The following sections present the results of two approaches in reviewing the literature. The first approach presents the data for impairment across all behavioral areas, counting the number of studies with each study counted once at the lowest BAC for which impairment was found. Most studies, however, reported on more than a single measure and, in fact, several reported findings for multiple driving skills across different behavioral areas (i.e., vigilance and divided attention and psychomotor skills). In addition, several studies reported tests of performance in different behavioral areas and at different BACs. The second analysis, which focused on specific behavioral areas, examined reports of more numerous behavioral tests across BACs.

3.1. Overall Impairment

The following analysis is based on 109 of 112 reviewed studies. Three studies were not included. The following is a brief discussion on the rationale for the exclusions.

Willumeit, et al., (1984) described their apparatus as a driving simulator, but it is better described as a tracking device. Subjects moved a light signal in a horizontal plane to coincide with the appearance of a light stimulus. The light signal appeared in one of 50 possible blocks along a horizontal scale. Each appearance of the step signal was preceded by one of two arrows to indicate the direction of stimulus appearance. The study treatments were alcohol, two benzodiazepines, and a beta-blocker. The three drugs and a placebo were administered with and without alcohol for eight treatments total. At 0.05 g/dl BAC, there was no difference between alcohol and placebo treatments. Ten mg diazepam also failed to impair in comparison to placebo. Since this analysis focused on the BAC threshold for impairment, studies in which impairment was not found at any BAC were excluded, whether the result was due to instrument insensitivity, flawed methodology or other cause.

A study by McMillen, et al. (1989), which is described as a study of risk taking in a driving simulator, also was dropped from the analysis. The simulator was a video driving game (Sego, Model 100), and dependent measures were number of lane changes, cars passed, and time at maximum speed during a 4.5 minute drive. A mean BAC of 0.07 g/dl had no effect on any response measure. These results are at variance with studies which report alcohol effects on risk taking (Cohen, et al.,1958; Light and Keiper,1969; Fromme,1997). Similar research has reported that alcohol affects speed selection. Since McMillen, et al. (1989) reported no alcohol effects at the tested BACs, this study was not included in the analysis on the assumption that the measures obtained with the video driving game were insensitive to alcohol.

Finally, one other study was not included, although it did report a response measure sensitive to the effects of alcohol. Yesavage and Leirer (1986) examined the aftereffects of alcohol ingestion. Although other studies of aftereffects include data obtained before subjects' BACs dropped to zero, this one did not. Since it only measured performance at zero BAC, the results do not pertain to the issue of BACs at which alcohol impairment first appears.

Figure 1, based on 109 studies, shows the number of studies reporting impairment by the lowest BAC at which impairment appeared. Note that the BAC categories used here are slightly different than those Figure1. Chart of Number of studies reporting impairment (109), by loweat BAC at which impairment was found. in the Moskowitz and Robinson study. Here each BAC category ends with a 9 (e.g., 0.020-0.029 g/dl) whereas Moskowitz and Robinson used BAC categories ending in zero(e.g., 0.021-0.030 g/dl). Twenty-seven percent of the studies reported impairment by 0.039 g/dl, 47% by 0.049 g/dl and 92% by 0.079 g/dl. The impairment appeared in one or more of the response variables examined in the study. As shown in Tables B1 and B2 in Appendix B, impairment was reported by more studies and for lower BAC than in the 1988 review by Moskowitz and Robinson. The difference may be accounted for in two ways. First, pre-1988 studies included very few which examined more than one BAC. If impairment is reported only for a single selected BAC, no inference can be drawn about alcohol effects at lower BACs. In this review of more recent literature, the majority of studies have examined multiple BACs, which permits the identification of lower BAC at which impairment appears. Second, the methods and instruments used by researchers in this past decade have improved.

