Driver Mental Workload

L. R. Zeitlin

Transportation Research Board, 1998 Annual Meeting

1

A MICROMODEL FOR THE OBJECTIVE ESTIMATION OF DRIVER MENTAL WORKLOAD FROM TASK DATA LAWRENCE R. ZEITLIN,1 The City University of New York Observation of driver performance during a 36,000 mile field study of van pooling indicates that a good estimate of mental workload can be made from an analysis of objective performance data alone. Drivers traversed a mix of rural secondary roads, limited access expressways, high density limited access urban drives, and downtown city streets on a daily commute from upstate New York to New York City. Data included roadway characteristics, time, traffic density, speed, weather, brake applications, subsidiary task performance, and subjective difficulty ratings. Driving workload had two components, a steady state load dictated by roadway conditions, speed, and traffic density and a transient load determined by the degree of uncertainty in the driving situation. Brake actuations represent the uncertainty inherent in the driving situation while the log2 of the speed is a first approximation of the steady state information processing load imposed by tracking requirements of vehicle control. Unpredictability of traffic appeared to be the major determinant of perceived difficulty. Workload homeostasis occurred as drivers modified their performance to keep workload within a comfortable range. An objective workload index, Workload = Constant * (Brake actuation rate + log2 kph) based on this micromodel of driver behavior predicts subjective driving difficulty. An ANOVA shows that the workload index distinguished between roadway types at the p < .0001 level of significance (F = 58.101). The workload index correlates at r = .741, 18 df, with the subjective driving difficulty ratings and at r = .808, 18 df, with the mental workload estimates of the best subsidiary task. 1 Requests

for reprints should be sent to Lawrence R. Zeitlin, CUNY, 12 Brook Lane, Peekskill, NY

10566 Running title: OBJECTIVE ESTIMATE OF MENTAL WORKLOAD Key words: driver behavior, mental workload, subsidiary task, objective estimates

L. R. Zeitlin

Transportation Research Board, 1998 Annual Meeting

2

INTRODUCTION A significant problem in the study of automobile driving, as well as other perceptual-motor tasks, is the determination of the mental workload imposed upon the operator. The concept of mental workload is based on the conjoint assumptions that the act of responding to stimuli takes a finite amount of effort and that the total amount of effort an individual can expend is limited. For most vehicle control tasks, moderate increases in task difficulty may produce few observable changes in error rate as the operator attempts to hold performance constant by allocating more resources to the task. When all available resources are committed, any further increase in difficulty results in a sharply increased error rate. The workload that can be handled before performance begins to deteriorate defines maximum capacity for a particular individual. Because of the slow increase in error below the break point, the fraction of total capacity allocated to the task is difficult to determine from an examination of error rate alone. The allowable mental workload sets the limit for the information processing and vehicle control tasks that can be imposed on an operator, thus an estimate of workload facilitates good system design. Four approaches have been used for determination mental workload in the sub critical zone. The first is to simply ask the operator how hard a particular task is. This introspective method is the basis for such rating scales as the SWAT or Subjective Workload Assessment Technique (Reid, Shingledecker, and Eggemeier, 1981; Reid and Nygren, 1988), the TLX or Task Load Index (Hart and Staveland, 1988) and the Overall Workload (OW) scale (Vidulich and Tsang, 1987). At the conclusion of a task, the operator is asked to rate task difficulty on one or more dimensions with the results often combined into a single workload estimate. While very useful in controlled situations, rating scales are usually administered ex post facto, permitting time and operator self image considerations to moderate the results. The latter phenomenon, loosely dubbed the Chuck Yeager effect, tends to constrict ratings since few experienced vehicle operators, particularly young males, are willing to admit that a task is near their maximum level of tolerance (Zeitlin, 1995). The second approach to workload estimation provides an estimate of primary task workload and unallocated or spare capacity by measuring the performance degradation on a subsidiary task designed to absorb unallocated resources relevant to the primary task.

