- Email: [email protected]
An alternative to standard drinks as a measure of alcohol consumption
Journal of Substance Abuse 12 (2000) 67 ± 78
An alternative to standard drinks as a measure of alcohol consumption Lee Ann Kaskutas*, Karen Graves Alcohol Research Group, 2000 Hearst Avenue, Suite 300, Berkeley, CA 94709, USA
Abstract Despite the field's longstanding concern with underreporting of alcohol consumption, traditional survey questions encourage error because respondents often must calculate their number of drinks based on standard drink sizes that often do not match their own drinking style. This study considered how often respondents' self-defined drink sizes matched a `standard' drink size based on approximately 12 g of ethanol for six different beverages. We also studied whether respondents could accurately judge the size of their drinks. Subjects were recruited and interviewed at urban prenatal clinics, health clinics, and via snowball referrals and community outreach in Los Angeles and the San Francisco Bay Area. Because of the urgency of accurate measurement of consumption during pregnancy, urban pregnant women from the groups most at risk for Fetal Alcohol Syndrome, Native Americans (n = 102) and African Americans (185), were targeted. A small comparison group of urban pregnant white women (n = 34) was included. One-hour in-person interviews were conducted. Self-defined drink sizes were determined for each beverage consumed, using models and photographs of vessels. Frequent drinkers and the majority of women who reported drinking higher alcohol content beverages reported drinking larger-than-standard drink sizes. The median size of a malt liquor drink among the daily drinkers was almost three times as large as the standard, their fortified wine drinks were four times the standard, and their spirits drinks were six times the standard size. The majority of drinkers of each beverage were unable to accurately judge the size of their drinks, underestimating the number of fluid ounces by about 30%. Although the vessels methodology used here must be refined and tested further on other populations (e.g., men, nonpregnant women, and all ethnic groups), results suggest that determination of risk levels should be based on survey data that takes into consideration the beverage mix and the actual size of respondents' alcohol drinks. D 2001 Elsevier Science Inc. All rights reserved. Keywords: Measurement; Methodology; Consumption; Drink size; Pregnancy
* Corresponding author. Tel.: +1-510-642-1751; fax: +1-510-642-7175. E-mail address: [email protected] (L.A. Kaskutas) 0899-3289/00/$ ± see front matter D 2001 Elsevier Science Inc. All rights reserved. PII: S 0 8 9 9 - 3 2 8 9 ( 0 0 ) 0 0 0 4 2 - 0
68
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
1. Introduction The validity of self-reported alcohol consumption, especially in terms of underreporting, is a longstanding methodological concern of researchers (Ernhart, Morrow-Tlucak, Sokol, & Martier, 1988; Midanik, 1982; Rosett & Weiner, 1984; Sokol, Martier, & Ernhart, 1985). Coverage of alcohol consumption in national surveys is quite low, with only a third (Rogers & Greenfield, 1999) to a half (Armor, Polich, & Stambul, 1978) of the yearly alcohol sales volume generally accounted for by extrapolations based on respondents' answers about their drinking in the past 12 months (also see Lemmens, 1994). Experts believe that errors in drink size represent a likely contributor to this undercoverage of consumption, especially for beverages that do not come in drink-size containers (Dawson, 1998). For example, Dawson has reported that the average drink of wine is a medium size wine glass of about 8 oz, twice the `standard' of 4 oz presumed in most surveys (Dawson, Grant, & Chou, 1995). Yet, it is common practice to expect respondents to think about their drinking in terms of standard drinks, even though some have never heard of the term `standard drink' and frequently pour larger-than-standard servings of beer, wine, and spirits (Carruthers & Binns, 1992). To address this concern in part, some studies (such as NLAES, NIAAA's National Longitudinal Alcohol Epidemiologic Survey) now simply ask respondents how many ounces are in their standard drinks of beer, wine, and liquor (Dawson et al., 1995; Williams, Proudfit, Quinn, & Cambell, 1994), on the assumption that respondents either know or are able to accurately estimate their drink sizes for different alcoholic beverages. Significantly higher estimates of heavier drinking among women (but not among men) have been found when beverage-specific data based on respondent-defined drink size were gathered this way (Williams et al., 1994). This underreporting effect among women is compounded by their preference for higher alcohol content beverages (wine and spirits) (Dawson, 1993). To minimize the error associated with estimating or remembering the number of ounces in one's alcoholic drinks, another innovation is to interview subjects at home and have them pour water into the actual glasses from which they drink alcoholic beverages (Carruthers & Binns, 1992; Lemmens, 1994). This has the drawback of being limited to household surveys (and only directly applicable to drinking done at home, with those glasses), but is a powerful methodology, accounting for 12% more drinking among women and 6% more for men. As above, the greatest discrepancy comes from the stronger beverages, spirits (26% higher using self-pour) and fortified wines (14% higher) (Lemmens, 1994). In order to avoid the need for interviewing respondents in their homes, a variation on the latter approach used a variety of actual vessels in a laboratory protocol and setting, showing respondents life-size models of typical glasses, bottles, and cans for drinking beer, wine, wine coolers, and spirits (Russel et al., 1997). For each beverage type, respondents chose the vessel they usually drink from; back-up photographs of additional vessel shapes and sizes were available when the vessel models were insufficient. Respondents also were asked how high they pour, if they drank from a glass. Among moderate (one drink per day) drinkers, about twice as many ounces of absolute alcohol were reported overall when respondents defined their drink size using vessels (Nochajski et al., 1999), with the greatest difference for spirits (over four times more ounces).
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
69
Because of concern regarding accurate assessment of women's drinking before and during pregnancy, the senior author (Kaskutas) modified Russell's laboratory protocol (above) for use in the field. Using vessel models expanded to include malt liquor (which often comes in `standard' 16- and 40-oz containers), samples of pregnant African American, Caucasian, and Native American women were recruited and interviewed in diverse urban settings including busy prenatal clinics, Native American cultural events, and health vans (Branco, Witbrodt, & Kaskutas, 1998). We report here on the results of that study, focusing on questions about their drinking during the 12 months prior to knowing they were pregnant. In this paper we compare the self-selected portion size to a standard size for each beverage type consumed, and study whether self-selected vessel size is related to frequency of drinking or ethnicity; we also evaluate the ability of the respondent to estimate the volume of their self-selected portion sizes. 2. Materials and methods 2.1. Sample The study was designed to gather comparable data about drinking before and during pregnancy among understudied urban Native American and African American women, with a small comparison group of pregnant Caucasian women. We report here on drinking during the year prior to pregnancy. The study was conducted in Los Angeles and the San Francisco Bay Area between July 1996 to July 1997. The sample was recruited in health and prenatal clinics and was augmented by snowball sampling and outreach to women not necessarily seeking prenatal care. As much as possible, ethnicity of respondents and interviewers were matched. A $15 incentive was given. In-person interviews 1 h in length were conducted with a total of 102 Native Americans, 185 African Americans, and 34 Caucasians (74% response rate). Methods and procedures followed were in accord with the standards of the Public Health Institute's Institutional Review Board for the Protection of Human Research Subjects. Additional details of the sampling and recruitment strategy may be found in Branco et al. (1998). Of the 321 pregnant urban women interviewed, 221 women (69%) reported prepregnancy alcohol use and are included in the analysis presented here. The ethnic breakdown of the drinking subsample is 56% African American, 34% Native American, and 11% Caucasian. A majority (62%) reported being single (53% had never married and 9% were separated or divorced); 38% had not completed high school; and the average age of the interviewees was 26 years. 2.2. Measures Frequency of consumption for the 12 months before pregnancy was assessed by asking respondents how often they had any alcoholic beverage (Clark & Hilton, 1991). For use here, their open-ended responses were grouped into three categories: daily, weekly, and infrequent.
