A primary purpose of lighting a roadway at night is to increase the visibility of the roadway and its immediate environment, thereby permitting the driver to maneuver more efficiently and safely. The visibility of an object is that property which makes it discernible from its surroundings. This property of an object depends on a combination of the following factors: (1) the differences in luminance, hue, and saturation between the object and its immediate background (contrast); (2) the angular size of the object at the eye of the observer; (3) the luminance of the background against which it is seen; and (4) the duration of the observation.

Historically, two complementary measures of lighting system performance have been employed: (1) illuminance, or the amount of light from an installation incident upon a given surface of interest (visibility target) in the roadway environment, and (2) luminance, or the amount of reflected light returned to the driver’s eye from the visibility target.

The optimal design of highway lighting systems incorporates photometric properties of light sources, lighting geometry, targets, road conditions, road surfaces, and surroundings. These are the physical properties of a lighting system that are important for defining the visual stimulus. Once the stimulus is defined, the visibility of targets can be calculated using models that are based on psychophysical data on the visual processing of spatial (angular size of target), temporal (exposure duration) and spectral (brightness and color) information. Visibility models must also incorporate age-related changes in visual processing efficiency that have pronounced effects on target visibility.

Visibility factors are extremely important in the design of highway lighting. Illuminance criteria have been proven to be inadequate predictors of the effectiveness of lighting systems. Although the visibility of targets is typically directly proportional to illuminance (all other variables held constant), there are too many intervening variables that determine the visual stimulus and the efficiency with which that stimulus is processed by the visual system. Even if visibility criteria are used in the design of lighting systems, this is not always predictive of lighting system effectiveness, when accident rates are used as the measure of effectiveness (MOE). For example, there have been some reports of accident rates increasing after installation of fixed lighting systems (Gordon and Schwab, 1979). Although the reasons for this are not well-understood, it is hypothesized that some lighting systems can actually reduce the average contrast of targets even though they meet lighting specifications based on pavement illuminance, set prior to 1982 (Keck, 1989).

Lighting system design dictates not only the amount of light provided by an installation, but also its distribution on the pavement and the amount of glare experienced by drivers. In addition, light distribution critically affects the contrast of targets viewed by drivers. As a case in point, a lighting system in which luminaire height is low and angular coverage is high produces a wide zone over which a reversal in contrast polarity occurs.

Negative contrast results when the luminance from behind is higher than that from the front. When the target is placed slightly beyond the distance of the luminaire providing the luminance from behind (i.e., more light is directed to front of target), the contrast polarity changes to positive (i.e., luminance from front is higher than background). In this example, where luminaire height was the variable of interest but the spacing of the luminaires was not considered, the distance over which this contrast reversal takes place is large because the change in luminance with distance is gradual. In an alternate example, luminaire spacing could be varied; and, for any given luminaire height, closer spacing would also tend to produce larger zones over which targets undergo contrast reversal. Clearly, the design of lighting systems must take the interplay of many factors into account, since different combinations of lamp type, luminaire housing (cutoff for glare control), and height and spacing of luminaires all contribute to the overall system output and jointly determine the visibility of targets at a specified location in the roadway. Thus, lighting designs that produce short, abrupt contrast reversal zones are recommended. This recommendation is contrary to existing lighting uniformity standards that call for maximum changes in illuminance of 3:1 (average-to-minimum luminance in commercial areas) to minimize visual demands as well as psychological and physiological stress (FHWA, 1983). Optimizing the tradeoff between these variables will be discussed in more detail in following sections. This example illustrates the importance of understanding and applying visibility factors in the design of roadway lighting.

The justification for highway lighting is in terms of a cost savings due to accident reduction. Although estimates vary, the savings can be enough to pay for a lighting installation in a few years. Estimates by Box (1989) indicated that lighting can reduce the ratio of night-to-day accidents by as much as 14 percent of total accidents. In a more recent analysis by Griffith (1994), the safety benefit was found to be much higher, with an accident reduction of 32 percent for (property-damage-only accidents). As a result of this study, which is discussed in more detail later in this section, Griffith estimated that the cost benefit of roadway lighting between 1985 and 1990, for roadways included in the analysis, was about $6,000,000.

The cost benefits cited by previous studies is typically in terms of property damage. Perhaps more important is the reduction in fatality rates for drivers, as well as pedestrians and cyclists. In a quasi-experimental analysis of the Fatal Accident Reporting System (FARS) data base, Owens and Sivak (1993) used a unique approach to isolate the contribution of visibility factors to accidents. In this accident analysis, they compared accidents that occurred during lighter (i.e., higher visibility) and darker (i.e., lower visibility) periods of twilight. A key assumption was that factors other than light condition (such as traffic density, driver-age mix, fatigue and alcohol consumption) were relatively stable during these light/dark periods of twilight and thus not confounded in the analysis. The main conclusion from the analysis was that poor visibility is related to fatal accidents involving pedestrians and cyclists and bears little relation to other types of fatal accidents. This was attributed to the widespread use of marker lights and reflective materials on vehicles and fixed obstacles.