Subsidiary tasks have been used by

researchers in the estimation of spare capacity and primary task workload for four decades (Bahrick, Noble, and Fitts, 1954; Brown and Poulton, 1961; Brown, 1962; Brown, 1964; Mackworth, 1959; Zeitlin and Finkelman, 1969). The nature of the resource pool to be allocated between primary and subsidiary tasks is still unclear. Most research prior to the mid-1970s hypothesized either a single input-output channel of limited capacity (Broadbent, 1971) or an undifferentiated mental resource to be allocated according to subject

L. R. Zeitlin

Transportation Research Board, 1998 Annual Meeting

3

set and task demand (Kahneman, 1973; Moray, 1967; Sanders, 1979). Subsequent research proposes multiple input-output channels to a single executive process (Badderly, 1986), a hybrid model (Kantowitz and Knight, 1976), or multiple largely independent resources (Wickens, 1984). So many subsidiary tasks have been utilized, in such specific contexts, that the worker who desires guidance in the selection of a subsidiary task is discomfited by the, often conflicting, bulk of the literature (Johannsen, Moray, Pew, Rasmussen, Sanders and Wickens, 1979; Moray, 1982; Wierwille, Rahimi, and Casali, 1985; Wickens, 1992). Multiple resource conceptualizations require a clear identification of the resources utilized by both primary and subsidiary tasks and an apriori specification of their interaction. Such information is largely lacking for many of the real world activities for which estimates of mental workload would be useful. In these activities, resource demand may vary both in quantity and kind from moment to moment as a function of the dynamics of the situation. Thus automobile driving may shift from a low frequency tracking task to a probabilistic decision making task to a disjunctive reaction time task within a few seconds. This adds to the burden of the researcher who wishes to estimate mental workload in an applied setting. The third approach to mental workload estimation assumes that information processing involves central nervous system activity and that manifestations of this activity produce physiological consequences as task loading nears maximum capacity. A number of physiological measures have been used to infer workload, including variability in cardiovascular response (Sokolov, Podachin, and Belova, 1983), pupillary response (Beatty, 1982), event related brain potentials (Kramer, Wickens, and Donchin, 1983), and the traditional lie detector measures of GSR and respiration rate. Physiological measures permit continuous data collection during task performance and estimation of overall task loading, but generally are intrusive, requiring expensive equipment and an unusual degree of cooperation on the part of the vehicle operator. The final approach is measurement of driver performance on the primary task, inferring workload from directly observable operator responses and control actions. The obvious difficulty is that similar tasks, both demanding resources below the maximum capability limit, may present almost identical responses and actions, yet one may be more difficult than the other. Further, one operators experience and cognitive appraisal of a situation may differ substantially from that of another operator, with consequent differences in the perceived task difficulty, yet the control actions may be identical. Because of these problems, Derrick (1988) contends that workload estimation by this method is difficult to achieve. Still, the economical, non intrusive, and real time workload estimation advantages of the task performance method are significant enough that in the isolated cases where it can be used it offers a

L. R. Zeitlin

Transportation Research Board, 1998 Annual Meeting

valuable alternative to one of the more indirect methods.

4

Such a situation presented itself

during the course of a 36000 mile, four year long study on commuter van pools. The data, originally gathered to explore the efficacy of the subsidiary task method of workload estimation, permitted the generation of a micromodel of driver performance, and the development of an index of mental workload based on directly observable task performance measures. Commuting to New York City from the surrounding suburbs is uncertain at best. In 1987, responding to anticipation of a transit strike, several large New York corporations subsidized employee car and van pools. One of the author's graduate students, in his capacity as Director of Transportation Services for a large insurance company, made it possible for the author to conduct a long term field evaluation of two subsidiary tasks previously explored in simulation experiments (Zeitlin and Finkelman, 1975).

