RESULTS
Table of Content


This section describes the data reduction and analysis procedures applied in this study, and presents results in two areas. To begin, the classification of driving errors resulting from the present study is developed, from two sources. First, a procedure for translating DMV examiner scoresheet entries into categories and events describing each subject's errors related to intersection negotiation is detailed; then, specifics of the reduction and coding of behavioral errors videotaped during the test drives are explained. The distributions of errors, so defined, are then presented according to each route type (standard/low familiarity vs home area/high familiarity) on which subjects were tested in this study.


To conclude this section, a weighted error score as a criterion variable denoting driving competency (on the standard exam route) is defined, then multiple regression and correlation analyses document the following relationships of interest in this study. Initially, the prediction of driving competency (weighted error score) by the functional measures obtained using the MultiCAD protocol is reported. Then, the relationship between behavioral errors coded from the videotaped observations and the weighted error score adopted as the overall measure of competency during the on-road exam is analyzed.


CLASSIFICATION OF INTERSECTION DRIVING ERRORS
Table of Content


A central objective of the research was to create an older driver intersection error classification. A two-level classification was conducted. First, a classification was developed from the data coded from the videotaped observations of driving performance during the on-road examinations; exhaustive review of these data identified not only behaviors thought to be associated with driving errors, but also subjects' levels of exposure to situations where such errors would be most critical. This classification thus provides a description of intersection maneuver errors and errors of observation, and their associated probabilities of occurrence, for both the standard and home area routes.


A second classification was developed based on supplemental data reduction from DMV scoresheets. The DMV examiners' scoresheets for the standard area exam and the home area exam, provided by CA DMV, included a standard checklist of possible errors plus a comments section. The examiners provided rich descriptions of maneuver and observation errors, in addition to marking the errors on the standard checklist. However, while such error frequency counts could be reduced from the standard checklist, they were not exposure-based (i.e., it is unknown to what extent errors were not scored because the opportunity was not present, when the same behavior in another situation would be recorded as an error, nor is it known how many times a given error occurred--only whether it was scored or not scored for each subject on each route). Therefore, it was meaningless to compare simple error counts for drivers who passed versus failed the exams, and comparisons of frequencies among the error types would not be reliable. Instead, the most useful approach in creating a classification of errors from the DMV score sheets, was to determine what proportion of drivers committed each error, for each exam route.


Together, these approaches to the analysis and summary of intersection driving errors provide a comprehensive picture of the behaviors evidenced by the present sample of older (referred) drivers. Procedures for reducing data from the videotapes and scoresheets are described below, followed by the driver error taxonomies.


Videotape Data Reduction Procedures


Each road test produced video data for the three views (forward traffic scene, driver's face, and the rearview traffic scene with accelerometer); participants who completed both the standard and home area exams generated six videotapes of data. However, several participants did not complete the standard route exam because the examiners terminated the test for reasons of safety; and, others who completed the standard exam were not given the opportunity to take the home area exam, because of their hazardous performance on the standard route test. Video equipment difficulties also contributed to missing data for some participants. Videotape observations were available for 62 subjects on the standard exam route and 51 subjects on the home area exam route.


Data were reduced from each videotape, and were coded and entered into an electronic (Microsoft Access) database. Specialized video equipment and software were used so that time code information could be sent directly from the video playback deck to the database. This way time code data could be recorded quickly and without the risk of keying

errors. All data were initially recorded as times at which behaviors occurred. In this database, each case represented a single intersection traversed. These data were subsequently analyzed using the Statistical Package for the Social Sciences (SPSS) to generate counts of errors and opportunities for errors. In some cases, the fact that a time was recorded for a behavior signified that an error had taken place (e.g., if a time was recorded for "cuts corner when making turn," an error had occurred). In other cases, it was necessary to compare variables and use formulas to identify errors (e.g., if a driver made a lane change and there were no visual observations in the direction of the lane change within a certain

time prior to beginning the lane change, an error had occurred).


Although all subjects traversed the same route during the standard exam, differing traffic situations and signal phases reduced the opportunities for errors to occur for some subjects at some intersections. In addition, some subjects did not complete the standard exam due to extremely hazardous performance, and thus had less opportunity to commit errors. Comparing error frequencies for intersection types, error types, or route type would therefore be meaningless. To normalize error frequency, a count of errors in addition to a count of opportunities to commit an error was entered into the spreadsheet for each subject at each intersection. If an error was observed from the videotape, a "1" was entered into the database for that subject, for the particular error, for a particular intersection. A "0" was entered if the opportunity to commit the error was present, but no error was committed. Using frequency of errors committed and number of opportunities that presented themselves for an error to occur, an error probability was calculated. Once errors and opportunities for errors had been determined for each intersection, a file was created which contained scores for each error type, for each subject, for each route type, aggregated across all intersections. This file was then matched with data from the other measures for analysis.


From the videotape showing the forward view camera, the following data were entered into the database:


Intersection type/geometry.

Type of traffic control.

Direction of intended maneuver.

Time driver left prior intersection.

Time when driver changed lane prior to the left turn at the intersection.

Time driver completed lane change.

Time of brake light activation of vehicles slowing/stopping ahead and point when driver reaches lead vehicle.

Minimum headway while slowing/stopping for traffic signal.

Time when driver reaches traffic signal.

Time of traffic signal change.

Driver stops before, during, and through intersection (if any).

Available gap openings and closures with crossing traffic at the intersection.

Driver swings wide prior to making turn.

Time driver begins crossing/merging maneuver, then stops.

Time driver enters intersection entrance.

Driver swings wide when making turn.

Driver "cuts" corner when making turn.

Driver uses incorrect lane (i.e., right or straight through lane) to cross intersection.

Driver enters far lane (instead of near lane) when turning and stays in far lane.

Time driver is out of intersection (turn completed).

Driver changes lane after making turn.

From the videotape that showed the driver's face and top portion of the steering wheel, the following data were entered into the database:


Driver makes visual checks to side view mirror to the left.

