do alcohol warning labels influence men's and women's attempts to deter others from...

9
Do Alcohol Warning Labels Influence Men’s and Women’s Attempts to Deter Others from Driving When Intoxicated? Tammy W. Tam and Thomas K. Greenfield Alcohol Research Group, Public Health Institute, Emeryville, California Abstract Research on the federally mandated alcohol warning label has found mixed results, but some findings are consistent with a modest influence on precautionary behaviors to reduce drinkers’ self-reported drunk driving. We hypothesized that warnings would also influence the likelihood of intervening to deter others’ driving after drinking. Using data from 1,376 adult drinkers in a U.S. national survey, a conceptual model reflecting effects of exposure to the label’s drunk driving message on taking actions to avert another’s driving under the influence was tested in a structural equation modeling framework. For men and women, in structural models with drinking and handling of alcoholic beverages potentially affecting both message recall and intervening, the predicted relationships were found between message recall and actions to deter another’s drinking driving. This finding suggests that an important preventive effect of the alcohol warning label may be to legitimate collateral attempts to avert another’s drunk driving. C 2010 Wiley Periodicals, Inc. Keywords: Warning labels; Alcohol policy; Drunk driving 1. INTRODUCTION In November 1989, a federal law requiring alcohol con- tainer warning labels (Public Law No. 100-690, 1988) took effect. From approximately 6 months before this, in Summer, 1989, until approximately 5 years after, in Summer, 1994, five U.S. national telephone surveys were conducted by the Alcohol Research Group (ARG; Greenfield, 1997). The survey series, which included an Ontario, Canada control site, was part of a feder- ally mandated evaluation of the new law (Greenfield, Graves, & Kaskutas, 1993, 1999). Congress intended Correspondence to: Tammy W. Tam, Alcohol Research Group, Public Health Institute, 6475 Christie Ave., Suite 400, Emeryville, CA 94608. Phone: 510-597-3440; e-mail: [email protected] Received: 30 October 2008; revised 31 March 2009; accepted 15 April 2009 View this article online at wileyonlinelibrary.com. DOI: 10.1002/hfm.20239 that the warning label serve to remind the public of a number of the risks associated with alcohol con- sumption. The mandated container label reads: “GOV- ERNMENT WARNING: (1) According to the Surgeon General, women should not drink alcoholic beverages during pregnancy because of the risk of birth defects. (2) Consumption of alcoholic beverages impairs your ability to drive a car or operate machinery, and may cause health problems.” The multiyear study was de- signed as a pre-repeated-post, control-intervention- group quasi-experiment (Cook & Campbell, 1979) to examine potential effects of the natural experiment (Greenfield, 1994; Greenfield & Kaskutas, 1998). Anal- yses of data gathered shortly after implementation found modest potential effects on individuals’ precau- tionary behaviors that might reduce their likelihood of driving while intoxicated (DWI; Greenfield et al., 1993). More recent quasi-experimental comparisons have shown mixed results. There was some indication that in the United States there was a significant rela- tionship between seeing the label and deciding not to 538 Human Factors and Ergonomics in Manufacturing & Service Industries 20 (6) 538–546 (2010) c 2010 Wiley Periodicals, Inc.

Upload: tammy-w-tam

Post on 11-Jun-2016

214 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Do alcohol warning labels influence men's and women's attempts to deter others from driving when intoxicated?

Do Alcohol Warning Labels Influence Men’s andWomen’s Attempts to Deter Others fromDriving When Intoxicated?Tammy W. Tam and Thomas K. Greenfield

Alcohol Research Group, Public Health Institute, Emeryville, California

Abstract

Research on the federally mandated alcohol warning label has found mixed results, but some findingsare consistent with a modest influence on precautionary behaviors to reduce drinkers’ self-reporteddrunk driving. We hypothesized that warnings would also influence the likelihood of intervening todeter others’ driving after drinking. Using data from 1,376 adult drinkers in a U.S. national survey, aconceptual model reflecting effects of exposure to the label’s drunk driving message on taking actions toavert another’s driving under the influence was tested in a structural equation modeling framework. Formen and women, in structural models with drinking and handling of alcoholic beverages potentiallyaffecting both message recall and intervening, the predicted relationships were found between messagerecall and actions to deter another’s drinking driving. This finding suggests that an important preventiveeffect of the alcohol warning label may be to legitimate collateral attempts to avert another’s drunkdriving. C© 2010 Wiley Periodicals, Inc.

Keywords: Warning labels; Alcohol policy; Drunk driving

1. INTRODUCTION

In November 1989, a federal law requiring alcohol con-tainer warning labels (Public Law No. 100-690, 1988)took effect. From approximately 6 months before this,in Summer, 1989, until approximately 5 years after,in Summer, 1994, five U.S. national telephone surveyswere conducted by the Alcohol Research Group (ARG;Greenfield, 1997). The survey series, which includedan Ontario, Canada control site, was part of a feder-ally mandated evaluation of the new law (Greenfield,Graves, & Kaskutas, 1993, 1999). Congress intended

Correspondence to: Tammy W. Tam, Alcohol Research Group,Public Health Institute, 6475 Christie Ave., Suite 400,Emeryville, CA 94608. Phone: 510-597-3440; e-mail:[email protected]

Received: 30 October 2008; revised 31 March 2009; accepted15 April 2009

View this article online at wileyonlinelibrary.com.