L. R. Zeitlin

Transportation Research Board, 1998 Annual Meeting

5

FIELD TRIAL Method Overview. Drivers in two commuter van pools performed two subsidiary tasks, delayed digit recall and random digit generation, while travelling a mix of rural secondary roads, limited access expressways, high density urban drives, and downtown city streets on a daily commute from upstate New York to New York City. Roadway, traffic, and weather data were recorded along with subsidiary task performance. The field trial lasted from December, 1987 through August, 1992. Four different vehicles were used during the trial, a 1986 Dodge Caravan, a 1987 Plymouth Voyager, a 1988 Dodge Caravan SE, and a 1991 Chevrolet Astrovan LT. Subjects. The van pools consisted of six adult male members at a time. Driving responsibility rotated on a daily basis. All members lived in Northern Westchester county and worked in midManhattan. The drivers each had at least ten years of driving experience and were fully familiar with the route. Over the four year duration of the study, the membership in each pool changed as the original members moved, changed jobs, changed working hours, or went on vacations or sabbaticals. By August, 1992, twenty drivers had completed 9 full trip trials of each task over the four roadway types, a total of 2,880 two minute task samples. Subsidiary tasks. Two previously validated subsidiary tasks were used (Zeitlin and Finkelman, 1975). The first, delayed digit recall, presented the driver a sequence of random digits from 1 to 9 at two second intervals for a two minute period. The driver was required to say the digit before the last one presented during the inter-digit interval. Errors were scored for digits missed or omitted. This task, first used by Mackworth (1959), has the advantage that difficulty can be easily adjusted without modifying the stimulus set by requiring the subject to respond with the digit two or even three places back. The second task, random digit generation, required the driver to call out digits from 1 to 9 in random order at a self paced rate for a two minute period. In general, randomness decreases as workload increases, the subjects appearing to respond in a stereotyped or automatic manner to lessen resource demand (Tune, 1964; Zeitlin and Finkelman, 1969). Subsidiary task performance was evaluated by determining the "randomness" of the digit sequence using the evaluation criteria suggested by Wagenaar (1972). Subjective task difficulty ratings. Sheridan (1980) suggests that operator ratings are the most direct indicators of workload. Global overall workload ratings are the least intrusive of all techniques because they can be administered during performance of a task without adversely influencing the operator. A modified Overall Workload (OW) scale (Vidulich and Tsang, 1987) requiring an assessment of driving difficulty on a 10 point scale during each two minute test period was adopted for the experiment.

L. R. Zeitlin

Transportation Research Board, 1998 Annual Meeting

6

Multidimensional rating scales such as the SWAT and the TLX were not used because of the impracticability of obtaining the required multiple subscale ratings with relatively untrained personnel during the commute. A recent study by Hill, Iavecchia, Byers, Bittner, Zaklad and Christ (1992) confirms that the OW scale is as sensitive as the multidimensional scales, has higher user acceptance, is easier to complete and requires substantially less time for training, preparation, and data reduction. Roadway characteristics. Both commutes started at 7 a.m. in Northern Westchester county, one in Yorktown Heights, the other in Peekskill. Members were picked up by 7:30 and the designated driver took his position. Each van travelled rural secondary roads before entering the limited access road system. Posted speeds on the rural roads ranged from 48 to 72 kph (30 to 45 mph). Traffic density was moderate, 2 to 4 cars visible. During the warm months, foliage screened driveways. Depending on starting location, the average one way distance on secondary roads was 24 to 28 kilometers (15 to 18 miles). After leaving the secondary roads, the vans entered the NY Thruway, a high speed, limited access expressway. Posted speed was 89 kph (55 mph) but traffic exceeded the limit by 8 to 16 kph (5 to 10 mph). During the commuting hours, traffic density was moderate, 4 to 8 cars visible but all moving in the same direction. Thruway distance was approximately 26 kilometers (16 miles). At the city limits, the Thruway changed into a high density, limited access urban drive. The vans crossed the East River to the FDR Drive. Posted speeds on both roads are 80 kph (50 mph) but actual speed ranged from 100 kph (62 mph) to a bumper to bumper crawl depending on traffic conditions. Observed traffic density was extremely high during commuting hours, twenty or more cars visible. Limited access urban drive distance was 23 kilometers (14 miles). Downtown city streets comprised the final lap. Bidirectional traffic, cross traffic, pedestrian traffic, buses, taxicabs, and traffic lights required constant alertness and mandated low speeds and repeated brake applications. Depending on destination, city street distance was 5 to 8 kilometers (3 to 5 miles). The rural road, expressway, urban drive, city street sequence was repeated in reverse order every evening. Design limitations. Roadway type is the only variable that can be considered independent in the classical experimental sense. Sequence of presentation, time of day, and traffic conditions, including speed and density were largely determined by the invariant nature of the commute. A rough form of ABBA counterbalancing was provided by the reverse presentation of conditions on the homeward leg. Weather conditions, a factor which could be expected to influence driver workload, varied by season, but the long duration of the experiment had the effect of randomizing it over subjects. For safety reasons, data was not collected during extremely bad weather.

