Effects of Different Alcohol Dosages on Steering Behavior in Curve Driving

Abstract

Objective:

The aim of this article is to explore the detailed characteristics of steering behavior in curve driving at different alcohol dosages.

Background:

Improper operation of the steering wheel is a contributing factor to increased crash risks on curves.

Method:

The experiments were conducted using a driving simulator. Twenty-five licensed drivers were recruited to perform the experiments at the four different breath alcohol concentration (BrAC) levels. The steering angle (SA), steering speed (SS), steering reversal rate (SRR), and peak-to-peak value of the steering angle (PP) were used to characterize the steering behavior. The vehicle’s speed and the number of lane exceedances per kilometer were also used to examine the driving performance.

Results:

The SSs on the 200 m (χ²(3) = 20.67, p < .001), 500 m (χ²(3) = 22.42, p < .001), and 800 m (χ²(3) = 22.86, p < .001) radius curves were significantly faster for drivers under the influence of alcohol compared with those given a placebo. There were significant effects of alcohol on the SRR and PP on the 200 m, 500 m, and 800 m radius curves.

Conclusion:

For all of the curves, the SS, SRR, and PP had a tendency to increase as the BrAC increased. The large PP at a high BrAC, accompanied by the high speed, SS, and SRR, resulted in a high probability of lane exceedance. The use of measures of SS, SRR, and PP aided in the improvement of the accuracy of the intoxication detection for the different types of curves.

Application:

The most important application is to provide guidance for detecting alcohol-impaired-driving.

Keywords

driver impairment driver behavior traffic safety alcohol steering behavior

Introduction

Steering is one of the basic skills involved in driving. A driver controls the trajectory of a vehicle by steering it. A driver is constantly perceiving and judging the situation, as well as adjusting the steering wheel to negotiate a curve and to stay within or change lanes. Steering behavior has attracted a great deal of attention. Researchers have attempted to analyze and describe steering behavior in terms of the features of steering movement. Generally, the features or measures of steering wheel movement are grouped into the following three classes: time domain features, frequency domain features, and state space features (Krajewski, Sommer, Trutschel, Edwards, & Golz, 2009). Time domain features include the standard deviation of the steering angle, the steering velocity (Peters, Kloeppel, & Alicandri, 1999), and the steering rate (Verwey, 2000). Frequency domain features include the spectral flux, the high frequency component of the steering angle (Östlund, Nilsson, et al., 2004), and the steering entropy (Nakayama, Futami, Nakamura, & Boer, 1999). Krajewski et al. (2009) used a three-dimensional state space and a reconstructed phase space to describe steering behavior. Compared with the frequency domain features and with the state space features, the time domain features require less computational power and are more straightforward and easy to understand.

Alcohol is one of the factors that affects steering behavior. In the U.S., alcohol-related crashes contribute to approximately 31% of all traffic fatalities (National Highway Traffic Safety Administration, 2009). A report showed that alcohol was a vital factor in approximately 10,300 traffic fatalities and that approximately 28 alcohol-related deaths occur every day (National Highway Traffic Safety Administration, 2013). In China, alcohol-related crashes have been shown to contribute to 34.1% of all road crashes (Li, Xie, Nie, & Zhang, 2012). Numerous studies have shown that alcohol intoxication impairs driving performance (Keall, Frith, & Patterson, 2000; Linnoila, Erwin, & Ramm, 1980; Liu & Ho, 2010; Moskowitz, Burns, Fiorentino, Smiley, & Zador, 2000; Moskowitz & Robinson, 1988; SRA, 1989; Compton, 1991; Snyder, 1992; Zhao, Zhang, & Rong, 2014). Alcohol is particularly deadly on curves. Glennon, Neuman, and Leisch (1985) reported that a crash on curved roads was more severe than that on straight roads. Chen, Rakotonirainy, Loke, and Krishnaswamy (2007) showed that alcohol increased the risk of crashing on horizontal curves. The reasons for this are many. Compton (1991) showed that alcohol affected steering and brake control ability. Seehafer, Huffman, and Kinzie (1968) found a significant increase in steering reversals with moderate levels of alcohol (blood alcohol concentration [BAC] 0.05%–0.07%). Huntley, Kirk, and Perrine (1972) reported that high levels of alcohol (BAC 0.085%–0.116%) increased the amount of coarse steering reversals. However, Huntley and Perrine (1971) showed that alcohol had no significant effect on steering reversals. Allen, Stein, and Hogue (1982) reported that steering and speed control behavior were impaired by alcohol at a BAC of 0.10%. SRA (1989) showed that a BAC of 0.08% impaired the accuracy of steering, braking, and speed control. Rakauskas et al. (2008) showed there was a significant difference in steering variation between alcohol (BAC of 0.073%) and the placebo. Fillmore, Blackburn, and Harrison (2008) demonstrated steering rate was faster under alcohol (BAC of 0.065%) compared with the placebo. Verster et al. (2009) reported that steering error was impaired by alcohol at a breath alcohol concentration (BrAC) at or above 0.05%. Van Dyke and Fillmore (2015) found that a moderate level (BAC of 0.065%) of alcohol impaired the steering rate. Drew, Colquhoun, and Long (1958) indicated that low doses (BAC as low as 0.020%) could impair steering accuracy. Martin (1971) showed that low levels of alcohol (BAC 0.045%–0.051%) achieved better driving efficiency than a BAC of 0.000%, as measured by fine steering reversals. Linnoila et al. (1980) showed that a low level of alcohol (BAC 0.035%) had an effect on drivers’ ability to drive a vehicle. In summary, it can be seen that the findings of the effects of alcohol on steering behavior are inconsistent at low levels of alcohol. The causes of these inconsistencies are varied and include the time and methods used for measuring BAC, the testing methods, the testing devices, and the individual differences among subjects. Huntley (1970) stated that in addition to alcohol, driving experience, personality, drinking experience, and the nature of the driving tasks can also affect driving performance. Of course, the driver’s state, such as fatigue or drowsiness, can also affect driving performance. However, in general, impairment appears at moderate to high levels of alcohol. In particular, fully eight studies find an effect of alcohol on steering behavior (Allen et al., 1982; Fillmore et al., 2008; Huntley et al., 1972; Rakauskas et al., 2008; Seehafer et al., 1968; SRA, 1989; Van Dyke & Fillmore, 2015; Verster et al., 2009) whereas only one study finds no effect of alcohol on steering behavior (Huntley & Perrine, 1971) at moderate to high levels of alcohol. In other words, there is clear evidence that steering behavior is significantly impaired by alcohol.