Driver makes visual checks to side view mirror to the right.

Driver makes visual checks to blind spot to the left.

Driver makes visual checks to blind spot to the right.

Driver makes visual checks to inside rearview mirror.

Driver makes visual checks to oncoming traffic.


From the videotape that showed the rear camera view, the following data were entered into the database:


Starting point of deceleration of driver's vehicle to intersection.

Decelerations greater than 0.3 g.

Accelerations greater than 0.3 g.

Maximum lateral acceleration and magnitude of change.

First intersecting traffic which crosses road that driver was on before making turn.

Following vehicle reaches the point at which driver completed a lane change.


To determine the probability of occurrence of each error type, errors were aggregated across all intersections within each exam route. Table 2 presents the classification of driver maneuver errors and errors of observation as reduced from the videotaped data, and the probability of occurrence for the standard exam and the home area exam. Where the occurrence of an error is operationally defined in terms of the commission or omission of an action, within a definite timeframe or in relation to other specific actions or events, the defining conditions are noted explicitly in Table 2. Time/distance values adopted as thresholds for the scoring of errors in this formidable task were based on the understanding of relevant human factors and traffic safety design principles by the research team.


Table 2. Classification and associated probability of occurrence of intersection maneuver errors (unshaded) and errors of observation (bold and shaded), for the standard and home area exams.
Table of Content

Description of Error Exam Route
Standard

(n=62)

Home Area

(n=51)

Infringes on others' right-of-way when changing lanes (lane change made with less than 2 s between vehicles) 0.90 0.57
Fails to observe behind within 5 s prior to beginning deceleration for intersection 0.87 0.96
Fails to look to the sides while in intersection 0.75 0.75
Fails to check right mirror within 5 s prior to right lane change 0.73 0.77
Fails to look to the sides during approach to intersection (within 5 s prior to entering intersection) 0.36 0.44
Fails to check right blind spot within 5 s prior to right lane change 0.35 0.33
Fails to check left mirror within 5 s prior to left lane change 0.31 0.35
Fails to check either right mirror or right blind spot 0.30 0.23
Fails to check left blind spot within 5 s prior to left lane change 0.29 0.37
When lane change is necessary to cross intersection, changes lanes too close to intersection (less than 5 s between lane change and entry into intersection) 0.19 0.12
Fails to check to the left (upstream) within 5 s prior to entering intersection when turning right from a stop or yield sign (to check for potential conflict vehicles/make gap judgments) 0.17 0.15
Fails to check to the right (downstream) within 5 s prior to entering intersection when turning right from a stop or yield sign (to look for pedestrians or a traffic queue in the intended travel path) 0.15 0.09
Deceleration greater than -.3g (abrupt or panic stop) 0.15 0.29
Rejects a safe gap (during a gap selection task, a subject arrives at an intersection and the first car to cross the subject's path is greater than 10 s away) 0.13 0.06
Lateral acceleration greater than +/- .3g during turns 0.10 0.13
Changes lanes prematurely in anticipation of left turn (crosses solid paint line to get into turn lane) 0.08 0.10
Acceleration greater than +.3g 0.08 0.13
Fails to check either left mirror or left blind spot 0.07 0.10
Enters far lane during turn 0.04 0.06
Swings wide after turning 0.02 0.03
Accepts an unsafe gap (during a gap selection task, a subject negotiates an intersection with less than 6 s between his/her intersection exit and the first car to reach the subject's path) 0.02 0.03
Cuts across lane of intersecting roadway during turn 0.01 0.01
Turns into oncoming traffic lane or median strip 0.007 0.002
Uses incorrect lane to cross intersection 0.005 ---
Fails to come to complete stop at stop sign 0.004 0.001
Stops for no apparent reason 0.003 ---
Hits object 0.002 0.001
Drifts into another lane on straight driving section 0.001 0.001
Backs up after stopping at an intersection 0.001 ---
Swings wide before turning 0.001 0.001
Changes more than one lane at a time 0.001 ---



The maneuver error with the highest probability of occurrence on both test routes was infringing on the right-of-way of other drivers when changing lanes. The probability of an unsafe lane change increased substantially on the unfamiliar route. A possible explanation for a lower incidence for the familiar route could be that the drivers, rather than the examiners, determined where they would drive. If a driver planned to turn at a particular intersection, and knew ahead of time what lane he/she should be in to execute the maneuver, the driver could change lanes as far away from the intersection, and at any particular time he/she was ready to execute the lane-change maneuver. There are many anecdotal reports of older drivers who drive in the left lane for several blocks in preparation for a left-turn maneuver at an intersection to avoid the requirement to make a lane change closer to the intersection. On the standard route, the examiner directed all maneuvers, and thus, there was more opportunity for a lane change to be executed without as much preparation time as was available for drivers on the home area route. It should be noted that the opportunity to commit this error was only present for a small number of drivers (5 to 7), but the high error rate indicated that this error occurred almost every time it was possible.


The next maneuver error listed, changing lanes too close to the intersection, occurred more often on the standard route (in 19 percent of the opportunities presented) than on the home area route (in 12 percent of the opportunities). The same factors noted in the paragraph above are possible explanations for this behavior.


Next, errors of observation, of several types, occurred with relatively equal probabilities for both route types. It should be noted that many more opportunities for these error types to occur were present and many more drivers contributed to these errors than for the lane-change maneuver error described above. Drivers almost always (87 to 96 percent of the time) failed to observe the traffic situation behind them prior to decelerating for an intersection. Drivers also frequently failed to look to the sides while in the intersection (75 percent of the time on each route). There were 883 occurrences on the area route by 41 drivers, and 931 occurrences by 52 drivers on the standard exam route of failing to check traffic while in the intersection; since videotaped data were available for only 67 subjects (across both routes), this indicates that failure to observe to the sides while traversing an intersection is a behavior that was common for a significant majority of drivers in this study. The fact that these error types occurred with similar frequencies on both routes, given the fact that the poorest performers who committed multiple critical or hazardous errors were eliminated from the subset of drivers who took the area test (i.e., 18 subjects failed the standard area exam and were not given the opportunity to take the home area exam), further indicates that "good" drivers and "bad" drivers alike commit these kinds of errors.