L. R. Zeitlin

Transportation Research Board, 1998 Annual Meeting

7

Because of the adventitious nature of the driver sample, the use of lightly trained observers, and the varying conditions of the commute, the study was not conducted with the rigor generally expected in laboratory based research. Data for some drivers were collected over a period of months, for others, nearly four years. No control was possible over most conditions influencing the driving workload. Data collection Subsidiary task performance. The driver was requested to perform the assigned subsidiary task for a two minute period during each of the four roadway conditions on the inbound commute. The process was repeated on the outbound commute. The same subsidiary task was used for all roadways of a given commute. Subsidiary tasks were alternated at each complete rotation of drivers, about every two weeks. An event counter was wired into the brake light circuit of each van, incrementing the count each time the brake was depressed. At the conclusion of each two minute subsidiary task, the driver was asked to rate the difficulty of that portion of the drive on a 10 point scale. Roadway and traffic data. In addition to initiating the task, the observer noted the roadway being travelled, time of day, traffic conditions including vehicle density and estimated speed, weather conditions, driver's difficulty rating, and the starting and ending count on the event counter. Observer notations were made during and at the conclusion of each two minute trial. Observers. Initially, the observer was either the author or a graduate student. Several months into the experiment, van pool members participated as observers using prepared data forms. Inter-observer reliability was assessed by comparing the speed and traffic density estimates on known stretches of roadway. The estimates of all observers were similar.

RESULTS Observer data on roadway traffic density, average speed, and brake actuations per minute for all drivers is summarized in Table 1. Despite the fact that the roadways clearly vary in all dimensions contributing to driver workload, the overall driving performance of the van drivers was relatively error free during the course of the experiment. The data included in this study was collected on 180 commuting days over a four year period. During that time the vans travelled over 36,000 miles in rush hour conditions. A total of five serious driving errors were recorded for the cases included in the data set, three resulting in moving violation summonses (two for failure to observe stop signs, one for speeding) and two in minor paint scraping incidents in dense traffic. While it is certainly true that an external observer might have noted other occasional deviations from prudent driving practice, no consequences resulted from these errors. This low error rate indicates that the mental workload imposed by the daily commute was, for the

L. R. Zeitlin

Transportation Research Board, 1998 Annual Meeting

8

most part, below the maximum capacity level of the drivers who participated in the trial. The low error rate also precludes any simple estimate of subsidiary task intrusion into the primary task, however it appeared that driving behavior was unaffected. This observation is in line with the findings of Kantowitz and Casper (1988) who report that in actual flying and driving, operators are aware of the negative consequences of decreased primary task performance and do not allow subsidiary tasks to intrude on a primary task.

TABLE 1. OBSERVED ROADWAY CHARACTERISTICS Roadway VehicleDensity Av. Speed Type Density std. dev. Rural road 2.7 .92 53.3 Expressway 6.7 2.01 Urban drive 17.2 7.71 City street 10.1 2.45

Speed (kph) 3.96 100.9 66.9 19.0

Brake Brake std. dev. per minute std. dev. 2.74 .73 8.59 .78 .53 5.61 3.30 1.31 1.76 8.49 2.26

