IV. RESULTS
A. Blood Alcohol Concentration
Table 1 presents the mean BACs measured immediately before and after each test battery. As can be seen in the table, the mean for each battery was within 0.002% BAC of the levels specified in the design of the experiment. The mean of 0.098% measured at battery 2 is for moderate and heavy drinkers only. Batteries 3 through 7 include all Ss.
Table 1 | |||
Mean BACs (%) | |||
Test Battery | Pre-Battery | Post-Battery | Battery Average |
1 | 0.000 | 0.000 | 0.000 |
2 | 0.102 | 0.094 | 0.098 |
3 | 0.082 | 0.073 | 0.078 |
4 | 0.063 | 0.055 | 0.059 |
5 | 0.044 | 0.035 | 0.040 |
6 | 0.024 | 0.015 | 0.020 |
7 | 0.001 | 0.000 | 0.001 |
B. Sensitivity of Study Measures
It was necessary to determine whether the response measures for both SIM and DAT had proved to be sufficiently sensitive for detection of the differential effects of alcohol as a function of age, gender and drinking practice. The placebo scores were examined for this purpose and, in fact, demonstrated that the measures were capable of detecting differences between S classifications on these three variables. Although the results are of considerable interest, their presentation is deferred to a subsequent publication to avoid distraction from the main thrust of the experiment.
C. Ethanol Clearance Rate
The ethanol clearance rate is the BAC decline over time after alcohol absorption from the intestinal tract is complete. The rate varies as a function of age, gender, and drinking practices. The rate for male Ss in this experiment was 0.0149% and the rate for female Ss was 0.0184%. The rates by age group were 0.0156% for Ss ages 19-20, 0.0156% for Ss ages 21-24, 0.0168% for Ss ages 25-50, and 0.0183% for Ss ages 51-69. The ethanol clearance rates for Ss as a function of drinking practices varied from 0.0157% for light drinkers to 0.0165% for moderate drinkers and 0.0176% for heavy drinkers.
The ethanol clearance rate shows an increase with greater frequency of alcohol consumption. This variability is due to the stimulation by alcohol of the production of a liver enzyme. Although the clearance rate for the moderate drinkers in this experiment was found to be very close to that which was anticipated, the rate for light drinkers was higher and the rate for heavy drinkers was lower than expected. The finding strongly suggests an under-representation of the lighter drinkers of the light-drinking category and of the heavier drinkers of the heavy-drinking category. The latter result may be the consequence of the SCRI practice of excluding alcoholics from alcohol experiments.
D. Sequence and Order Effects
Ss in this experiment received a placebo treatment and an alcohol treatment on test days separated by one week. In this repeated measures design, it is necessary to examine the data to determine whether the sequence of treatments affected the results. Was the effect of alcohol different for Ss who received placebo first and alcohol second in comparison to Ss who received treatments in the reverse order? A difference would indicate a sequence effect.
An additional question asks whether the average performance on test day one differs from the average performance on test day two, perhaps due to the difference in practice on the tests. A difference would indicate an order effect. It should be noted, however, that given the training sessions, such differences would be small, and the number of Ss per cell would limit the power of the test to detect such effects.
1. Sequence Effect
Half the Ss received treatments in the sequence placebo-alcohol, and half the Ss received treatments in the sequence alcohol-placebo. The mean scores for the two sequences were examined with statistical tests for each response measure at each BAC and across BACs. Table AP-I-3 presents the statistical analysis. Twelve response measures for SIM and DAT were examined. None of the tests for overall sequence effects were statistically significant. Two of 64 tests at separate BACs were significant at the .05 level. The finding of only two statistically significant tests of 76 total tests (12 across BACs plus 64 at separate BACs) suggests that the two were random occurrences, and it is concluded that there is no evidence of a sequence effect that might influence data analysis.
2. Order Effects
All response measures were examined for an order effect. That is, did the average performance score on test day 1 differ from the average performance score on test day 2? Figures AP-I-1a and AP-I-1b present the mean change in score from baseline for each measure for the two test days. Each figure has two lines, one representing the difference scores for each test time for day 1 and one representing the difference scores for day 2. Recall that a difference scores is the post-treatment test score minus the pre-treatment test score. For each line half the Ss are under placement treatment, and half are under alcohol treatment.
Table AP-I-4 presents the results of a statistical analysis, which controlled for the variables of alcohol, battery and the alcohol X battery interaction, and examined whether the mean difference scores on days 1 and 2 differed. Five of the nine response variables were not statistically significant, but four were. Three of the four significant measures were worse on day one and one was worse on day two. These somewhat contradictory results clearly cannot rule out the possibility of an order effect. Since the treatments in the study were counterbalanced, however, the existence of an order effect would have no influence on the analysis of the alcohol, age, gender and drinking practice variables or their interactions. Such an effect would only limit statements about the impairment by alcohol of each individual.
