Journalof Studieson Alcohol, Vol. 52, No. 5, 1991

Effects of Minimum Drinking Age Laws on Alcohol Use, Related Behaviors and Traffic Crash Involvement among American

Youth:

1976-1987'

PATRICKM. O'MALLEY, PH.D.,ANDALEXANDERC. WAGENAAR,PH.D.t Institutefor Social Research,Universityof Michigan, Ann Arbor, Michigan 48106-1248

ing age in severalstateson ratesof fatal crashes.Time-seriesresults were comparedwith findingsfrom self-reportdata. The major findings include:higherminimumdrinkingageswere associated with lower levels of alcohol use amonghigh schoolseniorsand recent high schoolgraduates,evenafter multivariatecontrols;lowerlevels of alcoholusewere observedacrossa numberof demographic variables;the lowerlevelsof usepersistedinto the early 20s, evenafter all respondents were of legal age; and loweredinvolvementin alcohol-related fatal crashesamongdriversunder21 appeareddue to loweralcoholconsumption rates--in particular,lessdrinkingin bars

ABSTRACT. This studyhas two separatebut related purposes:(1) to delineatecross-sectional differencesamongU.S. high schoolseniorsand youngadultsthat may be due to variationsin recentyears in state-levelminimumdrinkingage laws and (2) to examinethe effects of recentchangesin minimumdrinking age laws on alcohol consumptionand other relevantattitudesand behaviors.Analyses usedexistingdata collectedby the Monitoringthe Futureproject,an ongoingstudyinvolvingannual,nationallyrepresentative surveysof high schoolseniorsand annualfollow-upsurveysby mail of recent graduates.A separate,coordinatedstudyusedtime-seriesanalysesof officialreportsto examineeffectsof increases in the minimumdrink-

or taverns. (J. Stud. Alcohol 52: 478-491, 1991)

HISRESEARCH hadtwoseparate butrelated pur-

poses: (1) to delineate cross-sectionaldifferences among U.S. high school seniorsand young adults that may be attributed to variations in state-levelminimum drinkingage laws, and (2) to examinethe effectsof recent changesin minimum drinking age laws on alcoholconsumption, alcohol-relatedautomobilecrashesand other health-related behaviors.i

The first purpose is intended to show whether there wereindeedany differencesin drinkingbehaviorsin states with a high versusstateswith a low minimumdrinking age. It has sometimesbeen assumedor assertedthat differenceswere minimal; the presentanalysesare intended to provideempiricaldata on the question,usingas much data as are availableto addressthe issue.The secondpurposeaddresses a ratherdifferentquestion:What happened in statesthat raised their minimum drinking ages in re-

Received:February23, 1990. Revision:July 15, 1990. *This researchwas supportedby researchgrant AA05928 from the National Institute on Alcohol Abuse and Alcoholism.The self-report datawerecollectedunderresearchgrantDA01411fromthe NationalInstituteon Drug Abuse(Lloyd Johnstonand JeraldBachman,principal investigators).

*Alexander C. Wagenaar is withtheDivisionof Epidemiology, School of Public Health, Universityof Minnesota, Minneapolis,Minn.; data analyses werecompleted whilehe waswith the Universityof Michigan's TransportationResearchInstitute. 478

sponseto federalinitiatives(as comparedto stateswith a constantminimum drinking age)? Both purposesare intendedto addressa more generalquestion:What are the effectsof minimumdrinking age laws on the behaviorof youngAmericans? Between 1970 and 1988, there were major fluctuations

in minimum drinking agelawsin theUnitedStates.:' In the early 1970s,manystatesloweredthe minimumagefor drinkingfrom 21 to 18, 19 or 20. All of thosestateshave raised the minimum age in subsequentyears. Some of these raiseswere impelled largely by beliefs that there were significantincreasesin the incidenceof alcoholrelated traffic crashesinvolvingthe newly enfranchised drinkers.A major impetusfor raisingthe age amongremainingstateswas a federallaw passedin 1984 that requiredstatesto haveenacteda minimumdrinkingageof 21 by October,1986,or facethe lbssof a portionof federal highwayfundsbeginningin 1987. Almostall of the statesthat did not havea minimumage of 21 at that time subsequently passedlegislationto raise the age to 21 by 1987,andby 1988all haddoneso (DistilledSpiritsCouncil, 1989).

Althoughby 1988all statesprohibitedpurchase of alcoholic beveragesof any kind by anyoneunderage 21, as recentlyas 1982only 14 statesprohibitedpurchase of alcoholicbeverages of any kind by anyoneunderthe ageof 21, while the others varied considerablyin permitting

O'MALLEY

AND

purchaseof at least sometypesof alcoholto personsbetween the agesof 18 and 20. In thosestateswhere purchaseof alcohol at age 18 was permitted, a significant proportionof high schoolseniorswere eligible. An important questionthat the current analysesaddressis: What effectsdid thesediffering minimumdrinking ageshaveon the behaviorsand attitudesof America's high schoolseniors?The answersto thisquestionhavesignificantpolicy implicationsbecausethe debate about the wisdom of a minimum age of 21 is likely to continue,particularlyin othercountrieswith lower legal ages. The first purposeof the research(to examinecrosssectionaldifferences)is accomplishedby examinationof self-reportdata from annual surveysof high schoolseniors on their own alcohol-relatedbehaviors,primarily consumption.The researchalso providesinformationon the extent to which minimumdrinking age laws affect a broadarrayof othermeasures; morespecifically,thereare measuresavailableon: (1) circumstances or settingof alcohol use;(2) degreeand durationof intoxication;(3) use of otherpsychoactivesubstances; and (4) gradeat first use of alcohol. In additionto the annualsurveysof seniors, subsamples of eachseniorclassare followedup after high schoolgraduation.The datasetthusprovidesan opportunity to investigate theeffectsof differentminimumdrinking age laws on youngAmericansas they make the important transitionsfrom high schoolstudentsto youngadults,and as they movethrougha part of the life cycle duringwhich they are at high risk for alcohol-related problems. The secondpurposeis to look at the effect of changes in minimumdrinkingages.Before-and-after dataare avail-

WAGENAAR

479

creasein the minimum drinking age, and has stimulated numerous questions concerning the effectsof raiseddrinking ages on alcohol consumptionand drinking-driving patternsamongyoungpeople.(See Wagenaar,1983, for a detailed discussionof literature and issues.) Research to

date, however,has not clarified the interveningmechanisms through which the change in law causesreduced crash involvement.For example:Are youth drinking to excesson fewer occasions? Are youth drinking on fewer occasions(with aboutthe sameamountconsumedper occasion)?Are youthdrinkingaboutthe sameamountwith about the same frequency,but less often driving after drinking?The presentstudy is designedto help answer thesequestionsby assessing the effectsof the minimum drinkingage on individualbehaviors,as well as on aggregate crashfrequencies. Method

The self-reportdata for this reportcomefrom the Monitoring the Future project, an ongoing study of young Americans,conductedby the Institutefor SocialResearch at the Universityof Michigan.The studydesignhasbeen described in detail elsewhere (Bachman et al., 1987;

Johnstonet al., 1989); briefly, it involvesnationallyrepresentativesurveysof each high schoolseniorclass, beginningin 1976, plus follow-upsurveysmailedeachyear to a subsetof eachseniorclasssample.A technicalreport that includes much more detail about methods and find-

ings of the presentresearchis availablefrom the authors (O'Malley and Wagenaar,1989).

able in a total of 26 states that raised their minimum

drinking agesbetween1976 and 1987. In addition,effects of changesin minimum drinking ages on official crash statisticsin specificstatesare comparedwith effectson self-reportdata. Previousresearchon minimumdrinking age laws has dealtprimarily with the effect of changesin thoselaws on the frequencyof traffic crashes.There is much less re-

Samplesand surveyprocedures

search on the extent

ing for nearly all of the nonrespondents. Schoolsare requestedto participatefor 2 consecutiveyears;thus, each yearhalf of the schoolsare participatingfor the first time and half are participatingfor the secondtime. When a sampledschoolis unwillingto participate,a replacement schoolis selected,controllingfor factorssuchas population density,geographicregion, size, racial composition

to which

variations

in minimum

drinkingagesamongstatesare associated with differences in individual behaviorsand attitudes. In support of a higher minimum drinking age, it has been argued that, "as a group, peopleaged 18-20 are extraordinarilyprone to... violent crime and other formsof sociallydestructive activity, and it is simply foolishto exacerbatethese tendenciesby legalizing drinking for this age group" (Moore and Gerstein, 1981, pp. 77-78). In the case of motor vehicle crashes,the evidenceis quite clear that thesetendenciesare in fact exacerbatedby drinking. However,as Vingilis and DeGenova(1984) notedafter reviewing the literature, resultsare not as clear in the case of other alcohol-relatedproblems. The researchto date has demonstratedthat significant reductions

in automobile

crash involvement

follow

an in-

A three-stagenationalprobabilitysampleresultsin selfcompletedquestionnaire administrations in about130 high schools(approximately110 public and 20 private) and yields about 17,000 respondentsper year. The response rate is about 83% of all seniors, with absenteesaccount-

and other relevant factors.