70
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
The methodology for determining self-defined drink size was patterned after Russell's method of showing vessels and photos premarked with letters that correspond to the number of fluid ounces (subsequently referred to simply as ounces) when filled to different levels (Russell et al., 1997). For example, the letter X indicated 2 oz, A indicated 4 oz, B indicated 6 oz, etc.; and a number of different glass shapes appropriate for drinking wine, beer, etc. had the letter X written at the 2-oz level, and so on. Respondents were not told the number of ounces associated with a given letter. Diverse drinking vessels, as well as alcohol container shapes commonly sold in California urban retail settings, were included for beer, malt liquor, wine, wine coolers, fortified wine, and spirits. Respondents simply picked the can, bottle, or glass they mainly used for a given beverage; and for glasses, they were additionally asked to specify ``About how high would you fill it? Just tell me the letter.'' For containers other than glasses Ð that is, bottles and cans Ð there was a single letter on the vessel corresponding to the number of ounces the container held (as sold). Facilitated by these vessels and by photographs of other types of containers, consumption was assessed using respondentdefined drink sizes for each alcoholic beverage they consumed. To determine whether subjects had an accurate understanding of how big their drinks were, we also asked respondents how many ounces they thought were in the drink they had just indicated. 2.3. Analysis The statistical methods consisted of nonparametric (chi-square, Kruskal±Wallis) procedures conducted using SPSS (SPSS, 1996). For each beverage consumed, we compared the respondent's self-selected portion size to a standard size for that beverage such that the ethanol content (approximately 0.5 fl. oz (12 g) per drink) was similar across all beverages. Thus, the standard sizes used here are: 12 fl. oz for beer and wine coolers; 8 oz for malt liquor; 4 oz for wine; 3 oz for fortified wine; and 1 oz for spirits. A `conversion ratio' was calculated post hoc for each beverage that a respondent consumed, which compared her self-defined drink size to the standard size for that beverage. For example, assume that a respondent selected a 16-oz vessel for her beers, and that for wine she pointed to a glass filled to the 5-oz mark. Her drinks, then, are larger than the respective standard drink sizes, by a ratio of 16 divided by 12 for beer (since the standard size for beer is 12 oz, but her beers are 16 oz); and by a ratio of 5/4 for wine (standard wine is 4 oz, and her wine drink is 5 oz). Thus, when she drinks a beer, she is actually having 1.33 standard drinks (16 oz divided by 12 oz) and when she drinks wine, she is having 1.25 standard drinks (5 oz divided by 4 oz). We refer to the latter numbers as the beverage-specific `conversion ratios.' In this sample, there were individuals who indicated large self-selected portion sizes, with 11% of the beer drinkers reporting 40 oz as their drink size and 10% of the liquor drinkers indicating drink sizes in excess of 12 oz. Due to concern with such (potential) outliers, and because this survey methodology is new, we believe that a median value may represent a better measure of central tendency, as it is less dependent upon extreme values. Thus, we show median conversion ratios (i.e., the conversion ratio for women at the median drink size for each beverage) in the analysis presented below.
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
71
In addition to the conversion ratio comparing self-defined drink size to standard drink size, we also calculated an estimation ratio comparing women's judgment of her indicated drink size vs. the true (actual) size of her indicated drink (for each beverage consumed). Again, the median is our focus here. The median was obtained by computing this estimation ratio for each woman, then taking the median of these ratios. 3. Results A synopsis of the general drinking characteristics of the sample provides context for our analysis of self-defined drink size. Approximately 60% of the 221 prepregnancy drinkers reported drinking at least weekly, and 28% indicated that they had consumed alcohol on a daily basis during the year prior to pregnancy. Half of the sample reported weekly or more frequent consumption of three or more drinks per occasion (3+), and almost one third reported five or more (5+) at that frequency. Further, a quarter of the women reported daily drinking at the 3+ level and 16% at the 5+ daily level. Thus, this is a sample of frequent, somewhat heavy drinkers. They tended to select a variety of alcoholic beverage types: Beer was the most frequently mentioned, and fortified wine the least often consumed alcoholic beverage by this group of urban minority women. There were very few women who drank only one beverage type. 3.1. Self-defined drink sizes The median self-selected drink sizes in this sample are as follows: 22-oz malt liquor; 12-oz beer; 12-oz wine cooler; 8-oz fortified wine; 5-oz wine; 2-oz spirits. The large serving sizes observed for malt liquor and fortified wine reflect the fact that 24 women (25% of the malt drinkers) selected 40 oz as their drink size, and for fortified wine seven women selected a drink size of greater than 16 oz (results not shown). Table 1 reports the proportion of women in this sample whose self-selected drink sizes were smaller than, the same as, or larger than the standard size for each beverage consumed. For those having larger Ð or smaller Ð than standard drinks, the magnitude of difference between self-selected and standard size is reflected in the median conversion ratios shown for each beverage. About half of the beer and wine drinkers, three-fourths of the spirits drinkers, and almost all of the fortified wine and malt liquor drinkers indicated portion sizes that were greater than the assigned standard size. Only among wine cooler drinkers did the majority (88%) select a serving size that was identical to the standard size of 12 oz. For all beverage types, the median self-selected drink size among those who drank more than the standard ranged from 1.5 to 4 times that of the standard size. Among all respondents who consumed the beverage, the median values for the self-selected drink size for wine, spirits, malt, and fortified wine, are 1.3±2.8 times that of the standard portion size, suggesting that for this sample of urban minority women, alcohol intake is being underestimated using methodology based on standard drink size. We further examined whether there was any difference in magnitude of self-selected vessel size among women who were frequent drinkers (weekly and daily) as compared to those
b
15 8 1 4 12 7
Less than standard (%) 38 42 24 2 0 88
Standard (%)
Limited to women who have completed both measures. Conversion ratio = self-select portion size/standard portion size.
12 4 1 8 3 12
Beer Wine Spirits Malt liquor Fortified wine Wine cooler
a
Standard size (fl. oz)
Beverage 48 50 75 94 88 5
More than standard (%)
Percentage whose self-defined drink size was:
Comparison of self-selected vessel size to standard sizea
Table 1 Conversion ratios and proportions selecting standard vs. nonstandard drink sizes
0.63 0.50 na 0.84 0.67 0.50
Less than standardb 1 1 1 1 1 1
Standard
1.90 1.50 4.00 2.81 3.00 2.80
More than standardb
1.0 1.3 2.0 2.8 2.7 1.0
Overallb
Median conversion ratios when self-defined drink size was:
72 L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
73
Table 2 Do frequent drinkers have larger drink sizes than less frequent drinkers? Conversion ratios for frequent and infrequent drinkersa,b Median ratio
Median ratio
Median ratio
Beverage
Infrequent
Weekly
Daily
Significance of K ± W testc
Beer (n = 122) Wine (n = 62) Spirits (n = 116) Malt liquor (n = 96) Fortified wine (n = 34) Wine cooler (n = 102)
1.00 1.00 2.00 1.50 1.30 1.00
1.00 1.50 4.00 2.00 2.00 1.00
1.90 1.50 6.00 2.81 4.00 1.00
.000 .039 .000 .001 .019 .046
a b c
Limited to women who have completed both measures. conversion ratio = self-select portion size/standard portion size. Because of multiple comparisons, a P value of .008 was deemed statistically significant.
