Effects of blood-alcohol concentration (BAC) feedback on BAC estimates over time.

Abstract

This study examines the effects of self-tested blood alcohol concentration (BA C) feedback, from personal hand-held breathalyzers, on the accuracy of BAC estimation. Using an e-mail prompted web-based questionnaire, 19 participants were asked to report both BAC estimates and subsequently measured BAC levels over the course of 27 days. Results from the 14 subjects who reported drinking during that time period suggest that BAC estimation improves over the first four drinking events, only when controlling for amount of alcohol consumed. BAC estimate accuracy was found to decrease as number of drinks and measured BAC increased. Participants were more likely to over-estimate their BAC's than to under-estimate them but this trend was much more pronounced for light drinkers than for heavy drinkers. There were no additional effects of heavy/light drinker status on estimate accuracy, beyond the effects of BAC at time of measured event.

INTRODUCTION

Department of Transportation projection statistics indicate that 249,000 people were injured, and 16,645 died in alcohol-related crashes in the U.S. in 2004 (National Highway and Transportation Safety Administration [NHTSA], 2005). Although rates for these incidences have been decreasing over the past several years, the decreases are predominantly in crash rates for those with blood alcohol concentration (BAC) levels between .01 and .07. For those with BAC levels at or above .08, the decline has been minimal. Trends also show that young drivers 21-34 years of age are disproportionately represented in alcohol related crashes and represent the highest rate of driving with BAC levels above .08% (NHTSA, 2004).

Although 1.5 million drivers were arrested for driving while impaired in the U.S. in 2002 (NHTSA, 2004), very few impaired drivers actually get stopped and even fewer are convicted. Only 4% of surveyed drivers aged 16-20 who reported driving impaired in 2002 and 2003 were actually stopped and booked for DUI offences (Beitel, Sharp, & Glauz, 2000).

The most effective policies directed at the reduction of injury and death from drunk driving in the past several years have been those that increase the legal drinking age, increase enforcement and punishment of drunk drivers, and adopt lower per se drinking and driving BAC limits (Greenfield, 1998; Hingson, Heeren, & Winter, 1994; Mann, 2002; Nan, Van Houten, Rolider & Jonah, 1993; Robertson, Rich, & Ross, 1973; Shults et al., 2001; Voas & Holder, 1997; Wagenaar, O'Malley, & LaFond, 2001). These per se laws state that a specific amount of alcohol in the blood indicates evidence of impaired driving ability. Recent federal policies have mandated the adoption of a drinking and driving BAC limit of .08% (grams of alcohol in 100 ml of blood) for all states. In addition, all 50 states and the District of Columbia have adopted very low (.01-.02%), or zero-tolerance, limits for underage drivers (Governor's Highway Safety Association, 2003). Although epidemiological research concerning car crashes and alcohol use shows that lower limits on per se laws are effective in reducing alcohol-related injuries and deaths (Hingson, Heeren, & Winter, 1994; Shults et al., 2001; Wagenaar, O'Malley, & La Fond, 2001), the association between measured BAC and impairment is not straightforward.

The amount of alcohol it takes to impair one's driving ability depends on several factors. Because BAC measures the percentage of alcohol in the bloodstream, it controls for factors that affect the absorption level of consumed alcohol such as type and timing of drinks, food consumption, body weight, sex, and individual physiological differences. Laboratory studies concerning drinking and driving impairment generally test reaction time, tracking ability, and vision loss at various BAC levels. Such studies indicate that significant impairment begins at the lowest measurable BAC and increases steadily with higher BAC's (Greenfield, 1998; Hutt, 1997; Levinthal, 2002; Mann, 2002; Moskowitz & Burns; 1990).

Although impairment generally increases with BAC level, this association is dependent on other individual factors. First, tolerance to the effects of alcohol, which increases with habitual alcohol consumption, can play a large part in determining how impaired one is at a given BAC level. Those who drink heavily, more regularly, and have longer drinking histories are likely to experience impairment at higher BAC levels than those who are light, or inexperienced drinkers (Levinthal, 2002; Moskowitz & Burns, 1990; Nicholson, Wang, & Mahoney, 1994; Thombs, Olds, & Snyder, 2003). This is illustrated in findings which show that individuals are moderately accurate in estimating their impairment levels but not accurate in estimating their BAC levels (Lewis, Merz, Hays, & Nichols, 1995; Nicholson, Wang, & Mahoney, 1994).