Although there have been numerous studies on the effect of alcohol on steering behavior, few studies have explored the effect that alcohol has on the detailed characteristics of steering behavior (e.g., the speed, frequency, and amplitude of steering movement) at different BAC levels, especially on curves. Therefore, it is of interest to explore the detailed characteristics of steering behavior in curve driving at different BAC levels. A clearer understanding of how alcohol affects steering behavior could potentially aid in the improvement of accurate methods to detect alcohol-impaired driving. Therefore, this study used a driving simulator to investigate the effect of different alcohol dosages on steering behavior on different curved roads. In this article, steering behavior was characterized by the steering angle, steering speed, steering reversal rate, and peak-peak value of the steering angle. In addition, the vehicle’s speed and the number of lane exceedances per kilometer were also used to examine the driving performance.

Materials and Methods

Equipment

The simulator (Figure 1) used in this study was technically similar to a Toyota passenger car, with the gas pedal, safety belt, brake pedal, and steering wheel identical to a real car. There were pneumatic rods and four motors beside the four wheels that could independently provide pitch and roll cues to the driver. In other words, the simulator has 3 degrees of freedom, namely, pitch (tilting forward and backward), roll (tilting side to side), and elevating (moving up and down). A real-time, 3D virtual scenario was shown on 2D displays located in front of the simulator. The scenario provided a 130° field of view, with two side mirrors and one rear view mirror on which images were projected. The driver sat on the left-hand side of the simulator. The subject had a full view of the road scene. The simulator can record driving parameters, such as the vehicle’s position, the lane position, the speed, and the steering angle. The data sampling frequency was 30Hz. The simulator has been well calibrated and validated in our previous studies (Ding, Zhao, Rong, & Ma, 2013; Guan, Zhao, Qin, & Rong, 2014; Wu et al., 2016; Zhao, Li, Han, Zhang, & Rong, 2015). In addition, a total of 200 drivers in each driving experiment evaluated the validity of the simulator through questionnaires. The results are shown in Table 1. Almost all drivers thought that the visual scenarios, the steering wheel, and the accelerator were quite similar to the real conditions. A blowing-type BrAC sensor (EnviteC AlcoQuant 6020) was used to test the drivers’ BrAC levels. The sensor was calibrated by the Beijing Traffic Management Bureau.

Figure 1.

The simulator.

Table 1:

Results of Subjective Evaluation of the Driving Simulator (1 = not real at all; 10 = very real)

	Steering wheel	Brake	Accelerator	Gear	Speed perception	Scenario
Mean	7.9	7.5	7.9	7.5	7.7	8.1

Participants

Nagoshi, Wilson, and Rodriguez (1991) found that male drivers affected by alcohol-impaired-driving were more impulsive than females. Mayhew, Donelson, Beirness, and Simpson (1986) and Zador (1991) showed that young drivers were more likely to crash than older drivers at the same BAC level. Woodall et al. (2004) stated that the study of young drivers alcohol-impaired-driving was the key to improving traffic safety. Accordingly, 25 young, healthy male drivers between 20 and 35 years of age were recruited by advertisements. The average age of the subjects was 25, with a standard deviation of 4.1. All the drivers had held a valid driver’s license for more than 3 years. Individuals who self-reported irregular circadian rhythm, physical or mental illness, or drug use were excluded. All the participants had the ability to consume enough alcohol to achieve a BrAC level of higher than 0.09%. All the drivers agreed and signed their informed consent before they participated in the experiments. Subjects received 400 RMB (about $61) for their participation.

Scenario

The experimental scenario (Figure 2) was designed in accordance with the road design criteria. All the roads in the scenarios were divided highways with two 3.75-m-wide lanes in each direction separated by a fence. The lane boundaries were clearly delineated using white lines. These road boundaries provided the drivers with some visual cues and helped them maintain their position within the lane. The driving scenario consisted of straight and curved road segments. The straight road segments and curved road segments were alternated in the driving scenario. Two adjacent curves were connected by a tangent section, which was 1,200–1,500 m long. The curve warning signs were put in the driving scenarios at distances from 50 to 100 m before the curve entry. The speed limit signs (80 km/h) were put alongside the road. The total travel distance was approximately 14 km, and the drive typically required approximately 10 min. There were three curves bending to the left with radii of 200 m, 500 m, and 800 m and three curves bending to the right with radii of 200 m, 500 m, and 800 m. The participants were required to drive in the nearside lane for the duration of the drive. While each subject was driving, no other vehicles were present in the scenario.

Figure 2.

The scenario.

Experimental Design and Procedure

According to Chinese traffic laws, a BrAC level between 0.02% and 0.08% is regarded as drink driving, and a BrAC level of 0.08% or above is regarded as alcohol-impaired driving. At the same time, the dosage of alcohol that was consumed in many studies (Moskowitz & Fiorentino, 2000) was 0%–0.09%. Therefore, 0.03% (a unit more than 0.02%) and 0.09% (a unit more than 0.08%) were selected as the two levels. The 0.06% between the low (0.03%) and high (0.09%) BrAC levels was selected as another level. The level of 0.00%, i.e., the sober state, was selected as another level. Accordingly, there were four BrAC levels (0.00%, 0.03%, 0.06%, and 0.09%). Each subject drove the simulator at the four different BrAC levels at 2-day intervals on days 1, 3, 5, and 7 of the study, respectively, to avoid the residual effects of each alcohol dose. The order of the four different BrAC levels for each subject was randomized to avoid any effect of order. Each subject did not know the level he had selected.