E. Alcohol Effects Analysis
One hundred sixty-eight Ss were tested six (or seven) times at two sessions, one with a placebo treatment and one with an alcohol treatment (Table AP-I-5). The DAT provided three response measures: reaction time to peripheral signals, number of incorrect responses to peripheral signals, and error on the tracking task in central vision. The SIM provided six response measures: lane deviation variability, speed variability, number of collisions, number of times over the speed limit, reaction time to peripheral signals, and number of incorrect responses to peripheral signals. Additionally, two performance indices were created by combining all measures for the DAT into one composite score and combining all measures for the SIM into another composite score. These two composites were also combined to create a single index of overall performance.
Figure 2 presents the average raw scores for the three DAT measures, and Figure 3 presents the average raw scores for the six SIM measures. In both figures the scores are shown by battery (seven for DAT, six for SIM) and by treatment condition. Battery 1 is the pre-test or the pre-treatment test. Battery 2 is the first post-treatment test at 0.10% BAC when only moderate and heavy drinkers were tested. Light drinkers were first tested post-treatment at battery 3 when all Ss were tested at 0.08% BAC. The mean BAC for battery 4 was 0.06%. For battery 5 it was 0.04% and for battery 6 it was 0.02%. The BAC was 0.00% for battery 7 when the only test was DAT.
The raw scores, which do not take pre-treatment performance into account, reveal impairment of all DAT and SIM measures at all positive BACs in comparison to performance in the placebo condition. If the curves are mentally adjusted so that the placebo and alcohol curves begin at the same point, it will be seen that the differences in SIM scores are even greater than they appear in the figures. In all comparisons, the adjustment produces greater separation of the alcohol and placebo curves. In order to take into account the variability of the pre-test scores on the two test days, the comprehensive statistical analysis used an impairment score, as described below.
Note in the succeeding tables, that the measure "number of incorrect responses to peripheral signals" for both DAT and SIM appears in two forms. The measure is tabled both as number of errors and as percent errors. The percent measure was generated, because there was a statistical question as to whether the measure, number of errors, would be normally distributed. As it developed, the results of statistical tests were the same for both measures but rather than re-create the tables both are included. Note, however, that in all subsequent discussions, the response error measure is counted only once for DAT and once for SIM.
1. Impairment Scores
An impairment score was created for the statistical analysis of the alcohol effect. An impairment score is defined as the performance score on the alcohol treatment day at a given BAC minus the comparable placebo score minus the differences in the pre-treatment test scores on the two test days. Thus, the impairment score takes into account the time-of-day factor at which testing occurred under the two treatments, and it also takes into account variations in a S's overall performance from day to day.
Ss were tested at BACs from 0.08% to 0.00%. Moderate and heavy drinkers only (n=112) were also tested at 0.10% BAC (Table AP-I-1). Impairment scores were created for each S at each BAC for the nine original single response measures and three composite scores. The advantage of the latter is that they take a larger slice of the performance information into account and, therefore, provide a more stable measure.
Table 2 shows the percent of Ss whose impairment score was poorer under alcohol than under placebo. Table 3 shows the results of the statistical test of the null hypothesis, which states that fifty percent of the Ss would have performed worse under alcohol if alcohol had no effect. To be redundant, the null hypothesis states that there is no difference between active and placebo treatments, and that by chance half the scores will be poorer on the alcohol treatment day, and half the scores will be poorer on the placebo treatment day.