Other errors of observation centered around checking mirrors and blind spots prior to changing lanes. Drivers failed to check their right mirrors prior to changing lanes to the right approximately three-quarters of the time for both routes, but only failed to check either the right blind spot (head check) or the right mirror 20 to 30 percent of the time. Drivers were slightly more likely to check the right blind spot on the home area route than on the standard route. Conversely, drivers were less likely to check the left blind spot (head check) prior to a left lane change on the home area route than on the standard route.


Drivers were almost twice as likely to execute a hard braking maneuver on the home area route (in 29 percent of the opportunities) than on the standard route (in 15 percent of the opportunities). This may have reflected varying expectancies across the two test routes; alternatively, drivers may have been more vigilant about observing lead vehicle performance on the standard exam, after having completed the MultiCAD battery on the same day.


Finally, the videotaped data were aggregated across error types to explore the relationship between error rates and intersection traffic control and operations, as a function of route familiarity level. Table 3 presents the mean error rate across all observation and maneuver errors, for the standard and home area routes, by drivers' movements and type of traffic control at intersections encountered on the test routes. These results indicate that (relative) route familiarity level had little to no effect on error rates exhibited at signalized intersections, where through maneuvers accounted for the highest proportion of errors. At stop-controlled intersections, through maneuvers again accounted for the highest proportion of errors; these errors were committed in 12 to 13 percent of the opportunities, regardless of route familiarity. Left-turn and right-turn errors at stop-controlled intersections, although infrequent, were committed in different proportions as a function of route familiarity. Slightly higher percentages of errors were observed for left-turn maneuvers on the standard route compared to the home area route, but the error rate for right-turn maneuvers on the more familiar (home area) route was almost twice that on the standard, less familiar route.


At yield-controlled intersections, left-turn errors occurred equally often (in about 7 percent of the opportunities) for both route types. However, for right turns, errors occurred on the standard route at almost twice the rate of those occurring on the home area route. This may have resulted from drivers "knowing what to look for" as a result of experience in familiar areas.


At uncontrolled intersections on the standard route, errors occurred during right-turn maneuvers in approximately one-fourth of all opportunities that existed for errors to occur. In contrast, on the familiar route, right-turn errors only occurred in 9 percent of the opportunities presented. Again, it appears that when drivers know where and when to look for potential conflicts, and are cognizant of the intersection demands, they can approach and negotiate an intersection with fewer errors.


DMV Scoresheet Data Reduction Procedures


As described earlier, the DMV scoresheets list the scored maneuvers, and the types of errors that could occur. If a driver commits an error at a scored location, a mark is made on the scoresheet next to the error, but multiple errors of the same type are only tallied if the maneuver is required at multiple locations. Certain errors were elaborated upon in the "comments" section of the scoresheet. For example, "unnecessary stop" is listed as an error for the four left and right turn approaches, but in several instances, the examiners wrote "driver stopped at 'stop ahead' pavement markings." Another example would be descriptions of the "intervention by examiner" critical error, such as "driver drove straight for pedestrian crossing the road; I had to intervene."


To create a classification of driver errors, errors from each subjects's standard area exam and home area exam scoresheets were entered into a table that contained simple descriptions of the scored maneuvers by subject number and route type. If an examiner provided descriptive information about particular errors, this information was included in the table of errors. A count of the number of subjects who made each error was made, then the frequency was translated into percent of subjects who committed each error, based on the number of subjects who took each drive test. The results of this data reduction activity are






Table 3. Mean error rate* (percent) based on analysis of video observational data, across all error types and subjects, for standard (unfamiliar) and home-area (familiar) routes, by intersection control type and direction of movement.
Table of Content


Route Type Intersection Control Type and Direction of Movement
Signalized Stop Yield Uncontrolled
Through

Mvmt.

Left- Turn

Mvmt.

Right- Turn

Mvmt.

Through

Mvmt.

Left- Turn

Mvmt.

Right- Turn

Mvmt.

Through

Mvmt.

Left- Turn

Mvmt.

Right- Turn

Mvmt.

Through

Mvmt.

Left- Turn

Mvmt.

Right- Turn

Mvmt.

Standard 19.8 11.6 8.8 12.9 9.4 4.5 --- 7.9 13.6 --- 9.5 26.0
Home Area 18.9 11.4 9.6 12.4 7.6 8.3 20.0 7.4 7.5 14.3 6.9 9.2


*error rate was calculated by dividing the number of errors that occurred by the number of opportunities/situations that allowed for an error to occur.

presented in Table 4. Errors are presented in the following general categories: scanning, compliance with traffic control devices, lane use, speed control, reaction to other traffic/hazards, and use of vehicle controls and auxiliary equipment. Also, unlike the classification produced from the videotaped observations, the results presented in Table 4 include but are not limited to intersections. The errors recorded at sites other than intersections have been preserved for this presentation in the broad interest of documenting older driver difficulties and, in some cases, because of their presumed generalizability to intersection as well as non-intersection locations (e.g., "disregarded pavement markings").


It should be noted that 80 subjects took the standard exam (28 passed and 52 failed) and 61 subjects took the home area exam (25 passed and 36 failed). Two subjects who completed the MultiCAD test battery did not take either road test; one of these subjects was judged by the examiner to be too visually impaired to drive, and the other subject was unable to get his vehicle into safe driving condition in order to participate in road testing. Because it was hypothesized that drivers might demonstrate greater competency in familiar areas (and DMVs have the authority to restrict drivers to driving within a specific radius from home, rather than remove all driving privileges), drivers who failed the standard exam were permitted to take the home area exam, unless the standard exam was terminated due to extremely hazardous performance. The examiners terminated the standard exam for 17 subjects whose performance was unduly hazardous. For two additional subjects who failed the standard exam, one was unable to schedule an appointment to take the home area exam within the study period, and the other started the home area exam, but was unable to complete it because of the mountainous terrain and the lack of intersecting roadways.