drinking with less regularity (infrequent drinkers) (Table 2). Again, we applied the ratio of self-selected size to standard size to make the comparison, using a nonparametric test (Kruskal±Walls) to assess differences in self-selected portion size among women classified as infrequent, weekly, and daily drinkers. Women who drank daily were more likely to select portion sizes that were larger than their less frequent drinking counterparts. This observation was noted for most beverage types but only reached statistical significance for beer, spirits, and malt liquor. When translated into volume (oz) of the serving size, the differences between infrequent and frequent drinkers become more pronounced. For example, malt liquor drinkers who drank infrequently reported a median serving size of 12 oz, but among those who drank weekly, the median serving size for malt liquor was 16 oz; and for the daily drinkers, the median serving size became a 22-oz malt liquor. No statistically significant differences in self-selected portion sizes for any beverage type were observed between African Americans vs. Native Americans (Caucasians were not included in the comparison because of their small numbers; result not shown). 3.2. Judgment of drink size As a measure of respondents' ability to estimate the volume of their drink size, they were also asked to indicate their estimate, in ounces, of the volume of the portion size they had just selected. Results of the comparison between the estimated size and true size are presented in Table 3. Few women correctly estimated the volume of the portion size that they had selected. The beverages estimated with the most accuracy were beer, malt liquor, and wine coolers; but even here, only about two-fifths of the women were able to correctly judge the volume of their self-selected portion size. The tendency was for women to underestimate portion size. The proportion of respondents who underestimated their selfselected drink size ranged from 43% to 55%, with the median estimation ratio (of the estimated self-selected size divided by true size) ranging from 0.6 for spirits to 0.8 for fortified wine. To determine whether those indicating the standard size for a given beverage were more knowledgeable of the number of ounces in that serving, we selected those individuals who
74
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
Table 3 Do respondents know how big their drinks are? Comparison of estimated size to true size of self-selected vessela Percentage whose estimate was:
Median estimation ratios when estimate was:
Beverage
Less than true size
Same as true size
More than true sizeb
Less than true size
Same as true size
More than true sizeb
Beer Wine Spirits Malt liquor Fortified wine Wine cooler
44% 55% 43% 46% 50% 43%
44% 28% 21% 42% 30% 41%
12% 18% 36% 12% 20% 17%
0.7 0.7 0.6 0.7 0.8 0.7
1 1 1 1 1 1
1.5 1.3 2.0 1.3 1.8 1.3
a
Limited to women who answered both sets of items on self-selected vessel size. Median estimation ratio = respondent's estimated size/true size of vessel. Size shown is for women at the median size for each category. b
had chosen as their self-selected portion size the assigned standard size, and examined their estimates of portion size (results not shown). The proportion of respondents identifying the correct volume improved for all beverages excepting wine, where the tendency was still to underestimate the portion size. Interestingly, the heavier drinkers of standard drinks were no better at estimating portion size than the less frequent drinkers. 4. Discussion The results of this pilot study are provocative and raise the question of ``how well do current standard measures of alcohol estimate intake, particularly in minority populations?'' While results do not suggest that researchers abandon the `standard drinks' formulation, they do argue strongly for alternative methods that can illuminate the research question at hand. Our findings of a discrepancy between self-defined and standard portion sizes are consistent with other studies that have examined the issue of portion size (Carruthers & Binns, 1992; Lemmens, 1994; Wilson, 1981). Generally, individuals are consuming more of a specific beverage than is presumed. Our study focused only on women, but other surveys have found that the effect of the difference between actual and standard size of drinks on overall consumption is greater for women that men, because women consume other alcoholic beverages besides beer which is the dominant beverage for men (Dawson, 1993; Lemmens, 1994). The present study further confirms the result of Lemmens (1994) in that the greatest differences between self and standard sizes are for those beverages with higher volume percentage of ethanol that have smaller standard serving sizes, namely fortified wine and spirits. The strength of the agreement between the present and prior studies provides face validity for the `vessel' methodology used here; that is, the use of a set of vessels and photos premarked with letters that correspond to given volumes. This methodology has the advantage of being transportable for use in diverse field (vs. justin-home interview) settings.
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
75
In addition to the larger discrepancy for higher alcohol content beverages, our study also found that for the more frequent (daily) drinkers, selected beverage sizes were four to six times larger than the standard sizes. This adds to previous knowledge in the literature regarding underestimates of consumption for moderate drinkers (average daily volume of one standard drink), among whom ethanol intake was doubled when they were allowed to self-define their drink size using vessels (Nochajski et al., 1999). Thus, the impact of the difference between self-defined and standard drink size on overall consumption of alcohol is greatest among the heaviest groups of drinkers, when definitions of heavy drinking are based on frequency of consumption or strength of beverage consumed. This has implications for undercoverage of alcohol intake, in that discrepancies between actual and standard drink sizes for the heavier drinkers contributes to more undercoverage than if this occurred among less heavy drinkers. There is little conformity on the definition of standard serving sizes in general; for example, a standard glass of wine can vary from 3 to 6 oz (Turner, 1990). Some researchers have defined a serving size based on the amount of absolute alcohol (standard alcohol units) delivered, and have attempted to keep alcohol content relatively constant across beverage types (Miller, Heather, & Hall, 1991). However, the international alcohol community has not been able to reach consensus on the definition of an alcohol unit (Miller et al., 1991). In the US, drinking guidelines have been set, which recommend drink sizes that deliver approximately 14 g (International Center for Alcohol Policies (ICAP), 1996). Yet, in a review of 27 studies (Turner, 1990), the standard drink used on average provided 12 g (about 0.5 oz) of ethanol, as did the more recent Project MATCH (Miller & Del Boca, 1994). In surveys conducted at the Alcohol Research Group (Clark & Hilton, 1991; Greenfield, Graves, & Kaskutas 1999; Greenfield, Midanik, & Rogers, 2000), the standard drink sizes as defined for beer, wine and liquor provide 0.4 oz of ethanol (for spirits) to 0.53 oz (for beer). In order to deliver a similar amount of ethanol, a standard serving of malt liquor must be set (as we did here) at 8 oz, a size not sold. Since ethanol content also varies within a beverage type (Martin & Nirenberg, 1991) especially among the stronger malt liquors and fortified wines that are popular in some areas, determining the size for standard drinks will inevitably involve some error. In our sample, the most frequently selected size for malt liquor was 40 oz rather than our presumed 8 oz standard malt liquor; and the modal size of drink for fortified wine was 6 oz instead of the 3 oz assumed in the literature. However, for the other beverages, the most frequently selected size was actually the standard size: about a third of the beer drinkers have 12 oz beers; two-fifths of the wine drinkers indicated a 4 oz pour of wine and twice that proportion of wine coolers' drinkers have 12 oz coolers; and about a fourth of the spirits drinkers have 1-oz drinks. Thus, the extent of error really depends on the beverage mix of the population under study (and does not appear to vary by ethnicity, when controlling for beverage type). Given our experience with this study, we believe that when more accuracy is needed (as with studies assessing risk, or striving to more fully account for alcohol sales data), it may prove prudent to abandon the use of standard drink definitions when measuring alcohol consumption, and to instead allow respondents to simply indicate their portion sizes up front, at the beginning of the consumption series of questions, for each beverage they consume.