Individuals' expectancies regarding alcohol's effects may also determine impairment level. Those with expectancies of higher impairment may actually feel more impaired, and try harder to compensate for the expected impairment, than those who do not expect impairment (Fillmore, Roach, & Rice, 2002; Williams, 1991). These effects may add to the discrepancies between BAC and individuals' impairment.

Despite these discrepancies between actual BAC and individual impairment thresholds, the legal, economic and measurement difficulties involved in establishing individual thresholds for impairment make it likely that drunk-driving laws will continue to focus on standard per se BAC limits. However, individuals rarely have access to the means to determine their own BAC level.

The most common advice given to drinkers is to estimate their BAC level using charts that give BAC levels that correspond to various body-weight/sex/drinking pattern combinations. Several studies have investigated individuals' ability to estimate their BAC after drinking and found moderately accurate estimates when determining gross estimates of under/over legal limits but only among light drinkers. The accuracy is much lower when estimating actual BAC levels, and especially poor at higher levels of alcohol consumption (Carey & Hustad, 2002; Cox, Quillian, Gressard, Westerman, Gonder-Frederick, & Canterbury, 1995; Lansky, Nathan, Ersner-Hershfield, & Lipscomb, 1978; Russ, Harwood, & Geller, 1986; Thombs, Olds, & Snyder, 2003). Under typical circumstances, individuals estimate their BAC level and make driving decisions based on those (possibly impaired) perceptions.

The effects of intoxication itself on the ability to estimate BAC have been noted by several researchers. Carey and Hustad (2002) and Russ, Harwood and Geller, (1986) found that correlations between actual and estimated BAC were not significant at higher BAC levels. On the other hand Williams (1991) found accurate estimates only at a mid-range of BAC levels: between .05 and .10, with inaccurate estimates above and below that range. Similarly, Thombs et al. (2003) found accurate estimates at mid-ranges of BACs of .05 to .07. These findings suggest that those most likely to be impaired are the least likely to make accurate BAC estimations.

Lack of experience with both drinking and driving may also lead to faulty judgments regarding drunk driving capabilities, putting younger drivers at especially high risk for drunk driving incidents (Williams, 1991). However, most studies which test whether alcoholics, or heavy drinkers, differ from light or moderate drinkers in their abilities to estimate BAC find that heavy drinkers are less accurate than light or inexperienced drinkers, in estimating BACs. Since none of these studies control for BAC at the time of estimation, it is still unclear whether the inaccuracies are due to tolerance to the effects of alcohol or higher intoxication among heavy drinkers at the time of testing.

Because heavy or experienced drinkers are more likely to have increased tolerance to alcohol's effects, it is suggested that they may be more likely to under-estimate their actual BAC, whereas light drinkers, with low tolerance, may be more likely to over-estimate their BAC. However, findings regarding the direction of mis-estimation of BAC scores are contradictory. Russ, Harwood and Geller (1986) found that heavy drinkers over-estimated and light/medium drinkers under-estimated BAC's whereas Thombs, Olds and Snyder (2003) found the opposite; heavy drinkers under-estimated and light drinkers over-estimated their BACs. Cox et al. (1995) found that light drinkers over-estimated, Levinthal (2002) found that heavy drinkers under-estimated and Williams (1991) found that a sample of bar patrons all tended to over-estimate.

In summary, findings generally suggest that the ability to estimate BAC decreases as intoxication increases and that the effects of participant's heavy/light drinking status on BAC estimates are unclear. The present study will explore the effects of BAC feedback from personal breathalyzers, on BAC estimation accuracy over time, among heavy and light drinkers.

Hypotheses

1) Estimate accuracy of BAC will improve over time with feedback.

2) Estimate accuracy of BAC will decrease as number of drinks in creases.

3) Heavy drinkers will be more accurate in their estimates and more likely to under-estimate than over-estimate their BAC's while light drinkers will be less accurate in their estimates and more likely to over-estimate than under-estimate their BACs.

METHOD

Participant selection

The data collection methods were pre-tested on a convenience sample of four participants. The pre-test group included two nontraditional college students and two non-students. Because the pretest went smoothly the pre-test participants remained in the study and were included in all analyses. The pre-test group participation was staggered one week ahead of the recruited sample but all other conditions were the same as for the recruited sample. These pre-test data were shifted one week in the analyses to match "time in the study" with other participants.