Table 4. Classification of driving errors from DMV score sheets, and associated percent of sample committing each error, as a function of on-road exam route type.
Table of Content


Driving Error Exam Route
Standard

(n=80)

Home Area (n=61)
Scanning
Failure to look left and right at through intersections (stares straight ahead) 76% 85%
Failure to check traffic on approach to turns 54% 43%
Failure to check traffic when changing lanes or merging 69% 57%
Failure to check traffic when pulling to and from curb 63% 62%
Failure to look left when turning right (to check for approaching traffic) 5% 0%
Attempted to run blind intersection without looking left or right 5% 0%
Backs up using mirror(s) only 5% 0%
Backs up with no look at all

8% 3%
Compliance with Traffic Control Devices
Failure to come to complete stop at stop sign 53% 57%
Stops over limit lines 45% 28%
Ran red light (went through, turned left, or u-turn) 5% 7%
Ran stop sign 6% 3%
Made illegal left or right turn 0% 3%
Wrong-way maneuver (entered parking lot in exit only driveway; turned left on left side of island) 6% 2%
Slow reactions to stop signs and red lights 1% 0%
No reaction to flashing signal at railroad crossing 0% 2%
Sat at green light waiting to make right turn 8% 2%
Sat through most of green light, waiting for green arrow (where there was no green arrow phase) 1% 0%
Disregarded pavement markings (lane lines) in parking lot 0% 2%
Stops for no reason (e.g., in middle of intersection, mid-lane on approach to turn, at stop limit on green light for right turn, at uncontrolled right turns, or before pulling over to park) 39% 26%
Stopped at "Stop Ahead" sign or pavement marking 13% 2%
Unsure of right-of-way, creating confusion 1% 3%
Lane Use
Turns too wide or too short 46% 26%
Executes left turn from center or right lane, ignoring left-turn lane 4% 13%
Executes left turn from left side of double yellow line 5% 2%
Pulls into left turn lane late 1% 2%
Completes left turns in opposing traffic lane (on wrong side of street) 9% 7%
Drives in center left turn lane after completing left turn 0% 2%
Executes right turn from outside lane (or too far from curb) 4% 7%
Drives in shoulder, parking lane, or bike lane after completing right turns 1% 5%
Used turn-only lane for through maneuver 0% 2%
Changes more than 1 lane at a time 1% 3%
Drives in far right of lanes (or in parking or bike lanes), confusing other drivers 10% 3%
Drives on left lane lines (on raised pavement markers) 1% 0%
Straddles lanes/drifts in and out of lanes 10% 15%
Speed Control
Brakes before changing lanes or at other unnecessary/inappropriate time 19% 8%
Traverses intersections too fast 3% 10%
Changes lanes too quickly 1% 2%
Consistently drives too slow (e.g., 20 mi/h on 45 mi/h boulevard; 10-15 mi/h under speed limit) 24% 5%
Consistently drives too fast (5-10 mi/h over speed limit) 4% 15%
Does not coordinate accelerating and braking smoothly 1% 2%
Reaction to Other Traffic/Hazards
Unsafe left turn gap acceptance (near collision) 22% 15%
Unsafe right turn gap acceptance (in front of approaching cross traffic); near collision 16% 8%
Slow reactions to cross traffic (several attempts to pull out were aborted) 1% 0%
Accelerated toward (or no response to) vehicle stopped ahead in same lane 0% 5%
Infringes on others right-of way when changing lanes (near miss) 8% 23%
Struck object (curb when backing, median after turning left, or object in parking lot) 18% 0%
Near miss (pedestrian or car) other than during gap acceptance 16% 20%
Follows too close 0% 2%
Unsafe passing maneuver 0% 3%
Approached road work area by going into opposing left lane instead of around on the right side 0% 2%
Use of Vehicle Controls and Auxiliary Equipment
Failure to use turn signals for turning, lane changing, or merging 65% 20%
Erratic Steering 4% 2%
Forgot car was in reverse 1% 0%
Frequently/always fails to cancel turn signal 0% 2%
Could not locate turn signals or defroster; had to try every accessory 1% 0%

As might be expected, different kinds of errors were noted at each level of the classification. With respect to the videotaped driving observations, there were times when data were not available for a particular view, either because the camera's field of view did not include a target/event of interest, or because of equipment malfunction. The human observer (examiner), however, experienced no such restrictions. Further, an understanding of subjects' behavior in context was available to the examiners that was not available during the coding of the videologs. At the same time, the videolog provided the opportunity "after the fact" to code errors on a micro level that may have gone unnoticed by examiners who were looking for gross commissions/omissions in subjects' behavior during the road test, and greater reliability in the scoring of performance could be achieved with the coded video observations than may have been possible between examiners. Also, it may be noted that objective criteria were consistently applied in the coding of all subjects' behaviors for the purpose of defining errors in this classification; e.g., a precise interval (2 s) was applied as a minimum clearance ahead of an adjacent-lane vehicle to define a "safe" maneuver when the subject changed lanes.


Examination of Table 4, reveals that scanning errors were committed by the largest proportion of drivers. A majority (over 75 percent) of the subjects on each test route failed to check traffic when traversing (through) intersections, but a higher percentage failed to do so on the home area route. A large proportion of subjects also failed to check traffic on the approach to intersections, but slightly more subjects failed to do so on the standard route. Failure to check traffic when changing lanes or pulling to and from the curb was recorded for approximately 60 percent of the subjects on each test route. Compliance with traffic control devices also was a problem for drivers on both routes. Slightly over half of the subjects who took each on-road exam failed to stop completely at stop signs. A large proportion of drivers also failed to stop behind the limit lines at intersections; a larger percentage of drivers committed this error on the standard area route than on the home area route. A greater proportion of drivers committed errors such as running a stop sign, performing a wrong-way maneuver, stopping at a green light to turn right, and stopping for no reason on the standard route than on the familiar home area route. With the exception of stopping for no reason, the differences in the percentages are relatively small. On the other hand, a larger percentage of drivers ran a red light, made an illegal turn, and did not react to a flashing railroad crossing signal on the familiar route than on the standard route; again, however, the difference in percentages is small (1 or 2 subjects).