76
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
This would enable post hoc conversion by the analyst to a common metric (be that a standard alcohol unit or beverage-specific standard drink sizes), which could be clearly defined at the data analysis stage. This seems more reliable than expecting respondents to convert their self-defined serving sizes into standard sizes, especially across beverages. This recommendation is reinforced by current work underway at ARG using protocol analysis, in which respondents were asked to `think aloud' as they answered the graduated frequency series of questions (Clark & Hilton, 1991). One confused respondent asked for a calculator; another asked for help figuring out how many times 12 oz (of beer) goes into 40 oz (his standard beer). Note, however, that we are not suggesting that future studies simply ask respondents how large their drinks are. As shown in Table 3, our sample of minority pregnant urban women were not very accurate in judging the number of ounces that the letter on the vessel or photograph represented. Thus, some form of vessels probably must be used. We appreciate that the vessels methodology will increase the administration time for alcohol surveys. At least among the population under study here, the underreporting bias seems to be dependent upon the mix of beverages consumed by the participant, which varies. Thus, it seems important that alcohol researchers at least conduct preassessment inquiries to determine the mix of beverages consumed by the population of interest, and when possible, the distribution of drink sizes for each. For example, most studies of the general population do not collect data on malt liquor, which proved to be an important beverage type for these urban minority women. Since malt liquor comes in larger containers (than beer) and has higher alcohol content (than beer), overall consumption will be underestimated for populations where malt liquor is a beverage of choice (as appears to be the case for urban African American pregnant women (Graves & Kaskutas, 1999; Kaskutas, 1999) and urban African American men (Brown Taylor, 1999)). Measuring alcohol consumption is no easy task and may best be approached by using a multiplicity of strategies Ð among which would be collecting beverage-specific information, using respondent defined drink sizes, and in a way such that drinking patterns are distinguishable and can be separated from general measures of consumption. One final concern is respondent's ability to estimate the size of their drinks without visual aids like vessels or photographs. We found in our study that even the women who drank the standard size of a particular beverage were just as often wrong about the size of those drinks as were the women whose drinks were larger than the standard. Additional findings from this study, focusing on drinking during pregnancy among this sample, also are available (Kaskutas, in press) and are informed by the methodology presented here. In closing, although the challenges are many, increased accuracy of alcohol survey data would allow for more refined determination of risk levels, and for intervention strategies to be tailored to the needs of specific population subgroups. Acknowledgments This work was supported by a Center grant from the National Institute of Alcohol Abuse and Alcoholism to the Alcohol Research Group (#P50 AA05595), supplemented by the Office of Minority Health.
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
77
References Armor, D. J., Polich, J. M., & Stambul, H. B. (1978). Alcoholism and treatment. New York: Wiley. Branco, E. S., Witbrodt, J., & Kaskutas, L. A. (1998). Technical report for the DEER project. Berkeley, CA: Alcohol Research Group. Brown Taylor, D. (1999). Knowledge, attitudes and malt liquor beer drinking behavior among African American men in south central Los Angeles. Los Angeles, CA: California School of Professional Psychology. Carruthers, S. J., & Binns, C. W. (1992). The standard drink and alcohol consumption. Drug and Alcohol Reviews, 11, 363 ± 370. Clark, W. B., & Hilton, M. (Eds.). (1991). Alcohol in America: drinking practices and problems. Albany, NY: State University of New York Press. Dawson, D. A. (1993). Patterns of alcohol consumption: beverage effects on gender differences. Addiction, 88, 133 ± 138. Dawson, D. A. (1998). Measuring alcohol consumption: limitations and prospects for improvement. Addiction, 93, 965 ± 968; discussion 969 ± 977. Dawson, D. A., Grant, B. F., & Chou, P. S. (1995). Gender differences in alcohol intake. In: W. A. Hunt, & S. Zakhari (Eds.), Research monograph no. 29: stress, gender, and alcohol-seeking behavior ( pp. 3 ± 21). Bethesda, MD: National Institutes of Health, National Institute on Alcohol Abuse and Alcoholism (NIH Publication No. 95-3893 Edition). Ernhart, C., Morrow-Tlucak, M., Sokol, R., & Martier, S. (1988). Underreporting of alcohol use in pregnancy. Alcoholism: Clinical and Experimental Research, 12, 506 ± 511. Graves, K., & Kaskutas, L. A. (1999). Impact of ethnicity and drinking frequency on alcoholic beverage choice. Paper presented to the 127th annual American Public Health Association meeting. Greenfield, T. K., Graves, K. L., & Kaskutas, L. A. (1999). Long-term effects of alcohol warning labels: findings from a comparison of the United States and Ontario, Canada. Psychology and Marketing, 16, 261 ± 282. Greenfield, T. K., Midanik, L. T., & Rogers, J. D. (2000). A ten-year national trend study of alcohol consumption 1984 ± 1995: is the period of declining drinking over? American Journal of Public Health, 90, 47 ± 52. International Center for Alcohol Policies (ICAP). (1996). Safe alcohol consumption: a comparison of nutrition and your health: dietary guidelines for Americans and sensible drinking. In: ICAP reports I. Washington, DC: International Center for Alcohol Policies (ICAP). Kaskutas, L. A. (1999). Understanding drinking during pregnancy among urban Native and African Americans: health messages, risk beliefs, and how we measure consumption (invited presentation). Paper presented to the 22nd Annual Meeting of the Research Society on Alcoholism. Kaskutas, L. A. (2000). Understanding drinking during pregnancy among urban American Indians and African Americans: health messages, risk beliefs, and how we measure consumption. Alcohol Clin Exp Res, 24, 1241 ± 1250. Lemmens, P. H. (1994). The alcohol content of self-report and `standard' drinks. Addiction, 89, 593 ± 601. Martin, C. S., & Nirenberg, T. D. (1991). Alcohol content variation in the assessment of alcohol consumption. Addictive Behaviour, 16, 555 ± 560. Midanik, L. T. (1982). The validity of self-reported alcohol consumption and alcohol problems: a literature review. British Journal of Addiction, 77, 357 ± 382. Miller, W. R., & Del Boca, F. K. (1994). Measurement of drinking behavior using the form 90 family of instruments. Journal of Studies on Alcohol (Suppl. 12), 112 ± 113. Miller, W. R., Heather, N., & Hall, W. (1991). Calculating standard drink units: international comparisons. British Journal of Addiction, 86, 43 ± 47. Nochajski, T. H., Russell, M., Peirce, R. S., Chan, A. W. K., Freudenheim, J., Muti, P., Carosella, A. M., Trevisan, M. Assessment of alcohol consumption: quantity-frequency vs. semi-quantitative food frequency. Paper presented to the annual meeting of the Research Society on Alcoholism, June 26 ± July 1, 1999. Rogers, J. D., Greenfield, T. K. Does under-reporting of drinking affect survey estimates of the distribution of alcohol consumption? Paper presented to the Research Society on Alcoholism, June 26 ± July 1, 1999. Rosett, H. L., & Weiner, L. (1984). Alcohol and the fetus: a clinical perspective. New York: Oxford Univ. Press.
78
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
Russell, M., Marshall, J. R., Trevisan, M., Freudenheim, J., Chan, A. W. K., Markovic, N., Vana, J. E., & Priore, R. L. (1997). Test ± retest reliability of the cognitive lifetime drinking history. American Journal of Epidemiology, 146, 975 ± 982. Sokol, R. J., Martier, S., & Ernhart, C. (1985). Identification of alcohol abuse in the prenatal clinic. In: N. C. H. M., H. M. Chaos (Ed.), Early identification of alcohol abuse: proceedings of a workshop, October 31 ± November 1, 1983 ( pp. 209 ± 227). Rockville, MD: National Institute on Alcohol Abuse and Alcoholism (NIAAA research monograph no. 17). SPSS. (1996). SPSS for windows: base system user's guide, release 6.0. Chicago, IL: SPSS. Turner, C. (1990). How much alcohol is in a `standard drink'? An analysis of 125 studies. British Journal of Addiction, 85, 1171 ± 1175. Williams, G. D., Proudfit, A. H., Quinn, E. A., & Campbell, K. E. (1994). Variations in quantity ± frequency measures of alcohol consumption from a general population survey. Addiction, 89, 413 ± 421. Wilson, P. (1981). Improving the methodology of drinking surveys. Statistician, 30, 159 ± 167.