Thesesamplingprocedures do not resultin a rigorously representativesample within each of the states in the study.The samplesare drawn so as to be nationallyrepresentative,including all geographicregions, levels of populationdensity,typesof schoolsand so on; more precisely,the designis suchthat the base-yearsamplesare representativeof each of the four major geographic regions (Northeast, North Central, West and South).

480

JOURNAL

OF STUDIES

ON ALCOHOL/SEPTEMBER

Althoughit is not the casethat the samplein anyone state is assuredlyrepresentative of that state, it is the casethat, as one aggregatesacrossstates,one approaches a representative samplefor the aggregateset of states.Thus, for example, we believethat data aggregated acrossthe several stateswith an unchangedminimumage of 21 fairly accurately representall seniorswho live in stateswith a consistentminimum drinking age of 21. Similarly, there are datafrom 21 statesthat increasedtheir minimumdrinking age from 18 to 19, 20 or 21. In the aggregate,this should be a goodsamplefrom which to draw inferencesaboutthe effect that those increases have had.

In addition to the senior-yeardata collection, annual follow-upsurveysare obtainedby mail. From eachgraduating class, 2,400 of thosewho participatedas seniors are selectedfor follow-up and randomlydivided into two equal-sizedgroups.Membersof one group are invited to participatein the first year after graduationand every 2 years after that; those in the other group are invited to participatein the secondyear after graduationandevery2 yearsafter that. This procedureresultsin individualparticipantsbeingsurveyedon a 2-yearcycle, beginningeither 1 or 2 yearsafter graduation.Respondents are paid

$5 for eachfollow-upparticipation.The follow-upsamplesare drawnso as to be largelyself-weighting;however, becausethe primary focusof the originalstudyis on drug use, usersof illicit drugsare over-sampledfor follow-ups (by a factorof threeto one). Weightsare usedin analyses to adjustfor the differentialselectionprobabilities. These follow-up procedureswere initiated beginning with the follow-up of 1978; the 1977 follow-up of the classof 1976 differedfrom all succeeding follow-upsin that respondents were not paid for participation,so the responserate in that year was lower. Other than this exception, responserates generally exceed 75% of the senior-year participantsselectedfor follow-up.Thus, the follow-up samplestypically representat least 63% (75% of the initial 83%) of the initial targetsampleof seniors. An extensivediscussion of the likely effectsof excluding absentees anddropoutsfrom the seniorsurveysis available elsewhere (Johnston et al., 1989).Data from follow-upre-

spondents arealsoweighted to takeaccount of attrition. 3 Drug-use measures

Use of alcoholand of variousillicit drugsduring the respondent'slifetime, the last 12 monthsand during the last30 daysare measured by questions havinga standard, closed-ended format with sevenresponsealternativesas follows: (1) 0 occasions,(2) 1-2 occasions,(3) 3-5, (4) 6-9, (5) 10-19, (6) 20-39 and (7) 40 or more occasions. An additionalquestionaboutheavyuse of alcoholasks respondents how manytimes in the last 2 weeksthey had five or moredrinksin a row; response alternativesrange from 1 (none) to 6 (10 or more times). The "current"

1991

drinking measures(last 30 days and last 2 weeks) are most relevant to the current research and are the focus of

resultspresentedhere. Analysesof fatal crash data Data on fatal motor vehicle

crashes in all states from

1976 through 1986 were obtained from the United States National HighwayTraffic SafetyAdministrationFatal Accident Reporting System (FARS). Box-Jenkinsand BoxTiao (Box and Jenkins, 1976; Box and Tiao, 1975) timeseries methodswere employedto control for long-term trends and seasonalcycles and to estimate changesin crash rates beginning the first month after minimum drinking age laws were changed.At a conceptuallevel, the analytic strategyinvolvesexplainingas much of the variancein each variable as possibleon the basisof its pasthistory,beforeattributingany of the varianceto other variables, such as changesin drinking age laws. The intervention-analysis approachis particularly appropriate for the presentstudybecausethe objectivewas to identify significantshifts in fatalities associatedwith changesin drinkingage laws, independent of observedregularitiesin the historyof each variable. Without thesemethods,incorrectconclusionscan be made. For example,a change

in injuriescould be falsely attributedto a specificintervention, when in fact it is entirely consistentwith a preexistingmulti-yearcyclein fatalities.In short,controlling for baseline trends and cycles with time-series models produces more accurate estimates of the effects of drinking-agelegislation. All time serieswere logarithmicallytransformedto reduceheteroscedasticity beforeparameterswere estimated. All resultspresentedare basedon final modelsthat were carefullyevaluatedto ensure:(1) low correlationsamong parameterestimates,(2) significantnoisemodel parameters meetingrequirementsfor invertabilityand stationarity, (3) insignificant residual autocorrelationsand (4) parsimonious modelsaccountingfor a high proportionof total time-series

variance.

An intervention

variable

was

addedto eachmodelto measurean abrupt,permanenteffect of drinking-law changes.Full discussionof timeseriesmethodsappliedto evaluationof minimumdrinking agepoliciescan be foundin Wagenaar(1983), and more detailson the presentresultscan be found in O'Malley and Wagenaar(1989). Results

Cross-sectional analyses

Minimum drinkingage differences:1976-81. We begin by comparingstates that maintaineddifferent, but unchanged,minimumdrinkingagesthroughoutthe period from 1976to 1981.Table1 providesa listingof minimum

O'MALLEY

AND

TABLE 1. Effective datesof changesin minimumdrinking ages by states, 1976-87

State

Typeof change Effectivedate

Comments

Alabama Arizona Connecticut

19 to 21 19 to 21 18 to 19

10/1/85 1/1/85 7/1/82

Connecticut Connecticut Delaware Florida Florida

19 20 20 18 19

10/1/83 9/1/85 1/1/84 10/1/80 7/1/85

Georgia Georgia Georgia

18 to 19 19 to 20 20 to 21

9/1/80 9/30/85 9/30/86

Illinois Iowa Iowa

19 to 21 18 to 19 19 to 21

1/1/80 7/1/78 9/1/86

Beer & wine G G

Kansas

18 to 21

7/1/85

G, Beer

Maine Maine

18 to 20 20 to 21

10/24/77 7/1/85

Maryland

18 to 21

7/1/82

Massachusetts Massachusetts

18 to 20 20 to 21

4/16/79 6/1/85

Michigan

18 to 21

12/21/78

Minnesota

18 to 19

9/1/76

G

Minnesota

19 to 21

9/1/86

G

Mississippi

18 to 21

10/1/86

Beer & wine

Montana Nebraska Nebraska

18 to 19 19 to 20 20 to 21

1/1/79 7/19/80 1/1/85

G G

New Hampshire New Hampshire New Jersey New Jersey

18 to 20 20 to 21 18 to 19 19 to 21

5/24/79 6/1/85 1/2/80 1/1/83

G G G

New York New York

18 to 19 19 to 21

12/4/82 12/1/85

North Carolina North Carolina Ohio

18 to 19 19 to 21 18 to 19

10/1/83 9/1/86 8/19/82

Beer

Ohio

19 to 21

7/31/87

G, Beer

Oklahoma

18 to 21

9/22/83

Beer

Rhode Rhode Rhode South

18 19 20 18

Island Island Island Carolina

to to to to to

to to to to

20 21 21 19 21

G* G

G G

WAGENAAR

481

TABLE2. Numbersof statesand seniorsin the Monitoring the Future surveyswith consistent minimumdrinkingages, 1976-81

Minimum

1976-8 l

drinking

# of

# of

% of

age

states

seniors

seniors

16 3 10 29

41,569 4,137 33,429 79,135

52.5 5.2 42.2 100.0

Low (18 some) Mixed (19-20) High (21 all) Total

G

constantenvironmentwith enoughsamplecasesfor analysis. Three groupsof statesare distinguished(although only the two extremegroupsare large enoughfor most analysispurposes):(1) those with a minimum drinking ageof 18 for at leastsomealcoholicbeverages throughout

G

the entire 1976 to 1981 interval; (2) those with a mini-

G, Beer & wine

mum ageof 21 for all alcoholicbeverages in that interval; and (3) the remaining stateswith any other unchanged minimumage law. Table 2 summarizesthe overall numbersof participants providingdataby the threecategories. Only Maine and Minnesotachangedminimum drinking age category between 1976 and 1978, while 11 states (Florida, Georgia, Illinois, Iowa, Massachusetts, Michigan, Montana, New Hampshire,New Jersey,Rhode Island and Tennessee)changedtheir minimum drinking age