DOI: 10.1002/hfm.20239

that the warning label serve to remind the public ofa number of the risks associated with alcohol con-sumption. The mandated container label reads: “GOV-ERNMENT WARNING: (1) According to the SurgeonGeneral, women should not drink alcoholic beveragesduring pregnancy because of the risk of birth defects.(2) Consumption of alcoholic beverages impairs yourability to drive a car or operate machinery, and maycause health problems.” The multiyear study was de-signed as a pre-repeated-post, control-intervention-group quasi-experiment (Cook & Campbell, 1979) toexamine potential effects of the natural experiment(Greenfield, 1994; Greenfield & Kaskutas, 1998). Anal-yses of data gathered shortly after implementationfound modest potential effects on individuals’ precau-tionary behaviors that might reduce their likelihoodof driving while intoxicated (DWI; Greenfield et al.,1993). More recent quasi-experimental comparisonshave shown mixed results. There was some indicationthat in the United States there was a significant rela-tionship between seeing the label and deciding not to

538 Human Factors and Ergonomics in Manufacturing & Service Industries 20 (6) 538–546 (2010) c© 2010 Wiley Periodicals, Inc.

Page 2: Do alcohol warning labels influence men's and women's attempts to deter others from driving when intoxicated?

Tam and Greenfield Alcohol Warning Labels

drive after drinking (Greenfield, 1997; Greenfield et al.,1999).

This article examines the potential impact that therecipient of a product warning might have on the indi-vidual seeing the message, to encourage taking actionsto avert harms that could involve others’ risk behav-ior. In studying impacts of the alcohol warning, thesecond author, as principal investigator of the warn-ing label study described earlier in text, conceived thenotion that the warning could legitimate interventionsthat might be aimed at deterring another person fromengaging in a warned-about risk—specifically, to detera friend, relative, or acquaintance from proceeding todrive when the person in question was seen as beingunder the influence. From the viewpoint of human fac-tors research on product warnings, this is a relativelyunique framing of an important empirical question.Many warnings might engage the efforts of viewers tointervene so as to reduce others’ risks. In the case ofplastic bags, buckets, and hot-tub cover lock downs,which warn about suffocation or drowning dangers,this conceptualization of encouragement to take pre-cautions on behalf of others is clear, but often aimedat reducing risks to children or infants. In the case ofwarnings on stepladders, a spouse might urge his orher partner to heed the warning not to stand on thetop step. From the viewpoint of human factors researchon warning labels, this consideration of actions takento deter other adults from risky behavior has not beencommon. Specifically with regard to alcohol warningresearch, it is entirely new. We developed a set of ques-tions to capture particular behaviors intended to stopanother person from driving after having had too muchto drink, and hypothesized that these may be engagedin more by those tending to recall the alcohol warning’sdriving impairment message.

This study used a subset of the U.S. cross-sectionaldata collected in the 1993 and 1994 study years, ap-proximately 3.5 to 4.5 years after the law’s imple-mentation. In this longer-term phase of the research,improved, sensitive measures (Greenfield, 1994) weredeveloped to assess impacts that might accrue over timeas a greater proportion of the drinking public becameexposed to the label’s specific messages, such as thedriving impairment message. For example, by 1994,it was estimated that between 35% and 45% of maleheavy drinkers saw the warning label and recalled itsdrunk driving message (Greenfield & Kaskutas, 1998).The hypothesized impact we focus on here was notdirectly on the drinking driver, but instead on his or

her collaterals—friends, acquaintances, or significantothers. We hypothesized that those persons seeing andrecalling the warning label’s driving impairment mes-sage would be more likely than others, all things equal,to intervene so as to deter another person from drivingwhen intoxicated. Epidemiological research has recog-nized that certain individuals may want a heavy drinkerto drink less (Room, Greenfield, & Weisner, 1991), butoften “hold back,” sometimes perceiving it to be sociallyinappropriate to attempt to “control” others’ behavior.We believed that the warning label message might serveto legitimate a socially challenging intervention, so in-creasing the likelihood of reporting behaviors designedto reduce the likelihood of an inebriated person get-ting behind the wheel. Early in the news media’s cov-erage of the alcohol warning label (Lemmens, Vaeth,& Greenfield, 1999), a related instance was reported inwhich two waiters were fired from a Seattle restaurantafter confronting a drinking patron who was obviouslypregnant by giving her a copy of the label with itsother main message that pregnant women should notdrink. In just such a way, we believed, friends and ac-quaintances of someone about to risk for DWI mightintervene to discourage driving or provide alternatives.We reasoned that those persons who had seen the la-bel’s drunk driving message might be emboldened toact so as to deter another from drinking and driving,hopefully with less dire personal consequences thansuffered by the waiters.