The remaining participants were selected from responses to recruitment posters and college classroom announcements for participation in the study. Posters announced that participants were being sought for a social drinking study and that they could earn up to $25.00 and be entered into a $100.00 lottery. Interested participants were contacted by telephone and screened for participation. Participant screening criteria required that participants be over 21 years of age (with proof of age required), be social drinkers, not pregnant, have no history of serious drinking problems or trouble with the law, be daily e-mail users, and be available and willing to complete daily e-mail drinking reports for about 4 weeks. Participants were offered one dollar per daily report, regardless of whether or not they drank that day, for up to $25.00, as well as one chance in a $100.00 lottery for each daily report. In addition, participants were given the option to keep their breathalyzers (a $75.00 retail value) instead of receiving the cash compensation for the study. It was thought that this reward system would encourage daily reporting and discourage loss of the breathalyzers before the study was completed. This recruitment yielded 15 participants. The total sample size with recruited participants and the pre-test group was 19.

Data Collection

Participants attended orientation sessions in which baseline questionnaires were completed. These included items on basic demographics, drinking history, risk-taking history, body weight, and attitudes regarding drinking. Participants were then given a short presentation regarding drinking and blood alcohol concentrations. Handouts with charts for estimating BAC from body weight, sex, and drinks per hour, as well as the state legal limits for drinking and driving were distributed to the participants. The study was then explained to the participants and they were shown how to use the small digital handheld breathalyzers (Viper Technologies, Model CA-2000/DT) and instructed to pick up their breathalyzers in one week, after the control week was over. The participants were then shown how to use the daily web-reporting questionnaire. E-mail addresses were collected, code names were assigned, and proof of age was collected. The e-mail addresses were compiled in a group mailing list and messages were sent to the list automatically at 5:00 a.m. every morning reminding the participants to complete the report for the previous day's drinking. The e-mail messages included a link to the daily web report questionnaire.

The first seven days of the study constituted the control week in which participants did not yet have breathalyzers and so reported only BAC estimates based on the drinking charts, without an actual BAC measure. This period was intended to familiarize the participants with the process and to provide a baseline drinking comparison week. The participants then picked up their breathalyzers and used them for the actual BAC measures in the following three weeks. The last week of the study coincided with finals week at the University.

The daily web report questionnaire first asked for codename and date, and whether or not the participant had drunk the previous day. (Previous day was defined as any time before they went to bed the night before). If participants had not drunk the previous day they were asked to skip directly to the "submit" button at the bottom of the questionnaire. If participants had drunk the previous day they were asked for up to six reports for that day. Each report asked for the time, amount drunk since last report, a BAC estimate and the subsequently measured BAC value for that time. Where possible, items were formatted with drop-down selection menus and a "clear" button was included at the bottom of the questionnaire to allow for last minute changes. A "thank-you" screen appeared when a report was successfully submitted and data were transferred directly to a database file with each daily submission listed as a case.

Analysis

Sample descriptives. Sample descriptive analyses first draw on data from the baseline survey. These analyses present summary statistics for age, sex, past drinking habits including quantity, frequency, and mean age at first drink. In addition, a dichotomous heavy/light drinking variable was constructed and all of the characteristics listed for the descriptive analyses were computed for each of these two groups. Determination of coding for "light" and "heavy" drinking categories used information from the baseline questionnaire. A score was created from the typical number of drinks drunk on a drinking day, divided by body weight. Scores were split at the median and participants with median and above scores were designated as heavy drinkers and below median scoring participants were designated as light drinkers. Median drinking group status was extrapolated from daily diary information for one participant who did not provide the necessary baseline information. With the exception of one relatively high-weight female (2.5 drinks = light drinker), one high-weight male (4.5 drinks = light drinker), and one low-weight male (4.0 drinks = heavy drinker) this measure corresponded with traditional definitions of heavy drinking (average of > 2 drinks per occasion for females, > 4 drinks per occasion for males). Because the added information regarding body weight was considered, this measure offers more accuracy than the traditional quantity measure.

Next, summary statistics were calculated from the daily reports. This analysis includes the mean quantity and frequency of drinking for the participants for each of the four weeks in the study, as well as mean maximum BAC levels for the three breathalyzer weeks of the study. Again, these statistics are presented for light drinkers and heavy drinkers as well as all drinkers combined.