G

Beer & wine

19 20 21 19

7/1/80 7/1/81 7/1/84 1/1/84

South Carolina South Carolina South Dakota

19 to 20 20 to 21 18 to 19

1/1/85 9/14/86 7/1/84

Tennessee Tennessee

18 to 19 19 to 21

6/1/79 8/1/84

G

Texas Texas Vermont

18 to 19 19 to 21 18 to 21

9/1/81 9/1/86 7/1/86

G

Virginia Virginia Virginia Washington,D.C. West Virginia West Virginia

18 to 19 19 to 21 20 to 21 18 to 21 18 to 19 19 to 21

7/1/81 7/1/85 7/1/87 9/1/86 7/1/83 7/1/86

Wisconsin Wisconsin

18 to 19 19 to 21

7/1/84 9/1/86

category between 1979and1981. 4 Resultsshowclearlythat statesallowing18-year-olds to purchasealcoholhave higher rates of alcoholuse among high schoolseniorsthan statesthat limit purchaseto those aged•21 or over. Figure 1 showsthe data for the 7categorymeasureof alcoholuse in the last 30 days;the vertical scale in Figure 1 representsapproximatelyone standarddeviation (1.6). As can be seen, the differences,

Beer & wine Beer & wine Beer & wine Beer

while perhapsnot large in absoluteterms, are by no means inconsequential.Combined across the entire 6 3.4'

•3.0 ................................................................................................................. Beer G

2.8................ 2.6...............

G, Beer & wine 2.4.................................................................................................................

2.2............................................................................................................... G

* G = Grandfather clause included.

Note: The followingsourcesof informationwere used:Bonnie (1985); Distilled SpiritsCouncil(various);DuMouchelet al. (1987); Insurance Information Institute (various); National Safety Council (1985); Wagenaar(1983).

1.8 1•76 1•77

1•78 1•79

1•80 1•1

Year of Data Collection

l-a-MDA: 18• MDA: 21] drinking age changesduring the 1976 to 1987 interval. Use of the shorter1976-81intervalprovidesa relatively

F•GURE 1. 81

Alcoholuse(30-daymean)by minimumdrinkingage, 1976-

482

JOURNAL

OF STUDIES

ON ALCOHOL

years, the mean for the 18-age statesis 2.834 and the mean for the 21-age states is 2.605; the differenceof 0.229 is about 14% of a standarddeviation. (The data from the third groupof states[thosestatesallowingsome alcoholpurchasefor 19-20 year-olds]are much less stable, becauseof the relativelysmallnumbersof cases,and are thereforenot displayedin the graph.) Minimumdrinkingage differences:1976-87. An important issueis whetheroveralltrendsin alcoholuseduring the period from 1976 to 1987 are attributable to the changingminimumdrinking ages.This issueis relevant becausea gradual decline in alcohol use was observed amonghigh schoolseniorsnationallybetween1980 and 1987 (Johnstonet al., 1989, p. 12), and a naturalquestion is whetherthe decline is attributableto the changesin minimumdrinkingage laws, or to other factors. In order to addressthis question,we categorizedall statesaccordingto their recent(1976-87) history of minimum drinking age law. The categorizationresultedin two conceptuallyimportantgroupsof states:(1) statesthat changedfrom 18 to 21 during the 1976-87 interval--an alternativeway of statingthe criterionfor inclusionin this categoryis that it includesall statesthat allowedpurchase of at least some form of beveragealcoholat age 18 at somepoint in the 1976-87 interval; and (2) stateswith a minimum age of 21 throughout.The numberof respondentsin the "18 at sometime" groupis at least8,500 per year and in the "constant21" group is 4,000 or more. The residual category, all others (most of these were statesthat changedfrom 19 or 20 for somealcoholto 21), is very small, with as few as 1,220 respondents per year; consequently, the data for the "all others" categoryare understandably unstableand are thereforenot considered

/ SEPTEMBER

1991

standard deviation (1.6). From these data, one can see

clearly that from the mid-1970sthroughthe early 1980s there were differences

between states with

a minimum

drinkingage of 18 at sometime and stateswith a constant minimum age of 21. Since the early 1980s, all of the former stateshaveraisedtheir minimumdrinkingage to 21. As can be seen,the differencebetweenthe two groups of stateshas been essentiallyeliminated. This provides additionalevidencethat the minimumdrinking age does havean effect on the frequencyof drinkingby high school seniors.Note for examplethat there is a distinctlyhigher rate of drinkingin 1979amongschoolslocatedin statesthat had a minimum drinking age of 18 at some time, compared to the schoolsin the constant-21states.By 1986, this differencehas all but disappeared.Becauseexactly the samestatesare involved in both years, the difference is highly unlikely to be due to other "cultural" factors associatedwith differencesin minimum drinking ages. Figure2 showsthat the time trendsin alcoholusein the constant-21statesare not monotonic.There was a slight rise in the late 1970sthrough1981, then a gradualdecline in the early 1980s, followed by a stabilizationfrom 1985 to 1987. These data make it clear that the overall declines

in alcohol use that were observedin the early to mid1980swere not attributablesolely to changesin stateminimum drinkingage. In particular,the fact that therewere substantial declines in the constant-21

states refutes that

possibility.It is also the case, however,that someof the overalldeclineappearsdue to the effect of changesin the laws, because the states that increased their minimum

drinking agesshowedlarger declines. Multivariate

controls. These cross-sectional

differences

observedbetweenstateswith varying minimumdrinking agesare potentiallyconfounded by differentialpopulation further. characteristics. Variations in minimum drinkingagescould Figure 2 displaysmeansfor the measureof 30-day alreflect these more fundamental differences. For example, cohol use; the vertical scale is again approximatelyone stateswith high levelsof religiosity,on average,mightbe more inclinedto prohibit alcoholto youngerindividuals, 3.4 and the level of religiositymay be the more fundamental determinantof teenagers'drinking. The data in Figure2 argue againstthis position, becausethe same statesare •'--_. 3.0 ................................................................................................................

.3.2J ................................................. involved and the differences

197619771978197919801981f982 19831984198519861•87 Year of Data Collection

-a-MDA:18 atSome'13me--a=MDA:21 Throughout J FIGURE2. 1976-87

Alcohol use (30-day mean) by minimum drinking age,

in alcohol use diminished

fol-

lowing changesin states' laws. Obviously, changesin stateminimumdrinking age laws imposedby federalaction would not result in changesin such fundamental characteristics. Nevertheless,it may be usefulto address the issueas to whether the law is an importantdeterminantof behaviorin anotherway, specifically,by controlling statisticallyfor selectedindividual-levelcharacteristics, includingreligiosity,truancy, race and sex, to determine if minimumdrinking age categorystill "explains" some variance.

Resultsof suchanalysesconfirm that minimumdrinking age laws make a difference.Even after controllinga number of relevant variables, the states with a minimum

drinking age of 18 show higher alcohol use. The effects

O'MALLEY

AND

T^BLE3. Regressionanalysis:Backgroundvariablesrelatedto alcohol usein high schoolclassesof 1976-81 in stateswith no changein minimumdrinkingage Prevalence of

Prevalence of

alcoholuse in

heavydrinking in

Independent

last 30days last 2weeks

variables

B"

Sex (M = 1, F = 2) Race(W=0, B = 1)

-0.086 -0.188

I•b -0.094* -0.143'

College plans(No = 0, Yes = 1) -0.030 -0.033* # parentsin household(0,1,2) -0.005 -0.007 Parental education(10-60) 0.002 0.051' Religiosity(10-40) -0.010 -0.199'

B"

I•b

-0.188 -0.178

-0.193' -0.128'

-0.081 -0.082* -0.015 -0.017 0.000 0.013' -0.009 -0.164'

WAGENAAR

483

measureas dependentvariable leads to similar conclusions, with a higher proportionof varianceexplained.) Similarly,the corresponding coefficientpredictingprevalenceof heavydrinkingis .028, implyinga 2.8% higher prevalenceof that behavior,after controls.These effects are only very slightly--if at all--smaller than the zeroorder effects (i.e., the effectsbefore controlsfor other fac-

tors). Clearly, the observedminimumdrinkingage effect

is notonethatis easilyexplained by otherfactors. 7

'•Thevaluesin this columnare unstandardized regression coefficients.