Traffic density, as estimated by the number of vehicles visible through the front window, was dependent both on the nature of the roadway and the time of day. The commute, starting at the beginning of rush hour, encountered progressively greater traffic as the roadways converged on New York City. Peak density was observed on the urban limited access drives. The number of vehicles observed varied greatly from moment to moment as traffic was slowed by bottlenecks and construction sites. City street vehicle density was largely regulated by traffic lights and pedestrian movement and varied substantially less than did the density on urban drives. Speed limits were posted on all roadways, but actual driving speed appeared to be determined by the interaction of roadway characteristics, traffic density, weather, and the driver's comfort level. At the gross observational level, speed adjustment appeared to be the primary way of maintaining a comfortable workload. Drivers exceeded the posted limit on rural roads and expressways, dropped below the posted limits on urban drives and city streets. The number of brake actuations was also determined by roadway characteristics, traffic density, and weather. Unlike speed control, the driver has little choice about braking. A high rate of brake actuation reflects a similarly high degree of uncertainty and unpredictability in the driving situation. Expressway driving, with good visibility, gentle curves, and constant speed allowed minimal use of the brake. City driving, by contrast, required brake actuation on an average of every seven seconds.

L. R. Zeitlin

Transportation Research Board, 1998 Annual Meeting

9

The subjective assessment of task difficulty confirms the direct observations. Table 2 summarizes driving difficulty ratings and subsidiary task measures for the different roadways. The driving difficulty rating is the mean value of all ratings for a given condition, taken just after the completion of the subsidiary task. The ratings are based on a 1 to 10 scale where 1 is referenced to an extremely easy driving experience and 10 is referenced to an almost impossible, anxiety provoking experience. The ratings are completely subjective, but the low standard deviations attest to considerable agreement. Rural driving and expressway driving were viewed as relatively unstressful. Urban driving was near the midpoint of the range. City driving was rated both stressful and unpleasant. The estimates of driving difficulty are relatively constricted, ranging from 2.74 for rural roads to 6.15 on city streets despite wide changes in traffic density. It appears that a workload compensatory process was employed, drivers adjusting behavior, primarily speed, to keep difficulty within a tolerable range.The author observed that the drivers contribution to the intra-vehicle conversation generally decreased as the driving difficulty increased.

TABLE 2. DRIVER RESPONSE DATA Roadway Difficulty Rating Digit recall Type Rating Std. dev. Task errors Rural road 2.74 .718 2.60 Expressway2.76 .715 1.54 Urban drive4.22 1.320 5.28 City street 6.17 1.843 9.49 1 Index

Recall Random1 Workload2 Std. dev. Task Index Index .823 .165 8.509 .643 .197 7.432 .892 .083 9.366 1.572 .055 12.756

= probability of "true" random sequence. index = (brake actuation per minute) + log2 (kph)

2 Workload

An ANOVA shows that the differences in perceived difficulty ratings of the four roadway types were significant at p < .0001 level (F = 38.247). Difficulty ratings did not differentiate between rural and expressway driving because the mean difference in rated difficulty was only .02 on a 1 to 10 scale. The conservative post hoc Scheffe F-test showed that ratings did differentiate significantly between all other roadway comparisons at the p < 0.05 level. Errors made on the delayed digit recall task correlate (r =.784, 18 df) with subjective assessments of driving difficulty. An ANOVA shows that the digit recall error differences for each roadway type were significant at p < .0001 level (F = 504.15). The Scheffe F-test showed that all driving regimes could be clearly differentiated by the delayed digit recall error score at the p < 0.05 level. The obtained data