Table 2 | ||||||
Measurement |
BAC (%) | |||||
.00 |
.02 | .04 | .06 | .08 | .10 | |
DAT Reaction Time | 50 | 50 | 63 | 70 | 80 | 90 |
DAT Number Incorrect (%) | 39 | 37 | 45 | 54 | 58 | 71 |
DAT Number Incorrect (#) | 41 | 38 | 44 | 54 | 58 | 70 |
DAT Tracking Error | 53 | 60 | 68 | 77 | 79 | 83 |
SIM Reaction Time | . | 52 | 58 | 72 | 72 | 78 |
SIM Number Incorrect (%) | . | 52 | 54 | 63 | 72 | 73 |
SIM Number Incorrect (#) | . | 53 | 56 | 65 | 73 | 78 |
SIM Speed Deviation | . | 55 | 55 | 61 | 64 | 66 |
SIM Lane Deviation | . | 70 | 77 | 88 | 90 | 88 |
SIM Collisions | . | 43 | 56 | 65 | 72 | 75 |
SIM Times Over Speed Limit | . | 57 | 65 | 78 | 79 | 86 |
DAT Performance Index | 52 | 56 | 67 | 82 | 83 | 91 |
SIM Performance Index | . | 68 | 71 | 84 | 88 | 96 |
DAT+SIM Performance Index | . | 68 | 79 | 88 | 92 | 93 |
The scores in Table 2 were tested for statistical significance with a two-tailed binomial distribution test. If the null hypothesis were true and alcohol and placebo treatments were equal in effects, half the impairment scores would be positive and half would be negative. Examining the binomial distribution for 168 Ss reveals that the probability is less than .05 that as many as 58 percent of the Ss would show a positive impairment score if the null hypothesis were true, or conversely that as few as 42 percent would show a negative impairment score. For 112 Ss (the number tested at 0.10% BAC), the probability is less than .05 that as many as 60 percent would have an alcohol score worse than the placebo score. Thus, the null hypothesis is rejected if as many as 58 or 60 percent of the Ss exhibit poorer performance under alcohol.
Table 3 gives the exact probability for tests of the null hypothesis for each response measure at each BAC. Beginning with 0.02% BAC, two of the nine single response measures and two of the three composite measures showed statistically significant poorer performance under alcohol.
Table 3 | ||||||
Measurement |
BAC (%) | |||||
.00 | .02 | .04 | .06 | .08 | .10 | |
DAT Reaction Time | .938 | 1.000 | .001 | .001 | .001 | .001 |
DAT Number Incorrect (%) | .004 | .001 | .165 | .355 | .045 | .001 |
DAT Number Incorrect (#) | .016 | .001 | .123 | .355 | .031 | .031 |
DAT Tracking Error | .395 | .014 | .001 | .001 | .001 | .001 |
SIM Reaction Time | . | .537 | .031 | .001 | .001 | .001 |
SIM Number Incorrect (%) | . | .643 | .280 | .001 | .001 | .001 |
SIM Number Incorrect (#) | . | .440 | .123 | .001 | .001 | .001 |
SIM Speed Deviation | . | .217 | .217 | .003 | .001 | .001 |
SIM Lane Deviation | . | .001 | .001 | .001 | .001 | .001 |
SIM Collisions | . | .064 | .123 | .001 | .001 | .001 |
SIM Times Over Speed Limit | . | .090 | .001 | .001 | .001 | .001 |
DAT Performance Index | .588 | .123 | .001 | .001 | .001 | .001 |
SIM Performance Index | . | .001 | .001 | .001 | .001 | .001 |
DAT+SIM Performance Index | . | .001 | .001 | .00 | .001 | .001 |
At 0.04% BAC, five of the single response measures and all three composite scores show statistically significant alcohol impairment. At 0.06% BAC eight of the nine measures and all composite scores show statistically significant alcohol impairment. At 0.08% and 0.10% BAC all single and composite scores show statistically significant alcohol impairment.
As suggested by the graphs and supported by the statistical analyses, the overwhelming majority of Ss were significantly impaired by alcohol on some important measures beginning at 0.02% BAC, the lowest level tested. The number of Ss who were impaired by alcohol increased as BACs increased. Also, in general the magnitude of the impairment increased with increasing BAC (Figures 2 and 3).
Several of the single response variables showed a slight deviation from the finding of maximum impairment at 0.10% BAC in that the greatest impairment occurred at 0.08% BAC. To investigate this phenomenon, which initially was believed to be due to the light drinkers being first tested at 0.08% BAC, raw scores were examined separately for light, moderate, and heavy drinkers. For all drinking practices groups, greater impairment on some responses occurred at the second post-treatment testing. That is, for moderate and heavy drinkers, there was more impairment on a few of the responses at 0.08% than at 0.10%, and for light drinkers there was more impairment on some variables at 0.06% than at 0.08%.
It should be noted that since the behavioral tests began at the same time for all Ss, the moderate and heavy drinkers were tested at 0.10% at the time of day when light drinkers were tested at 0.08%. It is possible, therefore, that a time-linked factor increased impairment at the second post-treatment test. Perhaps some source of stimulation offset impairment at the first post-treatment test, or possibly a circadian interaction produced a greater decrement at the next test time. The issue cannot be resolved from the data, but the analyses make it clear that it was the order of testing rather than BAC that caused the variation in the magnitude of impairment. For the majority of measures, the expected relationship of greater impairment with higher BAC was found. Figures AP-I-3a to AP-I-10b, which present raw DAT and SIM scores separately for light, moderate and heavy drinkers, illustrate that for most measures the greatest impairment occurred at the highest BACs.