The experiments were performed between 2:00 pm and 4:30 pm to avoid the effect of sleepiness, because the recruitment survey showed that the subjects were not drowsy after 2:00 pm according to their sleep cycle. All the subjects were required to sleep well before the sessions and to abstain from alcohol or drugs. The alcohol dose of each subject was calculated according to Watson’s study (Watson, 1989). The dose for the BrAC level was estimated according to Equation (1) (Watson, 1989),

\begin{array}{l} Alcohol dose (g) = \frac{10 \times BAL \times TBW}{0.8} + \\ 10 \times MR \times (DDP + TPB) \times \frac{TBW}{0.8} \end{array}

(1)

where BAL is the expected BAC level, TBW is the total body water, MR is the metabolic rate (generally 0.015 g/100 mL/h), $D D P$ is the duration of the drinking period, and $T P B$ is the time to peak BAL (generally 0.5 h). The TBW for men is as follows (Watson, 1989):

\begin{array}{l} {Men}^{'} s TBW = 2.447 - 0.09516 \times Age + 0.1074 \times \\ Height (cm) + 0.3362 \times Weight (kg) \end{array}

(2)

The dose was calculated according to the equations above before each subject drank. For the alcohol condition, Chinese liquor (46% alcohol content) was mixed with water up to a volume of 500 mL. Each subject did not know how much alcohol he drank. The placebo consisted of a volume of 500 mL of water. The experimental procedure was as follows:

First, a participant was instructed on the operation of the simulator and the experimental tasks. Then, the participant practiced operating the simulator for approximately 5 min and completed an initial familiarization task to become familiar with the experimental procedures.

Second, the participant consumed the dose of alcohol over 20 min. Approximately 15 min after consuming, his BrAC level was measured every 5 min.

Third, as soon as the participant reached the target BrAC level, the experiment was performed. The participant was required to drive smoothly to pass through the curves. Once a crash occurred, the subject was asked to drive again.

Finally, the BrAC level was also measured after completion of the session. The participant was required to complete a questionnaire to ensure he was not fatigued.

Measures

Four measures, namely, the steering angle (SA), steering speed (SS), steering reversal rate (SRR), and peak-to-peak value of the steering angle (PP), were used to characterize steering behavior. The SA refers to the angle of the absolute position (radians). The angle is positive if rotating in a counter-clockwise direction. The angle is negative if rotating in a clockwise direction. The SS refers to the speed of the steering wheel movement (radians/second). Steering speed is a scalar quantity and has a magnitude only. The SRR is the number of steering reversals that are larger than a finite angle in a given time period (reversals/second). The angle is also called a “gap,” across which the reversals are measured. A change in the steering reversal that is larger than the gap is regarded as one steering reversal. The gap sizes used in the literature generally vary between 0.5–10° (Mclean & Hoffmann, 1975; Sinelnikova, Lee, Reimer, Mehler, & Coughlin, 2015). Four different gap sizes of 1°, 5°, 10°, and 15° were tested in this study. The SRRs with four different gap sizes are shown in Table. 2. According to a comprehensive analysis of the overall effect and to pairwise comparisons, a gap size of 10° was selected in this study. The PP is the difference between the maximum and the minimum values of the steering angle in a given time period (30 s). The vehicle’s speed (VS) was also selected as one measure. In addition, the number of lane exceedances per kilometer (NLE) was used to examine the performance of lateral vehicle control. Thus, there were six measures, namely, VS, SA, SS, SRR, PP, and NLE.

Table 2:

Summary of the Results for SRR With Four Different Gap Sizes

Radius	Gap(°)	BrAC				Overall Effect
Radius	Gap(°)	0.00%	0.03%	0.06%	0.09%	Overall Effect
200 m	1	0.81(0.23)	0.88(0.21)	0.88(0.24)	0.99(0.23)*	χ²(3) = 10.745, p = .013
	5	0.47(0.16)	0.60(0.20)	0.60(0.20)	0.70(0.21)*	F(3,63) = 8.969, p < .001
	10	0.31(0.15)	0.42(0.18)	0.44(0.20)	0.52(0.19)*^∆	F(3,63) = 9.557, p < .001
	15	0.23(0.14)	0.34(0.17)	0.35(0.19)	0.42(0.17)*	χ ² (3) = 16.964, p < .001
500 m	1	0.76(0.17)	0.80(0.24)	0.83(0.30)	0.92(0.23)	χ²(3) = 4.964, p = .174
	5	0.43(0.16)	0.50(0.23)	0.55(0.27)	0.63(0.20)*^∆	F(3,63) = 5.740, p = .002
	10	0.19(0.13)	0.29(0.22)	0.36(0.25)*	0.43(0.20)*^∆	F(3,63) = 9.118, p < .001
	15	0.09(0.08)	0.18(0.19)	0.25(0.22)	0.31(0.19)*^∆	χ ² (3) = 25.844, p < .001
800 m	1	0.69(0.19)	0.90(0.30)*	0.81(0.29)	0.90(0.22) *	F(3,63) = 4.761, p = .005
	5	0.30(0.14)	0.53(0.31)*	0.50(0.28)*	0.60(0.21) *	F(3,63) = 8.741, p < .001
	10	0.10(0.10)	0.27(0.29)*	0.31(0.26)*	0.37(0.21) *	χ ² (3) = 20.127, p < .001
	15	0.03(0.05)	0.17(0.25)*	0.20(0.20)	0.24(0.19) *	χ ² (3) = 18.925, p < .001

Note. A repeated-measures analysis of variance (ANOVA) was used to conduct the significance tests. Nonparametric tests (the Friedman test and the Wilcoxon test) were applied if the data did not fit a normal distribution. Overall significant difference is shown in bold. Values in parentheses indicate the standard deviation. BrAC = breath alcohol concentration.

∆

p < (0.05/6) significant difference from 0.03%.

p < (0.05/6) significant difference from 0.00%.

Data Analyses

Three subjects did not complete all the sessions due to simulator sickness. There were 22 subjects who performed the whole session. SPSS 20.0 and SAS 9.4 were used for the analysis. The independent variables were the BrAC (four levels), curve radius (three levels), and turning direction (two levels), which were within-subject factors. The dependent variables were the VS, SA, SS, SRR, PP, and NLE. A Shapiro–Wilk test was used to test whether the data followed a normal distribution. A repeated measures ANOVA (two-tailed, p < .05) was used to conduct the significance tests. Nonparametric tests (the Friedman test and the Wilcoxon test) were applied if the data did not fit a normal distribution. If an overall effect of alcohol exists, it is meaningful to perform the pairwise comparisons to identify differences between the groups. Bonferroni’s correction was used to adjust for the pairwise comparisons.