Using meta-analytic techniques, Elvik (1995) combined the results of 37 accident studies that assessed the safety benefit of highway lighting. Meta-analysis is an analytical tool that allows reviewers to combine the results from independent studies to determine if there is an overall effect of lighting. Each study is treated as an individual data point. From the 37 studies included in the meta-analysis, 142 results were included. Of the 142 results, 97 were not statistically significant at a=0.05. However, the effect of public lighting depended on accident severity and type of accident. Fatal accidents were reduced by 65 percent due to public lighting whereas nighttime injury and property-damage-only accidents were reduced by 30 percent and 15 percent, respectively. Roadway lighting appears also to have a large safety benefit for pedestrians and intersections, which agrees with the conclusions of Owens and Sivak (1993). The 15 percent reduction in accidents for property-damage-only accidents does not agree with Griffith's (1994) 32 percent estimate, which specifically isolated property-damage-only accidents.

CIE (1990) reports that road accidents at night are disproportionately higher in number and severity compared to daytime. Data from 13 OECD countries showed that the proportion of fatal nighttime accidents ranged between 25 and 59 percent (average value of 48.5 percent). Estimates of vehicle kilometers traveled during the hours of darkness ranged from 17 to 32 percent (average of 25 percent). Although factors such as increased alcohol usage, fatigue, and overrepresentation of young drivers in night traffic contribute to the problem, the major factor is darkness, demonstrated in accident studies in which the effect of all factors has been taken into account (CIE, 1990). In an evaluation of 62 lighting and accident studies from 15 countries, 85 percent of the results showed lighting to be beneficial, with approximately one-third of the results statistically significant. CIE (1990) concludes, therefore, that roadway lighting is successful; however, the installation of lighting cannot be expected to result in a reduction in accidents if there is a major non-visual problem at any particular site.

Lipinski and Wortman (1976) found that illumination resulted in a 45 percent reduction in the night accident rate and a 22 percent reduction in the night accident/total accident ratio at rural at-grade intersections. The 445 intersection data years were categorized according to the following two variables: presence or absence of illumination, and presence or absence of channelization. Channelization is frequently used in connection with rural intersection improvements, and illumination and channelization improvements are frequently undertaken at the same time. Channelization was therefore included as a variable because of the effect it has on the roadway environment. The mean night accident/total accident ratio when both lighting and channelization were present was lower than lighting without channelization, channelization without lighting, and no lighting/no channelization locations. It was concluded that the simultaneous introduction of channelization and illumination at locations experiencing a high number of nighttime accidents should be encouraged.

Similarly, Walker and Roberts (1976) found a 52 percent reduction in nighttime accidents at 47 intersections, in a 6-year before and after study. The analysis of the effect of lighting on channelized intersections (n = 28) versus nonchannelized intersections (n =19) showed a significant reduction in the accident rate for channelized intersections after lighting was installed. Before lighting was installed, the channelized intersections performed in a less satisfactory manner than the non-illuminated, non-channelized intersections. Although lighting improved the performance of both intersection geometries, the difference in accident rates favored the channelized intersections. An analysis of 21 intersections with one or more routes entering in one direction but departing in another direction showed the highest before-lighting accident rate. Accident history after the installation of lighting for the 21 turning intersections showed a significant reduction in accident rate, resulting in no significant difference between accident rate for lighted turning intersections and lighted non-turning intersections. Lighting also reduced the accident rate for four-legged intersections, and reduced the number of accidents (but not significantly) for three-legged T and Y intersections. Complex intersections, which were also associated with increased numbers of lights (6 to 9 and 10 to 15) showed significant reductions in nighttime accidents, although even less-complex intersections with only 3 to 5 lights showed night improvement with the addition of lighting. The greatest improvement in the nighttime accident rate was shown for intersections with a traffic volume of 3,500 vehicles/day or greater.

Stark (1992) cites research done by Wortman (1972), who conducted an investigation of rural at-grade intersection illumination to develop warrants for lighting. It was determined that rural intersections should be considered for lighting if the average number of nighttime accidents (N), per year, exceeds one-third of the average number of day accidents (D). All the accident data available since the date of the last modification to the intersection should be used when calculating these averages. If N is greater than D/3, the likely average benefit should be taken as N-D/3 accidents per year. It is further recommended that illumination be provided whenever an intersection is channelized. In this investigation, it was found that states with rural intersection illumination programs designed their installations to provide for 0.6 to 1.5 maintained horizontal footcandles (6 to 16 lux). AASHTO (1984, 1994) notes that lighting of spot locations in rural areas should be considered whenever the driver is required to pass through a section of road with complex geometry and/or raised channelization, and may require higher levels than those specified in the Informational Guide to Highway Lighting (AASHTO, 1984).

