Use of a virtual reality driving simulator as an alcohol abuse prevention approach with college students.

Dear Editor:

A study by Hingson, Heeren, Zakocs, Kopstein and Wechsler (2002) reported alarming estimates of the number of U.S. college students in the age range 18-24 in 1998 who experienced alcohol-related problems: over 1400 students died (including motor vehicle accidents), over 2 million of 8 million college students drove under the influence, and over 3 million of 8 million college students rode with a drunk driver. A variety of approaches have been employed to try to reduce the risks associated with the consumption of alcohol among young people, including promising new approaches that include some type of experiential learning. An example of such an approach is the "Road Ready Teens" videogame developed by several groups (e.g., Mothers Against Drunk Driving; MADD, 2003) to increase teenagers' awareness and understanding of driving risks, including driving while drunk. An evaluation of this game by The University of Michigan Transportation Institute showed that playing the game increased a group of twenty-five teenagers' perceptions of personal driving risk (Bingham & Shope, 2003).

Researchers at the University of Missouri-Rolla (UMR) developed a personal computer (PC) based virtual reality (VR) driving simulator that included a simulation of the effects of driving while drunk. The "Road Ready Teens" video game also included a drunk driving component and participants in the evaluation study conducted by the University of Michigan Transportation Institute reported that the game was a more engaging mechanism in comparison to written materials for communicating with teenagers. However, some of the young people expressed the criticism that they thought video games were "for fun" and not "for education" (Bingham & Shope, 2003).

In contrast, the UMR driving stimulator was clearly designed and presented to participants as an educational tool and not as a game experience.

Our study attempted to evaluate the VR Driving Simulator in order to: (1) establish the internal validity of the drunk driving simulation by comparing participants' performances under a "sober" condition to a "driving drunk" condition; and (2) assess the external validity of the simulator by asking subjects about their future likelihood of engaging in drinking and driving and their general expectancies regarding the use of alcohol.

Participants in our study were 86 undergraduate students enrolled in Introductory Psychology classes at the University of Missouri-Rolla. They completed the study as part of their course requirements. Thirty-one students participated in the driving simulator condition and 55 students in a control group condition. Consistent with the campus composition, there were 51 males and 35 females and the average age was 20.9 years. On a background information sheet, students were asked the number of times in the past two weeks that they had four or more drinks (if female) or five drinks (if male) at a single setting. For purposes of analysis in our study, this provided an index of "binge drinking."

The virtual reality driving simulator was restricted to a single desktop computer and screen to make it more portable and affordable in future research and educational endeavors. The plug and play capability of the hardware in this setup was also an important consideration in its creation. The virtual simulator consisted of two parts, the static world (scene data) and the moving entities. The scenario development was carried out using an event based triggering method. In the simulation, the image responded to input from the steering wheel, accelerator, and brake. There was also appropriate accompanying visual and sound feedback. The synchronization was provided by a vehicle dynamics model, the core component of the driving simulation system. The vehicle model implemented in this system consisted of different mathematical models for various parts of the vehicle such as steering, engine, driveline, differential, aerodynamics, and tires. A scoring system constantly monitored the driver's performance during the experiment.

To equate the drivers in their skill levels before beginning the experiment, participants in the VR simulator condition completed a driver training task. In the training task, participants completed a driving course that familiarized them with the apparatus and control devices. They had to complete the training course without losing more than a total of four points (using a scoring system to be described later) before they could start the actual driving tests. In the actual experiment, there were two driving tests: a "sober driving" task and a "drunk-driving" task where the controls were specifically modified to simulate the effects of driving while drunk. To control for order effects, the order of the drunk versus sober driving tests was decided randomly by a coin flip.

Moskowitz et al. (2000) also showed that longer reaction times are associated with consuming alcohol and are related to delays in the decision to act and uncertainty about what action to take. Based on this concept, it was possible to create a similar situation by implementing reaction time lags in the control devices of the simulator. This was achieved by creating a time lag between the moment of making a response using the hardware of the simulator and the corresponding change in the response of the simulator. Moskowitz et al. (2000) showed that the average increase in reaction time for individuals consuming alcohol in the age group 21-24 is 0.5 seconds. Thus a delay of 0.5 seconds was added to the controls of the simulator including steering, brake pedal, and accelerator pedal under the "drunk driving" condition.