Survey waves 4 and 5 (1993, 1994) collected dataon five questions to tap the domain of relevant behav-iors involved in interventions to deter another fromdrunk driving, such as simply asking someone not todrive, offering to drive the person home, having theperson stay over, or suggesting that he or she take ataxi, bus, or subway, among others (see Measures).With these measures, we proposed a conceptual modelthat examined the direct and indirect relationshipsamong alcohol consumption, container handling, mes-sage recall, and such deterrence-oriented interven-tions. The model includes the following hypothesizedrelationships:

1. Warning label message recall, alcohol con-sumption, and container handling will havedirect positive effects on intervention.

2. Higher consumption level and frequent con-tainer handling will be associated with higherlikelihood of label message recall.

Human Factors and Ergonomics in Manufacturing & Service Industries DOI: 10.1002/hfm 539

Page 3: Do alcohol warning labels influence men's and women's attempts to deter others from driving when intoxicated?

Alcohol Warning Labels Tam and Greenfield

InterventionAlcohol

Consumption

Container

Handling

Recall drunk

driving message

Asked someone

not to drive

Offered to drive

someone home

Asked someone

to take a taxi, etc.

Tried to takesomeone’s

car keys

Asked someone

to stay at your

home

.96*

.88*

.67*

.86*

.77*

.08*

.28*

.18*

.08*

.29*

.05

InterventionAlcohol

Consumption

Container

Handling

Recall drunk

driving message

InterventionAlcohol

Consumption

Container

Handling

Recall drunk

driving message

Asked someone

not to drive

Offered to drive

someone home

Asked someone

to take a taxi, etc.

Tried to takesomeone’s

car keys

Asked someone

to stay at your

home

.96*

.88*

.67*

.86*

.77*

Asked someone

not to drive

Offered to drive

someone home

Asked someone

to take a taxi, etc.

Tried to takesomeone’s

car keys

Asked someone

to stay at your

home

.96*

.88*

.67*

.86*

.77*

Asked someone

not to drive

Offered to drive

someone home

Asked someone

to take a taxi, etc.

Tried to takesomeone’s

car keys

Asked someone

to stay at your

home

.96*

.88*

.67*

.86*

.77*

.08*

.28*

.18*

.08*

.29*

.05

.08*

.28*

.18*

.08*

.29*

.05

Figure 1 Model predicting intervention to deter others’ drunk driving. All coefficients are standardized. ∗ p < 0.05.

3. Higher consumption level will be associatedwith more frequent container handling thatwill have an indirect effect on intervention (seeFigure 1).

The model also controls for the effects of gender, age,ethnicity, and education on the relationships detailedin the model.

Using structural equation modeling (SEM) tech-niques, this study asks the following questions:

1. How well does the proposed model relatingrecall of the label’s drunk driving message toactions taken to deter others’ drunk driving fitthe data?

2. Are the relationships between the warning la-bel message and intervening to deter drunkdriving the same for men and women?

Regarding the last issue, we know that moreyoung men than women drink heavily (Greenfield,Giesbrecht, & Kavanagh, 1996; Greenfield & Rogers,1999b), especially beer (Rogers & Greenfield, 1999),and are more often stopped for DWI (Berger & Snor-tum, 1985). Young men, in particular, often overesti-mate the quantity of alcohol that they can drink be-fore driving is impaired (Greenfield & Rogers, 1999a;Hennessy & Saltz, 1990). There is a dearth of empiricaldata on intervening with collaterals to prevent peo-ple from drinking and driving; we do not know, forexample, whether women or men are more likely to as-sertively intervene to deter acquaintances, friends, and

lovers from drinking and driving. Although womenhave sometimes been regarded as less assertive thanmen (Fukuyama & Greenfield, 1983), their frequentrole as “navigators”(in couples at least), their poten-tially greater interest in health warnings, and a strongerhealth consciousness (Kaskutas & Greenfield, 1997)might suggest that the influence of warning messageson women’s willingness to intervene might exceed thatof men. The present study aimed to explore this di-mension of preventive behavior and examine the rela-tionships between message recall and intervening forU.S. men and women.

2. METHODS

2.1. Sample

Data came from the ARG’s long-term Impacts of Al-coholic Beverage Warning Labels project (Greenfield,1997; Greenfield & Kaskutas, 1998), using institution-ally approved voluntary consent procedures. The U.S.national data were collected approximately 3.5 and 4.5years after the implementation of Public Law No. 100-690 (1988) in November 1989. Cross-sectional sam-ples were drawn by random digit dialing (RDD) meth-ods in 1993 (n = 1,026) and 1994 (n = 1,016) from48 contiguous states; respondents were interviewed inJuly and August each year using computer-assisted tele-phone interview (CATI) techniques and closely com-parable instruments, with response rates of 63% and

540 Human Factors and Ergonomics in Manufacturing & Service Industries DOI: 10.1002/hfm

Page 4: Do alcohol warning labels influence men's and women's attempts to deter others from driving when intoxicated?