3.2. Impairment, by Behavioral Areas

The following analysis is based on all 112 reviewed studies.Figure 2 summarizes the number of test results Figure2. Chart of Summary of behavioral tests results (556). Aftereffects are reported at zero BAC. by BAC for all the tasks examined.Note that these are not the number of studies, but the total number of tasks across experimental conditions for all studies. In some cases, impairment was reported at BACs as low as 0.009 g/dl. By the time subjects reach BACs of 0.030 g/dl, the number of impaired behavioral areas is greater than the number not impaired. As BACs increase, the number of areas showing impairment also increase. Clearly, the measurement of impairment at very low BACs requires highly sensitive measures. Also, as will be seen later, some behavioral areas are far more sensitive to the effects of alcohol than others. Even within a given area, there was considerable variation in the BAC at which impairment was first reported. As previously noted, it is relevant that experimental procedures vary greatly since procedures affect the sensitivity of response measures. The following are comments on the results within each behavioral areas.

3.2.1. Driving and Flying: On the Road and Simulators (Figure 3, Table 1 in Appendix A).
Twenty-five studies produced a total of 50 behavioral tests. Impairment was reported for BACs below 0.01 g/dl. As shown in Figure 3, nearly all driving and flying simulator studies or on-the-road studies of driving reported impairment by alcohol. The lowest BAC at which impairment was found (0.001 g/dl) was reported by Morrow et al. (1990) who, using a flying simulator, required subjects to integrate information about aircraft, traffic, and weather conditions; maneuver the aircraft along a dynamic flight path; and maintain radio communications.

McMillen et al. (1989) reported the highest BAC at which impairment was not found (0.070 g/dl). The methodological and instrumental problems in that study were discussed in the preceding section.

In the decade since the Moskowitz and Robinson report was published, the sensitivity, reliability and face validity of driving and flying simulators have improved. Driving simulators now present scenarios which better reflect the mental workload of actual driving, which may account for their increased sensitivity to alcohol. Note that subjects in the Morrow et al. study, which reported impairment at 0.001 g/dl, were required to perform multiple tasks simultaneously in a divided attention paradigm.

3.2.2. Divided Attention (Figure 3, Table 2 in Appendix A)
Eighteen studies of divided attention yielded 52 behavioral tests. In general, experimental tasks aimed at measuring the ability to divide attention are sensitive to alcohol effects, beginning at BACs of 0.005-0.010 g/dl. Divided attention tests require subjects to perform two tasks concurrently, and most use a central tracking task and a peripheral visual search task. This approach is appropriate since it models the divided attention characteristics of driving; tracking can be considered analogous to maintaining lane position and visual search corresponds to monitoring the environment. Roehrs et al.(1994) used this configuration and measured impairment at BACs as low as 0.005 g/dl.

A few divided attention tasks use apparatus which requires subjects to simultaneously monitor number displays in central and peripheral vision. This approach appears less sensitive than the

combined tracking-visual search task, possibly due to the similarity of the two number tasks and the lack of a continuous component (such as tracking).

3.2.3. Drowsiness (Figure 3, Table 3 in Appendix A)
Although wakefulness is not a measure of skills performance, it is an essential requirement for safe driving. Sleep, or more accurately drowsy driving due to sleep loss or deprivation, has been identified as a contributing variable to crashes, and its potential interaction with alcohol is of import because most alcohol-related crashes occur at night, when drivers are more likely to need sleep.

Figure3. Charts of Behavioral test results, by behavioral area.

Six studies of the effects of alcohol on drowsiness produced 13 behavioral test reports of which 11 showed impairment. The findings were obtained with two tests, the multiple sleep latency test (MSLT) and the repeated test of sustained wakefulness (RTSW). The MSLT is a highly standardized measure of physiological sleep tendency. Subjects are connected to polysomnographic equipment, and are given the opportunity to fall asleep at regular intervals. Sleep latency is a measure of elapsed time from when the subjects are told to fall asleep to the occurrence of the first epoch of any sleep stage. The RTSW also measures physiological sleep tendency, but in this test subjects are instructed to resist falling sleep. It has not been validated as extensively as the MSLT.

In general, wakefulness tests were found to be very sensitive to the effects of alcohol. The time to fall asleep was shorter with BACs of 0.010 g/dl and higher, except for two instances, one time at 0.021 g/dl and the other at 0.034 g/dl. The latter result was obtained with the RTSW, arguably the less sensitive of the two test.