Regarding lane use, approximately twice as many drivers turned too wide or too short on the standard area route than on the home area route, and more drivers on the standard route drove too close to the curb or in bike/parking lanes than on the home area route. However, a larger percentage of drivers on the home area route executed a left turn from the wrong lane, drove in the center lane after completing a turn, and executed a right turn from an outside lane.


Comparing speed control errors, a larger percentage of drivers on the standard route drove too slow, and a larger percentage of home area route drivers drove too fast. More drivers applied their brakes before changing lanes when driving on the standard route than when they drove on the home area route. Driving too slow and braking before changing lanes are characteristic of older drivers who, in unfamiliar locations, compensate for slower information processing capability by reducing their speeds.


Considering errors committed when reacting to other traffic, higher percentages of drivers made unsafe left- and right-turn gap acceptance errors when driving on the standard route than when driving on the home area route. Familiarity with the operational characteristics of intersections, and/or negative past experience in their home areas, may explain these differences. Similarly, the only incidences of striking an object occurred on the standard area drive test. However, a higher percentage of drivers on the familiar route had a near miss with another vehicle when changing lanes or with a pedestrian, than on the standard route. This could be explained by the "looked but did not see" phenomenon, where familiarity may reduce vigilance for certain tasks.


In the use of vehicle controls category, a much greater percentage of drivers did not use their turn signals when driving on the standard route than when they drove on the home area route. This difference could be the result of drivers planning their own course when driving in their home area, and knowing farther in advance where and when they were going to turn. Conversely, not having advance information about the route configuration--allowing anticipation of where to turn--may have contributed to an overload condition on the less familiar route, resulting in the "shedding" of turn signal activation as an unnecessary or low priority task.


CORRELATION AND REGRESSION ANALYSES
Table of Content


To determine the efficacy of the MultiCAD tests in predicting on-road driving performance, correlational analyses were performed to determine the strength of the relationship between each test and a weighted error score on the standard exam. As described earlier, the test examiners used a standard form to record when "structured maneuvers at predesignated points on the route were performed unsatisfactorily" (Janke and Hersch, 1997). Examples of structured maneuver errors are "inadequate traffic check," "poor lane position," and "turns too wide or too short." A subset of errors defined as critical driving errors were listed in a separate section of the DMV score sheet. These are serious errors; under normal testing circumstances (i.e., other than a research situation), a driver's test would immediately be terminated. Critical errors included: examiner intervention; driver strikes object; drives up/over curb/sidewalk; drives in oncoming traffic lane; disobeys sign/signal; dangerous maneuver; inappropriate reaction to school bus; inappropriate reaction to emergency vehicle; inappropriate speed; inappropriate auxiliary equipment use; turn from improper lane. A subset of critical errors was also defined by Janke and Hersch as hazardous errors, with the belief that these errors are predictive of driving impairment. These included "dangerous maneuver" and "examiner intervention."


A weighted error score, designated MSCORE by Janke and Hersch (1997), thus serves as the primary criterion (dependent) variable for these analyses. It was calculated by adding the total number of errors on the standard exam (regardless of severity) to twice the sum of critical and hazardous errors. Since hazardous errors are a subset of critical errors, and critical errors are a subset of total errors, this scheme weighted hazardous errors by a factor of five and other critical errors by a factor of three.


Table 5 presents the simple Pearson product-moment correlations between each test in the MultiCAD battery and the weighted error score (MSCORE), and their probability levels. A total of 82 subjects completed the MultiCAD battery; of this number, 26 were cognitively impaired and 56 were cognitively unimpaired as per classification criteria of Janke and Hersch. Due to missing cells in the correlation matrix, the N's involved in these analyses ranged from 36 to 79. A correction was applied in the reported analyses. Also, the correlation matrix indicated that intercorrelations between measures ranged up to .84. Therefore, it cannot necessarily be concluded that significant relationships between isolated measures and the weighted error score criterion which follow connote functional deficits exercising separate influence on safe driving behavior.


As evidenced by the shaded entries in Table 5, multiple significant relationships exist between visual performance and driving competency, predominantly when visual performance is measured as time to respond. Response accuracy significantly related to MSCORE much less often. This pattern of results suggests that significant correlations between driving performance and the acuity measures are likely due to a choice-reaction time element, rather than to acuity in and of itself.


Table 5. Correlations between MultiCAD variables and MSCORE, with significant relationships (p<.05) shaded, and in bold font.