Tam and Greenfield Alcohol Warning Labels

54%, respectively. These rates are similar to those ofother recent telephone surveys (Frey, 1989). Subjectswithin a household who were 18 years old or older wereselected by the most recent birthday technique. This ar-ticle examines only the data collected from 1,376 menand women who reported drinking at least once a year(“drinkers”).

Design effects for the surveys were modest. Increasedstandard errors due to the RDD sampling design aroseprimarily from the unequal probabilities of selectioncaused by more than one telephone number and/ormore than one adult in a household. Calculations in-dicated increases in sample variance of 29% and 33%,respectively, each year in the United States due to thesefactors and weighting for sample representativeness bygender and age (Greenfield et al., 1999). In statisticalcomparisons of proportions, design effects were ac-counted for by appropriate weighting (Greenfield &Kaskutas, 1998). The cross-sectional data from the 2years were pooled because statistical differences in keyvariables were not seen. Rates and percentages reportedhere are based on the weighted data, whereas the sam-ple sizes reported are unweighted.

2.2. Measures

The ARG warning label project assessed the followingin each respondent: demographic variables, exposureto the warning label and recall of its contents, alcoholconsumption, alcohol container purchases and han-dling, and behaviors relevant to the investigation of po-tential effects. Specific measurement approaches havebeen detailed elsewhere (Greenfield et al., 1993, 1999;Kaskutas & Greenfield, 1992).

2.2.1. Label and Message Recall

Exposure to the label was assessed in each year’s sur-vey by the question: “Now thinking about the last 12months, have you seen any labels on bottles or cans ofbeer, wine, or liquor?” After the label awareness ques-tion(s), respondents were asked a series of questionsabout what the warning labels said. The question onmessage recall of drunk driving—“Did the warninglabel say anything about drunk driving?”—was usedhere. The dichotomous measure indicating recall ofthis message seen on the warning label during the last12 months was used in the analyses.

2.2.2. Interventions to Deter Drunk Driving

The dependent variables, actions taken to deter anotherperson’s drunk driving, were assessed by the followingquestions:

Have you asked someone not to drive?Have you offered to drive someone homeyourself?Have you asked someone to take a taxi, bus, orsubway?Have you tried to take someone’s car keys?Have you asked someone to stay at your home?

These items represent a variety of preventive stepsthat hosts or friends might take. In the interview, theyfollowed a series of items on drinking and driving aswell as precautionary behaviors that the respondentmight undertake to avoid his or her own drunk driv-ing (Greenfield & Kaskutas, 1993). The immediatelypreceding items were two questions on driving impair-ment given certain levels of consumption, so that thecontext for asking about interventions to deter drunkdriving was clear to respondents. Response categoriesfor each item were binary, indicating an affirmativeresponse to the questions (1 vs. 0).

2.2.3. Container Handling

Purchase and handling of alcoholic beverages were as-sessed by two items, respectively: “In the last 30 dayshave you bought any alcoholic beverages in a store?”and “In the last 30 days have you opened a bottle or canof any alcoholic beverage?” For use in the analyses, athree-category “container handling” measure was con-structed indicating negative response to both questions(0) or an affirmative response to one (1) or both (2)questions.

2.2.4. Alcohol Consumption

Average number of drinks per day was assessed by thequestion: “On those days when you drink wine, beer, orhard liquor, how many drinks do you usually have perday?” (Participant was required to provide the actualnumber of drinks.) A dichotomous measure of usualamounts of three of more drinks in a day was used inthe analysis to indicate higher quantity consumption.This amount clearly exceeds the moderate drinkingguidelines established by the Federal Department of

Human Factors and Ergonomics in Manufacturing & Service Industries DOI: 10.1002/hfm 541

Page 5: Do alcohol warning labels influence men's and women's attempts to deter others from driving when intoxicated?

Alcohol Warning Labels Tam and Greenfield

Agriculture (Dufour, 1999), which are no more thantwo drinks a day for men and one drink a day forwomen. To avoid confounding measurement differ-ences with gender, we retained the 3+ drinks level forboth men and women (Graham, Wilsnack, Dawson,& Vogeltanz, 1998), representing more than light (butnot necessarily heavy) drinking.

2.3. Analysis

The chi-square test of independence was conductedto examine gender differences in demographic charac-teristics, alcohol consumption, recall of drunk drivingmessage, handling of alcoholic beverage, and interven-tions to deter others’ drunk driving. Fisher’s exact test(Wickens, 1989) was used to compare contingency ta-bles with small cell sizes. Multiple comparisons wereadjusted by using the Bonferroni method.

In the proposed model, the five variables measur-ing actions taken to deter another’s drunk driving wereanalyzed as indicators of a latent construct of interven-tion. Our analyses first examined the reliability, valid-ity, and factor structure of the measurement model forthis latent construct of intervention by confirmatoryfactor analysis (CFA; Bollen, 1989). Because the fiveobserved variables were dichotomous, the models wereestimated by weighted least square using the latent vari-able SEM program Mplus (Muthen & Muthen, 1998–2007). The testing of gender differences in the measure-ment relationship was conducted using multiple-groupanalysis.