L. R. Zeitlin

Transportation Research Board, 1998 Annual Meeting

10

reaffirms the sensitivity of the delayed digit recall task as an indicator of operator loading as reported in earlier studies (Finkelman and Glass, 1970; Zeitlin and Finkelman, 1975; Finkelman, Zeitlin, Filippi, and Friend, 1977). The random digit generation task was relatively insensitive, capable of discriminating only between the low average workload and high average workload situations. Direct observation of the driver's performance provided the reason for this insensitivity. During all of the driving regimes, the driving task appeared to be composed of two components, a steady state driving workload dictated by roadway conditions, speed, and traffic density and a transient workload determined by the degree of uncertainty and unpredictability in the driving situation. While the overall driving task may have been of moderate difficulty as a whole, there were occasional moments of high difficulty. The integrative nature of the random digit generation subsidiary task tended to bury these moments of high difficulty, averaging them over the entire duration of the digit sequences. These observations of task composition prompted the development of a metric combining steady state and transient conditions in an attempt to provide an objective estimate of driver workload for the range of driving conditions encountered in this study. A micromodel of driver performance was conceptualized which assumed that overall load was the sum of the information processing requirements for handling uncertainty and tracking. From this model, an index of workload was derived by adding the number of brake actuations per minute during the test period to the log2 of the observed speed. Brake actuations represent the uncertainty inherent in the driving situation and are proportional to the information processing required to reduce that uncertainty (Shannon and Weaver, 1949) while the log2 of the speed is a first approximation of the steady state information processing load imposed by the tracking requirements of the driving regime (Elkind and Sprague, 1961). The final formulation of the index was, Workload = Constant * (Brake actuation rate + log2 kph). The constant is included for scaling purposes only and was kept at C = 1 for this experiment. The results of a calculation of workload based on the above index, averaged across all drivers for each roadway type, are listed in Table 2. An ANOVA showed that the workload index distinguished between roadway types at the p < .0001 level of significance (F = 58.101). The workload index for all subjects averaged over all conditions correlates well with the subjective driving difficulty ratings (r = .741, 18 df) and at a slightly higher level with errors on the digit recall task (r = .808, 18 df). Figure 1 illustrates difficulty ratings, digit errors, the random digit probabilities and the workload index for the four roadway types. Standard or Z scores were computed to provide the convenience of plotting all measures on the same scale although there is no assurance that the responses were normally

L. R. Zeitlin

Transportation Research Board, 1998 Annual Meeting

11

distributed given the uncontrolled conditions of a field trial. The workload index, calculated from directly observed driver actions, tracks the subjective difficulty rating and the best subsidiary task measure quite well.

6

4

Z Z Z Z

Z SCORES

2

difficulty digit errors work index random digits

0

-2

-4 RURAL

X'WAY

URBAN

CITY

ROADWAY Figure 1. Driver response measures by roadway type.

DISCUSSION This experiment confirmed the fact that a workload index, derived from directly observable task performance data, Workload = Constant * (Brake actuation rate + log2 kph), is a parsimonious metric combining transient (brake actuations) and steady state information processing (log2 kph) conditions and offers a simple, but directly observable, estimate of subjective driving difficulty. The r = .741 correlation of this metric with the rating of driving difficulty indicates

L. R. Zeitlin

Transportation Research Board, 1998 Annual Meeting

12

that in this limited case a subjective measure of workload can be predicted from an analysis of the objective data. However with d = .549, nearly half the variability in the predicted rating of driving difficulty is due to factors other than those used in the workload index. The workload index could well be augmented by the consideration of vehicle characteristics, traffic density, driver experience and the environment. Alternatively, the subjective ratings themselves may fail to fully reflect real differences in the mental workload. The micromodel of driver behavior, however, is both parsimonious and logical and could well serve as the starting point for more inclusive and complex formulations. The reason why a useful workload index could be derived from directly observable task data lies in the nature of the experiment itself. The long duration of the study had the effect of minimizing the effect of day to day uncontrolled variation in driving conditions. The same driving tasks were repeated over and over on the same roadways. Drivers were measured repeatedly and the results averaged. Driver responses, as well as subjective difficulty ratings, were a function of specific driving regimes hence the correlation between the two was high. The lesson to be learned from this study is not to use objective performance measures to estimate mental workload imposed by different man-machine systems but rather to compare the workloads imposed by the same system under different conditions with results averaged over many trials. The moderate dissociation between the subjective difficulty rating, the objective characteristics of the driving tasks, and the subsidiary tasks appears to be due to several factors. Most driving regimes underload the driver to a greater or lesser extent. By the time a driver has reached the midpoint of a driving career, he or she has over 4000 hours behind the wheel and has had ample experience with most contingencies that arise in the normal course of vehicle operation. Unless the driver experiences unusual conditions, driving becomes a virtually automatic procedure requiring only surface attention. Yeh and Wickens (1988) point out that under such light task loading, performance measures and subjective measures of workload tend to dissociate. Further, the elicitation of ratings from untrained subjects in a public social setting, introduces a number of modifying variables such as self-image, peer pressure, experience, status and risk tolerance. There is a strong inference of a form of workload homeostasis operating in this operator controlled, self paced driving task. Unlike tasks which are either system controlled or fixed paced, the driver is able to control enough parameters of the driving task to hold the workload at an acceptable level. The driver may vary speed, modify attention to information sources (road signs, radio), or modify participation in non-essential activities (conversation, smoking). Drivers were observed trying to increase their overall workload under easy driving conditions nearly as often as they tried to decrease their workload under difficult conditions. Indeed, this may be the general rule for all situations not under external control.