BAC estimate accuracy. There were a total of 420 reports from 18 participants, 141 of which represented drinking days. Of these, 40 events occurred in the control week before participants were given their breathalyzers and do not contain information on actual measured BAC. The final analyses were based on fifty daily drinking reports with measured BAC's from the 14 participants who drank and reported during that time period.

In order to explore for associations between number of drinks and BAC accuracy, Pearson's correlation coefficients were calculated between "total number of drinks" at the last or only report for the day, and the BAC accuracy from the last or only report. BAC accuracy was calculated as the absolute value of the difference between the estimated BAC and the subsequent measured BAC for the last, or only, report of the day. Accuracy then represents the gap between the BAC estimate and the actual BAC, regardless of direction, with higher values representing more inaccuracy. The use of the last BAC measure of the day allowed for more variability in the "number of drinks" score than would reports from earlier in the day. Similar correlations were calculated separately for the heavy and light drinking groups.

In the analyses regarding improvement of BAC estimates over time we were interested in the accuracy of the first guess of each drinking day over time. Because individual drinking days occurred at different intervals throughout the study for each participant (ranging from zero to fourteen drinking day reports during the study), the "day in study" did not reflect the amount of experience with the breathalyzers for each individual participant. In order to get consistent indicators of breathalyzer experience over time, data were collapsed from "day in study" to breathalyzer events, with participant's first reported drinking day coded as event one, second reported drinking day as event two, etc. These analyses also used the first, or only, report of the day because estimates from this report were likely to represent the first breathalyzer use of the day and were less likely to be compromised by the effects of higher alcohol consumption. These accuracy measures were calculated as the absolute value of the difference between the estimated and subsequently measured BAC for the first or only, report of the day. Summary statistics of accuracy by event day are presented for heavy drinkers, light drinkers, and all drinkers combined. In order to determine any patterns in over- and under-estimation of BAC's, a frequency of over-, under-, and correct estimates by event and drinker type are presented as well.

Improvement in BAC estimate accuracy was also tested with multiple regression analyses. The dependent variable was "inaccuracy" measured as the absolute difference between estimated BAC and actual BAC for the first report of the day, and included reports for the participants' first four drinking events. The independent variable, "time" corresponded to the event number, coded 1 through 4, with 1 representing the first event. This method was used to detect a linear increase in accuracy over the first four events. Next, heavy/ light drinker status was added to the equation to see if drinker status affected estimate accuracy. Finally, BAC at time of measured events was added as a control to determine the effects of feedback over time, independent of the effects of intoxication. This also provided a test for the effects of drinker status on estimate accuracy, independent of intoxication level at time of measured event.

RESULTS

Baseline Sample Descriptives

The baseline sample data are arranged with the individual as the unit of analysis and are summarized for all participants, as well as separately for the "heavy" drinkers and "light" drinkers (see Table 1). Results show that the light drinkers are somewhat older than the heavy drinkers. Gender distribution is consistent across groups. Mean age of first drink is about one year younger among the heavy drinkers. Mean number of drinks per occasion is 2.61 for light drinkers and 5.56 for heavy drinkers. Mean number of drinking days per week is 2.72 for light drinkers and 3.00 for heavy drinkers.

Daily Reporting Descriptives

The daily report analyses are conducted with "report" as the unit of analysis. Percentages in these analyses were based on valid percent and do not include the approximately 60-67% of the reports referring to non-drinking days (see Table 2). Results concerning drinking frequency for all participants show a mean of 4.47 drinking days over the 20 days with the breathalyzers; 2.56 for light drinkers, 6.20 for heavy drinkers. Drinking quantity results show a mean of 4.67 drinks per drinking day for all participants; 2.34 for light drinkers and 5.82 for heavy drinkers.

For all participants, mean maximum BAC levels for the last measure of the day decreased after the first week with the breathalyzer and increased during final exam week (Table 3). Looking at these rates separately for light and heavy drinkers, analyses reveal that heavy drinkers show a slight decrease from the second to the third week and an increase during the finals week. Light drinkers showed a substantial decrease in BAC from week two to week three, and an increase during finals week.