An alternative to standard drinks as a measure of alcohol consumption Lee Ann Kaskutas*, Karen Graves Alcohol Research Group, 2000 Hearst Avenue, Suite 300, Berkeley, CA 94709, USA
Abstract Despite the field's longstanding concern with underreporting of alcohol consumption, traditional survey questions encourage error because respondents often must calculate their number of drinks based on standard drink sizes that often do not match their own drinking style. This study considered how often respondents' self-defined drink sizes matched a `standard' drink size based on approximately 12 g of ethanol for six different beverages. We also studied whether respondents could accurately judge the size of their drinks. Subjects were recruited and interviewed at urban prenatal clinics, health clinics, and via snowball referrals and community outreach in Los Angeles and the San Francisco Bay Area. Because of the urgency of accurate measurement of consumption during pregnancy, urban pregnant women from the groups most at risk for Fetal Alcohol Syndrome, Native Americans (n = 102) and African Americans (185), were targeted. A small comparison group of urban pregnant white women (n = 34) was included. One-hour in-person interviews were conducted. Self-defined drink sizes were determined for each beverage consumed, using models and photographs of vessels. Frequent drinkers and the majority of women who reported drinking higher alcohol content beverages reported drinking larger-than-standard drink sizes. The median size of a malt liquor drink among the daily drinkers was almost three times as large as the standard, their fortified wine drinks were four times the standard, and their spirits drinks were six times the standard size. The majority of drinkers of each beverage were unable to accurately judge the size of their drinks, underestimating the number of fluid ounces by about 30%. Although the vessels methodology used here must be refined and tested further on other populations (e.g., men, nonpregnant women, and all ethnic groups), results suggest that determination of risk levels should be based on survey data that takes into consideration the beverage mix and the actual size of respondents' alcohol drinks. D 2001 Elsevier Science Inc. All rights reserved. Keywords: Measurement; Methodology; Consumption; Drink size; Pregnancy
* Corresponding author. Tel.: +1-510-642-1751; fax: +1-510-642-7175. E-mail address: [email protected] (L.A. Kaskutas) 0899-3289/00/$ ± see front matter D 2001 Elsevier Science Inc. All rights reserved. PII: S 0 8 9 9 - 3 2 8 9 ( 0 0 ) 0 0 0 4 2 - 0
68
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
1. Introduction The validity of self-reported alcohol consumption, especially in terms of underreporting, is a longstanding methodological concern of researchers (Ernhart, Morrow-Tlucak, Sokol, & Martier, 1988; Midanik, 1982; Rosett & Weiner, 1984; Sokol, Martier, & Ernhart, 1985). Coverage of alcohol consumption in national surveys is quite low, with only a third (Rogers & Greenfield, 1999) to a half (Armor, Polich, & Stambul, 1978) of the yearly alcohol sales volume generally accounted for by extrapolations based on respondents' answers about their drinking in the past 12 months (also see Lemmens, 1994). Experts believe that errors in drink size represent a likely contributor to this undercoverage of consumption, especially for beverages that do not come in drink-size containers (Dawson, 1998). For example, Dawson has reported that the average drink of wine is a medium size wine glass of about 8 oz, twice the `standard' of 4 oz presumed in most surveys (Dawson, Grant, & Chou, 1995). Yet, it is common practice to expect respondents to think about their drinking in terms of standard drinks, even though some have never heard of the term `standard drink' and frequently pour larger-than-standard servings of beer, wine, and spirits (Carruthers & Binns, 1992). To address this concern in part, some studies (such as NLAES, NIAAA's National Longitudinal Alcohol Epidemiologic Survey) now simply ask respondents how many ounces are in their standard drinks of beer, wine, and liquor (Dawson et al., 1995; Williams, Proudfit, Quinn, & Cambell, 1994), on the assumption that respondents either know or are able to accurately estimate their drink sizes for different alcoholic beverages. Significantly higher estimates of heavier drinking among women (but not among men) have been found when beverage-specific data based on respondent-defined drink size were gathered this way (Williams et al., 1994). This underreporting effect among women is compounded by their preference for higher alcohol content beverages (wine and spirits) (Dawson, 1993). To minimize the error associated with estimating or remembering the number of ounces in one's alcoholic drinks, another innovation is to interview subjects at home and have them pour water into the actual glasses from which they drink alcoholic beverages (Carruthers & Binns, 1992; Lemmens, 1994). This has the drawback of being limited to household surveys (and only directly applicable to drinking done at home, with those glasses), but is a powerful methodology, accounting for 12% more drinking among women and 6% more for men. As above, the greatest discrepancy comes from the stronger beverages, spirits (26% higher using self-pour) and fortified wines (14% higher) (Lemmens, 1994). In order to avoid the need for interviewing respondents in their homes, a variation on the latter approach used a variety of actual vessels in a laboratory protocol and setting, showing respondents life-size models of typical glasses, bottles, and cans for drinking beer, wine, wine coolers, and spirits (Russel et al., 1997). For each beverage type, respondents chose the vessel they usually drink from; back-up photographs of additional vessel shapes and sizes were available when the vessel models were insufficient. Respondents also were asked how high they pour, if they drank from a glass. Among moderate (one drink per day) drinkers, about twice as many ounces of absolute alcohol were reported overall when respondents defined their drink size using vessels (Nochajski et al., 1999), with the greatest difference for spirits (over four times more ounces).
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
69
Because of concern regarding accurate assessment of women's drinking before and during pregnancy, the senior author (Kaskutas) modified Russell's laboratory protocol (above) for use in the field. Using vessel models expanded to include malt liquor (which often comes in `standard' 16- and 40-oz containers), samples of pregnant African American, Caucasian, and Native American women were recruited and interviewed in diverse urban settings including busy prenatal clinics, Native American cultural events, and health vans (Branco, Witbrodt, & Kaskutas, 1998). We report here on the results of that study, focusing on questions about their drinking during the 12 months prior to knowing they were pregnant. In this paper we compare the self-selected portion size to a standard size for each beverage type consumed, and study whether self-selected vessel size is related to frequency of drinking or ethnicity; we also evaluate the ability of the respondent to estimate the volume of their self-selected portion sizes. 2. Materials and methods 2.1. Sample The study was designed to gather comparable data about drinking before and during pregnancy among understudied urban Native American and African American women, with a small comparison group of pregnant Caucasian women. We report here on drinking during the year prior to pregnancy. The study was conducted in Los Angeles and the San Francisco Bay Area between July 1996 to July 1997. The sample was recruited in health and prenatal clinics and was augmented by snowball sampling and outreach to women not necessarily seeking prenatal care. As much as possible, ethnicity of respondents and interviewers were matched. A $15 incentive was given. In-person interviews 1 h in length were conducted with a total of 102 Native Americans, 185 African Americans, and 34 Caucasians (74% response rate). Methods and procedures followed were in accord with the standards of the Public Health Institute's Institutional Review Board for the Protection of Human Research Subjects. Additional details of the sampling and recruitment strategy may be found in Branco et al. (1998). Of the 321 pregnant urban women interviewed, 221 women (69%) reported prepregnancy alcohol use and are included in the analysis presented here. The ethnic breakdown of the drinking subsample is 56% African American, 34% Native American, and 11% Caucasian. A majority (62%) reported being single (53% had never married and 9% were separated or divorced); 38% had not completed high school; and the average age of the interviewees was 26 years. 2.2. Measures Frequency of consumption for the 12 months before pregnancy was assessed by asking respondents how often they had any alcoholic beverage (Clark & Hilton, 1991). For use here, their open-ended responses were grouped into three categories: daily, weekly, and infrequent.