L. R. Zeitlin

Transportation Research Board, 1998 Annual Meeting

13

REFERENCES Bahrick, H. P., Noble, M., and Fitts, P. H. (1954). Extra task performance as a measure of learning a primary task. Journal of Experimental Psychology, 48 (4),

198-203.

Badderly, A. D. (1986). Working memory. New York: Oxford University Press. Beatty, J., (1982). Task-evoked pupillary response, processing load, and the structure of processing resources. Psychological Bulletin. 91, 276-292 Broadbent, D. E. (1971). Decision and Stress. New York: Academic Press. Brown, I. D., and Poulton, E. C. (1961). Measuring the "spare mental capacity" of car

drivers by a

subsidiary task. Ergonomics, 4 (1), 35-40. Brown, I. D. Measuring the "spare mental capacity" of car drivers by a subsidiary auditory task. (1962). Ergonomics, 5, 247-250. Brown, I. D. (1964). The measurement of perceptual load and reserve capacity. Transactions of the Association of Industrial Medical Officers. 14, 44-49. Derrick, W. L. (1988). Dimensions of operator workload. Human Factors. 30(1), 95-110 Elkind, J. I., and Sprague, L. T. (1961). Transmission of information in simple manual

control

systems. IRE Transactions on Human Factors in Electronics, HFE-2. 58-60. Finkelman, J. M., and Glass, D. G. (1970). Reappraisal of the relationship between noise and human performance by means of a subsidiary task measure. Journal of Applied Psychology. 54, 211213. Finkelman, J. M., Zeitlin, L. R., Filippi, J. A., and Friend, M. A. (1977). Noise and driver performance. Journal of Applied Psychology. 62, 6, 713-718. Hart, S. G., and Staveland, L. E. (1988). Development of a NASA-TLX (Task Load Index): Results of empirical and theoretical research. In. P. A. Hancock and N. Meshkati (Eds.) Human Mental Workload (pp. 139-183). Amsterdam: North-Holland. Hill, S. G., Iavecchia, H. P., Byers, J. C., Bittner, A. C., Zaklad, A. L., and Christ, R. E. (1992). Comparison of four subjective workload rating scales. Human Factors. 34 (4), 429-439. Johannsen, G., Moray, N., Pew, R., Rasmussen, J., Sanders, A., and Wickens, C. D. (1979). Final report of experimental psychology group. In N. Moray (Ed.), Mental Workload: Its Theory and Measurement. (pp. 101-114) New York:Plenum. Kahneman, D. (1973). Attention and Effort. Englewood Cliffs, NJ: Prentice Hall. Kantowitz, B. H., and Knight, J. L. (1976). Testing tapping time-sharing: I. Auditory secondary task. Acta Psychologica, 40, 343-362.

L. R. Zeitlin

Transportation Research Board, 1998 Annual Meeting

14

Kantowitz, B. K., and Casper, P. A. (1988). Human workload in aviation. In E. L. Weiner, and D. C. Nagel (Eds.), Human Factors in Aviation (pp. 157-187). San Diego,

CA: Academic Press.