(Reported by Janke and Hersch, 1997).
Table of Content


Measure r with MSCORE Nominal p
Static acuity accuracy @ 20/40 .0855 .457
Static acuity accuracy @ 20/80 -.0799 .487
Static acuity accuracy @ 20/200 -.0048 .966
Static acuity response time @ 20/40, correct trials .3395 .004
Static acuity response time @ 20/80, correct trials .4230 .000
Static acuity response time @ 20/200, correct trials .1970 .090
Dynamic acuity accuracy @ 20/40 -.1418 .219
Dynamic acuity accuracy @ 20/80 -.1211 .294
Dynamic acuity accuracy @ 20/200 -.2283 .046
Dynamic acuity response time @ 20/40, correct trials .3092 .010
Dynamic acuity response time @ 20/80, correct trials .3256 .005
Dynamic acuity response time @ 20/200, correct trials .3297 .004
Static contrast sensitivity accuracy @ 20/40, high contrast .0519 .654
Static contrast sensitivity accuracy @ 20/40, low contrast -.2477 .030
Static contrast sensitivity accuracy @ 20/80, high contrast -.0582 .613
Static contrast sensitivity accuracy @ 20/80, low contrast -.1513 .189
Static contrast sensitivity response time @ 20/40, high contrast, correct trials .1666 .181
Static contrast sensitivity response time @ 20/40, low contrast, correct trials .1926 .240
Static contrast sensitivity response time @ 20/80, high contrast, correct trials .3884 .001
Static contrast sensitivity response time @ 20/80, low contrast, correct trials .0747 .561
Dynamic contrast sensitivity accuracy @ 20/40, high contrast -.0705 .548
Dynamic contrast sensitivity accuracy @ 20/40, low contrast .0643 .586
Dynamic contrast sensitivity accuracy @ 20/80, high contrast -.2575 .024
Dynamic contrast sensitivity accuracy @ 20/80, low contrast -.2030 .081
Dynamic contrast sensitivity response time @ 20/40, high contrast, correct trials .0401 .782
Dynamic contrast sensitivity response time @ 20/40, low contrast, correct trials -.2059 .180
Dynamic contrast sensitivity response time @ 20/80, high contrast, correct trials .2466 .049
Dynamic contrast sensitivity response time, @ 20/80, low contrast, correct trials -.0947 .500
Mean brake response time with visible brake lights, correct trials

(12 trials)

.0861 .457
Proportion error, trials with visible brake lights .2801 .013
Mean brake response time with no visible brake lights, correct trials

(3 trials)

-.0238 .841
Proportion error, trials with no visible brake lights .1994 .080
Mean brake response time to threats at 15 degrees, correct trials

(2 trials)

.1891 .144
Proportion error, threats at 15 degrees .2430 .043
Mean brake response time to threats at 30 degrees, correct trials (1 trial) .1181 .429
Proportion error, threats at 30 degrees .1675 .163


Specifically, response time for correct responses to static acuity targets at 20/40 and 20/80; to dynamic acuity targets at 20/40, 20/80, and 20/200; to high-contrast static contrast sensitivity targets at 20/80; and to high-contrast dynamic contrast sensitivity targets at 20/80 was significantly correlated to driving competency, operationalized using the weighted error score measure. As would be expected, these correlations were all positive, i.e., increasing response times were associated with higher error scores. Similarly, dynamic acuity response accuracy (using a 20/200 target) and static contrast sensitivity accuracy (using a 20/40, low contrast target) were significantly, though inversely, related to MSCORE: as response accuracy decreased driving errors increased for these isolated relationships.


A different pattern of results was demonstrated for the relationships between MultiCAD tests measuring perceptual capability and MSCORE. It was response accuracy--rather than response time--that best predicted on-road driving performance in these analyses (see Table 5). For the angular motion sensitivity tests, only the proportion of errors on trials where the lead vehicle brake lights were activated correlated significantly with weighted error score. On the useful (functional) field of view tasks, only the proportion of errors on trials where a threat entered from the periphery, at a 15-degree angle of eccentricity, was significantly correlated with the weighted error score criterion.


A further goal was to determine whether any of the tests measuring functional ability discriminated the cognitively impaired referral drivers from the cognitively unimpaired referral drivers. Cognitive impairment was determined in this case using information provided on the referral or medical evaluation forms. Not surprisingly, the cognitively impaired drivers exhibited a general trend of inferiority in road test performance (e.g., higher weighted error score) on the standard exam, compared to that of the cognitively unimpaired drivers. On the MultiCAD tests, a significantly higher mean error score was demonstrated for cognitively impaired versus cognitively unimpaired drivers responding to a lead vehicle braking, with visible brake lights. Specifically, the mean error score for the cognitively impaired drivers on this test was 0.473 (errors were made on 47 percent of the trials of this type); this was more than twice that of the cognitively unimpaired drivers, who demonstrated a mean error rate of 0.210. As per Janke and Hersch (1997) this difference, nominally significant at p<.001, was significant at the p<.05 level using a Bonferroni-type correction. As such, this measure was the strongest discriminator between cognitively impaired and cognitively unimpaired subjects in the test sample.


Four additional measures in the MultiCAD battery that Janke and Hersch (1997) reported as useful discriminators between these groups included: (1) brake response time for the pedestrian intersecting the driver's path at 30 in the driving video; (2) response time to the static, high contrast, low resolution (7 cycles/degree) contrast sensitivity targets; (3) accuracy of response to the dynamic, high contrast, high resolution (15 cycles/degree) contrast sensitivity targets; and (4) accuracy of response to the dynamic, high contrast, low resolution contrast sensitivity targets. The mean response time to the pedestrian target was 1.871 seconds for the cognitively impaired drivers, and 1.493 seconds for the cognitively unimpaired drivers. This difference was nominally significant (p=0.026). The mean response time for correct responses to the static, high contrast, low resolution contrast sensitivity targets was 2.053 seconds for cognitively impaired drivers, and 1.666 seconds for the cognitively unimpaired drivers. This difference was nominally significant (p=0.051). Regarding accuracy of response to the high contrast dynamic contrast sensitivity targets, the average score(1) for the high resolution targets was .231 for cognitively impaired drivers and .451 for cognitively unimpaired drivers (nominal p = 0.050); for the low resolution targets, average scores were .500 and .774 for the cognitively impaired and cognitively unimpaired drivers, respectively (nominal p = 0.023).


Finally, Table 6 reports the results of analyses which describe the relationship with MSCORE of the driving errors reduced from the videotapes of driving behavior recorded during the test drives, separately for errors of observation and maneuver errors. Unfortunately, finer analyses within these categories of videotaped driving errors were precluded due to missing data; a missing observation for any single behavior among the many component behaviors within an error category (see Table 2) resulted in all observations for that subject being excluded from entry into the multiple regression equation.


As apparent in Table 6, these relationships were quite weak. Overall, the model was not significant (F(3,51)=1.125), and accounted for only six percent of the variance in these data. It must be concluded that, while all drivers in the study sample committed behaviors that could be interpreted as errors according to the objective criteria adopted for reduction/coding of these observational data, such behaviors were only rarely and not systematically associated with the occurrence of the critical or hazardous errors toward which the MSCORE criterion variable is strongly weighted.