Results

BrAC Levels

The actual BrAC levels (mean ± SD) of the participants matched the target BrACs for 0.03%, 0.06%, and 0.09%, being 0.033 ± 0.002%, 0.064 ± 0.002%, and 0.092 ± 0.001%, respectively. No BrACs were detected in the placebo condition.

Measures

The VSs in all treatments met the normality assumption, whereas the SAs, SSs, SRRs, PPs, and NLEs violated the normality assumption. Accordingly, the VSs were analyzed using a repeated measures ANOVA. The detailed analysis was addressed in our previous study (Zhang, Zhao, Du, Ma, & Rong, 2014). To the best of our knowledge, the Friedman test does not treat the two factors symmetrically and it does not test for an interaction between them. Therefore, “turning direction” was excluded in the Friedman test when the effect of BrAC on the steering behavior at each curve radius was examined. Along this line, for each BrAC level, the mean of the absolute value of each dependent variable on left and right curves with the same radius was regarded as the value of the dependent variable on the radius curve. The VSs were also processed in this way to be consistent with other measures. The sample mean and sample standard deviation were computed for each measure at four BrAC levels on all of the curves. The results are summarized in Table 3.

Table 3:

Summary of the Results for Each Measure for Curves

Radius	Measure	BrAC				Overall effect
Radius	Measure	0.00%	0.03%	0.06%	0.09%	Overall effect
200 m	VS (km/h)	78(7)	81(7)*	80(9)	74(11)^∆	F(3,63) = 5.040, p = .003
	SA (rad)	0.38(0.22)	0.46(0.23)	0.58(0.33)*	0.67(0.53)*	χ ² (3) = 20.564, p < .001
	SS (rad/s)	0.22(0.21)	0.32(0.23)	0.42(0.38)*	0.48(0.27) *	χ ² (3) = 20.673, p < .001
	SRR (r/s)	0.31(0.15)	0.42(0.18)	0.44(0.20)	0.52(0.19)*^∆	F(3,63) = 9.557, p < .001
	PP(rad)	1.35(1.35)	1.64(1.02)	2.41(1.99) *	3.21(2.98) *	χ ² (3) = 18.600, p < .001
	NLE (times/km)	0.95(1.84)	0.77(1.11)	1.59(1.53)	1.73(1.32)	χ ² (3) = 9.688, p = .021
500 m	VS (km/h)	89(9)	96(16)	96(13)	98(14)*	F(3,63) = 3.039, p = .035
	SA (rad)	0.20(0.03)	0.25(0.11)	0.28(0.16)	0.28(0.11)	χ²(3) = 5.945, p = .114
	SS (rad/s)	0.11(0.05)	0.16(0.12)	0.21(0.17)*	0.24(0.14) *^∆	χ ² (3) = 22.418, p < .001
	SRR (r/s)	0.19(0.13)	0.29(0.22)	0.36(0.25) *	0.43(0.20) *^∆	F(3,63) = 9.118, p < .001
	PP (rad)	0.57(0.23)	0.74(0.42)	1.02(0.82) *	0.99(0.46) *	χ ² (3) = 17.564, p < .001
	NLE (times/km)	0.60(0.97)	0.82(1.03)	1.19(1.51)	1.51(1.97)	χ² (3) = 3.253, p = .354
800 m	VS (km/h)	91(12)	104(21)	101(19)	103(19) *	F(3,63) = 3.870, p = .013
	SA (rad)	0.14(0.03)	0.19(0.11)	0.19(0.10)	0.20(0.08) *	χ ² (3) = 10.691, p = .014
	SS (rad/s)	0.07(0.03)	0.15(0.14) *	0.17(0.13)*	0.20(0.12) *	χ ² (3) = 22.855, p < .001
	SRR (r/s)	0.10(0.10)	0.27(0.29) *	0.31(0.26) *	0.37(0.21) *	χ ² (3) = 20.127, p < .001
	PP (rad)	0.42(0.25)	0.57(0.38)	0.77(0.54) *	0.97(0.60) *^∆	χ ² (3) = 23.073, p < .001
	NLE (times/km)	0.38(0.87)	0.52(0.78)	0.79(1.00)	1.52(1.47) *	χ ² (3) = 14.566, p = .002

Note. A repeated-measures ANOVA was used to conduct the significance tests. Nonparametric tests (the Friedman test and the Wilcoxon test) were applied if the data did not fit a normal distribution. Overall significant difference is shown in bold. Values in parentheses indicate the standard deviation. BrAC = breath alcohol concentration, VS = vehicle’s speed, SA = steering angle, SS = steering speed, SRR = steering reversal rate, PP = peak-to-peak value of the steering angle, NLE = number of lane exceedances per kilometer.

∆

p < (0.05/6) significant difference from 0.03%.

p < (0.05/6) significant difference from 0.00%.

Among the six measures, the VSs, SSs, SRRs, and PPs on all of the curves were significantly affected by alcohol. As stated previously, the VSs have been analyzed in our previous study (Zhang et al., 2014). Therefore, the SSs, SRRs, and PPs were further analyzed in this study. The SSs on all of the curves, the SRRs on the 800m radius curve, and the PPs on the 500 m and 800 m radius curves followed a gamma distribution. The SRRs on the 200 m and 500 m radius curves followed a normal distribution. The PPs on the 200 m radius curve did not belong to the family of exponential distributions. The PPs on the 200 m radius curve followed a gamma distribution while they still did not follow a normal distribution after seven outliers (the data outside the outer fences of the box plot were considered to be outliers) were removed. The data that followed a gamma distribution were analyzed using the generalized linear mixed model (GLMM) with an identity link function, and the data that followed a normal distribution were done using the linear mixed model (LMM). A diagonal structure was selected as the covariance structure for repeated measures. The covariance structure for random effects was specified as a variance components structure. The results in Tables 4 and 5 show that a linear dose–response relationship is found for the SS, SRR, and PP on all of the curves. Estimates of covariance parameters tell us whether the null hypothesis (a random effect is not needed) is accepted (IBM Corporation, 2013). The statistics indicate that random intercepts and random slopes are not significant, which suggests that excluding the random effects is reasonable to explain the effects of alcohol on steering behavior.