The path model that examined the relationship be-tween categorical variables of recall of the drunk driv-ing message on the warning label and actions taken todeter others’ drunk driving was estimated by weightedleast squares with robust standard errors and the mean-and variance-adjusted chi-square test statistic, againusing Mplus. Model fit was evaluated using summarystatistics including model chi-square, root mean squareof approximation (RMSEA), and weighted root meansquare residual (WRMR). As a general guideline, amodel is considered to be a good representation ofthe data when the chi-square/degree of freedom ratiois less than two (Newcomb, 1994), RMSEA values areless than .05 (Browne & Cudeck, 1993; Yu & Muthen,2001), or the WRMR value is less than .9 (Yu & Muthen,2001). The significance, direction, and magnitude ofthe regression paths in the model were also examinedto evaluate how well the data fit the conceptual model.

Significance of coefficients was evaluated as z statisticsat p < 0.05.

3. RESULTS

3.1. Bivariate Analyses

There were no significant gender differences in de-mographic characteristics including age, income, andeducation (Table 1). A higher proportion of Hispanicand ‘Other’ ethnicities were found in the male sample(p < 0.05), however. Men reported usually drinking ahigher quantity of drinks in a day (33.5% vs. 13.8%for women), purchased and opened alcoholic bever-ages more often, and were more likely to recall seeingthe drunk driving message on beverage containers thanwere women (p < 0.01). With regard to actions takento deter drunk driving, however, no gender differenceswere found in any of the five measures.

3.2. CFA

CFA was conducted with the full sample of drinkers(n = 1,376) to evaluate the unidimensional modelrepresenting interventions to deter another’s drunkdriving. Table 2 shows the estimated validity and reli-ability of the measures. The proposed unidimensionalmodel demonstrated excellent fit to the data (χ2 =6.56,df = 5, p = 0.25; RMSEA = .02; WRMR = .63). Theaverage off-diagonal absolute standardized residualwas .02, indicating small discrepancies betweenthe observed and estimated covariance among thevariables. This analysis shows that the five measuresare good observed indicators of a latent construct ofintervention. The reliability of the five measures as in-dicated by the squared multiple correlation coefficient(R2; Bollen, 1989) ranged from a low of .42 to a highof .94. The validity of each measure with respect to theintervention factor was evaluated by the magnitude ofthe direct structural relationship between the factorindicators and the latent factor (Bollen, 1989). Theunstandardized validity coefficients (ranging from .67to .91) indicated acceptable validity of the measures.Overall, the item “asked someone to take a taxi, bus, orsubway” has the least validity and reliability among thefive measures. Additional multiple-group analyses totest for gender differences (data not shown) suggestedthat the relationship between the latent factor ofintervention to deter another’s drunk driving and

542 Human Factors and Ergonomics in Manufacturing & Service Industries DOI: 10.1002/hfm

Page 6: Do alcohol warning labels influence men's and women's attempts to deter others from driving when intoxicated?

Tam and Greenfield Alcohol Warning Labels

TABLE 1. Distribution of Demographic Characteristics andOther Measured Variables by Gender

Women Men p(n = 645) (n = 731) Value

Age 0.4618—25 17.4% 19.0%26—30 11.5 11.831—40 24.7 27.641—60 29.5 26.561 or older 17.1 15.0

Ethnicity 0.04White 84.3 81.9Black 8.3 6.5Hispanic 3.9 5.7Other 3.5 6.0

Income 0.08Less than $20,000 14.5 11.1$20,000–$59,999 56.5 55.1$60,000 or more 28.9 33.8

Education 0.79High school or less 42.5 43.3Some college 57.5 56.7

Alcohol Consumption(drinks per day) 0.00Less than 1 drink 3.0 2.11–2 drinks 51.3 33.92–3 drinks 31.9 30.53–4 drinks 8.6 13.94–5 drinks 1.7 7.0More than 5 drinks 3.5 12.6

Recalled drunkdriving messageon beverage container 20.8 31.1 0.00

Container Handling 0.00Did not open or buy 37.2 20.6Opened or bought 20.6 20.2alcoholic beverageOpened and bought 42.2 59.2alcoholic beverage

Intervention to DeterDrunk DrivingAsked someone 48.7 45.8 0.29not to driveOffered to drive someone 47.3 49.7 0.38home yourselfAsked someone to take a 12.9 16.6 0.06taxi, bus, or subwayTried to take someone’s 21.7 21.8 0.98car keysAsked someone to 37.7 40.6 0.26stay at your home

TABLE 2. Estimated Coefficients and Reliability of ItemsMeasuring Intervention to Deter Drunk Driving (n = 1,376)

Measures b β R2

Asked someone not to drive 1.00 .968 .937Offered to drive someone home .911 .882 .778yourselfAsked someone to take a taxi, .671 .650 .423bus, or subwayTried to take someone’s car keys .897 .868 .753Asked someone to stay .796 .770 .593

at your home

Note. b: unstandardized factor loadings; β: standardizedfactor loadings; R2 : estimated reliability.

each of the five observed measures did not statisticallydiffer across gender.

