noted be-fore, some investigators have combined QFquestions with a question about heavy (“binge”)drinking, but that is sub-optimal due to severalproblems with that dichotomous measure (9).A weakness of the present study is that theSCANB survey did not happen to include QFmeasures against which to compare the CDSmeasure. This can also be addressed in futureresearch.

Gruenewald and his colleagues (16,17) havedeveloped a QF measure that asks respondentsa series of questions about their alcohol con-sumption during the past 28 days, first report-ing on how many days they had at least onedrink, then on how many days they had morethan one drink, three or more drinks, and six ormore drinks. From this information, three di-mensions of alcohol consumption can be re-ported: drinking frequency, mean drinks peroccasion, and a drinking variance estimate.Each individual’s drinking pattern is describedas a point in three-dimensional space definedby these variables. Improving on simpler QFmeasures, this scheme captures atypical or ex-treme drinking episodes. However, while thismeasure might have scientific advantages, itdoes not capture the concept of problem drink-ing through a simple, one-dimensional indexandthereforewouldbedifficult for thepublic tounderstand.

Another limitationof thepresentstudyis thatthe study sample included only college stu-

44 SUBSTANCE ABUSE

Dow

nloa

ded

by [

McM

aste

r U

nive

rsity

] at

01:

23 2

1 D

ecem

ber

2014

dents. Remember that the CDS involves nor-malizing the response distributions for eachquestion and adding the z-scores. This means,then, that the subjectivemeaningof theCDS, asa continuous scale, or the CDS/Q1-Q4 scores istied to the population being studied, in this casea sample of students from 32 four-year collegesand universities. Looking at a given study sam-ple, individualscanbecompared tooneanotherin terms of their CDS or CDS/Q1-Q4 scores,but they cannot be compared to members ofother samples.

Correcting this deficiency would require es-tablishing U.S. national norms for the compos-ite drinking scale. Once that were done, a givenCDS score would have real world meaning rel-ative to a standard distribution of CDS scores,just asan IQ scorehasmeaningagainst thestan-dard IQ distribution. For example, peoplewouldcometounderstandthataQ4scoresigni-fies a pattern of heavier drinking (and is morelikely to be associated with alcohol-relatedproblems) than a Q3 score, just as they under-stand that an IQ of 120 signifies greater intelli-gence than an IQ of 110.

Additional research would then be needed todeterminewhichofseveralscoringoptions–theCDS/Q1-Q4 based on quartiles, another point-based system (e.g., one based on deciles), or acontinuous CDS score–would work best. It isvery likely that a CDS/Q1-Q4 score tied to thenational norms would be simple enough for thepublic to understand, but in practice a four-point scale might not be fine-grained enough toserve research purposes.

In conclusion, a composite drinking scaleappears to be a viable measure of college stu-dent drinking with several scientific advan-tages over the heavy (“binge”) drinking mea-sure used in many studies (2). Additional workisneededto identify theoptimalsetofquestionsto include in such a measure, not only for col-lege students but perhaps also for the generalU.S. population. In order to make comparisonsacross study samples, as well as to improvepublic understanding of the data, it would benecessary to establish national norms for thecomposite drinking scale. This would be a ma-jor undertaking, but absent this effort, the fieldwill continue to be saddled with inadequatemeasures that give an incomplete or distortedpicture of problem drinking.

REFERENCES

1. Allen JP, Wilson VB (eds.) Assessing AlcoholProblems: A Guide for Clinicians and Researchers (2nded.). Bethesda, MD: National Institute on Alcohol Abuseand Alcoholism; 2003.

2. O’Malley PM, Johnston LD. Epidemiology of al-cohol and other drug use among American college stu-dents. J Stud Alcohol 2002; S14:23-39.

3. Wechsler H, Austin SB. Binge drinking: Thefive/four measure. J Stud Alcohol 1998; 59:122-123.

4. Flegal KM. Agreement between two dietarymethods in the measurement of consumption. J Stud Al-cohol 1990; 51:408-414.

5. Redman S, Sanson-Fisher RW, Wilkinson C,Fahey PP, Gibberd RW. Agreement between two mea-sures of alcohol consumption. J Stud Alcohol 1987;48:104-108.

6. Ellickson PL, McGuigan KA, Adams V, BellRM, Hays RD. Teenagers and alcohol misuse in theUnited States: By any definition, it’s a big problem. Ad-diction 1996; 91:1489-1503.

7. Dimeff LA, Baer JS, Kivlhan DR, Marlatt GA.Brief Alcohol Screening and Intervention for CollegeStudents (BASICS): A Harm Reduction Approach. NewYork: Guildford; 1999.

8. Fitzgerald JL, Mulford H. Self-report validity is-sues. J Stud Alcohol 1987; 48:207-211.

9. DeJong W. Finding common ground for effec-tive campus-based prevention. Psychol Addict Behav2001; 15:292-296.

10. Perkins HW (ed.). The Social Norms Approachto Preventing School and College Age Substance Abuse:A Handbook for Educators, Counselors, and Clinicians.San Francisco, CA: Jossey-Bass; 2003.

11. Cronbach LJ. Coefficient alpha and the internalstructure of tests. Psychometrika 1951; 16:297-334.

12. Wechsler H, Davenport A, Dowdall G, MoeykensB, Castillo S. Health and behavioral consequences ofbinge drinking in college: A national survey of studentsat 140 campuses. JAMA 1994; 272:1672-1677.

13. Edwards P, Roberts I, Clarke M, DiGuiseppi C,Pratap S, Wentz R, Kwan I. Increasing response rates topostal questionnaires: Systematic review. BMJ 2002;324:1183-1191.

14. SAS. SAS/STAT User’s Guide: Version 8.2.Cary, NC: SAS Institute, Inc.; 2001.

15. Wechsler H. Alcohol and the American collegecampus: A report from the Harvard School of PublicHealth. Change 1996; July/August:20-25,60.

16. Gruenewald PJ, Mitchell PR., Treno AJ. Drink-ing and driving: Drinking patterns and drinking prob-lems. Addiction 1996; 91:1637-1649.

17. Gruenewald PJ, Nephew T. Drinking in Califor-nia: Theoretical and empirical analyses of alcohol con-sumption patterns. Addiction 1994; 89:707-723.

Huang et al. 45

Dow

nloa

ded

by [

McM

aste

r U

nive

rsity

] at

01:

23 2

1 D

ecem

ber

2014