Table 4:

Tests of Fixed Effects

Radius	Measure	Source	df1	df2	F	P
200m	SS^a	BrAC	1	86	23.205	< .001
	SRR^b	BrAC	1	86	24.172	< .001
	PP^a	BrAC	1	79^c	25.313	< .001
500m	SS^a	BrAC	1	86	23.868	< .001
	SRR^b	BrAC	1	86	34.569	< .001
	PP^a	BrAC	1	86	22.015	< .001
800m	SS^a	BrAC	1	86	31.989	< .001
	SRR^a	BrAC	1	83^d	34.169	< .001
	PP^a	BrAC	1	86	14.527	< .001

Note. BrAC was treated as a scale variable. SS = steering speed, BrAC = breath alcohol concentration, SRR = steering reversal rate, PP = peak-to-peak value of the steering angle, GLMM = generalized linear mixed model, LMM = linear mixed model.

The measure was analyzed using GLMM.

The measure was analyzed using LMM.

Seven outliers were removed.

Three records (the value of zero) were automatically excluded.

Table 5:

Estimates of Fixed Effects and Estimates of Covariance Parameters

Radius	Measure	Parameter	Estimates of Fixed Effects					Estimates of Covariance Parameters
Radius	Measure	Parameter	Estimate	Std. Error	df	t	p	Estimate	Std. Error	Z	p
200m	SS^a	Intercept	0.197	0.030	86	6.593	<.001	0.010	0.007	1.281	.200
		BrAC	2.661	0.552	86	4.817	<.001	2.184	2.415	0.904	.366
	SRR^b	Intercept	0.324	0.032	86	10.109	<.001	0.011	0.006	1.776	.076
		BrAC	2.211	0.450	86	4.916	<.001	0.984	1.364	0.721	.471
	PP^a	Intercept	1.034	0.089	79^c	11.585	<.001	0.088	0.063	1.392	.164
		BrAC	11.899	2.365	79^c	5.031	<.001	32.496	43.904	0.740	.459
500m	SS^a	Intercept	0.100	0.010	86	10.194	<.001	0.001	0.001	1.407	.160
		BrAC	1.543	0.316	86	4.885	<.001	1.861	2.136	0.871	.384
	SRR^b	Intercept	0.196	0.027	86	7.316	<.001	0.008	0.005	1.681	.093
		BrAC	2.587	0.440	86	5.880	<.001	2.962	1.603	1.848	.065
	PP^a	Intercept	0.531	0.045	86	11.941	<.001	0.023	0.018	1.266	.206
		BrAC	5.023	1.071	86	4.692	<.001	20.121	61.879	0.325	.745
800m	SS^a	Intercept	0.060	0.005	86	12.078	<.001	0.000	0.000	1.245	.213
		BrAC	1.598	0.282	86	5.656	<.001	1.552	0.810	1.915	.055
	SRR^a	Intercept	0.079	0.011	83^d	7.250	<.001	0.000 (redundant)
		BrAC	2.824	0.483	83^d	5.845	<.001	3.313	1.890	1.753	0.080
	PP^a	Intercept	0.412	0.045	86	9.128	<.001	0.000 (redundant)
		BrAC	4.964	1.302	86	3.811	<.001	19.601	12.437	1.576	0.115

Note. SS = steering speed, BrAC = breath alcohol concentration, SRR = steering reversal rate, PP = peak-to-peak value of the steering angle, GLMM = generalized linear mixed model, LMM = linear mixed model.

The measure was analyzed using GLMM.

The measure was analyzed using LMM.

Seven outliers were removed.

Three records (the value of zero) were automatically excluded.

Similarly, “BrAC level” was excluded in the Friedman test when the effect of the turn direction on the steering behavior at each curve radius was examined. The results showed that the NLE (z = −2.658, p = .008) was significantly affected by the turning direction for the 200 m radius curve. The SA (z = −2.873, p = .044, SS (z = −2.906, p = .044, and SSR (F(1,21) = 7.005, p = .015) were significantly affected by the turning direction for the 500 m radius curve. The SS (z = −2.191, p = −0.28) and SSR (z = −2.451, p = .014) were significantly affected by the turning direction for the 800 m radius curve. Overall, the effects of the turning direction on the measures on all the curves were inconsistent.

Detection Accuracy

As mentioned in the introduction, a clearer understanding of how alcohol affects steering behavior could aid in the improvement of the accurate detection of alcohol-impaired-driving. To illustrate this, a comparative experiment of intoxication detection was conducted. Fisher discriminant analysis was used to identify the driver’s state (sober: 0%, intoxication: 0.09%). According to the number of features, four cases were tested. The first only regarded VS as the feature. The second only regarded SA as the feature. There were two features (VS and SA) in the third case. Five features (VS, SA, SS, SRR, and PP) were regarded as the features in the fourth case. The results showing the detection accuracies are shown in Table 6. We can see that the detection using five features achieved the highest accuracy (80.75%).

Table 6:

Detection Accuracy

Radius	Feature
Radius	VS	SA	VS, SA	VS, SA, SS, SRR, PP
200 m	62.50%	61.35%	62.50%	76.10%
500 m	69.80%	73.40%	73.20%	84.30%
800 m	70.55%	76.45%	73.95%	81.85%
Average	67.62%	70.40%	69.88%	80.75%

Note. VS = vehicle’s speed, SA = steering angle, SS = steering speed, SRR = steering reversal rate, PP = peak-to-peak value of the steering angle.

Discussion

The results suggest that a smaller radius results in a lower speed, a finding that corresponds to the results of other studies (Winsum & Godthelp, 1996). The vehicle’s speed has been shown to have a curvilinear relationship with the curve radius (Kanellaidis, Golias, & Efstathiadis, 1990). The smaller the radius is, the lower the speed. The driver compensates for the steering angle by reducing the vehicle’s speed. Meanwhile, as the BrAC level increases, the vehicle’s speed has a tendency to increase. Along this line, some research has reported that intoxicated drivers drive at higher speeds because they are impulsive and risky (Nagoshi et al., 1991). From Table 3, we can see that almost all the subjects were speeding, even if driving sober. One of the reasons for this could be that subjects drive faster in simulators than on the road. Keith et al. (2005) stated that people drive faster in simulators than in the field. Ding, Zhao, and Rong (2014) showed that the speed in the simulator was higher than the speed on the road, and the difference in these two speeds (simulator and on road) was consistent. Another reason for this speeding could be attributed to the alcohol. A surprising phenomenon is that the vehicle’s speed decreased at a BrAC of 0.09% on the 200 m radius curve. One possible reason is concerned with the increased crash at a BrAC of 0.09% on the 200 m radius curve. The subjects were asked to drive again once a crash occurred. Accordingly, the subjects deliberately reduced speed to avoid a crash.