Finally, Ss were tested on DAT only when their BACs returned to 0.00%. Three of the four DAT response measures showed no alcohol effect; the percent of Ss with more impairment after alcohol was roughly equal to the percent of Ss with more impairment after placebo. There was, however, an unusual effect for the remaining measure, number of errors in detecting peripheral signals. In a statistically significant deviation, performance was better than what would be expected. Whether this result reflects some time-linked factor or a rebound effect cannot be determined. Since it was only one of four response measures, it also cannot be predicted that it would occur upon retest.
2. Age, Gender and Drinking Practice Effects
As discussed above, these data indicate that alcohol, even at 0.02% BAC, produces impairment in some important measures in the majority of Ss. This section considers whether impairment by alcohol varies as a function of age, gender or drinking practices.
The original design for this study would have supported a complete factorial analysis of variance with age, gender and drinking practices as the main effect and with the alcohol treatments nested within each cell of the factorial design. All Ss were to have received sufficient alcohol to achieve a peak 0.11% BAC. As discussed in an earlier section, however, it was determined that many light drinkers probably would experience nausea or more severe effects at that level of alcohol. As a consequence, the peak BAC for light drinkers was set at 0.09%. It was further determined that a standard analysis of variance of main effects, which would sum the effect of a variable across all levels, might obscure small effects which occurred at certain BACs and not at others. It was decided, therefore, to simplify the analysis by examining each alcohol level as a separate factorial design. This analysis also removed the problems of the interactions with the different BACs, which would have required another dimension in the factorial design
The five factorial designs for statistical analysis of the impairment scores at the five BACs from 0.02% to 0.10% are shown in the matrix below. With only heavy and moderate drinkers at 0.10% BAC, the design is a 4 (Age) X 2 (Gender) X 2 (Drinking Groups) factorial. With all Ss and three drinking groups at 0.08%, 0.06%, 0.04%, and 0.02%, the design is 4 X 2 X 3.
BAC | DESIGN |
0.10% | Age (4) X Gender (2) X Moderate and Heavy Drinkers (2) |
0.08% |
Age (4) X Gender (2) X Light, Moderate and Heavy Drinkers (3) |
Table 4 summarizes the mean impairment score for each response variable within age, gender, and drinking practice. These data, which are across all BACs, are not the basis of the statistical significance tests. They merely provide an overview of the variability in impairment by alcohol with the three groupings. The scores are in the original response measure dimension, adjusted for baseline; for example, reaction time is in seconds. The figures in Appendix II show the original impairment scores at each BAC for each of the single response variables by each of the categories within age, gender, and drinking practices. The figures in Appendix III show the impairment scores at each BAC for the three DAT, SIM, and DAT+SIM composite scores.
Table 4 | ||||||||||
Measurement |
Subject Groups | |||||||||
Age |
Drinking |
Gender | All | |||||||
19-20 | 21-24 | 25-50 | 51-69 | Light | Mod | Heavy | Female | Male | ||
DAT Reaction Time | 0.50 | 0.37 | 0.53 | 0.33 | 0.43 | 0.30 | 0.36 | 0.34 | 0.53 | 0.43 |
DAT Number Incorrect (%) | 1.65 | 1.72 | 1.43 | 0.76 | 1.94 | 0.73 | 1.16 | 1.05 | 1.73 | 1.39 |
DAT Number Incorrect (#) | 0.41 | 0.41 | 0.57 | 0.25 | 0.54 | 0.29 | 0.29 | 0.24 | 0.58 | 0.41 |
DAT Tracking Error | 0.42 | 0.37 | 0.29 | 0.21 | 0.22 | 0.35 | 0.28 | 0.40 | 0.25 | 0.32 |
SIM Reaction Time | 0.30 | 0.26 | 0.24 | 0.06 | 0.21 | 0.14 | 0.21 | 0.22 | 0.22 | 0.22 |
SIM Number Incorrect (%) | 6.66 | 5.03 | 4.71 | 1.04 | 4.42 | 1.61 | 5.47 | 4.05 | 4.67 | 4.36 |
SIM Number Incorrect (#) | 0.72 | 0.64 | 0.57 | 0.08 | 0.58 | 0.20 | 0.55 | 0.50 | 0.50 | 0.50 |
SIM Speed Deviation | 0.33 | 0.81 | 0.42 | 0.22 | 0.42 | 0.21 | 0.66 | 0.73 | 0.16 | 0.44 |
SIM Lane Deviation | 0.35 | 0.55 | 0.46 | 0.56 | 0.46 | 0.53 | 0.40 | 0.48 | 0.48 | 0.48 |
SIM Collisions | 2.73 | 3.74 | 3.15 | 6.50 | 3.67 | 4.06 | 3.46 | 4.16 | 3.90 | 4.03 |
SIM Times Over Speed Limit | 3.68 | 6.30 | 4.89 | 4.81 | 4.80 | 4.14 | 4.81 | 4.88 | 4.96 | 4.92 |
DAT Performance Index | 0.47 | 0.38 | 0.42 | 0.28 | 0.31 | 0.33 | 0.32 | 0.38 | 0.39 | 0.39 |
SIM Performance Index | 0.51 | 0.61 | 0.52 | 0.46 | 0.51 | 0.47 | 0.48 | 0.53 | 0.53 | 0.53 |