A standardized set of instructions was read to all subjects. After completing the driving tests, subjects were shown bar graphs of their driving mistakes and their performance on the auditory shadowing task under the sober versus drunk conditions. Students in the no treatment control group filled out the same questionnaires that all other participants completed but did not experience any type of educational program.

In order to make the driving task as realistic as possible, different scenarios were designed to include day-to-day situations. To maintain similarity between the sober and drunk driving tests, the two testing tracks were essentially the same except that the time and location of events in the two tracks were different. The two testing scenarios were designed such that the driver could complete either one in about ten minutes.

A participant was told to obey a maximum speed limit of 35 miles per hour throughout the driving test. The student was able to view the surroundings in the same way that someone would be able to do if they were sitting in the driver's seat in a real car (e.g., they could see the car dashboard, speedometer, and what was behind them in a rear view mirror). The participant experienced a series of events (e.g., traffic lights, people crossing the street, cars passing by, etc.) while driving--events which were designed specifically as part of the testing scenario. The events involved a triggering method based on the position of the driver's vehicle in the virtual world. The participant was guided through the environment with visual cues. A red arrow appeared in the lower left or right side of the screen indicating which direction he/she should turn at the next intersection; and the arrow would go off when the turn was completed.

While completing these driving tasks, participants were also required to simultaneously perform an auditory shadowing task. Research (Moskowitz, Burns, Fiorentino, Smiley, & Zador, 2000) has shown that a major negative effect of alcohol on driving skills is a reduction in the driver's ability to divide his/her attention (e.g., converse on a cell phone while paying attention to driving cues). The auditory shadowing task was employed in our study because it involved the same type of response and has similar time and attentional demands as talking on a cell phone. It was run on a different computer system from the simulator.

The application consisted of two audio threads played in the two ears of a set of headphones that drivers wore while completing the simulator test drives. The pre-recorded audio threads were made of different words that were spoken at random time intervals. The participant's task was to respond to the words that were spoken only in his/her right ear by repeating them aloud. Participants were told to ignore all words presented in the left ear. "White" noise was played in both ears continuously in order to avoid any interfering noise.

The performance measurements for the driving task involved keeping track of driving mistakes through the use of a continuously running performance measurement module in the background while a participant was performing one of the driving tests. The mistakes a driver could make included deviations in driving speed, number and type of collisions, failure to follow speed limits, failure to respond to traffic lights and traffic signs, and the driver's failure to keep their vehicle under control. At the end of their test drives, participants were shown two graphs. These graphs showed the total points lost in the two tests and the total time taken to complete the two tests.

The performance measurements for the auditory shadowing task also involved use of a performance measurement module that was continuously running in the background while a subject was performing one of the driving tests. The measurements taken included the response time for each word and the number of missed or wrongly responded to words. The performance measurement results on each of these tasks were also shown to the driver in terms of bar graphs.

As an initial step toward exploring the external validity of the VR simulator, two questionnaires were passed out to all participants, the Cognitive Appraisal of Risky Activities questionnaire (CARE; Fromme, Katz, & Rivet, 1997) and The Alcohol Expectancy Multi-Axial Assessment (AEMax; Goldman & Darkes, 2004). The CARE measures respondents' beliefs about the consequences of 30 risky activities with four subscales, i.e., possible negative consequences associated with each risky activity, possible positive consequences, expected involvement in these activities in the next six weeks, and actual involvement. However, in the current study, the 30 item subscale related to "actual involvement" was not used. Although all 30 items from each of three sub-scales of the CARE were given to subjects, in the present investigation only the five items having to do with alcohol were used later in the analysis of results. Research by Fromme and colleagues has shown good test-retest reliability for the CARE with values of Pearson's r that ranged from .51 to .79 and good criterion validity in terms of scores being significantly related to actual risk taking behaviors.