Althoughthe total amountof explainedvariance may seemsmall, the effectsof minimumdrinking age are far from trivial. To put them in more concreteterms, there are (very roughly)about9 millionyoungAmericansaged 17-19;the 2.8% differencein prevalenceof heavydrinking translatesto about 252,000 fewer Americanshaving had five or more drinks in a row in the prior 2-week period. The 5.6% differencein 30-dayprevalencetranslates to over half a million fewer youngAmericanshavinghad a drink in the prior 30-days.These effectsare substantively significantby any standard.(See Abelson, 1985,

øThevalues in thiscolumn arestandardized regression coefficients.

and Rosenthal, 1990, for discussionsabout the substantive

CDummyvariableswereusedfor region,andthereforeone region(North

significance of effectsthat explainvery little variance.) Minimumdrinkingage differences:Post-high school.It

Region" South ( = 1) Northeast ( = 1) West ( = l) Urbanicity(1-5) MDA: 18 (= 1) MDA: 19-20 (= 1) Constant

Percentvarianceexplained

-0.029 -0.031 0.023 0.021 -0.084 -0.075*

-0.036 -0.018 -0.099

-0.035* -0.015 -0.085*

0.020

0.048*

0.009

0.019

0.056 0.024

0.060* 0.013

0.028 0.008

0.029* 0.004

1.017

0.996

10.4%

11.1%

Central) was excluded. *p < .05 (for both standardizedand unstandardized coefficients).

Notes:The weightednumberof casesis approximately 66,000. Even assumingan unlikely large designeffect of 11, the effectiveN would be more than 6,000; the value used in calculatingsignificancelevels is 6,000. (A large numberof designeffectshave beencalculatedfor various statistics;a very few have been as large as 11.)

are perhapsnot large, but the data are consistent,as shownin Table 3, which displaysthe resultsof multiple linearregressions. For this phaseof the analysis,a correlation matrix was computed,basedon the data collected in 1976 through1981 in thosestatesthat did not change their minimum drinking age laws throughoutthe period.

Theresulting number of casesis 71,319.5 Because of the very large numberof cases,almostall regressioncoeffi-

cientsweresignificant atconventional significance levels. 6 The variablesselectedfor controllingare known,basedon previousresearch(Bachmanet al., 1981), to be relatedto alcoholuse. Theseincludesex, race, collegeplans, numberof parentsin thehousehold, averageparentaleducation, religiouscommitment,regionof countryand population density (or urbanicity); see Bachmanet al. (1981) or O'Malley and Wagenaar(1989) for definitionsof variables.The major finding can be statedvery succinctly; even after controllingfor a numberof importantdemographicfactors,minimumdrinkingage remainsa significant predictor--substantively, as well as statistically--of alcoholuse(Table 3). The unstandardized regression coefficient predictingprevalenceof 30-dayalcoholusefor individualsin a state with a minimum drinking age of 18 was .056. This canbe interpretedas sayingthat thereis a 5.6% higherprevalence(on an absolutescale)even after controllingall the listed factors.(Prevalencefiguresare providedfor easeof interpretation; use of the 7-category

is conceivablethat minimum drinking agesaffect the behaviorof high schoolstudentsmore than they affect the behaviorof older adolescents. In particular,someobservers have indicatedthat alcoholuse is so pervasiveamong college studentsthat minimum drinking age laws would haveor havehadvery little effecton their drinkingbehaviors (Georgeet al., 1989;Hughesand Dodder,1986). Accordingly, our first examinationof the follow-up data involved college-agerespondents.As reports from the Monitoringthe Futureprojecthaveindicated(Johnstonet al., 1989), alcoholuse is higher amongcollege students than amongotherhigh schoolgraduatesof similarage not in college. (This contrastswith most illicit drug use, whichis loweramongcollegestudents thanamongpeople of the sameage who are not in college.) Almost all college studentsare eligible to purchasealcoholif the minimum drinkingage is 18 or 19, but mostare not eligibleif the minimumdrinkingage is 21. Figure 3 displaysmean 30-day alcoholuse for follow-up respondents who were collegestudents as of 1 to 4 yearsafter high schoolgraduation.(Collegestudents are heredefinedas thoserespondents1 to 4 yearsafter high schoolgraduationwho report

attendinga 2- or 4-yearcollegefull-time in March of the surveyyear. These data are availableonly since 1980, when a minimum of four previoushigh schoolclasses werefirst includedin the follow-upsurveys.)Collegestudents are divided into two groups:those who, as high schoolseniors,were residingin stateswith a minimum drinkingage of 18 for at least somealcoholversusthose residingin stateswith a minimumdrinkingage of 21. Collegestudents who werehigh schoolseniorsin states with a minimumdrinkingage of 18 do indeeddrink more while in collegethan their counterparts who were high

484

JOURNAL

OF STUDIES

ON ALCOHOL

3.6

2.2

1980 1981 1482 1483 1484 19'85 19•6

1487

Year of Data Collection



MDA:18 as HS Senior --,a- MDA:21 as HS Senior

FIGURE3. Alcohol use (30-day mean) among college students1-4 yearsafter highschoolby minimumdrinkingagewhenrespondent wasa high schoolsenior

schoolseniorsin stateswith a minimum drinking age of 21. Similarly, othergraduatesof the sameage not attending college also drink more on averageif they were seniors in a state with a minimum drinking age of 18 (Figure 4). Unfortunately,it is not possibleto ascertaindefinitively where the follow-uprespondents were residingin the interval for which they were reportingtheir alcoholuse for the follow-upsurveysprior to 1986. If we .assumethat the majority of respondentsremain within their state of residencein high school,thenthe abovecomparisons are reasonable. Some "noise" or error is introduced, but it is likely not a systematicbias. It is possiblethat a higher proportionof collegestudents,as comparedto noncollege

/ SEPTEMBER

1991

students,were living in a differentstatethan the one they were in as high school seniors, but that should only dampen any differencesby introducingmore error. (We assumethat choiceof collegeis not muchaffectedby the minimum drinking age of the states involved.) In any case,mostof the differencesdisappeared by 1986-87,when virtually all stateshad a minimumdrinkingage of 21. An additionalquestionof considerableinterestis whether lower rates of drinking observedamong high school seniors who reside in stateswith a high minimumdrinking age disappearwhen those individualsreach age 21. This questionwas addressedby separatingrespondentsfrom the classesof 1976 to 1981 into two categories:(1) those who were residingin stateswith minimumdrinkingage of 18 throughoutthe period and (2) thosewho were residing in stateswith minimumdrinking age of 21 throughout(all otherswere excludedfrom analyses).Then, reportsof alcoholuse at ages21 through25 were comparedfor these two groups.The follow-updata were collectedfrom 1979 through 1988. (In this case, unlike with the college-age respondents,it makeslittle differencewhere the respondentsresideat follow-up,becauseall are enfranchised.) Perhapssurprisingly,the data show that there appears to be a lingeringeffect; evenafter everyonehas achieved eligibilityto purchaseall typesof beveragealcohol,those who werepreventedby law from suchpurchasebeforeage 21 appearto drink slightlylessalcohol. Figure5 displays the data for mean 30-day alcohol use; note that the horizontalaxis is age at administration,as opposedto year of administration,which was used in previousfigures.The data pointsfor the 18-yearolds are the meansacrossall respondents in the combinedclassesof 1976-81 measured in senioryear of high school,separatedaccordingto state minimumdrinkingage as of base-year(age-18statesversusage-21states).The datapointsfor age21 at administra-

3.6'

3.2.................................................................................................................

3.0 ................................................ 'z.......... 'z.......... ........... ....... 2.8,

2.6 .................................................................................................................

•2.2

2.4 .................................................................................................................

2.0

22. 1•0 19•1 1•?. 1•k3 1• Yc•

1•

1•6 1•7

of Daa Coll•6on

1.8 1'8 1'9 •0

'•1



:L:93i4

::;5

Ageat DataColleclion

-a-MDA:18 asHS Senior • MDA:21 asHS Senior FIGURE 4. Alcoholuse(30-daymean)amongnon-college students 1-4 yearsafterhighschoolby minimumdrinkingagewhenrespondent wasa high schoolsenior

[--a-MDA:18 asHS Senior • MDA:21 asHS Senior ] FIGURE5. Alcoholuse (30-day mean)by age and minimumdrinking age(classesof 1976-81 followedup in 1979-87)

O'MALLEY

AND

tion are meansacrossall follow-uprespondents (separated by base-yearminimumdrinkingage) from the classesof 1976-81participating3 yearsafter graduation;this would be in 1979 for the class of 1976, 1980 for the class of 1977 and so on. a