3.2.4. Vigilance Tasks (Figure 3, Table 4 in Appendix A)
Nine vigilance studies produced 18 behavioral test results, of which 16 showed alcohol impairment. None of the studies reviewed examined vigilance at BACs below 0.020 g/dl, and two studies reported that there was no impairment at BACs of 0.021 and 0.028 g/dl. Note that one of the studies (Gustafson, 1986) required the subjects to press a switch as rapidly as possible when a tone of 1000 hz at 90 db was presented. A tone of that magnitude might alert subjects and offset the effects of alcohol, particularly at low BACs. At BACs of 0.030 g/dl and above, impairment was reported consistently across all studies.

Vigilance studies which had been published in the literature at the time of the Moskowitz and Robinson review were considerably less likely to show impairment by alcohol. Clearly, since then, more sensitive measures of vigilance have been developed.

3.2.5. Tracking (Figure 3, Table 5 in Appendix A)
Eleven studies of tracking produced 23 behavioral test results. Overall, the results indicated that threshold of impairment varied as a function of the type of tracking task used.

3.2.6. Perception (Figure 3, Table 6 in Appendix A)
Twelve studies produced 35 test results. Studies in this category used tasks which differed widely in terms of information processing requirements. Tasks included time estimation, auditory signal detection, visual search, pattern recognition, and traffic hazard perception. The diversity of mental workload is believed to underlie the reported diversity in alcohol sensitivity. In general, the evidence indicates a lack of significant impairment of perceptual abilities below BACs of 0.080 g/dl, although there were some reports of impairment at lower BACs. The lowest impairing BAC (0.037 g/dl) was reported by Lapp et al. (1994), and the highest BAC at which impairment was not found (0.080 g/dl) was reported by Heishman et al., (1997). Interestingly, both of these findings were both obtained with time production/estimation tasks.

Moskowitz and Robinson found perception tasks to be far more sensitive to alcohol than the previous paragraph reports. However, the studies in that earlier review reported on the examination of behaviors quite unlike those examined by investigators in the more recent studies reviewed here.

3.2.7. Visual Functions (Figure 4, Table 7 in Appendix A)
Nineteen studies examined alcohol effects on visual functions, producing a total of 63 behavioral tests. In general, the reports of impairment were consistent at 0.030 g/dl and higher. Many different behavioral functions were included in this category, including visual acuity, contrast sensitivity, eye movements and ocular motor control. Visual acuity appears not to be affected by alcohol below 0.070 g/dl BAC, a finding which is consistent with earlier studies. On the other hand, contrast sensitivity (the ability to discern spatially distinct luminance differences) and tests of oculomotor control were impaired by alcohol at 0.030 g/dl. Mattila et al. (1992), for example, found that subjects' coordination of extra-ocular muscles was significantly affected at 0.026 g/dl.

3.2.8. Cognitive Tasks (Figure 4, Table 8 in Appendix A)
This category encompassed 31 diverse studies with 145 test reports. The tasks varied considerably in assumed information processing characteristics and in sensitivity to alcohol. So perhaps it is not surprising that the appearance of impairment by alcohol varied from as low as 0.005 g/dl to as high as 0.160 g/dl.

Figure 4. Charts of Behavioral test results, by behavioral area.

Studies of backward masking, a well established measure of information processing rate or perceptual speed, found impairment at 0.030 g/dl (Wilkinson, 1995). Digit-symbol substitution tasks, on the other hand, did not reliably show a deficit until BACs above 0.060 g/dl. Even higher thresholds applied to card sorting, grammatical reasoning, and the Sternberg memory task. Card sorting tests were not generally affected by BACs below 0.090 g/dl, although Lyvers & Maltzman (1991) reported impairment at BACs below 0.050 g/dl. Grammatical and mathematical

reasoning tests were not generally affected by BACs below 0.080 g/dl, although Heishman et al. (1997) and Kennedy et al. (1993) reported impairment at BACs of 0.025 g/dl and 0.060 g/dl, respectively. Memory tests, including the Sternberg memory tests, which require subjects to memorize a set of symbols (letters or numbers) and to later determine whether a short sequence of symbols contains the memorized set, were not affected by BACs below 0.060 g/dl. A notable exception was the results reported by Millar et al. (1995) who reported impairment at BACs below 0.020 g/dl in selective reminding tasks.