The results indicated the characteristic effects of alcohol on steering behavior. From Tables 3 and 5, we can see that the SS, SRR, and PP had a tendency to increase as the BrAC increased on all of the curves. For the SS, the slope on the 200 m radius curve is nearly twice as high as that on the 500 m or 800 m radius curve. As the BrAC increased, the SRR tended to increase. Seehafer et al. (1968) found a significant increase in the steering reversals with moderate levels of alcohol (BAC 0.05%–0.07%). Huntley et al. (1972) reported that high levels of alcohol (BAC 0.085%–0.116%) increase the number of coarse steering reversals. In our study, high levels of alcohol (BrAC 0.09%) significantly increased the SRR on the 200 m and 500 m radius curves. The low (BrAC 0.03%), moderate (BrAC 0.06%), and high levels of alcohol (BrAC 0.09%) significantly increased the SRR on the 800 m radius curve. It should be noted that at the low level of alcohol (BrAC 0.03%), there are significant increases in the SS and SRR on the 800 m radius curve. This means that the low level of alcohol impaired the steering performance on the 800 m radius curve. However, a significant effect of alcohol on the steering performance does not appear with a low level of alcohol (BrAC 0.03%) on the 200 m and 500 m radius curves. This finding could be explained by the idea of compensatory behavior (Christoforou, Karlaftis, & Yannis, 2012; Vogel-Sprott, 1992; Vollrath & Fischer, 2017). In “safety” driving situations, drivers feel safe and relaxed. Accordingly, the ability to react adequately could be destroyed. In “risky” driving situations, drivers may fear that an accident might occur and are then able to react adequately. The PP increases with the BrAC on all of the curves except the 500 m radius curve with the BrAC from 0.06% to 0.09%. The slope of the PP on the 200 m radius curve is about twice as high as that on the 500 or 800 m radius curve. Large amplitude oscillations of the steering wheel may result in driving outside of the lane boundaries, which is illustrated by NLE. Although lane exceedance does not necessarily cause an accident, it represents the relative risk for an accident to some extent. Some studies (Knappe, Keinath, Bengler, & Meinecke, 2007; Östlund, Peters, et al., 2004; Wierwille et al., 1996) have shown that lane exceedance can be used as a risk estimation. With the exception of the NLE on the 200m radius curve with a BrAC of 0.03%, under the same level of BrAC, the smaller the radius was, the higher the NLE. Under the same radius, the higher the BrAC was, the higher the NLE. In general, sober drivers smoothly operate the steering wheel to negotiate a curve, while the alcohol-impaired drivers operate the steering wheel abruptly and quickly. These operations are reflected by the increased SS, SRR, and PP. These three indicators can be used to detect alcohol-impaired driving on different types of curves.

The measures used in this study are time domain measures. Wang, Bao, Du, Ye, and Sayer (2017) used fast Fourier transform (FFT) to extract the frequency domain measures and showed that the results based on the frequency measures were more robust and consistent than that based on the time measures. However, Wang et al. (2017) noted that FFT may not be a good method to interpret the dynamic changes (e.g., lane changes) because FFT generally assumes a steady-state process. Krajewski al el. (2009) used three feature sets of steering wheel in the time, frequency, and state space domain to detect sleepy drivers and showed that time domain feature achieved the best classification results. Daza et al. (2011) showed the results obtained in the frequency domain were not better than those in the time domain. It seems that the accuracy of time domain features is higher than that of frequency domain measures, whereas the consistency of time domain features is lower than that of frequency domain measures.

Conclusion

The results suggested the following conclusions. The first conclusion was that steering behavior in curve driving is impaired by alcohol. The SS, SRR, and PP on all of the curves were significantly affected by alcohol. The second conclusion was that the SS, SRR, and PP on all of the curves have a tendency to increase as the BrAC increases. The large amplitude oscillations of the steering wheel at a high BrAC, accompanied by the high speed, SS and SRR, resulted in a high NLE. The third is that a smaller curve radius resulted in a larger SS, SRR, and PP with the same level of alcohol. These conclusions contribute to the understanding of the detailed effects of alcohol on steering behavior. These findings, if appropriately validated and expanded, have potential utility for developing additional effective ways to detect alcohol-impaired-driving.

Our study has some limitations. For the alcohol condition, although the Chinese liquor was diluted with water, it was not tasteless. The simulator used in this study only has 3 degrees of freedom. It is worth mentioning that effect sizes of identical measures in a simulator and on the road at the same BrAC level may be different. Further studies are needed before the conclusions can be used in practical systems. The mean of the two BrAC levels measured at the beginning and at the end of the driving task were considered as the BrAC level during the simulated driving process. While the average BrAC level may have represented a decline in some subjects, it reflected an increase in others during the driving task. These limitations, to some extent, might affect the results of the correlation between the measures and the BrAC levels. In addition, high-dose (more than 0.09%) alcohol effects on steering behaviors were not examined in this study. The relatively small sample size is another limitation of this study. All the subjects were healthy young male drivers who were not representative of the general driving population. Only three types of curve radii were considered in the study, which may not be sufficient to explore all the detailed, characteristic effects of curve driving. Accordingly, future studies will employ more roadway geometries and recruit female drivers to increase sample sizes.

Key Points

Steering behavior in curve driving was impaired by the alcohol.

The SS, SRR, and PP on all of the curves were significantly affected by the alcohol.

The SS, SRR, and PP had a tendency to increase as the BrAC increased.

Footnotes

Zhenlong Li was born in Shanxi, China, in 1976, and he received a PhD in control theory and control engineering from Chinese Academy of Sciences, Beijing, China. He is currently an associate professor in the College of Metropolitan Transportation of Beijing University of Technology, and his primary research interests are centered on traffic control and driving behavior.

Xuewei Li was born in Hebei, China, in 1993. She is a postgraduate student at Beijing University of Technology. Her primary research interest is centered on driving behavior.