Incidentally,Figure 5 indicatesa decline in use with age after age 21 or 22. But there is a potentialproblem with this apparenteffect: the age data are confoundedby year (or by cohort). For example,the groupsproviding data at age 21 were followedup in 1979 through1984, whereasthe groupsprovidingdata at age 25 were followedup in 1983through1988. If usewere lowerin general in the later yearsof the 1979to 1988period, thenone might expectto see a decline in use by age in Figure 5, even if there were no real "age" effect. For presentpurposes,of course,we are not trying to determinewhether thereis an age effect;instead,we are comparingtwo differentgroupsof respondents from the sameclassesin the sameyear of data collection, and thereforeage effectsare not a problem. In other analyses,however,we have demonstratedby using more completedata that in fact there probablyis a real ageeffect (i.e., a declineafter age 21 or 22) as suggested in Figure 5 (O'Malley et al., 1988). A point to note in the data displayedfor ages 21 through25 is that we are dealingwith muchsmallernumbersof caseshereas comparedto analysesof high school seniors;it is reassuringto seethat the findingsare consistent acrossthe differentclassesand slightlydifferentsets of states.(Weightednumbersof respondents in all cases are greater than 1,200.) Effectsof changesin minimumdrinkingage laws In the nextphaseof analyses,we examinedthe effect of law changesin groupsof statesthat changedtheir minimum age during the study period. Groupsof stateswere necessarybecausesamplesof schoolsin individualstates did not provide sufficientlystable data to allow accurate estimates.We comparedalcoholusein the samestatesbefore and after minimumdrinkingage law changes.Then, we incorporatedadjustmentsfor seculartrendsand other potentiallyconfoundingvariables. We report here data primarily for statesthat changed from a minimumdrinking age of 18 to a higherage (19, 20 or 21). An increasefrom a minimumage of 18 should be particularlyimportantfor high schoolseniors,mostof whom are under 19. Grandfatherclausescomplicateinterpretationof theseanalyses;nevertheless, by groupingseveral yearsprior to the changeand lookingat differencesa year or more subsequentto the change,importantimmediate effectsshouldbe evident.In the absenceof any immediate effect, there is an alternative explanationthat drinkingpatternscould havebeenestablishedwell before the end of senioryear, and thus effects of changedlaws would not showup in seniorclassesuntil someyearslater.

WAGENAAR

485

The fact that we did observe immediate

effects alleviates

this potentialproblem. Comparisonof alcohol use beforeand after minimum age change.Combinedacrossall statesthat increasedthe minimumdrinkingage from 18 (to 19, 20 or 21), there wasa 13.3%decreasein drinkingin the past30 days(expressed as a percentof the total standarddeviation,which is 1.6; seeFigure6). Furthermore,the decreaseappeared immediatelyafter the changein law, with very slightad-

ditionaldecreases in the succeeding 2 years. 9 For those who questionedwhethera changespecificallyfrom a minimum drinkingage of 18 to 21 wouldhaveany important effect,the data are particularlyclear:Therewas a decline of

28.2%

of

a standard

deviation

in

alcohol

use--

certainly animportant effect. •o Differentialeffectson subgroups.One of the areasthat the presentresearchdesignwas uniquelyable to address is that of differential effects of minimum drinking age changeson different subgroupsof seniors.For example, one might hypothesizethat minimumdrinking age laws would have strongereffectson rural high schoolstudents comparedto urbanstudents,perhapsbecauseof easieralcoholavailabilityin urbansettings.We examinedtrendsin alcoholuse for selectedsubgroupsover time, in order to gain a clearer understandingof the nature of differences. Subgroupswere stratifiedon the basisof sex (male, female); race (white, black); college plans (yes, no); religious commitment(low, high); and populationdensity (large SMSA, other SMSA, non-SMSA). Findings were generallyas one would expect,if one assumedthat minimumdrinkingage effectswouldbe fairly well distributed throughoutthe populationof high schoolstudents;that is, the size of the effectsis proportionalto the amountof alcohol use before the minimum drinking age increase. Thus,for example,white seniorsreportmuchmorealcohol 3.4'

•3.0 ..................................................................................................................

2.8................................................................................................................. 2.6.......................................................................

22............................................................................................................. 2.0 ................................................................................................................ 1.8

-3

:2

-h

i

k



Years Before and After MDA Was Raisc•l

I o Before MDA Raised • After MDA Raised I FIGURE6. Alcoholuse(30-daymean)beforeandafter minimumdrinking age was raised

486 TABLE 4.

JOURNAL

OF STUDIES

ON ALCOHOL

/ SEPTEMBER

1991

Mean alcoholuse(30-day) 3 yearsbeforeand after minimumdrinkingage was raised,by background and demographicfactors Mean alcohol use

Independent variables

Number ofcases

past 30days

Standard

Deviation

% standard

deviation

Before

After

Before

After

(all 6 years)

change

12,045 12,682

I l, 196 12,187

3.181 2.636

2.945 2.447

1.698 1.485

- 13.9% - 12.7%

20,844 2,752

19,444 2,618

3.025 1.927

2.809 1.867

1.604 1.312

- 13.5% -4.6%

10,778 12,809

9,450 13,042

3.060 2.783

2.800 2.599

1.700 1.524

- 15.3% - 12.1%

11,497

10,472

3.245

2.992

1.651

- 15.3%

13,223

12,985

2.606

2.438

1.525

- 11.0%

Sex

Male Female Race

White Black

CollegePlans Noncollegebound Collegebound Religiouscommitment Low

High Populationdensity LargeSMSA Other SMSA Non-sMSA

4,824

4,818

3.116

2.897

1.602

-- 13.7%

13,027 7,246

11,717 7,585

2.848 2.877

2.648 2.632

1.610 1.642

-- 12.4% -- 14.9%

use than do black seniors, so minimum drinking law changesshould--and do--show larger effects for white seniors;similarly, seniorsreportinga low degreeof religiouscommitmentshowedmore of a decreasein alcohol use after the increasedminimum drinking age compared to thosereportinga high degreeof commitment.Table 4 indicatesthe percentof a standarddeviationdifferencein 30-day meanfrequencyof drinkingfor the 3 yearsbefore and after a minimumdrinkingage increasefrom 18 to any higher age for the varioussubgroups.Overall, the differencesbetweensubgroupsare quite modestin size; there do not appearto be any major differencesby subgroup beyond that which would be expectedon the basis of amount of use.

Law change effects on variables other than alcohol use.In additionto recentalcoholconsumption,we examined measuresrelated to: (1) circumstances or settingof alcohol use; (2) degreeand duration of intoxication;(3) useof otherpsychoactive substances; and(4) gradeat first use of alcohol. These other measureswere analyzedthe sameway the consumptionitems were, with analysesrestrictedto statesthat changedtheir minimum age laws from 18 to some higher age, and to the 3-year periods immediatelybefore and after the change. O'Malley and Wagenaar(1989) providemoredetailedresults;the major findingsare summarizedhere. ß Circumstancesor settingsof alcohol use. Respondentswere askedhow oftenthey usedalcoholduringthe pastyear in certain settings(e.g., when they were alone, at home, at school, etc.). Noneof thesesettingsshoweda significantshift following the increasein minimum drinking age. One particularly interestingsettingaskedabout is drinking in cars. Some observers have been concernedthat increasingthe minimum drinking age might lead to somedisplacement of drinking by under-agepeoplefrom more public placessuchas bars and

tavernsto more private placeslike cars. But the data showno evidenceat all for an increasein the frequencyof drinkingin cars;therewas in fact a very slightdecrease(4% of a standard deviation)observed in the 3 yearsafter 18-year-olds weredisenfranchised comparedto the 3 prior years.One behaviorrelated to alcoholuse did showa dramaticeffect: the frequency of going to bars or tavernsdecreasedsharply(by 31% of a standarddeviation).The conclusionto be drawn is clear: High school seniors drink more in bars and taverns when the mini-

mum drinking age is 18. They drink much less in bars and tavernswhenthe minimumdrinkingage is higher. Degree and durationof intoxication.One putativeeffectof a lower minimumdrinking age has been that young drinkers learn to drink more responsibly. Perhaps,therefore,a higher drinkingage couldlead to more intoxicationamongunder-age drinkerswhenthey do drink. Basedon seniors'self-reportsas to how high they usuallyget and how long they usuallystay highwhentheydrink alcohol,thereis no effectat all of minimumdrinkingage--therewas essentially no shiftfollowing an increasedminimumage. (Eta valuescomparingthe means

for 3 yearsbeforeand3 yearsafterare .004and.002.)II Useof otherpsychoactive substances. Onepossibleeffectof an increased minimumdrinkingagefearedby somewasa shiftin usefrom alcoholto otherpsychoactive substances, particularly marijuana.In fact, we observed a declineof 11%of a standard deviationin mean30-daymarijuanausefollowingincreases in minimumdrinkingages.Much, but not all, of thisdeclinecan be attributedto the seculartrendfor marijuana,whichdeclined in prevalence throughout the 1980s.Butevenaftersubtracting out the decline (based on data from the constant-21states), thereremainsa furtherdecreasein marijuanausefollowingthe law change. This decreaseis not inconsistent with anotherpossibleindirect effectof a higherminimumdrinkingage, that is, a decrease in use of illicit substances. To the extent that alcohol

functionsas a gatewaydrug, one that facilitatesuseof other

pgychoactive substances, a declinein alcoholuseshouldproducelessuseof otherpsychoactive substances suchas marijuana.Becauseof the generaldeclininguseof marijuanaand