BAC Estimate Accuracy

Pearson's correlation analyses suggest that BAC accuracy estimate is associated with the number of drinks (Table 4). For all participants, the number of total drinks per day is positively correlated with the difference between actual and estimated BAC (r = .612, p < .01). When looking at heavy and light drinkers separately, there is a similar correlation for heavy drinkers (r = .580, p < .01), but no significant correlation for light drinkers.

Table 5 presents the BAC estimate accuracy for consecutive drinking events over the course of the study for both heavy and light drinkers. BAC estimate accuracy appears to improve over the first few events, for both groups. The pattern remains erratic thereafter, possibly because means were computed from fewer cases that were more likely to include heavy drinkers. Because heavy drinkers had more drinking events, the decline in accuracy after the first few events may simply be an artifact of the association between BAC estimate accuracy and higher intoxication levels of the heavy drinkers.

Both heavy and light drinkers are more likely to over-estimate their BAC's than to under-estimate them or estimate them correctly, but this pattern is much more pronounced among the light drinkers (Table 5). Among light drinkers, it appears that early estimates are likely to be over-estimates with increases in correct estimates over the first three events. There does not appear to be any clear trend among the heavy drinkers, although the overestimations (32%) remain somewhat more frequent than the under-estimations (23%) throughout. Correct estimates do not appear to increase over time among the heavy drinkers.

Regression analyses suggest that estimate accuracy improves over time only when controlling for BAC at time of measured event (Table 6). This analysis uses the first four events to examine the effects of time. All possible dichotomous time combination variables were tested to determine the point at which estimates no longer improved. Significant improvement occurred over the course of events one through four. Lack of significance after event four may be due to the small number of participants with more than four reported drinking days.

This analysis also suggests that the effects of heavy/light drinking status on estimate accuracy found in the first regression model were eliminated after controlling for BAC at time of measured event (Table 6). In other words, drinking history and possibly tolerance to the effects of alcohol, does not exert independent effects on BAC estimate accuracy beyond those of BAC at time of event.

DISCUSSION

One drawback to the diary method used here is the inability to determine whether or not participants drank on a day for which they made no report. Discrepancies between the baseline reported drinking patterns and the daily diary drinking patterns suggest that there may have been some under-reporting of drinking days in the diary data (Tables 1 and 2). The frequency of drinking days per week reported in the baseline is higher than those actually reported during the study, for all groups. The quantity of drinks per occasion were similar between the baseline and the daily reports, although the light drinkers drank fewer drinks in the study than they indicated on the baseline survey while the heavy drinkers drank more drinks in the study than they indicated in the baseline. The heavy/light drinking criteria established from the baseline questionnaire seems to reflect clear group drinking pattern differences in the daily reports; the heavy drinkers averaged 5.82 drinks per occasion whereas light drinkers averaged 2.34.

Weekly BAC levels suggest that drinking events decreased over time. The mean maximum daily BAC levels dropped after the first week with breathalyzer use, although more so for the light drinkers than the heavy drinkers. This pattern may be due to the novelty of having access to the breathalyzers during the first week and a "leveling off" of drinking after the participants became familiar with their BAC ranges and limits. Anecdotal evidence suggests that there was a lot of "experimenting" the first few days with breathalyzers, which may have resulted in higher drinking frequency and BAC levels early in the study.

The higher BAC's at the later events may reflect the preponderance of heavy drinkers in these events. Because of the limited number of participants who drank more than four times during the course of the study and the short duration Of the study itself, long-term effects of breathalyzer use on drinking habits are still unclear.

The most consistent finding in this study is that BAC estimate accuracy decreases as alcohol consumption increases. It is likely that cognitive impairment associated with drinking is responsible for this association. In fact, this association accounts for all of the variation in the findings regarding heavy/light drinking status and BAC estimate accuracy. It appears that drinking status, which is likely to be associated with tolerance to the effects of alcohol, has no independent effect on estimating BAC, beyond the effects of the amount of alcohol consumed at the time of the estimation.

Findings also suggest that although BAC estimate accuracy does improve over time, at least for the first four drinking events, this improvement is dependant on the amount of alcohol consumed. Improvement over time does not occur when number of drinks is high.