70
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
The methodology for determining self-defined drink size was patterned after Russell's method of showing vessels and photos premarked with letters that correspond to the number of fluid ounces (subsequently referred to simply as ounces) when filled to different levels (Russell et al., 1997). For example, the letter X indicated 2 oz, A indicated 4 oz, B indicated 6 oz, etc.; and a number of different glass shapes appropriate for drinking wine, beer, etc. had the letter X written at the 2-oz level, and so on. Respondents were not told the number of ounces associated with a given letter. Diverse drinking vessels, as well as alcohol container shapes commonly sold in California urban retail settings, were included for beer, malt liquor, wine, wine coolers, fortified wine, and spirits. Respondents simply picked the can, bottle, or glass they mainly used for a given beverage; and for glasses, they were additionally asked to specify ``About how high would you fill it? Just tell me the letter.'' For containers other than glasses Ð that is, bottles and cans Ð there was a single letter on the vessel corresponding to the number of ounces the container held (as sold). Facilitated by these vessels and by photographs of other types of containers, consumption was assessed using respondentdefined drink sizes for each alcoholic beverage they consumed. To determine whether subjects had an accurate understanding of how big their drinks were, we also asked respondents how many ounces they thought were in the drink they had just indicated. 2.3. Analysis The statistical methods consisted of nonparametric (chi-square, Kruskal±Wallis) procedures conducted using SPSS (SPSS, 1996). For each beverage consumed, we compared the respondent's self-selected portion size to a standard size for that beverage such that the ethanol content (approximately 0.5 fl. oz (12 g) per drink) was similar across all beverages. Thus, the standard sizes used here are: 12 fl. oz for beer and wine coolers; 8 oz for malt liquor; 4 oz for wine; 3 oz for fortified wine; and 1 oz for spirits. A `conversion ratio' was calculated post hoc for each beverage that a respondent consumed, which compared her self-defined drink size to the standard size for that beverage. For example, assume that a respondent selected a 16-oz vessel for her beers, and that for wine she pointed to a glass filled to the 5-oz mark. Her drinks, then, are larger than the respective standard drink sizes, by a ratio of 16 divided by 12 for beer (since the standard size for beer is 12 oz, but her beers are 16 oz); and by a ratio of 5/4 for wine (standard wine is 4 oz, and her wine drink is 5 oz). Thus, when she drinks a beer, she is actually having 1.33 standard drinks (16 oz divided by 12 oz) and when she drinks wine, she is having 1.25 standard drinks (5 oz divided by 4 oz). We refer to the latter numbers as the beverage-specific `conversion ratios.' In this sample, there were individuals who indicated large self-selected portion sizes, with 11% of the beer drinkers reporting 40 oz as their drink size and 10% of the liquor drinkers indicating drink sizes in excess of 12 oz. Due to concern with such (potential) outliers, and because this survey methodology is new, we believe that a median value may represent a better measure of central tendency, as it is less dependent upon extreme values. Thus, we show median conversion ratios (i.e., the conversion ratio for women at the median drink size for each beverage) in the analysis presented below.
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
71
In addition to the conversion ratio comparing self-defined drink size to standard drink size, we also calculated an estimation ratio comparing women's judgment of her indicated drink size vs. the true (actual) size of her indicated drink (for each beverage consumed). Again, the median is our focus here. The median was obtained by computing this estimation ratio for each woman, then taking the median of these ratios. 3. Results A synopsis of the general drinking characteristics of the sample provides context for our analysis of self-defined drink size. Approximately 60% of the 221 prepregnancy drinkers reported drinking at least weekly, and 28% indicated that they had consumed alcohol on a daily basis during the year prior to pregnancy. Half of the sample reported weekly or more frequent consumption of three or more drinks per occasion (3+), and almost one third reported five or more (5+) at that frequency. Further, a quarter of the women reported daily drinking at the 3+ level and 16% at the 5+ daily level. Thus, this is a sample of frequent, somewhat heavy drinkers. They tended to select a variety of alcoholic beverage types: Beer was the most frequently mentioned, and fortified wine the least often consumed alcoholic beverage by this group of urban minority women. There were very few women who drank only one beverage type. 3.1. Self-defined drink sizes The median self-selected drink sizes in this sample are as follows: 22-oz malt liquor; 12-oz beer; 12-oz wine cooler; 8-oz fortified wine; 5-oz wine; 2-oz spirits. The large serving sizes observed for malt liquor and fortified wine reflect the fact that 24 women (25% of the malt drinkers) selected 40 oz as their drink size, and for fortified wine seven women selected a drink size of greater than 16 oz (results not shown). Table 1 reports the proportion of women in this sample whose self-selected drink sizes were smaller than, the same as, or larger than the standard size for each beverage consumed. For those having larger Ð or smaller Ð than standard drinks, the magnitude of difference between self-selected and standard size is reflected in the median conversion ratios shown for each beverage. About half of the beer and wine drinkers, three-fourths of the spirits drinkers, and almost all of the fortified wine and malt liquor drinkers indicated portion sizes that were greater than the assigned standard size. Only among wine cooler drinkers did the majority (88%) select a serving size that was identical to the standard size of 12 oz. For all beverage types, the median self-selected drink size among those who drank more than the standard ranged from 1.5 to 4 times that of the standard size. Among all respondents who consumed the beverage, the median values for the self-selected drink size for wine, spirits, malt, and fortified wine, are 1.3±2.8 times that of the standard portion size, suggesting that for this sample of urban minority women, alcohol intake is being underestimated using methodology based on standard drink size. We further examined whether there was any difference in magnitude of self-selected vessel size among women who were frequent drinkers (weekly and daily) as compared to those
b
15 8 1 4 12 7
Less than standard (%) 38 42 24 2 0 88
Standard (%)
Limited to women who have completed both measures. Conversion ratio = self-select portion size/standard portion size.
12 4 1 8 3 12
Beer Wine Spirits Malt liquor Fortified wine Wine cooler
a
Standard size (fl. oz)
Beverage 48 50 75 94 88 5
More than standard (%)
Percentage whose self-defined drink size was:
Comparison of self-selected vessel size to standard sizea
Table 1 Conversion ratios and proportions selecting standard vs. nonstandard drink sizes
0.63 0.50 na 0.84 0.67 0.50
Less than standardb 1 1 1 1 1 1
Standard
1.90 1.50 4.00 2.81 3.00 2.80
More than standardb
1.0 1.3 2.0 2.8 2.7 1.0
Overallb
Median conversion ratios when self-defined drink size was:
72 L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
73
Table 2 Do frequent drinkers have larger drink sizes than less frequent drinkers? Conversion ratios for frequent and infrequent drinkersa,b Median ratio
Median ratio
Median ratio
Beverage
Infrequent
Weekly
Daily
Significance of K ± W testc
Beer (n = 122) Wine (n = 62) Spirits (n = 116) Malt liquor (n = 96) Fortified wine (n = 34) Wine cooler (n = 102)
1.00 1.00 2.00 1.50 1.30 1.00
1.00 1.50 4.00 2.00 2.00 1.00
1.90 1.50 6.00 2.81 4.00 1.00
.000 .039 .000 .001 .019 .046
a b c
Limited to women who have completed both measures. conversion ratio = self-select portion size/standard portion size. Because of multiple comparisons, a P value of .008 was deemed statistically significant.