O'MALLEY

AND

mostotherillicit substances throughoutthe 1980s,it wouldbe difficult

to assert a causal connection with the decrease in al-

WAGENAAR

TABLE 5. Effects of increasedminimum drinking age on alcoholinvolvedfatal crashes:Time-seriesmodelingresultsusingFARS data

cohol use. But it is clear that there was no increase in use of

illicit drugscoincidentwith the declinein alcoholuse. Ageof onsetof alcoholuse.Therewasa slightdecreasein the averageageof onset(or gradeat first use)after the increasein the minimumdrinkingage, a directionthat is not consonant with increaseddrinkingby youngeradolescents whenthereis a lower minimumdrinking age. Three yearsmay not be a long enoughtime period for an effect on age of onsetto be manifested,but comparisonsof cross-sectional differences(instead of before-aftercomparisons) alsofail to showan earlierageof onsetwith a lowerminimumdrinkingage.Forexample,in the 1976-81intervaltherewasa nonsignificantly higherageof onset amongthe statesthat permitted18-year-oldsto drink comparedto statesthat requiredage 21, and the sameis true for the shorter 1979-81 interval. This latter interval, 1979-81, is

particularlyimportantbecauseby then moststateswith a low minimumdrinkingage had permitted18-year-oldsto drink for a numberof years.Elevenof the 16 age-18statesthatprovided data in the 1979-81periodhad permittedalcoholpurchaseby 18-year-oldsas far back as 1969, and the other five had permitted suchpurchaseat least as early as 1974 (Bonnie, 1985).

Effectsof minimumage laws onfatal traffic crashes

A separateportionof this researcheffort was a coordinated analysis of official statisticson fatal automobile crashes, using the Fatal Accident Reporting System (FARS). We examinedboth self-reportand crashdata for a subsetof stateswith crash data availablefor 3 years before and after a change. Thirteen statesmet this requirement:Delaware, Florida, Georgia, Illinois, Maryland, Massachusetts,Michigan, Nebraska, New Jersey, Ohio, Oklahoma,Tennessee and Texas.Our objectivewas to comparefindings from analysesof crasheswith selfreport data for the samestatesduring the sametime period, to determine whether declines in crashes can be attributed to less use of alcohol.

Effectsof law changeson fatal crasheswere evaluated in terms of frequencies(unadjustednumbers of fatal crashes)and rates (numbersof fatal crashes,adjustedfor the relevantnumberof drivers). The rate is the more appropriatemeasureand is the focusof our discussion.Table 5 provides estimated effects on rates for four categoriesof crashes:(1) the mostimportantcategoryfor present purposes is single-vehicle nighttime (SVN) crashesinvolving a driver under 21 years old; (2) SVN crashesinvolvinga driver 21 yearsold or older; (3) daytime crashesinvolving a driver under 21 years old; (4) daytimecrashesinvolvinga driver21 or over.Becausethe involvementof alcohol in motor vehicle crashesis very difficult to measure, and because its measurement is

highly variable over time and acrossjurisdictions,SVN crashes

are

used

as an

indicator

for

alcohol-related

crashes.See Wagenaar(1983, pp. 42-43) for a more ex-

487

Single-vehiclenighttime

Typeof change

Daytime

Less than

21 and

Less than

21 and

21

older

21

older

Rate of crashes

Aggregate

- 15.4'

- 13.9'

- 13.3'

3-Year 2-Year l-Year

-26.3* - 18.6' -21.6*

- 17.7' -9.9* 0.6

-5.4

-30.8* -26.7* 2.1

- 15.5' - 13.5' - 13.0'

Grandfathered

-23.9*

- 16.6'

- 18.6'

- 11.9*

Nong'fathered

-20.0*

-3.5

- 11.6

- 12.3'

Aggregate

- 13.2'

3-Year 2-Year 1-Year

-27.8* -22.9* -24.8*

Frequencyof crashes -4.2

- 15.2'

-12.8' -4.3 3.3

-33.8* -30.2* 1.3

-5.5 -11.0' -7.5* -3.3*

Grandfathered

-26.0*

-6.9*

-20.8*

- 1.2

Nong'fathered

-23.2*

-2.1

- 12.4

-5.4*

*Changeestimateis significantlydifferentfrom 0 (p < .05).

tensivediscussionof this point. Multiple categorieswere examinedto increaseconfidencethat observedchangesin fatalities were due to changesin minimum drinking age laws and not to other coincidentalfactors.Changesin fatalities attributableto law changesshouldbe seenonly in SVN crashesinvolvingdriversin age groupsaffectedby the laws. Similar changeswere not expectedin other groupsbecausenone of the laws affectedlegal accessto alcoholfor thoseaged 21 and over, and becausealcohol involvementis significantlyless prevalent in daytime crashes.

The principalfinding is that there was a decline in single-vehiclenighttime fatal automobilecrashesamong drivers of under 21 years of age following increasesin minimum drinking age (Table 5). In each of the several (not mutuallyexclusive)categoriesof change,the rate of youthSVN fatal crashesdeclinedsignificantly.The largest rate changeoccurredamongstateswhosechangein minimumdrinkingage was 3 years(i.e., from 18 to 21); in these states, there was a decline of 26.3% in the rate

per licenseddriver of alcohol-involvedSVN fatal crashes involvingdriversunder21 (and a declineof 27.8% in the frequencyof suchcrashes),comparingthe 3 yearsbefore and the 3 yearsafter a law change,as estimatedby the time-series intervention models. States with a 2-year change(from 18 to 20 or from 19 to 21) and stateswith a l-year change(from 18 to 19, or from 19 to 20, or from 20 to 21) also showedsignificantdeclinesin crashrates (18.6% and 21.6%, respectively).For SVN crashesinvolvingdrivers21 and older, the estimateddeclineswere smaller:17.7% for the 3-yearchangestatesand 9.9% for the 2-year states.There wasno significantchangein crash rate for the l-year states.Thus, the law changeappearsto

488

JOURNAL

OF STUDIES

ON ALCOHOL

have affected the under-21 drivers specifically. Aggregated acrossthe severalstates(ignoringdistinctionsas to type of minimumdrinking age change,as well as whether there was a grandfatherclause), there was a decline of 15.4% in fatal crashratesinvolvingdriversunder21 comparedto a declineof only 5.4% involvingdrivers21 and older. Whether a minimum drinking age change was "grandfathered" or not seemedto make relatively little difference:youth SVN crashesdeclinedslightlylessin the stateswithout a grandfatherclause(20.0%) comparedto thosestateswith a grandfatherclause(23.9%), a nonsignificant difference.

Figure 7 comparesthe percentdecline in the rate of SVN fatal crashesinvolvingdriversunderage 21 across the type of law change, and adds the decline in selfreportedmean frequencyof 30-day alcoholuse (expressed as a percentof a standarddeviation) among high school seniorssurveyedin the same states, in the same years. There was an evident decline in self-reportedmean alcohol use during the sametime period in the same states. The aggregatedeclinebetweenthe 3-year-beforeversus3year-after periods is 13.8% of a standarddeviation, as comparedto 15.4% for rate of youth SVN crashes.These self-reportdata thereforesupportthe notion that the decline in SVN crashesfollowing increasesin minimum drinking ages is a direct result of loweredamountsof alcoholconsumption. The declinesin bothvariables(crashes and self-reporteduse) are greatestfor the 3-year changes (from a minimumdrinkingage of 18 to 21), which is entirelyreasonable. However,theself-reportdataindicatethat the declineswere distinctly smaller for the 1-year and 2year changes;the crashdata do not showcorrespondingly largedifferentials.The self-reportdata alsoshowa stronger effect in the stateswithout a grandfatherclause, as