It was expected that heavy drinkers, who were likely to have built up a tolerance to alcohol effects, would tend to under-estimate their BAC, whereas light drinkers, who are likely to experience stronger effects with smaller amounts of alcohol would be likely to over-estimate their BAC. Findings show that both groups are more likely to over-estimate their BACs than to under-estimate them. However, this tendency is much stronger among the light drinkers than the heavy drinkers. On the other hand, the failure of heavy drinkers to improve BAC estimates over time appears to be due to the association between estimate accuracy and actual BAC. In other words, the more the participants drink, the less accurate are their BAC estimates, and the heavy drinkers tended to report at higher BAC levels.

Limitations of this study include the possibility of reporting errors: under-reporting of drinking events, inaccurate reporting due to mis-use of breathalyzers, and recall errors. Anecdotal feedback indicated that the reporting methods and equipment were easy to use but that the amount of information requested on the daily reports was difficult to supply. Many participants reported only one time each drinking day: an option that was encouraged if they had difficulties with multiple time reports for a single day. Future studies would likely benefit from an abbreviated daily report questionnaire. As discussed earlier, response rate biases may exist because it is not known how often non-response days included drinking events. This limitation could affect the results concerning estimate accuracy over time, but would not likely effect results that are independent of time, such as the association between estimate accuracy and BAC.

Summary and Applications

The most significant finding of this study is that BAC estimation accuracy is negatively associated with amount of alcohol consumed. Findings from this study suggest that feedback from personal breathalyzer use resulted in increased BAC estimate accuracy, over the first four events, at lower BAC levels only. Findings regarding the direction of mis-estimation show that both light drinkers and heavy drinkers tended to over-estimate their BAC's, but this pattern was much more pronounced for the light drinkers. Furthermore, there were no independent effects of heavy/light drinking status on BAC estimate accuracy.

The finding with the strongest policy implications is the tendency, especially pronounced among light drinkers, to over-estimate BAC levels. These results may be due (but only in part) to the relative inexperience of the young drinkers in this study who may be less likely to have developed a tolerance to the effects of alcohol and are more likely to experience higher levels of impairment with fewer drinks. The implications of this pattern are important with respect to drunk driving laws, especially if they are indicative of a wider pattern in which many individuals who feel too impaired to drive are actually well within legal BAC limits for driving. Because the main deterrent from drunk driving is the probability of getting arrested, rather than the probability of causing harm to oneself or others (Robertson, 1973; Voas & Holder, 1997), these drivers may be likely to use the measured BAC level rather than personal perceptions of impairment, to make driving decisions.

Applications for these findings would be most useful in education and treatment programs for heavy drinkers or DUI offenders. This is the group that would most regularly attain BAC levels above the legal driving limit and would more often reach levels of high inaccuracy of BAC estimates. Feedback from breathalyzers for this group would illustrate the inaccuracy of their BAC estimates at typical drinking levels. Furthermore, this group would be the most like to reduce their chances of impaired driving if they refrained from driving with BAC levels over the legal limit.

The educational and treatment benefits of breathalyzer feedback among light drinkers are less clear. Light or inexperienced drinkers are more likely than heavy drinkers to be impaired at BAC levels below the legal limits. It is conceivable that feedback for these individual might encourage risky behavior if they interpreted a "legal" BAC level as "non-risky." The most helpful educational application for this group would be the dissemination of findings showing that BAC estimates become more inaccurate at higher levels of drinking. Additionally, as breathalyzer technology becomes more accessible it is important to educate drinkers about the lack of correspondence between legal BAC limits and actual impairment levels, especially among light or inexperienced drinkers.

Author Note

Susan Bullers, Women's Resource Center and Department of Sociology and Criminal Justice, University of North Carolina at Wilmington; Melissa Ennis, Department of Psychology, University of North Carolina at Wilmington.

This study was funded by a grant from the UNCW Crossroads Program. We wish to extend special thanks to Jessica Burtt for her help on the project.

Correspondence concerning this article should be addressed to Susan Bullers, Women's Resource Center and Department of Sociology and Criminal Justice, University of North Carolina, Wilmington, NC 28403. E-mail: [e-mail]. Phone: [phone].

References

Beitel, G. A., Sharp. M. C., & Glauz, W. D. (2000). Probability of arrest while driving under the influence of alcohol. Injury Prevention, 6, 158-161.

Carey, K. B., & Hustad, J. T. P. (2002). Are retrospectively reconstructed blood alcohol concentrations accurate? Preliminary results from a field study. Journal of Studies on Alcohol, 3, 762-766.