It is difficult to summarize the evidence concerning alcohol effects on cognitive tasks. As can be recognized, the tasks discussed above have little in common behaviorally, and some are complex and likely require more than one cognitive function.

3.2.9. Psychomotor Skills (Figure 4, Table 9 in Appendix A)
Eighteen studies of the effects of alcohol on psychomotor skills produced 57 tests, 33 of which found impairment and 24 did not. Because of the diverse nature of psychomotor skills and the tests of those skills, it is difficult to establish a threshold for alcohol effects. For that reason, the results have been further divided into three task groups: Finger tapping, body balance, and skilled physical tasks.

3.2.10. Choice Reaction Time (Figure 4, Table 10 in Appendix A)
Choice reaction time experiments use multiple stimuli and response possibilities, thereby placing a greater information processing load on subjects than simple reaction time. Fifteen choice reaction time studies produced 37 behavioral test results. Although most choice reaction time studies showed impairment by alcohol, it was only at 0.060 g/dl that there were more reports of impairment than of no impairment. By 0.080 g/dl, however, more than 80% of the studies reported evidence of complex reaction time impairment. Differences in findings are attributable to a wide range of stimulus and response conditions. There is no doubt that choice reaction time is more sensitive to the effects of alcohol than simple reaction time, but a variety of experimental methods under this single rubric leads to differing findings.

3.2.11. Simple Reaction Time (Figure 4, Table 11 in Appendix A)
Five studies with 20 test results at various BACs examined alcohol effects on simple reaction time. Moskowitz & Robinson (1988) concluded that simple reaction time is an insensitive measure. The experiments involve repetitive testing with a single known stimulus and a single known response. Subjects not only know where and what the stimulus is and what the single response option is, they may also be cued when a stimulus is about to occur. As Figure 4 suggests, most experiments using simple reaction time as a measure failed to show any alcohol effects. These tasks, of course, are unrelated to the reaction time demands of actual driving where it is rare for a drivers to know about the initiating stimulus in advance or to know what response will be required.

3.2.12. Critical Flicker Fusion (Figure 4, Table 12 in Appendix A)
In Critical Flicker Fusion (CFF) a subject indicates the threshold at which he/she perceives a flickering on/off light to be constant; that is, not flickering anymore. It has frequently been used in studies of psychoactive drugs. In seven studies at 18 BACs, CFF was an extremely insensitive measure for which impairment occurred only above 0.100 g/dl BAC. Continued use of this test to examine the effects of alcohol on driving related behavior is unwarranted, both because of its insensitivity to alcohol, but also because there is no known relationship to driving.

3.2.13. Aftereffects (Table 13 in Appendix A)
The research area of alcohol aftereffects emerged during the last decade. It examines the residual effects after a positive BAC has declined to zero. Aftereffects are distinguished from hangovers, which are experienced subjectively, and may affect performance without subjective reaction. Twelve studies examined aftereffects of alcohol and produced 25 behavioral test results, of which six reported impairment and 19 did not. A variety of measures included tracking, body sway, eye movements, simple reaction time, critical flicker fusion, symbol copying and others. In the reported impairment, however, only three response measures were used: the MSLT, a flight simulator, and a measure of angular motion.

Reported findings appear to be a direct function of the measures used to study aftereffects. Angular motion, for example, as studied by Ross et al. (1995) used unusually elaborate and complex testing equipment. Subjects were seated in a compartment which rotated clockwise until they reported that the sensation of motion had stopped. Starting from 3 rpm, thresholds for detection of right turns (acceleration of the compartment) and for detection of left turns (deceleration of the compartment) were determined for each individual subject. The subjects' task also included calling out the direction of the turn while depressing a yoke button until the turning ceased, maintaining constant altitude by observing the altimeter and vertical speed indicator, making appropriate yoke inputs, and monitoring for two numbers on a separate visual search task. A significant shift in the threshold of angular motion was observed after the subjects ingested small quantities of alcohol and after a return to zero BAC.

Most other studies used less sensitive measures. Although this area of study has no bearing on the issue of BAC limits, the findings of impairment as a consequence of aftereffects is a traffic safety issue which needs further study.