Kramer, A., Wickens, C., and Donchin, E. (1983) An analysis of the processing requirements of a complex perceptual motor task. Human Factors, 25 (6), 597-621. Mackworth, J. (1959). Paced memorizing in a continuous task. Journal of Experimental Psychology, 58, 206-211. Moray, N. (1967). Where is capacity limited? A survey and a model. Acta Psychologica, 27, 84-92 Moray, N. (1982). Subjective mental workload. Human Factors, 24, 25-40. ODonnell, R., and Eggemeier, F. T. (1986). Workload assessment methodology. In K. Boff, L. Kaufman, and J. Thomas (Eds.) Handbook of perception and human performance, Vol. II. (pp. 41-1 - 42-49). New York: Wiley. Reid, G. B., and Nygren, T. E. (1988). The subjective workload assessment technique: A scaling procedure for measuring mental workload. In. P. A. Hancock and N. Meshkati (Eds.) Human Mental Workload (pp. 185-213). Amsterdam: North-Holland. Reid, G. B., Shingledecker, C., and Eggemeier, T. (1981) Application of conjoint measurement to workload scale development. In R.Sugarman (ed.), Proceedings of the Human Factors Society 25th Annual Meeting. Santa Monica, CA: Human Factors Society. Sanders, A. F. (1979). Some remarks on mental load. In N. Moray (Ed.),

Mental Workload: Its

Theory and Measurement. (pp. 41-77) New York:Plenum. Shannon, E. C. and Weaver, W. (1949). The Mathematical Theory of Communications. Urbana, IL: University of Illinois Press. Sheridan, T. (1980). Mental workload - What is it? Why bother with it? Human Factors Society Bulletin. 23 (2), 1-2. Solokov, E. I., Podachin, V. P., and Belova, E. V., (1983) Emotional Stress and Cardiovascular Response, Moscow: Mir Publishers. Tune, G. S. (1964). A brief survey of variables that influence random digit generation. Perceptual and Motor Skills, 18, 705-710. Vidulich, M. A., and Tsang, P. S. (1987). Absolute magnitude estimation and relative judgment approaches to subjective workload assessment. In Proceedings of the Human Factors Society 31st Annual Meeting. (pp. 1057-1061). Santa Monica, CA: Human Factors Society. Wierwille, W. W., Rahimi, M. and Casali, G. (1985). Evaluation of 16 measures of mental workload using a simulated flight task emphasizing mediational activity. Human Factors, 27, 489- 502. Wagenaar, W. A. (1972). Generation of random sequences by human subjects: A critical survey of literature. Psychological Bulletin, 77 (1), 65-72.

L. R. Zeitlin

Transportation Research Board, 1998 Annual Meeting

15

Wickens, C. D. (1984). Processing resources in attention. In R. Parasuraman and R. Davies (Eds.) Varieties of attention (pp. 63-101). New York: Academic Press Wickens, C. D. (1992). Engineering Psychology and Human Performance. New York: Harper Collins Yeh, Y-Y., and Wickens, C. D. (1988). Dissociation of performance and subjective workload. Human Factors. 30(1), 111-120. Zeitlin, L. R., and Finkelman, J. M. (1969). A subsidiary task evaluation of the information processing load in a vehicular simulation. APA Experimental Publication System, 3, Ms. 106B. Zeitlin, L. R., and Finkelman, J. M. (1971). Comparison of random digit generation and delayed digit recall as subsidiary tasks measures of operator loading in man-machine systems. APA Experimental Publication System, 12, Ms. 458-12. Zeitlin, L. R., and Finkelman, J. M. (1975). Subsidiary task techniques of digit generation and digit recall as indirect measures of operator loading. Human Factors, 17 (2), 218-220. Zeitlin, L. R. (1995) Estimates of mental workload: A long-term field trial of two subsidiary tasks. Human Factors, 37 (3), 611-621

L. R. Zeitlin

Transportation Research Board, 1998 Annual Meeting

16

LAWRENCE R. ZEITLIN is Professor Emeritus of I/O psychology and head of the Human Factors Laboratory at Baruch College of the City University of New York. He received his A.B. degree from Harvard University in 1951 and his Ph.D. from Northwestern University in 1954. He worked in both human factors and systems engineering at the U.S. Army Medical Research Laboratory, the RCA Airborne Systems Laboratory, the Bendix Corporation, Dunlap and Associates, and Bell Labs. Since 1963 he has been at CUNY. He has served on a number of National Academy of Science - National Academy of Engineering panels on marine casualty, safety in the offshore oil industry, and transportation of hazardous materials. He is a member of the TRB Committee on Simulation. His current human factors related research involves risk assumption and workload consideration in industrial and transportation accidents.