Cox, D. J., Quillian, w. c., Gressard, C. F., Westerman, P. S., Gonder-Frederick, L.A., & Canterbury, R.J. (1995). The effects of blood alcohol levels on driving variables in a high-risk population: Objective and subjective measures. Journal of Alcohol and Drug Education, 40(3), 84-98.

Fillmore, M. T., Roach, E. L., & Rice, J. T. (2002) Does caffeine counteract alcohol-induced impairment? The ironic effects of expectancy. Journal of Studies on Alcohol, 63(1), 745-755.

Governor's Highway Safety Association. (2003) State Impaired Driving Laws. Retrieved on September 15, 2005 from http://www.ghsa.org/ html/stateinfo/laws/impaired_laws.html

Greenfeld, L. A. (1998). Alcohol and crime: An analysis of national data on the prevalence of alcohol involvement in crime. U.S. Department of Justice.

Hingson, R., Heeren, T., & Winter, M. (1994). Lower legal alcohol limits for young drivers. Public Health Reports, 109(6), 738-745.

Hutt, K. R. (1997). Selling Limits, Saving Lives." The Case for .08 BAC Laws. National Highway Traffic Safety Council: U.S. Department of Transportation.

Lansky, D., Nathan, P. E., Ersner-Hershfield, S. M., & Lipscomb, T.R. (1978). Blood alcohol level discrimination: Pre-training monitoring accuracy of alcoholics and nonalcoholics. Addictive Behaviors, 3, 209-214.

Levinthal, C. F. (2002). Drugs, Behavior, and Modern Society, 3rd ed. (pp. 185-233). Allyn and Bacon: Boston, MA.

Lewis, M. W., Merz, J. F., Hays, R. D., & Nicholas, R. (1995). Perceptions of intoxication and impairment at arrest among adults convicted of driving under the influence of alcohol. The Journal of Drug Issues, 25(1), 141-160.

Mann, R. E. (2002). Choosing a rational threshold for the definition of drunk driving: What research recommends. Addiction, 97, 1237-1238.

Moskowitz, H. and Burns, M. (1990). Effects of alcohol on driving performance. Alcohol Health and Research World, 14(1), 12-15.

National Highway Traffic Safety Administration. (2004). Traffic Safety Facts 2003:Alcohol. DOT 809 761. Washington DC: National Highway Traffic Safety Administration.

National Highway Traffic Safety Administration. (2005). National Center for Statistics, 2004 Projections: Alcohol. DOT 809 862. Washington DC: National Highway Traffic Safety Administration.

Nau, P. A., Van Houten, R., Rolider, A., & Jonah, B. A. (1993). The failure of feedback on alcohol impairment to reduce impaired driving. Journal of Applied Behavioral Analysis, 26, 361-367.

Nicholson, M. E., Wang, M., & Mahoney, B. S. (1994). Perceived Intoxication: Implications for alcohol education. Journal of Alcohol and Drug Education, 40(1), 115-125.

Robertson, L. S., Rich, R. F., & Ross, H. L. (1973). Jail Sentences for Driving While Intoxicated in Chicago: A Judicial Policy the Failed. Law and Society. Fall.

Russ, N. W., Harwood, M. K, & Geller, E. S. (1986). Estimating alcohol impairment in the field: Implications for drunken driving. Journal of Studies on Alcohol, 47(3), 237-240.

Shults, R. A., Elder, R. W., Sleet, D. A., Nichols, J. L., Alao, M. O., Carande-Kulis, V. G., Zaza, S., Sosin, D. M., & Thompson, R. S., & the Task Force on Community Preventive Services (2001). Reviews of evidence regarding interventions to reduce alcohol-impaired driving. American Journal of Preventative Medicine, 21 (4S), 66-88.

Thombs, D. L., Olds, R. S., & Snyder, B. M. (2003). Field Assessment of BAC Data to Study Late-Night College Drinking. Journal of Studies on Alcohol, 64, 322-330.

Voas, R. B., & Holder, H. D. (1997). The effect of drinking and driving interventions on alcohol-involved traffic crashes within a comprehensive community trial. Addiction, 92(6), 221-237.

Wagenaar, A. C., O'Malley, R M., & LaFond, C. (2001). Lowered legal blood alcohol limits for young drivers: Effects on drinking, driving, and driving-after-drinking behaviors in 30 states. American Journal of Public Health, 91(5), 801-804.