/ SEPTEMBER

1991

comparedto the stateswith a grandfatherclause.The crash data showthe opposite,a strongereffect in the grandfathered states.The self-reportdata seem more plausible; the less plausible findings for the crash data are very likely due to the fact that fatal crashesare a rare occurrenceand the attendanthigh stochasticvariancemakesit more difficult to discerndifferentialpatterns(particularly in small groupsof states). In addition to the lower alcohol use, there is consider-

ably less frequentgoingto barsor tavernsby high school seniorsfollowingan increasein minimumdrinkingage. It thereforeseemsplausiblethat drinkingin barscontributes disproportionately to involvementin automobilecrashes. One reason for this disproportionateinvolvement may haveto do with the fact that there is generallya very specific time for terminatingalcoholservicein bars.Patrons thenmustleave, typicallyby automobile,just at the point whenthey are likely to be most intoxicated. In orderto rule out otherpotentialexplanations for the declinein crashesfollowingan increasein minimumdrinking age, we conductedadditionalmultivariateanalysesutilizing the self-reportdata. If, for example,the numberof milesdrivenperweekwereto declineafterminimumdrinking agelaw changes(for reasonsnot causallyrelatedto the law change),that mightaccountfor the declinein crashes. We ran multiple linear regressions on the pooledbeforeafter data set, predictingalcoholuse from a dummyvariable capturingthe before versusafter time period, and includinga numberof potentialcompetingexplanations (includingsex, race, numberof parentsin home, parental education,collegeplans, religiouscommitment,truancy, numberof eveningsout per week and weekly income, in additionto the numberof miles driven per week). A significant effect of minimum drinking age remainedafter controllingall theseother variables. Discussion and Conclusions What can we conclude from the results of the various

Aggregat•

2-Year '

3-Year

•'fathere•l

1~Year

NotG'fatherd

Typeof MDALawChange

I•11Under 21SVN • 30Day Mean (ø/•tDev) i FIGURE7. Percentchangein fatal crashesand self-reportalcoholuseby type of law change(Fatal AccidentReportingSystemdata and selfreport data)

analysesdescribedabove?Perhapsthe principalconclusion is that a minimumdrinking age of 21 versusa minimum drinkingageof 18 doesindeedaffect the behaviorof high schoolseniors;it leadsto lowerconsumption of alcohol. It has been demonstratedconclusivelythat alcoholinvolvedhighwaycrashesdecline among the 18- to 20year-old population,and the presentresearchmakes it clear that the declineis directlya resultof lowerlevelsof consumption. And it alsoappearsthat a majorfactorin the reducedrate of crashesis that the under-21group spend lesstime in barsand tavernswhen the minimumdrinking age is 21. Another importantfinding is that the lower rates of drinking appearto continueas youngadultsmature, at leastthroughthe early twenties.Thus, the lowered rates of drinking in the 18-20 age range are not compensatedfor by a higherrate of drinking after enfranchise-

O'MALLEY

AND

ment is achieved, but in fact continue even after alcohol is

legally accessible. As with all social scienceresearchin a real-life, nonlab-

oratorysituation,it is difficult to makeindisputableinferences.Wheneveran effect is claimed, it is necessaryto rule out potentialalternativeexplanations. The mostcommon alternativeexplanationfor cross-sectional differences in behaviorsuchas drinkingby high schoolseniorsassociatedwith differentminimumagesis that stateswith different minimum ages also differ on other factors. On a similar issue,for example,Bentier (1981) cites California as being reputedto have lesstraditionalstandardsof religion (amongother things)thanotherstates,and he notes thatthisdifferencecouldserveas a competingexplanation for differencesin marijuanause that might otherwisebe attributedto differencesin the legal statusof marijuana. In the present research,the cross-sectionalanalyses showeda significantregression coefficientfrom minimum drinkingage to alcoholuse even after controllinga number of important individual-levelfactorsassociatedwith

WAGENAAR

489

stateschangedtheir laws in responseto externalforces,in this caseby federal action. The law changeswere therefore not merely indicatorsof existingculturalsentiment, nor wouldthey be expectedto bring aboutshiftsin other variables like religiosity or anti-alcohol attitudes. The clear effectsobservedin a variety of statesare very unlikely to be due to extraneousfactors. This

research

has also demonstrated

that

the

lower

single-vehiclenighttimecrash rates after the minimum age was raisedare accompanied by lowerratesof alcohol use and lower amountsof time spentin bars and taverns. The findingsof O'Donnell (1985) are importantin emphasizingthe significanceof drinkingin barsand taverns;in her review of the limited literature available, she found

ing ages,whereaswe are moreconcernedaboutmisattributing outcomes to an increase in minimum ages. A particularstrengthof the presentanalysesis that suchex-

that "resultssuggestthat approximatelyhalf of the intoxicateddriverson our highwaysdrink at licensedpremises, especiallybars, before driving" (p. 516), even though only abouta quarterto a third of alcoholicbeveragesare sold for on-premiseconsumption. As in mostsurveyresearch,this researchhas relied on self-reportdata, and it could be that the observeddeclines in reporteduse reflect only declinesin willingnessto report alcoholuse. Other reportsof the Monitoringthe Future projecthavemoredetaileddiscussions of the validity of self-reportdata (Johnstonet al., 1989;O'Malley, 1984; O'Malley and Wagenaar,1989), but a few commentsare in order here. Althoughthere is very likely somedegree of underreportingof illegal drug use on self-reportsurveys, mostresearchhas shownthat, in general,it is of a small magnitudein self-completed,confidentialquestionnaires in normal populations(Benson and Holmberg, 1985; Single et al., 1975; Smart, 1975). In the present study,thereare no objectiveindividual-leveldata to demonstrate indisputablythat alcohol use was lower in the stateswith high minimumdrinkingagesas opposedto the alternativehypothesis that use was simplymore underreported in the high-agestates.However,severalindirect measuressupportthe interpretationthat use was lower. One indirect indicatoris the seniors'reportsof friends' use. If the changein MDA law merely changedseniors' willingnessto reporttheir own use,thentheremightbe no changein their reports of their friends' use of alcohol; but, in fact, reportedfriends'usealsodeclined.Otherevidencecomesfrom attitudinalmeasures:perceivedrisk of harm associatedwith alcoholuse, seniors'own personal disapprovalof use and friends' perceiveddisapprovalof use all increasedafter the law change;if there were no changein actual alcohol-relatedbehavior,there would be no particularreasonfor thesevariousassociatedattitudes or beliefsto change.Finally, the FARS data also showed

traneous factors as use of other substances or amount of

decreases in alcohol-related fatal crashes, and these are

driving were statisticallycontrolledat the individuallevel, and variationsassociatedwith changesin minimumdrinking age laws remained.Also of considerableimportance in drawingcausalinferencesis the fact that many of the

clearly not likely to be due to changesin willingnessto self-reportillegal behaviors. Althoughthe empiricaldata seemclear in showinga salutaryeffectof a high minimumdrinkingage, it can be

alcohol use. If adolescents in certain areas tended to drink

lessbecausetherewerehigherlevelsof "communityreligiosity" or someotherindicatorof anti-alcoholsentiment, thesewould presumablybe capturedby individual-level variables that would serve as indicators of commitment to societal institutions.

The introduction

of variables such as

religious commitmentand grades should, if minimum drinkingageeffectswerespurious,leadto lesssignificant valuesfor the relevantregressioncoefficients.But there were essentiallyno differencesbetweenthe bivariate and multivariateassociations. The mostparsimonious explanation remainsthe mostobviousone:minimumdrinkingage laws do have an effect on behavior. Robert Straus has observed:

I havebeena bit concernedwith the extentto which we may apply rather simplisticexplanationsto the apparentrise in problemsdue to the changein drinking age, without taking into accountabout six other things that were happeningin societyat the sametime, suchas the generalizedincreasein consumptionof alcohol, the increasein use of a whole variety of substances by youngerpeople, the enormousincrease in the numberof vehicleslicensedto and drivenby younger people,and severalothers.(1984, p. 130)

Straus'sconcernswere that problematicoutcomesmight have been misattributed

to a decrease in minimum

drink-

490

JOURNAL

OF STUDIES

ON ALCOHOL

argued on other groundsthat minimum drinking ages shouldnot be setat 21 whenso manyother"adult" roles can be assumedat age 18. On the surface,it may seem unfair to many observersto allow 18-20 year olds to marry, to have children, to own cars, homesand firearms and to be financiallyand sociallyindependent, and yet to be legally prohibitedfrom drinking a glassof wine in a restaurant,or even a glass of champagneat their own wedding.A numberof observershave worried aboutthe effectthis seeminglyinconsistent situationmay have.By stretchingout adolescence (that is, by holdingback the time when full adulthoodis achieved),are we creating otherproblems? Theseissuesmay continueto be debated. The contributionof the presentresearchis to demonstrate that, whatever one wishes to make of other factors, there

is a clear specificeffect of a higherminimumdrinking age: there is lessdrinkingand consequently fewer fatalities. The effects on drinking are modest;nevertheless, modestdifferences in ratesof drinkingcanbe very important, particularly when thosedifferenceslead to lowered rates of fatal crashes. However, it should also be remem-

beredthat drinkingremainsa popularactivityamonghigh schoolseniors,evenwhenthe minimumdrinkingageis 21. The popularityof drinkingamongseniorsis not surprising. Alcoholuseis a very commonsocialpracticeamong adults, particularlyamongyoungadults, and that alone would tend to make it an attractiveactivity for adolescents.And enforcementof minimumdrinking age laws tends to be lax in most states. In addition, the use of al-

coholis heavilypromotedand glamorizedin commercials; the entire aura aroundthosecommercials is pleasurable, athletic,sexual,fun--all the thingsthat appealto youth. Consequently, manymore societalchangesare neededin additionto changes in minimumdrinkingagelawsif drinking amonghigh schoolseniorsis to be furtherreduced. Acknowledgments The authorsthankJeraldBachmanand Lloyd Johnstonfor useof the data and their advice and assistancein this research;Fredrick Streff and

RobertSchultzfor their participationin the analyses of the trafficcrash data; and JeromeHiniker and Dan Zahs for assistance in data management and analysesof the self-reportdata.