3.3. Structural Equation Model

Because the measurement properties of the interven-tion factor were invariant across gender, the proposedpath model was evaluated for the entire sample. Sum-mary statistics of model fit indicated reasonable fit ofthe model with the data (χ2 = 62.14, df = 28, p =0.0002, RMSEA = .03, WRMR = .86). Figure 1 showsthe estimated standardized coefficients for the mea-surement and structural parts of the model. Table 3summarizes the regression parameters for the sociode-mographic variables that are included in the model.

All of the factor loadings for the intervention fac-tor were substantial and significant. All but one of theestimated regression parameters were significant anddemonstrated the relationships as hypothesized. Usualconsumption of three or more drinks on a drinkingday (standardized β = .28), greater handling of al-coholic beverages (β = .08), and recall of the drunkdriving message (β = .08) all had significant positivedirect effects on the intervention factor. As plausible, inaddition to direct effects, the relationship between con-tainer handling and intervention is mediated by recallof the drunk driving message. Persons who purchasedor opened alcoholic beverages within the last 30 dayswere more likely to recall the drunk driving message onthe warning label (β = .18), which subsequently has apositive effect on intervention.

With regard to sociodemographic characteristics,men were more likely to usually consume three or more

Human Factors and Ergonomics in Manufacturing & Service Industries DOI: 10.1002/hfm 543

Page 7: Do alcohol warning labels influence men's and women's attempts to deter others from driving when intoxicated?

Alcohol Warning Labels Tam and Greenfield

TABLE 3. Standardized Regression Parameters and Critical Ratios (in Parentheses) for Sociodemographic Variables Pre-dicting Alcohol Consumption, Container Handling, Recall of Drunk Driving Messages, and Intervention to Deter Other’sDrunk Driving

Variable Alcohol Consumption Container Handling Recall Drunk Driving Message Intervention

Male .30∗ (8.50) .13∗ (4.13) .09∗ (2.60) −11∗ (−3.73)White −.05 (−1.82) .02 (.76) −.02 (−.48) −.07∗ (−2.60)College or more −.16∗ (−4.86) .13∗ (4.47) .05 (1.36) −.01 (−.23)Age under 40 .20∗ (6.07) .00 (−.03) .23∗ (6.84) .30∗ (9.51)

Note: ∗ p < 0.05.

drinks (β = .30), handle alcoholic beverages (β = .13),and recall the drunk driving message (β = .09), butwere less likely to take action to intervene on another’sdrunk driving (β = –.11) than were women (Table 3).Thus the SEM result differs somewhat from the bivari-ate results. Persons who were 40 years old or youngerwere more likely to usually drink heavily (β = .20),recall the drunk driving message (β = .23), and takeaction by intervening to deter another’s drunk driv-ing (β = .30). Ethnic minorities were more likely thanWhites to intervene (β = –.07). These findings indicatevarying effects of gender, age, and ethnicity on drunkdriving message recall and intervention.

4. DISCUSSION

We take this investigation as a specific example of thetype of analysis that should more often be undertakenin the general case to study how warning messagesmight potentially enhance a recipient’s propensity tointervene so as to mitigate another’s risk behavior. Con-ceptually this might be so by reducing the sense of socialinappropriateness of taking such actions, although toconfirm this hypothetical mechanism of action wouldrequire further research. First, we showed that a shortresearch scale could be developed to assess interven-tion to deter another from driving when perceptuallyimpaired by alcohol. The measurement models we de-veloped were invariant to gender, indicating that theshort scale we developed to capture intervention toprevent others’ drunk driving is applicable for menand women.

Of more substantive importance, a major finding in-volves the relationship observed between awareness ofthe warning label’s message that “consumption of alco-holic beverages impairs your ability to drive a car... ” (asgauged by seeing the label and recalling that it included

a message about drunk driving) and a scale made upof five interventions to deter another individual fromdriving after drinking. Within the structural model,which controls for several plausible influences on bothlabel exposure and precautionary driving-related be-havior (Greenfield & Kaskutas, 1993), the effect ofwarning labels on intervening with others perceivedto be at risk for DWI was revealed. To our knowledge,this is the first such empirical evidence suggesting anindirect role for warning labels in legitimating pro-social risk-reducing behavior aimed at other individu-als, of a kind noted anecdotally in earlier news reports(Lemmens et al., 1999).

Although recall of the label’s drunk driving message,other factors taken account of in the model, appearsto be associated with increased intervention efforts, asignificant direct effect was found between gender andintervention, indicating that women are more likelythan men to take action to deter another’s drunk driv-ing. As mentioned in the introduction, there are severalplausible reasons that this may be so. Although some-what stereotypic, women may be socialized to take careof others, and they may nevertheless be affected morethan male drinkers by authoritative health warnings.This legitimation may free them more to discouragedrunk driving by partners or guests. Clearly, what un-derlies this gender difference is speculative and callsfor further study. We believe that human factors stud-ies of this type could be productive in relation to manytypes of product warnings. Supervisors might be “au-thorized” by warnings on industrial equipment to takecorrective action; spouses might try to prevent theirpartners from a range of dangerous behaviors, “em-boldened” by recourse to an authoritative warning.Regarding alcohol, it is known that risk behaviors areengaged in more by men than women (Greenfield &Rogers, 1999a), and it appears, as here, that women

544 Human Factors and Ergonomics in Manufacturing & Service Industries DOI: 10.1002/hfm

Page 8: Do alcohol warning labels influence men's and women's attempts to deter others from driving when intoxicated?