The AEMax is a recently developed comprehensive questionnaire that assesses subjects' global positive and negative expectancies associated with the use of alcohol (Goldman & Darkes, 2004). We used the 24-item short version of the AEMax (eight octants measured with three items per octant). Research indicates that the AEMax is predictive of concurrent and future drinking behavior among college students and that it also significantly correlates with other measures for expectancy of alcohol effects.

The results of the study showed that the VR simulator had very good internal validity: all of the comparisons with regard to the differences between how participants performed under the "driving sober" versus "driving drunk" conditions were significant. Students showed clear decrements under the drunk driving condition with regard to both their driving performance and their ability to perform the auditory shadowing task. Moreover, there were no order effects. A comparison of scores obtained by participants who experienced the "sober-driving" condition first and those who underwent the "drunk-driving" condition first showed no significant differences in their scores on any of the measures used in this study.

Drivers made a significantly [F(1, 29) = 325.48, p < .0001] greater number of driving errors in the "drunk" than in the "sober" condition. The number of errors in the drunk condition (mean = 93.81 points) was, in fact, more than three times greater than that of the sober condition (mean = 27.19 points). Additionally, it took drivers significantly [F(1, 29) = 72.48, p < .0001] longer to complete the driving simulation task under the "drunk" condition (mean = 621.11 seconds) than under the "sober" condition (mean = 511.30 seconds).

A similarly impressive set of differences was found with regard to how subjects performed on the auditory shadowing task. The average response time was significantly greater [F(1, 29) = 6.57, p < .02] for the drunk condition (mean = 1.83 seconds) than for the sober condition (mean = 1.65 seconds) and the number of omitted words was also significantly [F(1, 29) = 24.13,p < .001] greater when participants were "drunk" (mean = 3.8 words) than "sober" (mean = 1.9 words).

The CARE and AEMax questionnaires were used to address the issue of the external validity of the effects on people of the VR driving simulator. To remove individual differences regarding the drinking habits of subjects prior to the start of the study, frequency of "binge" drinking, i.e., the number of times in the past two weeks a subject had four or more drinks (if female) or five (if male) at a single sitting, was used as a covariate. Similarly, age of respondent was also used as a covariate.

CARE scores were analyzed by means of a 2 (treatment versus control) X 2 (male versus female) multivariate analyses of variance (MANOVA) using binge drinking and age as covariates. Only binge drinking was significant [[LAMBDA], F(15, 66) = 5.73, p < .000]. None of the other variables, i.e., treatment versus control [[LAMBDA], F(15, 66) = 1.34, p < .20], male versus female [[LAMBDA], F(15, 66) = 1.34, p < .20], treatment/control X male/female [[LAMBDA], F(15, 66) = .75, p < .72] interaction, or age [A, F(15, 66) = 1.37, p < .19] were significant. This indicated that participation in the virtual reality experiment had little effect on changing expectancies or attitudes regarding drinking alcohol.

AEMax scores were also analyzed by means of a 2 (treatment versus control) X 2 (male versus female) multivariate analysis of variance (MANOVA) with age and binge drinking serving as covariates. While the binge drinking variable was significant [[LAMBDA], F(8, 73) = 4.63, p < .000], none of the others were, i.e., treatment versus control [[LAMBDA], F(8,73) = 1.34, p < .20], male versus female [[LAMBDA], F(8,73) = .69, p < .70], treatment/control X male/female [[LAMBDA], F(8, 73) = .56, p < .81], or age [[LAMBDA], F(8, 73) = 1.63, p < .13]. These results again indicated that participation in the virtual reality simulation, age of participant, and sex of participant had little impact on expectancies toward alcohol.

Results of this study with regard to the internal validity of the VR Driving Simulator were impressive: participants clearly committed more driving errors while under the "drunk driving" condition and also showed poorer performance on the divided attention task. As a concept, the VR simulator has exciting possibilities as an experiential education tool. It clearly avoids the complex legal and ethical problems involved in actually giving alcohol to humans, and especially to young people.