/ SEPTEMBER

1991

up panel has (when reweighted)the same senior-yearprevalenceas the totalsenior-year samplefor that classyear.For example,suppose 50% of the entire senior-yearsamplereportedusingmarijuanain senior year,but amongthoseparticipatingin a given follow-uppanel from that classonly 40% had (as seniors)reportedsuchuse. The follow-uprespondents who had beenusersin senioryear wouldbe weighted5/4, and follow-up respondentswho had been nonusers would be weighted5/6, thus creatinga 50% senior-yearusagerate for the follow-uppanel.The follow-upprevalence rateswouldthenbe derivedby applyingtheseweightsto follow-updata. 4. Note that no changein categoryis not quite the sameas no changein minimumdrinking age; if a state raised its minimumdrinkingage from 19 to 20, that would not lead to a differentpositionon the 3category measure.

5. Thesedata were weighted.Weightswereusedbecauseit seemslikely that the best estimatesof individual-levelpopulationcorrelations among the various control variableswould be providedby the weighteddata. 6. Use of even a very large designeffect of 11--the maximumestimated value observedin an extensiveset of calculationsof design effectsacrossa numberof dependentvariables--leavesmore than 6,000 as the effective N.

7. Inclusionof a controlfor seculartrend by addingdummyvariables for year of administrationin the regressionequationdid not alter the minimumdrinkingageregression coefficientsmorethantrivially, and controllingfor individualreligiouspreference, in additionto religious commitment,also had virtually no effect on the regressioncoefficients.

8. Respondents participating3 years after graduationwould include very few who were not yet 21, but this is obviouslynot an important factor,as indicatedin Figure 5. 9. Each data point in Figure6 is basedon at least7,300 seniorsin at least 49 schools.

10. For analyses of theeffectof the 18to 21 change,the minimumnumber of seniorsper year was 950. 11. The eta statisticis a measureof the proportionof variancein the dependentvariablethat is explainedby a categoricalindependent variable.

A value of .004 indicates that 0.4%

of the variance

in the

dependent variableis explainedby the 2-categoryvariablecomparing the meansbeforeand after the law change. References

ABELSON, R.P. A varianceexplanationparadox:When a little is a lot. Psychol.Bull. 97: 129-133, 1985. BACHMAN,J.G., JOHNSTON,L.D. AND O'MALLEY, P.M. Smoking, drinking,and drug useamongAmericanhigh schoolstudents: Correlatesand trends,1975-1979.Amer. J. publ. Hlth 71: 59-69, 1981. BACHMAN,J.G., JOHNSTON, L.D. AND O'MALLEY, P.M. Monitoring the Future:Questionnaire Responses from the Nation's High School Seniors, 1986, Ann Arbor, Mich.: Institute for Social Research, 1987.

Notes 1. The term "crash"

is used rather than the more common

term "acci-

dent" in orderto emphasizethat automobilecrashesare not simply random, unavoidable events.

2. The term "minimum drinking age" law is necessarilysomewhatimprecise;someof the lawsreferredto specifya minimumage for purchase of alcoholicbeverages,others specify a minimum age for consumptionor possession. In spite of its imprecision,"minimum drinkingage" seemsto be the term mostcommonlyusedin the literatureand in policy debates,and so is usedhere. 3. The procedureusedto estimateprevalencein the follow-upsamples is to reweightparticipatingfollow-uprespondents so that eachfollow-

BENSON,G. ANDHOLMBERG, M.B. Validity of questionnaires in population studieson drug use. Acta Psychiat.scand.71: 9-18, 1985. BENTLER,P.M. A multivariateview of marijuanadecriminalization research.Contemp.Drug Probl. 10: 419-433, 1981. BONNIE,R.L. Regulatingconditionsof alcoholavailability:Possibleeffectson highwaysafety.J. Stud. Alcohol, SupplementNo. 10, pp. 129-143, 1985.

Box, G.E.P. AND JENKINS,G.M. Time Series Analysis:Forecasting and Control (Revised edition), Oakland, Calif.: Holden-Day, Inc., 1976.

Box, G.E.P. ANDTIAO, B.C. Interventionanalysiswith applications to economicandenvironmental problems.J. Amer. statist.Assoc.70: 70-79, 1975.

O'MALLEY

AND WAGENAAR

DISTILLEDSPIRITSCOUNCILOF THE UNITEDSTATES,INC. Summaryof state laws and regulationsrelating to distilled spirits, Washington, D.C., 1983, 1985, 1987, 1989. DuMouCHEL, W., WILLIAMS,A.F. ANDZAr•OR,P. Raisingthe alcohol purchaseage: Its effectson fatal motorvehiclecrashesin twenty-six states.J. legal Stud. 16: 249-266, 1987. GEORGE, W.H., CROWE, L.C., ABWENDER,D. AND SKINNER, J.B. Effectsof raisingthe drinkingage to 21 yearsin New York Stateon self-reportedconsumption by collegestudents.J. appl. socialPsychol. 19: 623-635, 1989. HUGHES,S.P. ANDDODDER,R.A. Raisingthe legal minimumdrinking age:Short-termeffectswith collegestudentsamples.J. Drug Issues 16: 609-620, 1986. INSURANCEINFORMATIONINSTITUTE. InsuranceFacts, New York: Insurance Information Institute, 1983, 1984, 1985.

JOHNSTON, L.D., O'MALLEY, P.M. AND BACHMAN,J.G. Drug Use, Drinking and Smoking:National SurveyResultsfrom High School, College, and Young Adult Populations,1975-1988, Washington: GovernmentPrintingOffice, 1989. MOORE,M.H. AND GERSTEIN,D.R. (Eds.) Alcohol and Public Policy: Beyond the Shadow of Prohibition, Washington,D.C.: National AcademyPress, 1981. NATIONALSAFETYCOUNCIL.Policy Update, Chicago,I11.:Author,September 20, 1985.

O'DONNELL,M.A. Researchon drinkinglocationsof alcohol-impaired drivers:Implicationsfor preventionpolicies.J. publ. Hlth Policy6: 510-525, 1985.

491

O'MALLEY,P.M. Cigarettesmokingamonghigh schoolseniors:Did the rate decline? Prey. Med. 13: 421-426, 1984. O'MALLEY, P.M., BACHMAN,J.G. AND JOHNSTON,L.D. Period, age,

andcohorteffectson substance useamongyoungAmericans:A decadeof change,1976-1986.Amer. J. publ. Hlth 78: 1315-1321, 1988.

O'MALLEY, P.M. AND WAGENAAR,A.C. Minimum Drinking Age Laws:Effectson AmericanYouth.Monitoringthe FutureOccasional PaperNo. 28, Ann Arbor,Mich.: Institutefor SocialResearch,1989.

ROSENTHAL, R. How are we doingin softpsychology? Amer.Psychol. 45: 775-777, 1990.

SINGLE,E., KANDEL,D. ANDJOHNSON, B.D. The reliabilityand validity of druguseresponses in a largescalelongitudinalsurvey.J. Drug Issues5: 426-443, 1975.

SMART,R.G. Recent studiesof the validity and reliability of selfreporteddruguse, 1970-1974.Canad.J. Criminol.Correct.17: 326333, 1975.

STRAUS, R. Commentson alcohol,youth,and drunkdriving. In: GERSTEIN,D.R. (Ed.) Towardthe Preventionof Alcohol Problems:Government, Business,and Community Action, Washington,D.C.: NationalAcademyPress,1984, pp. 129-130. VINGILIS, E.R. AND DEGENOVA,K. Youth and the forbiddenfruit: Experienceswith changesin the legal drinkingage in North America. J. Crim. Just. 12: 161-172, 1984.

WAGENAAR, A.C. Alcohol, YoungDrivers, and Traffic Accidents:Effectsof Minimum-AgeLaws, Lexington,Mass.: D.C. Heath Co., 1983.

Effects of minimum drinking age laws on alcohol use, related behaviors and traffic crash involvement among American youth: 1976-1987.

This study has two separate but related purposes: (1) to delineate cross-sectional differences among U.S. high school seniors and young adults that ma...
2MB Sizes 0 Downloads 0 Views