Xiaohua Zhao is a professor in the College of Metropolitan Transportation of Beijing University of Technology. She received her PhD from Beijing University of Technology, and her primary research interest is in driving behavior.

Qingzhou Zhang is a postgraduate student at Beijing University of Technology. His primary research interest is centered on driving behavior.

References

Allen

R. W.

Stein

A. C.

Hogue

J. R.

(1982). The effects of alcohol and marijuana on driver control behavior in a driving simulator: Phase I (Technical Report No. DOT HS 806 414). Washington, DC: National Highway Traffic Safety Administration.

Chen

Rakotonirainy

Loke

S. W.

Krishnaswamy

(2007). A crash risk assessment model for road curves. In Proceedings of the 20th International Technical Conference on the Enhanced Safety of Vehicles (pp. 1–8). Lyons, France.

Christoforou

Karlaftis

Yannis

(2012). Effects of alcohol on speeding and road positioning of young drivers. Transportation Research Record Journal of the Transportation Research Board, 24(2281), 32–42.

Compton

R. P.

(1991). Alcohol limits for drivers: A report on the effects of alcohol and expected institutional responses to new limits (Report No. DOT HS 807 692). Washington, DC: National Highway Traffic Safety Administration.

Daza

I. G.

Hernandez

Bergasa

L. M.

Parra

Yebes

J. J.

Gavilan

. . . Sotelo

M. A.

(2011, October). Drowsiness monitoring based on driver and driving data fusion. International IEEE Conference on Intelligent Transportation Systems (pp. 1199–1204). New York, NY. doi:10.1109/ITSC.2011.6082907.

Ding

Zhao

Rong

(2014). Experimental research on the validation of the driving simulator. Research report. Beijing: College of Metropolitan Transportation, Beijing University of Technology.

Ding

Zhao

Rong

(2013). Experimental research on the effectiveness of speed reduction markings based on driving simulation: A case study. Accident Analysis & Prevention, 60, 211–218.

Drew

G. C

Colquhoun

W. P

Long

H. A.

(1958). Effect of small doses of alcohol on a skill resembling driving. The British Medical Journal, 2(5103), 993–999.

Fillmore

M. T.

Blackburn

J. S.

Harrison

E. L.

(2008). Acute disinhibiting effects of alcohol as a factor in risky driving behavior. Drug & Alcohol Dependence, 95(1–2), 97–106.

10.

Glennon

J. C.

Neuman

T. R.

Leisch

J. E.

(1985). Safety and operational considerations for design of rural highway curves (FHWA/RD-86/035). Washington, DC: Federal Highway Administration.

11.

Guan

Zhao

Qin

Rong

(2014). An explanation of how the placement of traffic signs affects drivers’ deceleration on curves. Safety Science, 68(4), 243–249.

12.

Huntley

M. S.

(1970). Effects of alcohol and fixation-task demands upon human reaction time to achromatic targets in the horizontal median of the visual field. (Doctoral dissertation). Retrieved from https://library.uvm.edu/collections/theses

13.

Huntley

M. S.

Jr. Kirk

R. S.

Perrine

M. W.

(1972, January). Effects of alcohol upon nocturnal driving behavior, reaction time, and heart rate. Paper presented at SAE Automotive Engineering Congress. Detroit, MI.

14.

Huntley

M. S.

Perrine

M. W.

(1971). Influences of alcohol on driving behavior in an instrumented car. Institute for Road Safety Research SWOV. Noordwigkerhout, The Netherlands.

15.

IBM Corporation. (2013). IBM SPSS statistics 23 documentation. IBM Corporation.

16.

Kanellaidis

Golias

Efstathiadis

(1990). Drivers’ speed behaviour on rural road curves. Traffic Engineering and Control, 31(7), 414–415.

17.

Keall

M. D

Frith

W. J

Patterson

T. L.

(2000). The influence of alcohol, age and number of passengers on the night-time risk of driver fatal injury in New Zealand. Accident Analysis & Prevention, 36(1), 49–61.

18.

Keith

Trentacoste

Depue

Granda

Huckaby

Ibarguen

Kantowitz

Lum

Wilson

(2005). Roadway human factors and behavioural safety in Europe (Publication FHWA-PL-05-005). Washington, DC: U.S. Department of Transportation.

19.

Knappe

Keinath

Bengler

Meinecke

(2007). Driving simulator as an evaluation tool: Assessment of the influence of field of view and secondary tasks on lane keeping and steering performance. Gastrointestinal Endoscopy, 66(1), 162–167.

20.

Krajewski

Sommer

Trutschel

Edwards

Golz

(2009, June). Steering wheel behavior based estimation of fatigue. In Proceedings of the Fifth International Driving Symposium on Human Factors in Driver Assessment Training & Vehicle Design (pp. 118–120). Big Sky, MT.

21.

Xie

Nie

Zhang

(2012). The drink driving situation in China. Traffic Injury Prevention, 13(2), 101–108.

22.

Linnoila

Erwin

C. W.

Ramm

(1980). Effects of age and alcohol on psychomotor performance of men. Journal of Studies on Alcohol and Drugs, 41(5), 488–495.

23.

Liu

Y. C.

C. H.

(2010). Effects of different blood alcohol concentrations and post-alcohol impairment on driving behavior and task performance. Traffic Injury Prevention, 11(4), 334–341.

24.

Martin

G. L.

(1971). The-effects of small doses of alcohol on a simulated driving task. Journal of Safety Research, 3, 21–27.

25.

Mayhew

D. R.

Donelson

A. C.

Beirness

D. J.

Simpson

H. M.

(1986). Youth, alcohol, and relative risk of crash involvement. Accident Analysis & Prevention, 18(4), 273–287.

26.

Mclean

J. R.

Hoffmann

E. R.

(1975). Steering reversals as a measure of driver performance and steering task difficulty. Human Factors: The Journal of the Human Factors and Ergonomics Society, 17(3), 248–256.

27.

Moskowitz

Burns

Fiorentino

Smiley

Zador

(2000). Driver characteristics and impairment at various BACs. (DOT HS 809 075). Washington, DC: US Department of Transportation National Highway Traffic Safety Administration.

28.