drinking with less regularity (infrequent drinkers) (Table 2). Again, we applied the ratio of self-selected size to standard size to make the comparison, using a nonparametric test (Kruskal±Walls) to assess differences in self-selected portion size among women classified as infrequent, weekly, and daily drinkers. Women who drank daily were more likely to select portion sizes that were larger than their less frequent drinking counterparts. This observation was noted for most beverage types but only reached statistical significance for beer, spirits, and malt liquor. When translated into volume (oz) of the serving size, the differences between infrequent and frequent drinkers become more pronounced. For example, malt liquor drinkers who drank infrequently reported a median serving size of 12 oz, but among those who drank weekly, the median serving size for malt liquor was 16 oz; and for the daily drinkers, the median serving size became a 22-oz malt liquor. No statistically significant differences in self-selected portion sizes for any beverage type were observed between African Americans vs. Native Americans (Caucasians were not included in the comparison because of their small numbers; result not shown). 3.2. Judgment of drink size As a measure of respondents' ability to estimate the volume of their drink size, they were also asked to indicate their estimate, in ounces, of the volume of the portion size they had just selected. Results of the comparison between the estimated size and true size are presented in Table 3. Few women correctly estimated the volume of the portion size that they had selected. The beverages estimated with the most accuracy were beer, malt liquor, and wine coolers; but even here, only about two-fifths of the women were able to correctly judge the volume of their self-selected portion size. The tendency was for women to underestimate portion size. The proportion of respondents who underestimated their selfselected drink size ranged from 43% to 55%, with the median estimation ratio (of the estimated self-selected size divided by true size) ranging from 0.6 for spirits to 0.8 for fortified wine. To determine whether those indicating the standard size for a given beverage were more knowledgeable of the number of ounces in that serving, we selected those individuals who
74
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
Table 3 Do respondents know how big their drinks are? Comparison of estimated size to true size of self-selected vessela Percentage whose estimate was:
Median estimation ratios when estimate was:
Beverage
Less than true size
Same as true size
More than true sizeb
Less than true size
Same as true size
More than true sizeb
Beer Wine Spirits Malt liquor Fortified wine Wine cooler
44% 55% 43% 46% 50% 43%
44% 28% 21% 42% 30% 41%
12% 18% 36% 12% 20% 17%
0.7 0.7 0.6 0.7 0.8 0.7
1 1 1 1 1 1
1.5 1.3 2.0 1.3 1.8 1.3
a
Limited to women who answered both sets of items on self-selected vessel size. Median estimation ratio = respondent's estimated size/true size of vessel. Size shown is for women at the median size for each category. b
had chosen as their self-selected portion size the assigned standard size, and examined their estimates of portion size (results not shown). The proportion of respondents identifying the correct volume improved for all beverages excepting wine, where the tendency was still to underestimate the portion size. Interestingly, the heavier drinkers of standard drinks were no better at estimating portion size than the less frequent drinkers. 4. Discussion The results of this pilot study are provocative and raise the question of ``how well do current standard measures of alcohol estimate intake, particularly in minority populations?'' While results do not suggest that researchers abandon the `standard drinks' formulation, they do argue strongly for alternative methods that can illuminate the research question at hand. Our findings of a discrepancy between self-defined and standard portion sizes are consistent with other studies that have examined the issue of portion size (Carruthers & Binns, 1992; Lemmens, 1994; Wilson, 1981). Generally, individuals are consuming more of a specific beverage than is presumed. Our study focused only on women, but other surveys have found that the effect of the difference between actual and standard size of drinks on overall consumption is greater for women that men, because women consume other alcoholic beverages besides beer which is the dominant beverage for men (Dawson, 1993; Lemmens, 1994). The present study further confirms the result of Lemmens (1994) in that the greatest differences between self and standard sizes are for those beverages with higher volume percentage of ethanol that have smaller standard serving sizes, namely fortified wine and spirits. The strength of the agreement between the present and prior studies provides face validity for the `vessel' methodology used here; that is, the use of a set of vessels and photos premarked with letters that correspond to given volumes. This methodology has the advantage of being transportable for use in diverse field (vs. justin-home interview) settings.
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
75
In addition to the larger discrepancy for higher alcohol content beverages, our study also found that for the more frequent (daily) drinkers, selected beverage sizes were four to six times larger than the standard sizes. This adds to previous knowledge in the literature regarding underestimates of consumption for moderate drinkers (average daily volume of one standard drink), among whom ethanol intake was doubled when they were allowed to self-define their drink size using vessels (Nochajski et al., 1999). Thus, the impact of the difference between self-defined and standard drink size on overall consumption of alcohol is greatest among the heaviest groups of drinkers, when definitions of heavy drinking are based on frequency of consumption or strength of beverage consumed. This has implications for undercoverage of alcohol intake, in that discrepancies between actual and standard drink sizes for the heavier drinkers contributes to more undercoverage than if this occurred among less heavy drinkers. There is little conformity on the definition of standard serving sizes in general; for example, a standard glass of wine can vary from 3 to 6 oz (Turner, 1990). Some researchers have defined a serving size based on the amount of absolute alcohol (standard alcohol units) delivered, and have attempted to keep alcohol content relatively constant across beverage types (Miller, Heather, & Hall, 1991). However, the international alcohol community has not been able to reach consensus on the definition of an alcohol unit (Miller et al., 1991). In the US, drinking guidelines have been set, which recommend drink sizes that deliver approximately 14 g (International Center for Alcohol Policies (ICAP), 1996). Yet, in a review of 27 studies (Turner, 1990), the standard drink used on average provided 12 g (about 0.5 oz) of ethanol, as did the more recent Project MATCH (Miller & Del Boca, 1994). In surveys conducted at the Alcohol Research Group (Clark & Hilton, 1991; Greenfield, Graves, & Kaskutas 1999; Greenfield, Midanik, & Rogers, 2000), the standard drink sizes as defined for beer, wine and liquor provide 0.4 oz of ethanol (for spirits) to 0.53 oz (for beer). In order to deliver a similar amount of ethanol, a standard serving of malt liquor must be set (as we did here) at 8 oz, a size not sold. Since ethanol content also varies within a beverage type (Martin & Nirenberg, 1991) especially among the stronger malt liquors and fortified wines that are popular in some areas, determining the size for standard drinks will inevitably involve some error. In our sample, the most frequently selected size for malt liquor was 40 oz rather than our presumed 8 oz standard malt liquor; and the modal size of drink for fortified wine was 6 oz instead of the 3 oz assumed in the literature. However, for the other beverages, the most frequently selected size was actually the standard size: about a third of the beer drinkers have 12 oz beers; two-fifths of the wine drinkers indicated a 4 oz pour of wine and twice that proportion of wine coolers' drinkers have 12 oz coolers; and about a fourth of the spirits drinkers have 1-oz drinks. Thus, the extent of error really depends on the beverage mix of the population under study (and does not appear to vary by ethnicity, when controlling for beverage type). Given our experience with this study, we believe that when more accuracy is needed (as with studies assessing risk, or striving to more fully account for alcohol sales data), it may prove prudent to abandon the use of standard drink definitions when measuring alcohol consumption, and to instead allow respondents to simply indicate their portion sizes up front, at the beginning of the consumption series of questions, for each beverage they consume.