With regard to freeways, several research studies have documented both that freeway interchanges experience a higher accident rate than the mainline and that urban freeway lighting has beneficial safety effects. Box (1972) found that the average night/day accident rate was 66 percent greater on unlighted freeways than on lighted freeways, and that night accidents could be reduced by an average of 40 percent by the illumination of an unlighted freeway. Cirillo (1968) found that the highest accident rate along a freeway occurred within 305 m (1,000 ft) of an exit ramp nose or an entrance ramp merging end. Cirillo also found a reduction in the number of interchange accidents as lighting intensity increased. Gramza, Hall, and Sampson (1980) evaluated the interchanges in the Interstate Accident Research (ISAR-2) data base at which lighting had been introduced during the 10-year study period in a before-after design. During the daytime, there were 83 accidents before lighting and 80 accidents after lighting. At nighttime, by comparison, there were 76 night accidents before lighting and 43 accidents after lighting. In an accident analysis conducted in the 1980's, Young (undated) found a 35 percent increase in nighttime interchange ramp accidents in Wisconsin when the lighting was reduced from complete to partial for seven major interchanges and from complete to none at the balance of interchanges studied. Taylor and McGee (1973) found a reduction in erratic maneuvers at exit lane drop sites in a before-after study, when the exit area was illuminated during the "after" period of data collection. Taragin and Rudy (1960) found that the manner of night use of speed change lanes, particularly the acceleration lane, improved with increased illumination, and that some beneficial results of illumination in the deceleration area were derived when illumination was used at the full level. They noted that the best service is provided when illumination is combined with roadside delineation.

Gramza, Hall, and Sampson (1980) conducted an accident analysis of 400 nighttime accidents which occurred at 116 interchanges between 1971 and 1976 in 5 States (Maine, Maryland, Minnesota, Texas, and Utah). In an analysis of the presence of high mast lighting at interchanges, versus no lighting or other kinds of interchange lighting, the presence of high mast lighting was found to significantly reduce total accident rates, total accidents involving fatalities and injuries, and accidents involving fatalities and injuries other than the vehicle-to-vehicle and vehicle-to-fixed object categories (e.g., accidents caused by striking pedestrians).

Next, it was found by Gramza et al. (1980) that although the number of lights at an interchange and the level of illumination had no significant effect on the total number of nighttime accidents, significant decreases in a variety of distinct accident types were found with increases in illumination. Increases in the illumination level—horizontal foot candles— (HFC) at interchanges were associated with significant reductions in two types of accidents: vehicle-to-fixed object accidents involving property damage; and vehicle-to-vehicle accidents involving fatalities and injuries. In addition, increases in the number of lights active at an interchange was found to significantly influence (reduce) the following two accident types: vehicle-to-fixed object accidents involving fatalities and other injuries; and other property damage accidents. The number of lights at an interchange ranged from none to 114, with an average of 16 active lights and a median of 10. Thirty-two percent of the interchanges were unlit. As lighting levels increased, accident rates decreased. Illumination ranged from 0.0 HFC to 1.0 HFC, with an average of 0.51 HFC for the lighted sections. These four accident types accounted for 61 percent of the accidents observed in the sample.

Since there were relatively few accidents per interchange per year, Gramza et al. (1980) employed a model to predict the frequency of occurrence of each accident type per year, assuming three levels of traffic volume (average nighttime traffic of 5,000, 7,500, and 10,000 vehicles) at certain types of interchanges and assuming varying levels of illumination or varying numbers of lights, holding the year (1975) constant. The predicted relationships between traffic volume, accident frequency and lighting for each of the conditions for which illumination or number of lights was found to have a significant effect on accidents are described below.

Vehicle-to-vehicle accidents per interchange which resulted in a fatality or injury averaged 0.37. A decrease of 0.5 horizontal foot candles (HFC) in lighting would result in 0.13 more of this type of accident per interchange. At an average of 5,000 vehicles traffic volume at night, a reduction of illumination from 0.5 HFC to 0.0 HFC results in an increase from 0.07 to 0.23 accidents per interchange in partial cloverleaf interchanges and an increase from 0.38 to 0.50 accidents per interchange in all other interchanges. Doubling the volume to 10,000 vehicles, the same illumination reduction would result in an increase of from 0.44 to 0.57 accidents per interchange year in partial cloverleafs and an increase from 0.75 to 0.87 accidents per interchange-year in all other interchanges.