Table 6. Multiple linear regression using aggregated videotaped measures to predict MSCORE (n=52).
Table of Content


Measure r with MSCORE Nominal p
Videotaped errors of observation 0.2392 0.672
Videotaped maneuver errors -0.1003 0.869


GENERAL DISCUSSION
Table of Content


This research produced an empirically-based, exposure-corrected classification of older drivers' errors during intersection negotiation; and, measured the relationships between performance on a battery of selected functional tests and driver competency, an hypothesized surrogate for crash risk. The present findings yielded a profile of negligent driving behavior at intersections that is highly descriptive, though only weakly predictive of crash risk, while underscoring both the potential and the limitations of functional capabilities testing in driver licensing or reexamination programs.


Data described herein give evidence that route familiarity--an assumed correlate and predictor of response expectancy--and driver intention (type of planned maneuver) interact to influence the likelihood of behavioral errors, but in different ways at intersections with different types of traffic control. Across all subjects, a common pattern and strikingly similar magnitudes of errors at signalized intersections were reduced from videotape for the standard ("unfamiliar") and for the home area ("familiar") test routes: the highest error rates were noted for movements straight through the intersection--roughly double those during turning movements--and slightly lower error rates were observed when right turns versus left turns were being performed. At stop-controlled intersections, on the standard test route, the highest error rates were again noted for through movements, and those observed during left turns again exceeded those during right turns. On the home area route, however, an error rate of eight percent (rounded) was observed during left and right turns alike. These key findings were displayed earlier in Table 3 on page 36.


It deserves mention that the demand on the driver to perform rapid, directed visual search behaviors for conflicts with unexpected entries into an intersection--where peripheral target detection and processing speed in a divided attention task combine to operationally define driving competency--is greatest for through movements, less for left turning movements, and lowest for right turns.


At intersections where traffic movements were neither protected nor prohibited (i.e., traffic control was always in a permissive state), the highest error rate was observed during performance of right turns on the standard test route. And in all cases, right turn error rates exceeded left turn error rates at these intersections. The opportunity for errors during through movements did not exist on the standard test route, either for uncontrolled intersections or for those marked only with a yield indication. Such opportunities did exist on the home area route--where driver expectancy was presumably strongest--and error rates during through movements were higher than during turning movements, as observed at the signalized and stop-controlled locations.


These data reinforce prolific anecdotal reports that traffic control devices have increased salience for older drivers. Again with reference to the data displayed in Table 3, signalization appears to outweigh all other factors in influencing drivers' intersection negotiation behaviors. At such locations, behavioral errors--when they occur--directly reflect the information processing demands for conflict avoidance under alternative maneuver scenarios. Results at stop-controlled intersections repeated this pattern, except in one respect: the increased expectancy associated with traversal of the home area test route (versus the standard route) was associated with an upturn in visual search errors in the lowest demand situation (right turns). Where there was a complete absence of traffic control devices, the most pronounced change in behavior observed (i.e., at uncontrolled intersections) was subjects' responses to (reduced) situational demand, particularly on the standard (less familiar) test route; performance under these conditions resulted in a peak in the rate of videotaped driving errors.


Older drivers would thus appear to exhibit greater competency (and thus experience lower risk) when traversing routes where traffic signals regulate movements at intersections. This conclusion applies to familiar and unfamiliar routes. Where older drivers are exposed to nonsignalized intersections, there would appear to be some benefit to restricting driving to frequently-traveled routes; at least, the relative exaggeration in difficulty with left turns documented in recent crash analyses (cf. Staplin and Lyles, 1991; Council and Zegeer, 1992) may diminish, to the point where such maneuvers are no more risky than any other movement.


One inevitable conclusion from these findings is that older and cognitively impaired drivers, like all drivers, commit many common errors both during the stage of information acquisition and in the execution of vehicle control movements that appear to have little bearing on the likelihood of crash involvement--or rather, that the variance that can be accounted for by differences in these behaviors will always be lower than that accounted for by situational factors. For example, almost all drivers failed to look both ways before entering intersections to execute a through maneuver during the green (permissive) phase, and instead, treated their movement as one that was protected. Such "common," or nondiscriminating errors are therefore poor candidates for the validation of screening indices, or for identifying individuals deserving one sort of intervention or licensing action from another. Dobbs (1997) similarly has advocated the segregation of nondiscriminating from discriminating or hazardous errors in the development and application of screening instruments for driving competency.


The present interest in functional testing--specifically, upon the relationship between physical and psychophysical response capabilities measured out of context, and driving competency (driving errors and error rate)--reflects a broad consensus that, while older drivers are overrepresented for certain crash types in exposure-based analyses, it is not age per se that governs crash risk. As the performance distributions for sensory function, and especially for perceptual and cognitive abilities flatten and extend in range for older cohorts of drivers, the need to control exposure for the most at-risk individuals while preserving as many mobility options as possible for all seniors grows more acute. Accordingly, there have been numerous efforts to develop and validate measures of functional status as predictors of crash risk in recent years (Ball, Owsley, Sloane, Roenker, and Bruni, 1993; Brown, Greaney, Mitchel, and Lee, 1993; Gianutsos, 1994; Hennessy, 1995; Keyl, Rebok, Bylsma, Turn, Brandt, Teret, Chase, and Sterns, in press; Johansson, Seidman, Kristoffersson, Lundberg, Lennerstrand, Hedin, and Viitanen, 1997; NPSRI, 1991; Stutts, Stewart, and Martell, 1998; Tallman, Tuokko, and Beattie, 1993; and Temple, 1989). In this research, a battery of measures including static and dynamic vision, perceptual and (divided) attention measures, and an index of neck flexibility for head rotation were examined in relation to driving competency (errors) demonstrated during on-road testing.