Moskowitz

Fiorentino

(2000). A review of the literature on the effects of low doses of alcohol on driving-related skills. (DOT HS-809 028). Washington, DC: National Highway Traffic Safety Administration.

29.

Moskowitz

Robinson

C. D.

(1988). Effects of low doses of alcohol on driving-related skills: A review of the evidence. (DOT HS 807 280). Washington, DC: National Highway Traffic Safety Administration.

30.

Nagoshi

C. T.

Wilson

J. R.

Rodriguez

L. A.

(1991). Impulsivity, sensation seeking, and behavioral and emotional responses to alcohol. Alcoholism: Clinical and Experimental Research, 15(4), 661–667.

31.

Nakayama

Futami

Nakamura

Boer

E. R.

(1999, March). Development of a steering entropy method for evaluating driver workload. In Proceedings of the SAE International Congress and Exposition (pp. 1–10), Detroit, MI.

32.

National Highway Traffic Safety Administration. (2009). Fatality Analysis Reporting System (FARS). [Online]. Retrieved from https://www-fars.nhtsa.dot.gov/

33.

National Highway Traffic Safety Administration. (2013). Traffic safety facts 2012 data: Alcohol-impaired driving. (DOT HS 811 870). Washington, DC: U.S. Department of Transportation, National Highway Traffic Safety Administration.

34.

Östlund

Nilsson

Carsten

Merat

Jamson

. . . Brouwer

(2004). HASTE Deliverable 2: HMI and safety-related driver performance. (Project GRD1/2000/25361 S12.319626). Human Machine Interface and the Safety of Traffic in Europe.

35.

Östlund

Peters

Thorslund

Engström

Markkula

Keinath

Horst

Juch

Mattes

Foehl

(2004). Driving performance assessment – methods and metrics. Deliverable 2.2.5 of the AIDE project, IST-1-507674-IP, SP2. Retrieved from http://www.aide-eu.org/pdf/sp2_deliv_new/aide_d2_2_5.pdf

36.

Peters

R. D.

Kloeppel

Alicandri

(1999). Effects of partial and total sleep deprivation on driving performance. (Publication No. FHWA-RD-90.09-00.096). Washington, DC: U.S. Department of Transportation, Federal Highway Administration.

37.

Rakauskas

M. E.

Ward

N. J.

Boer

E. R.

Bernat

E. M.

Cadwallader

Patrick

C. J.

(2008). Combined effects of alcohol and distraction on driving performance. Accident Analysis & Prevention, 40, 1742–1749.

38.

Seehafer

R. W.

Huffman

W. K.

Kinzie

M. D.

(1968). Effects of low level blood alcohol concentration on psychophysiological and personality measures under controlled driving conditions. National Safety Congress Transactions, 23, 100–107.

39.

Sinelnikova

Lee

Reimer

Mehler

Coughlin

J. F.

(2015). Predicting secondary task involvement and differences in task modality using field highway driving data. In Proceedings of the Eighth International Driving Symposium on Human Factors in Driver Assessment, Training and Vehicle Design (pp. 387–393). Salt Lake City, UT.

40.

Snyder

M. B.

(1992, October). Driving under the influence: A report to Congress on alcohol limits. (DOT HS 807 879). Washington, DC: National Highway Traffic Safety Administration.

41.

SRA Technologies. (1989). Improving understanding of alcohol impairment and BAC levels and their relationship to highway accidents. (DOT HS 807 433). Washington, DC: National Highway Traffic Safety Administration.

42.

Van Dyke

N. A.

Fillmore

M. T

. (2015). Distraction produces over-additive increases in the degree to which alcohol impairs driving performance. Psychopharmacology, 232, 4277–4284.

43.

Verster

J. C.

Wester

A. E.

Goorden

Van Wieringen

J. P.

Olivier

Volkerts

E. R.

(2009). Novice drivers’ performance after different alcohol dosages and placebo in the divided-attention steering simulator (DASS). Psychopharmacology, 204, 127–133.

44.

Verwey

W. B.

(2000). On-line driver workload estimation. Effects of road situation and age on secondary task measures. Ergonomics, 43(2), 187–209.

45.

Vogel-Sprott

, 1992. Alcohol tolerance and social drinking. learning the consequences. New York, NY: Guilford Press.

46.

Vollrath

Fischer

(2017). When does alcohol hurt? A driving simulator study. Accident analysis and prevention, 109, 89–98.

47.

Wang

Bao

Sayer

J. R.

(2017). A spectral power analysis of driving behavior changes during the transition from nondistraction to distraction. Traffic Injury Prevention, 18(8), 826–831.

48.

Watson

P. E.

(1989). Total body water and blood alcohol levels: Updating the fundamentals. Human Metabolism of Alcohol, 1, 41–58.

49.

Wierwille

Tijerina

Kiger

Rockwell

Lauber

Bittner

Jr. (1996). Heavy vehicle driver workload assessment. Task 4: Review of workload and related research. (DOT HS 808 467). Washington, DC: U.S. Department of Transportation.

50.

Winsum

W.V.

Godthelp

(1996). Speed choice and steering behavior in curve driving. Human Factors, 38 (3), 434–441.

51.

Woodall

W. G.

Kunitz

S. J.

Zhao

Wheeler

D. R.

Westerberg

Davis

(2004). The prevention paradox, traffic safety, and driving-while-intoxicated treatment. American Journal of Preventive Medicine, 27(2), 106–111.

52.

Zhao

Chen

Rong

(2016). Modeling the influence of chevron alignment sign on young male driver performance: A driving simulator study. Accident Analysis & Prevention, 95, 479–486.

53.

Zador

P. L.

(1991). Alcohol-related relative risk of fatal driver injuries in relation to driver age and sex. Journal of Studies on Alcohol and Drugs, 52(4), 302–310.

54.

Zhao

Han

Zhang

Rong

(2015). A generic approach for examining the effectiveness of traffic control devices in school zones. Accident Analysis & Prevention, 82, 134–142.

55.

Zhao

X. H.

Zhang

X. J.

Rong

(2014). Study of the effects of alcohol on drivers and driving performance on straight road. Mathematical Problems in Engineering. doi:10.1155/2014/607652.

56.

Zhang

Zhao

Rong

(2014). Effect of different breath alcohol concentrations on driving performance in horizontal curves. Accident Analysis & Prevention, 72, 401–410.