DAT+SIM Performance Index | 0.54 | 0.54 | 0.50 | 0.40 | 0.46 | 0.43 | 0.43 | 0.49 | 0.50 | 0.50 |
Table AP-I-6 presents the mean scores for each classification within each of the three main grouping factors. The scores have been standardized; that is, they have been transformed in terms of standard errors of the mean so that they have a common metric. In addition, the probability level appearing after each mean value indicates whether the mean impairment score within that category, when divided by the standard error of the mean of that category, is statistically significant. The tabled probability values are defined as follows: zero = probability less than .10, 1 = probability less than .05, 2 = probability less than .01, and 3 = probability less than .001.
Finally, Table AP-I-7 gives the test results for the mean effects of age, gender and drinking practices and their interactions. At each BAC, for each of the three main effects and four interaction terms, there were 48 statistical tests. Composite measures were excluded. Thus, there are nine single measure (6 from SIM, 3 from DAT) for five BACs plus three DAT measures at zero BAC for a total of 48 tests.
Table 5 summarizes the number of tests that were significant at the .05 level for each factor and the interactions. Six tests are significant for age, four for gender, five for drinking practices, two for the age X gender interaction, and five for the age X drinking practices interaction.
Table 5 | ||
Effect | Tests (Total Number) |
Tests p.05 (Number) |
Age | 48 | 6 |
Gender | 48 | 4 |
Drinking Practice | 48 | 5 |
Age X Gender | 48 | 2 |
Age X Drinking Practice | 48 | 5 |
Gender X Drinking Practice | 48 | 0 |
Age X Gender Drinking Practice | 48 | 0 |
Thus, of 336 statistical tests performed to evaluate differential alcohol effects as a function of age. gender, or drinking practices, only 22 reached the .05 significance level. Given random performance variability, some statistical tests will be significant by chance even if there were no true underlying performance differences as a function of the experimental variable. An experiment-wide judgment of the number of findings expected to be significant at the .05 level by chance is difficult, because in the repeated measures design the same Ss were used in all tests. An approximation, however, assuming independence of statistical tests and using Fisher's exact test, indicates that six positive significant tests out of 48 are required to reach at least a .05 level. Five significant tests only reach a .18 probability level.
Only the age variable approaches overall significance. Even within the age variable, however, six significant test among four response variables at three BACs occurred in no consistent pattern. It is concluded, therefore, that within the limits of the population represented by the study sample, there is no significant evidence that either age, gender, or drinking practice produces a differential response to the impairing effects of alcohol.
As noted earlier, no Ss were younger than 19 years of age nor over 70, nor did the sample include alcohol abstainers, heavy alcohol abusers, or alcoholics. Thus, the conclusions are limited by the sample, but the characteristics of the sample likely represent the characteristics of 80 - 90 percent of the driving public who will take a drink.
To re-state the finding, for the population represented by the study sample, which demonstrated impairment in driving skills beginning at 0.02% BAC, differences in age, gender, and drinking practices provide no mitigation of impairment. Had the experiment used many more Ss to greatly increase the power of the statistical tests, some of the small differences might have reached statistical significance. From a social point of view, that would be irrelevant to the study findings, because the actual differences would remain small in comparison to the overall effects of alcohol.
The tables and figures in the appendix support these conclusions. In a non-significant trend, the oldest drivers' response to alcohol appeared dissimilar to the response of the other three groups. There was, however, no consistent direction since the oldest drivers were least impaired on four measures and most impaired on two measures. Males and females split the measures on which they were more impaired with no evidence of any gender superiority. Among drinking practice groups, light drinkers showed a tendency toward more impairment, but it was small and non-significant. Moderate and heavy drinkers were indistinguishable in degree of impairment. Even if these trends had been statistically significant, they were so small as to be socially irrelevant.