Interestingly, the dimension of functionality--response accuracy versus latency--that was most predictive of driving competency in this research varied according to the domain of functional abilities being tested. For sensory (visual) ability, it was response time rather than accuracy of response that related most strongly to MSCORE; but, in the domain of attentional and perceptual skills assessment, where a maneuver or vehicle control decision (performed in the context of a simulated driving scenario) was the measure of interest, the strength of relationship with MSCORE for response correctness was superior to response latency. Other analyses of an expanded set of data collected at the Santa Teresa, CA DMV site, as reported by Janke and Hersch (1997), confirm this pattern of results. Also, McKnight and Lange (1997) found that for cognitive tasks included in their Automated Psychophysical Test (APT) higher correlations were attained for error measures than for time measures, for both elderly referral and volunteer subjects. Under actual operating conditions, a driver must perform all stages of processing preceding a vehicle control movement (i.e., detection/recognition, decision, response selection/initiation) successfully to avoid a crash, and slower response speed at one stage may be compensated for by greater efficiency at another stage, such that a crash is avoided. A demented driver, by contrast, can have excellent sensory function yet commit a decision error from which there is no recovery. Thus, when measured against "critical" or "hazardous" error criteria as applied by examiners in the field, it makes sense that the measure of performance on the test battery that best discriminated between cognitively impaired versus cognitively unimpaired subjects would be most efficacious.


Limitations both in the reliability and feasibility of functional status measurement were also highlighted by this research. The counterintuitive finding of lower response accuracy for 20/200 acuity targets than for 20/40 targets was noted earlier; this artifact of stimulus presentation order draws attention to a broad range of potential biases from practice and learning effects in the administration of vision tests, and perhaps tests in other domains as well. Such threats to reliability must be anticipated, and test methodologies carefully designed to counter them, in any formal screening program which affects licensing decisions.


Next, the MultiCAD battery of functional tests administered in the Santa Teresa CA DMV facility required an average time of approximately 40 minutes per subject(2). For a screening tool, this is excessive. Of course, the present intent was to explore relationships involving a wider set of measures than would be envisioned for adoption in any formal program--or would be administered to any single driver. Yet a practical constraint on time-per-driver for functional testing that could be limited to 30 minutes in many jurisdictions, and as little as 5-10 minutes in some(3), mandates that only those screening measures and test procedures that are highly sensitive and specific will be viable. Unfortunately, the present results do not permit such test selection.


Further questions emerging from this study concern the methodologies best suited to obtain functional status measures. There are alternative approaches to the measurement of an overlapping set of constructs, some of which rely on proprietary materials or technology and some of which are in the public domain. There is a clear need to systematically document the functional capabilities assessed by each battery, device, or program applied in published reports as a driver screening tool; identify their unique contributions as well as their shared measurement objectives and techniques; and contrast their respective strengths and weaknesses. Certainly, test reliability and predictive validity are key. Also, as a practical matter, the ability to "bundle" tests which meet these criteria--but are now administered using separate materials and/or apparatus--on a common platform offers obvious advantages for implementation in a DMV setting. And, to the extent that techniques for performing such functional assessments can utilize in-context measures of functional capability, with realistic test stimuli, public acceptance of their results as determinants of licensing actions is likely to be heightened.


To help realize the potential of functional screening for driving competency, and contribute to efforts now underway to update a national model for driver screening and evaluation in the U.S.(4), several additional guidelines may be suggested. With a specific focus on individuals referred to a DMV through any one of a growing network of public agencies, private caregivers, and friends and family members, and who are thus identified as posing a potentially disproportionate risk to themselves and others, a goal of testing only a set of "minimum qualifications requirements" in core competencies is indicated. By avoiding any attempt to precisely measure functional ability level and instead seeking only to insure that a performance threshold is met, economies in test administration time, standardization in test procedures, and the perceived equality of all those who meet the common criterion without regard to driver age all become more easily achieved. In addition, a decision model must be developed that allows only those screens that are most appropriate to the particular diminished capabilities of a particular older driver to be invoked--vision screens for individuals referred for low vision problems, cognitive screens for those suspected to suffer from dementia, and so forth.


Finally, a more general observation on efforts to employ functional screens to guide licensing actions by State agencies may be in order, i.e., a diminishing return in the attempt to refine functional status indicators such that the relative crash risk of individuals can be reliably discriminated seems likely. The classic study by Salthouse (1984) is instructive, which measured component behaviors of typing (e.g., interstrike interval between keys, reaction time, etc.) in an effort to correlate age and overall typing speed. Older typists showed significant declines on the component measures, but there was age equivalence in typing speed. It was demonstrated that the older subjects processed a larger string of characters in advance than younger subjects, and that this compensatory mechanism at the tactical level allowed them to maintain equivalent performance despite declining efficiency for the component processes of the task. The finding in Waldman and Avolio's (1986) meta-analysis that there was no consistent relationship between age and job performance carries a consistent message. For complex, expert, highly-practiced behaviors--such as driving--declines on component behaviors do not lead directly and unavoidably to degradation at a molar level of task performance. The number of times an individual failed to scan to the sides before moving straight through an intersection on a green signal (ball) was plainly unrelated to an aggregate index of driving competency as operationally defined in this research. Thus, beyond the detection of gross impairments at the sensory or cognitive level--which can often be discerned through direct observation using relatively quick and inexpensive procedures--the predictive value even of statistically-significant differences in component functional processes for "bottom line" measures of driving safety under actual field operating conditions may continue to disappoint.

1. In the Janke and Hersch analysis for measures scoring accuracy as 0 vs.1, subjects correct on at least 2 of the 3 trials at each stimulus level were scored 1; otherwise, they were scored 0.

2. MultiCAD was one test battery in an overall program of functional testing performed by the CA DMV which required 2-2 hours per subject.

3. American Association of Motor Vehicle Administrators (AAMVA) Survey of United States and Canadian Provinces, Draft Report, May, 1997.

4. cf. U.S.DOT Contract DTNH22-96-C-05140, "Model Screening and Evaluation Program."