76
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
This would enable post hoc conversion by the analyst to a common metric (be that a standard alcohol unit or beverage-specific standard drink sizes), which could be clearly defined at the data analysis stage. This seems more reliable than expecting respondents to convert their self-defined serving sizes into standard sizes, especially across beverages. This recommendation is reinforced by current work underway at ARG using protocol analysis, in which respondents were asked to `think aloud' as they answered the graduated frequency series of questions (Clark & Hilton, 1991). One confused respondent asked for a calculator; another asked for help figuring out how many times 12 oz (of beer) goes into 40 oz (his standard beer). Note, however, that we are not suggesting that future studies simply ask respondents how large their drinks are. As shown in Table 3, our sample of minority pregnant urban women were not very accurate in judging the number of ounces that the letter on the vessel or photograph represented. Thus, some form of vessels probably must be used. We appreciate that the vessels methodology will increase the administration time for alcohol surveys. At least among the population under study here, the underreporting bias seems to be dependent upon the mix of beverages consumed by the participant, which varies. Thus, it seems important that alcohol researchers at least conduct preassessment inquiries to determine the mix of beverages consumed by the population of interest, and when possible, the distribution of drink sizes for each. For example, most studies of the general population do not collect data on malt liquor, which proved to be an important beverage type for these urban minority women. Since malt liquor comes in larger containers (than beer) and has higher alcohol content (than beer), overall consumption will be underestimated for populations where malt liquor is a beverage of choice (as appears to be the case for urban African American pregnant women (Graves & Kaskutas, 1999; Kaskutas, 1999) and urban African American men (Brown Taylor, 1999)). Measuring alcohol consumption is no easy task and may best be approached by using a multiplicity of strategies Ð among which would be collecting beverage-specific information, using respondent defined drink sizes, and in a way such that drinking patterns are distinguishable and can be separated from general measures of consumption. One final concern is respondent's ability to estimate the size of their drinks without visual aids like vessels or photographs. We found in our study that even the women who drank the standard size of a particular beverage were just as often wrong about the size of those drinks as were the women whose drinks were larger than the standard. Additional findings from this study, focusing on drinking during pregnancy among this sample, also are available (Kaskutas, in press) and are informed by the methodology presented here. In closing, although the challenges are many, increased accuracy of alcohol survey data would allow for more refined determination of risk levels, and for intervention strategies to be tailored to the needs of specific population subgroups. Acknowledgments This work was supported by a Center grant from the National Institute of Alcohol Abuse and Alcoholism to the Alcohol Research Group (#P50 AA05595), supplemented by the Office of Minority Health.
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
77
References Armor, D. J., Polich, J. M., & Stambul, H. B. (1978). Alcoholism and treatment. New York: Wiley. Branco, E. S., Witbrodt, J., & Kaskutas, L. A. (1998). Technical report for the DEER project. Berkeley, CA: Alcohol Research Group. Brown Taylor, D. (1999). Knowledge, attitudes and malt liquor beer drinking behavior among African American men in south central Los Angeles. Los Angeles, CA: California School of Professional Psychology. Carruthers, S. J., & Binns, C. W. (1992). The standard drink and alcohol consumption. Drug and Alcohol Reviews, 11, 363 ± 370. Clark, W. B., & Hilton, M. (Eds.). (1991). Alcohol in America: drinking practices and problems. Albany, NY: State University of New York Press. Dawson, D. A. (1993). Patterns of alcohol consumption: beverage effects on gender differences. Addiction, 88, 133 ± 138. Dawson, D. A. (1998). Measuring alcohol consumption: limitations and prospects for improvement. Addiction, 93, 965 ± 968; discussion 969 ± 977. Dawson, D. A., Grant, B. F., & Chou, P. S. (1995). Gender differences in alcohol intake. In: W. A. Hunt, & S. Zakhari (Eds.), Research monograph no. 29: stress, gender, and alcohol-seeking behavior ( pp. 3 ± 21). Bethesda, MD: National Institutes of Health, National Institute on Alcohol Abuse and Alcoholism (NIH Publication No. 95-3893 Edition). Ernhart, C., Morrow-Tlucak, M., Sokol, R., & Martier, S. (1988). Underreporting of alcohol use in pregnancy. Alcoholism: Clinical and Experimental Research, 12, 506 ± 511. Graves, K., & Kaskutas, L. A. (1999). Impact of ethnicity and drinking frequency on alcoholic beverage choice. Paper presented to the 127th annual American Public Health Association meeting. Greenfield, T. K., Graves, K. L., & Kaskutas, L. A. (1999). Long-term effects of alcohol warning labels: findings from a comparison of the United States and Ontario, Canada. Psychology and Marketing, 16, 261 ± 282. Greenfield, T. K., Midanik, L. T., & Rogers, J. D. (2000). A ten-year national trend study of alcohol consumption 1984 ± 1995: is the period of declining drinking over? American Journal of Public Health, 90, 47 ± 52. International Center for Alcohol Policies (ICAP). (1996). Safe alcohol consumption: a comparison of nutrition and your health: dietary guidelines for Americans and sensible drinking. In: ICAP reports I. Washington, DC: International Center for Alcohol Policies (ICAP). Kaskutas, L. A. (1999). Understanding drinking during pregnancy among urban Native and African Americans: health messages, risk beliefs, and how we measure consumption (invited presentation). Paper presented to the 22nd Annual Meeting of the Research Society on Alcoholism. Kaskutas, L. A. (2000). Understanding drinking during pregnancy among urban American Indians and African Americans: health messages, risk beliefs, and how we measure consumption. Alcohol Clin Exp Res, 24, 1241 ± 1250. Lemmens, P. H. (1994). The alcohol content of self-report and `standard' drinks. Addiction, 89, 593 ± 601. Martin, C. S., & Nirenberg, T. D. (1991). Alcohol content variation in the assessment of alcohol consumption. Addictive Behaviour, 16, 555 ± 560. Midanik, L. T. (1982). The validity of self-reported alcohol consumption and alcohol problems: a literature review. British Journal of Addiction, 77, 357 ± 382. Miller, W. R., & Del Boca, F. K. (1994). Measurement of drinking behavior using the form 90 family of instruments. Journal of Studies on Alcohol (Suppl. 12), 112 ± 113. Miller, W. R., Heather, N., & Hall, W. (1991). Calculating standard drink units: international comparisons. British Journal of Addiction, 86, 43 ± 47. Nochajski, T. H., Russell, M., Peirce, R. S., Chan, A. W. K., Freudenheim, J., Muti, P., Carosella, A. M., Trevisan, M. Assessment of alcohol consumption: quantity-frequency vs. semi-quantitative food frequency. Paper presented to the annual meeting of the Research Society on Alcoholism, June 26 ± July 1, 1999. Rogers, J. D., Greenfield, T. K. Does under-reporting of drinking affect survey estimates of the distribution of alcohol consumption? Paper presented to the Research Society on Alcoholism, June 26 ± July 1, 1999. Rosett, H. L., & Weiner, L. (1984). Alcohol and the fetus: a clinical perspective. New York: Oxford Univ. Press.
78
L.A. Kaskutas, K. Graves / Journal of Substance Abuse 12 (2000) 67±78
Russell, M., Marshall, J. R., Trevisan, M., Freudenheim, J., Chan, A. W. K., Markovic, N., Vana, J. E., & Priore, R. L. (1997). Test ± retest reliability of the cognitive lifetime drinking history. American Journal of Epidemiology, 146, 975 ± 982. Sokol, R. J., Martier, S., & Ernhart, C. (1985). Identification of alcohol abuse in the prenatal clinic. In: N. C. H. M., H. M. Chaos (Ed.), Early identification of alcohol abuse: proceedings of a workshop, October 31 ± November 1, 1983 ( pp. 209 ± 227). Rockville, MD: National Institute on Alcohol Abuse and Alcoholism (NIAAA research monograph no. 17). SPSS. (1996). SPSS for windows: base system user's guide, release 6.0. Chicago, IL: SPSS. Turner, C. (1990). How much alcohol is in a `standard drink'? An analysis of 125 studies. British Journal of Addiction, 85, 1171 ± 1175. Williams, G. D., Proudfit, A. H., Quinn, E. A., & Campbell, K. E. (1994). Variations in quantity ± frequency measures of alcohol consumption from a general population survey. Addiction, 89, 413 ± 421. Wilson, P. (1981). Improving the methodology of drinking surveys. Statistician, 30, 159 ± 167.