Tam and Greenfield Alcohol Warning Labels

may more often attempt to prevent them from drunkdriving, especially when they recall the warning label’simpairment message.

4.1. Limitations

As with any such study, appropriate caveats and cau-tions are in order. Measurement of drunk driving mes-sage recall was based on a leading question about drunkdriving rather than asking respondents open-endedquestions about the warning label messages. A less “di-rected” measure of recall might increase or decreasethe observed correlation between salient recall andintervention leading to findings different from thosereported here. Additionally, the results are based onself-report surveys, involving cross-sectional, not lon-gitudinal, data. Relationships are truly associational,and direction of causality cannot be determined. Fac-tors other than warning label recall, such as previousengagement in drunk driving, might increase the like-lihood of message recall and subsequently increase acollateral’s attempt to reduce another’s drunk driving.This being, results are consistent with a small effectof the appropriate sign. We believe that the structuralmodel helps to shed light on the possible pathways ofinfluence deserving further research.

4.2. Conclusions

The potential that alcohol container warning labelshave to instigate harm-reducing social discourse by en-couraging drinkers to urge on each other caution, whenheavy drinking happens in conjunction with risky situ-ations, is important for prevention (Greenfield, 1997).In this case, greater awareness of the warning label’s rel-evant message may encourage interactions with peopleclose to one who appears to have drunk too muchyet proposes to drive. We observed that respondentsreporting greater recall of the alcohol warning label’sdriving impairment message, other factors being equal,tended to also say they had headed off their asso-ciates from driving when they thought they had toomuch to drink. (Of course we know nothing aboutthe success of these efforts.) Similar results have beenreported that are consistent with significant but smalleffects of the warning label, including sparking con-versations about several other topics on the warninglabel, notably drinking and pregnancy and drinkingand health, respectively (Greenfield et al., 1999). To-gether with the present finding, these may be viewedcontextually as collectively promoting more prudent

situational norms (Greenfield & Room, 1997) that maytend to discourage risk taking with alcohol. Our anal-yses support the possibility that government warningmessages on alcohol containers may enhance a collat-eral’s attempts to reduce another’s drinking and driv-ing to some demonstrable, if small, degree. Such in-fluences (or more properly associations) are consis-tent with Congressional intent, when enacting the fed-eral warning label legislation, that the label serves to“remind” drinkers of certain health risks of alcoholconsumption, like drinking when pregnant or whendriving (Greenfield et al., 1999; Kaskutas, 1995). Webelieve that the illustration provided here, outlining anindirect social mechanism by which labels may helppromote safer alcohol-related behavior, is among themore important products of the nationally mandatedwarning label policy and its evaluation. The examplealso suggests that research on other types of productwarnings might productively measure potential effectson various significant others (including supervisors)that might encourage them to reinforce a product user’scompliance with particular messages or reduce warnedrisks.

ACKNOWLEDGMENTS

Data were drawn from the Alcohol Research Group’sImpact of Alcoholic Beverage Warning Labels projectfunded by Grant No. R01 AA08557 from the NationalInstitute on Alcohol Abuse and Alcoholism. Analysesand writing were funded by this and National AlcoholResearch Center Grant No. P30 AA05595 to the AlcoholResearch Group, Public Health Institute. The authorsthank Dr. Lee Ann Kaskutas for her valuable input inthe development of the survey instrument, collectionof the data, and commenting on a draft of this article.

References

Berger, D. E., & Snortum, J. R. (1985). Alcoholic bever-age preferences of drinking-driving violators. Journalof Studies on Alcohol, 46, 232–239.

Bollen, K. A. (1989). Structural equations with latent vari-ables. New York: John Wiley & Sons.

Browne, M. W., & Cudeck, R. (1993). Alternative waysof assessing model fit. In K. A. Bollen & J. S. Long(Eds.), Testing structural equation models (pp. 136–162). Newbury Park, CA: Sage Publications.

Cook, T. D., & Campbell, D. T. (1979). Quasi-experimentation: Design and analysis issues for fieldsettings. Boston, MA: Houghton Mifflin.

Human Factors and Ergonomics in Manufacturing & Service Industries DOI: 10.1002/hfm 545

Page 9: Do alcohol warning labels influence men's and women's attempts to deter others from driving when intoxicated?

Alcohol Warning Labels Tam and Greenfield

Dufour, M. C. (1999). What is moderate drinking? Defin-ing “drinks” and drinking levels. Alcohol Health & Re-search World, 23(1), 5–14.