Williams, J. G. (1991). Experience with alcohol and ability to discriminate legal intoxication status: A field study. Addictive Behaviors, 16, 355-362.

Susan Bullers & Melissa Ennis

University of North Carolina, Wilmington

Table 1
Baseline Sample Descriptives

                          Light      Heavy      All
Variable                  Drinkers   Drinkers   Drinkers

Mean age                  30.89      22.00      26.44
Percent female            44.40      44.40      44.40
Mean age at first drink   16.78      15.89      16.33
Mean usual # of
  drinks/occasion         2.61       5.56       4.00
Mean usual # of
  drinking days/week      2.72       3.00       2.85
N (participants)          9          9          18

Table 2
Daily Report Descriptives

                          Light         Heavy         All
Variable                  Drinkers      Drinkers      Drinkers

Mean # of drinking
  days/wk (of 20 days)     .90 (2.56)   2.18 (6.20)   1.56 (4.47)
Mean # of drinks/
  occasion                2.34          5.82          4.67
N (participants)          45            96            141

Table 3
Mean maximum reported BAC per drinking day, by week in study.

               All drinkers        Heavy drinkers
               Mean (sd)      n    Mean (sd)        n

Control week   --                  --
Week two       .070 (.054)    35   .085 (.052)      23
Week three     .067 (.047)    34   .084 (.043)      25
Finals week    .095 (.073)    25   .125 (.068)      17
Total          .076 (.059)    94   .095 (.056)      65

               Light drinkers
               Mean (sd)        n

Control week   --
Week two       .042 (.047)      12
Week three     .020 (.014)       9
Finals week    .032 (.027)       8
Total          .032 (.034)      29

Table 4
Correlations Between Number of Drinks and BAC Guess Accuracy

Variables                                          r       n

Number of drinks by last estimate difference *   .612 **   68
  (all resp.)
Number of drinks by last estimate difference     .580 **   55
  (heavy drinkers)
Number of drinks by last estimate difference     .298      13
 (light drinkers)

* absolute value of (actual BAC--guess BAC)

** p<.01

Table 5. BAC Estimate Accuracy and Direction Over Time

Event
        Light Drinkers

        Mean      over   correct   under   n
        Diff. *   est.   est.      est.

1       .022      6      --        --      6
2       .012      3      2         --      5
3       .004      2      3         --      5
4       .010      2      2         1       5
5       .020      1      --        1       2
6
7
8
9
10
11
12
13
14
Mean    .014
Total             14     7         2       23
(%)               (61)   (30)      (9)     (100)

        Heavy Drinkers

        Mean      over   correct   under   n
        Diff. *   est.   est.      est.

1       .021      4      1         4       9
2       .013      5      --        2       7
3       .013      3      2         1       6
4       .017      2      --        4       6
5       .016      4      --        1       5
6       .022      4      --        1       5
7       .024      3      --        2       5
8       .020      3      --        1       4
9       .003      --     2         1       3
10      .033      2      --        1       3
11      .020      --     --        2       2
12      .015      1      --        1       2
13      .055      1      --        1       2
14      .025      1      --        1       2
Mean    .021
Total             32     5         23      61
(%)               (52)   (8)       (38)    (100)

* absolute value of (actual BAC--guess BAC) for first or only
measure of the day

Table 6
Regression of Estimate Inaccuracy on Time, Actual BAC,
and Drinker Status

                          Model I                     Model II

Variables          B        (se)     Beta      B       (se)      Beta
Constant         .027     (.006)**            .019    (.007) *
Time             -.005    (.002)     -.289   -.005    (.002)     -.280
Heavy Drinker                                 .013    (.005) *    .352
BAC
R                     .289                        .445
[R.sup.2]             .083                        .207
Adj. [R.sup.2]        .061                        .168

                          Model III

Variables          B        (se)      Beta
Constant          .014    (.006) *
Time             -.005    (.002) *    -.282
Heavy Drinker     .004                 .105
BAC               .231    (.066) **    .500
R                     .630
[R.sup.2]             .397
Adj. [R.sup.2]        .350

* p<.05, ** p<.01

"Estimate Inaccuracy" = absolute value of (estimated BAC--measured BAC)
"Heavy Drinkers" coded 1, light drinkers coded 0.
"Time" refers to first through fourth breathalyzer event days,
coded 1 through 4.