With regard to our initial attempt to assess the external validity of the VR driving simulator, the results were disappointing as no significant differences were found with regard to the likelihood of students engaging in future drinking and driving behavior, or for any changes in their expectancies regarding the use of alcohol. This failure to find significant results for this aspect of the study is similar to the results reported by The University of Michigan Transportation Institute in their evaluation of the "Road Ready Teens" videogame (Bingham & Shope, 2003). The videogame, which also included a drunk driving experience, similarly did not impact subjects' intentions to avoid future risky driving behavior. Moreover, the participants in this study commented on the fact that the videogame lacked realistic feedback about the consequences of making "real life" driving mistakes, e.g., the degree of injury, possible repair costs, and damage from hitting obstacles.

Recent research in JADE evaluating another experiential approach to alcohol education by Jewell, Hupp, and Luttress (2004) has also raised this issue, i.e., the possible need for making explicit connections between a subject's experiences and the real life consequences of alcohol abuse (connections which were also lacking in our VR experiential approach). Jewell and colleagues examined the effectiveness of the "Fatal Vision Goggles" as a prevention tool for changing attitudes toward drinking. The goggles are an eye glass device that simulates the effects of alcohol by distorting the wearer's perception with respect to vision and sense of balance (Innocorp, LTD., 1997). In an experimental study comparing a control group, the group wearing the goggles, and a group of onlookers, Jewell and colleagues reported that participants in the goggles condition showing a significantly greater change in attitudes. However, before any post-treatment measures were taken, all participants in this study viewed a videotape "that is typically used during drinking and driving prevention programs, which recounts the story of a parent whose son died in a collision caused by a drunk driver (p. 71)." It may be that with experiential approaches, the connection between a participant's experiences of disrupted functioning and the possible negative consequences of drinking and driving need to be explicitly made to the participants. Future research is needed to explore this possibility.

Overall, the present study's results were positive in showing that the virtual reality simulator could be a safe, ethical and effective approach to educating young people about the dangers of drinking and driving. More work is needed, however, to demonstrate the impact of the VR simulator on expectancies and future behavior and the need to include more connections with realistic consequences associated with drunk driving.

References

Bingham, C. R. & Shope, J. T. (2003). An Evaluation of the Road Ready Teens Video Game: Final Report. University of Michigan Transportation Research Institute, Ann Arbor, MI (UMTRI-2003-28).

Fromm, K., Katz, E. C., & Rivet, K. (1997). Outcome expectancies and risk-taking behavior. Cognitive Research and Therapy, 21,421-442.

Goldman, M. S., & Darkes, J. (2004) Alcohol Expectancy Multi-Axial Assessment: A memory network-based approach. Psychological Assessment, 16, 4-15.

Hingson, R. W., Heeren, T., Zakocs, R. C., Kopstein, A., & Wechsler, H. (2002). Magnitude of alcohol-related mortality and morbidity among U.S. college students ages 18-24. Journal of Studies on Alcohol, 3, 136-144.

Innocorp, LTD. (1997). The Straight Line, 1.

Jewell, J., Hupp, S., & Luttress, G. (2004). The effectiveness of fatal vision goggles: Disentangling experiential versus onlooker effects. Journal of Alcohol and Drug Education, 48, 3, p. 63-84.

MADD, Teen safety program launches video game that increases awareness of driving risks. Retreived October 17, 2003 from http://www.madd.org/news/0,1056,7243.00.html

Moskowitz, H., Burns, M. Fiorentino, D., Smiley, A., & Zador, P. (2000). Driver Characteristics and Impairment at Various BACs, Technical Report, U.S. Department of Transportation, contract #DTNH-22-95-C-05000.

Frances Haemmerlie Montgomery, Ming C. Leu, Robert L. Montgomery, Michael D. Nelson, and Mannish Sirdeshmukh, University of Missouri-Rolla

Correspondence concerning this article should be addressed to Frances Haemmerlie Montgomery, Department of Psychology, 114 Humanities/Social Sciences Building, 1870 Miner Circle, University of Missouri-Rolla, Rolla, Missouri 65409. Phone: [phone]. E-mail: [e-mail].