Frey, J. (1989). Survey research by telephone. NewburyPark, CA: Sage Publications.

Fukuyama, M. A., & Greenfield, T. K. (1983). Dimensionsof assertiveness in an Asian-American student popula-tion. Journal of Counseling Psychology, 30(3), 429–432.

Graham, K., Wilsnack, R., Dawson, D., & Vogeltanz,N. (1998) Should alcohol consumption measures beadjusted for gender differences? Addiction, 93, 1137–1147.

Greenfield, T. K. (1994). Improving causal inference fromnaturalistic designs: Self-selection and other issues inevaluating alcohol control policies. Applied BehavioralScience Review, 2(1), 45–61.

Greenfield, T. K. (1997). Warning labels: Evidence of harmreduction from long-term American surveys. In M.Plant, E. Single, & T. Stockwell (Eds.), Alcohol: Min-imizing the harm (pp. 105–125). London: Free Associ-ation Books.

Greenfield, T. K., Giesbrecht, N., & Kavanagh, L. (1996).Drinking patterns of young adults in the US andCanada: The policy implications. International Alco-hol Policy Conference, Toronto, Canada, May 4–8.

Greenfield, T. K., Graves, K. L., & Kaskutas, L. A. (1993).Alcohol warning labels for prevention: National surveyfindings. Alcohol, Health and Research World, 17(1),67–75.

Greenfield, T. K., Graves, K. L., & Kaskutas, L. A. (1999).Long-term effects of alcohol warning labels: Findingsfrom a comparison of the United States and Ontario,Canada. Psychology & Marketing, 16(3), 261–282.

Greenfield, T. K., & Kaskutas, L. A. (1993). Early im-pacts of alcoholic beverage warning labels: Nationalstudy findings relevant to drinking and driving behav-ior. Safety Science, 16, 689–707.

Greenfield, T. K., & Kaskutas, L. A. (1998). Five years’exposure to alcohol warning label messages and theirimpacts: Evidence from diffusion analysis. Applied Be-havioral Science Review, 6(1), 39–68.

Greenfield, T. K., & Rogers, J. D. (1999a). Alcoholic bev-erage choice, risk perception, and self-reported drunkdriving: Effects of measurement on risk analysis. Ad-diction, 94(11), 1735–1743.

Greenfield, T. K., & Rogers, J. D. (1999b). Who drinksmost of the alcohol in the U.S.? The policy implications.Journal of Studies on Alcohol, 60(1), 78–89.

Greenfield, T. K., & Room, R. (1997). Situational normsfor drinking and drunkenness: Trends in the U.S. adultpopulation, 1979–1990. Addiction, 92(1), 33–47.

Hennessy, M., & Saltz, R. F. (1990). The situational riski-ness of alcoholic beverages. Journal of Studies on Alco-hol, 51, 422–427.

Kaskutas, L. A. (1995). Interpretations of risk: The use ofscientific information in the development of the alcoholwarning label policy. The International Journal of theAddictions, 30(12), 1519–1548.

Kaskutas, L. A., & Greenfield, T. K. (1992). First effects ofwarning labels on alcoholic beverage containers. Drugand Alcohol Dependence, 31, 1–14.

Kaskutas, L. A., & Greenfield, T. K. (1997). The role ofhealth consciousness in predicting attention to healthwarning messages. American Journal of Health Promo-tion, 11(3), 186–193.

Lemmens, P. H., Vaeth, P. A. C., & Greenfield, T. K. (1999).Coverage of beverage alcohol issues in the print media inthe US, 1985–1991. American Journal of Public Health,89(10), 1555–1560.

Muthen, L. K., & Muthen, B. (1998–2007). Mplus user’sguide (4th ed.). Los Angeles: Muthen & Muthen.

Newcomb, M. D. (1994). Drug use and intimate rela-tionships among women and men: Separating specificfrom general effects in prospective data using structuralequation models. Journal of Consulting and ClinicalPsychology, 62, 463–476.

Public Law No. 100–690. (1988). Alcoholic Beverage La-beling Act of 1988. Washington, D.C.: 100th Congress,2nd session.

Rogers, J. D., & Greenfield, T. K. (1999). Beer drinkingaccounts for most of the hazardous alcohol consump-tion reported in the United States. Journal of Studies onAlcohol, 60(6), 732–739.

Room, R., Greenfield, T., & Weisner, C. (1991). Peoplewho might have liked you to drink less: Changing re-sponses to drinking by U.S. family members and friends,1979–1990. Contemporary Drug Problems, 18(4), 573–595.

Wickens, T. D. (1989). Multiway contingency tables anal-ysis for the social sciences. Hillsdale, NJ: Lawrence Erl-baum Associates.

Yu, C.-Y., & Muthen, B. (2001). Evaluation of model fitindices for latent variable models with categorical andcontinuous outcomes. Los Angeles, CA: Social ResearchDivision, Graduate School of Education & InformationStudies, UCLA.

546 Human Factors and Ergonomics in Manufacturing & Service Industries DOI: 10.1002/hfm