Vehicle-to-fixed object accidents per interchange which resulted in property damage averaged 0.064 accidents per interchange/year. A 0.5 reduction in illumination would increase interchange accidents by 0.29. For all interchanges, a reduction in illumination from 1.0 HFC to 0.5 HFC would result in an increase from 0.12 accidents per interchange to 0.41 accidents per interchange at a 5,000-vehicle traffic volume. At 10,000 vehicles, the predicted change would be from 0.82 to 1.11 accidents per interchange. For all traffic volumes, the accident rate decreased with increasing light levels.

Vehicle-to-fixed object accidents which resulted in fatalities or injuries averaged 0.38 per interchange across all observations. With regard to the effect of the number of lights on accidents per interchange, a reduction of 20 lights would result in an increase in 0.15 accidents per interchange. For a traffic volume of 5,000 vehicles, a reduction in the number of lights from 40 to 20 would increase accidents at diamond interchanges from none to 0.11 per interchange, accidents at cloverleafs from 0.56 to 0.71 per interchange, and all accidents at all other interchanges from 0.14 to 0.29 per interchange. At a traffic volume of 10,000 vehicles, the same reduction in number of lights would result in an increase of from 0.37 to 0.53 accidents per diamond interchange, from 0.97 to 1.12 accidents per cloverleaf, and from 0.55 to 0.70 accidents per other interchange design type.

For other property damage accidents, a 20-light reduction was predicted to yield a 0.12 increase in accidents per interchange. With a traffic volume of 5,000 vehicles, a reduction from 40 to 20 lights would result in an increase of from none to 0.04 accidents per urban interchange and from 0.34 to 0.46 accidents per non-urban interchange. With a traffic volume of 10,000 vehicles, the same reduction in lights would result in an increase from 0.47 to 0.59 accidents per urban interchange and from 0.88 to 1.00 accidents per non-urban interchange. For all traffic volumes, the accident rate decreased with increasing light levels.

The most recent study evaluating the effects of urban freeway lighting on highway safety was conducted using the Highway Safety Information System (HSIS) data, and compared the safety of continuously lighted urban freeways with interchange lighting only (Griffith, 1994). Study sections were located in the Minneapolis-St. Paul metropolitan area and included 88 km (55 mi) of urban freeway segments with continuous lighting, and 57 km (35 mi) of urban freeway sections with interchange lighting only. Accident experience tended to be higher in interchange areas than in non-interchange areas. Results showed that the total night/total day accident ratio for the sections with interchange lighting only was 12 percent higher than the total night/day accident ratio for sections with continuous lighting. This finding was significant at the 0.05 level of significance. For interchange areas the night/day accident-rate ratios were statistically equal for continuously lighted sections and sections with interchange lighting only. By comparison, the night/day accident rate for non-interchange areas was 18 percent higher for sections with interchange lighting only than it was for continuously lighted sections. It was also found that the illumination of an unlighted urban freeway between interchange areas could theoretically reduce night accidents by 16 percent; the relative benefit of urban lighting between interchanges was associated primarily with property-damage-only accidents.

Accident studies that have investigated age-related benefits of highway lighting are of particular interest in this review. In an analysis of day and night accident rates (involvements per million miles), Massie and Campbell (1993) suggest that the nighttime accident involvement rate is much higher for younger drivers relative to daytime accident involvement. Drivers of all ages combined experienced 10.37 fatal involvements per 100 million miles at night and 2.25 during the day in 1990 (Massie and Campbell, 1993). In this analysis, the difference between daytime and nighttime fatal rates was found to be more pronounced among the younger age groups than the older ones, with drivers age 20 to 24 showing a nighttime rate that is 6.1 times the daytime rate and drivers age 75 or older showing a nighttime rate only 1.1 times the daytime rate. Data from 1985 show that drivers over age 64 have 15 to 30 percent of fatal accidents in darkness, whereas drivers as a whole have 50 to 60 percent of their accidents in darkness. The lower percentage of nighttime accidents of older drivers may be due to a number of exposure factors, e.g., through self-restriction, this group is driving less at night (NHTSA, 1987). In fact, the 1983 Nationwide Personal Transportation Study (NPTS) showed that only 14 percent of miles driven by those over 65 were at night compared to 26 percent for younger drivers. This may also represent a self-selection process in which those older drivers with low glare susceptibility and better-than-average nighttime contrast sensitivity are choosing to drive at night (i.e., the "most fit" among the older age cohort). It may also be the case that older drivers who drive at night compensate for any visual decrements by driving slower or engaging in other self-regulating behaviors that tend to minimize their overall workload.

Another approach to justifying the requirement for fixed roadway lighting is through a comparison of visibility relative to that provided by vehicle headlights. Improved headlight design has advantages over fixed roadway lighting systems in terms of cost per mile, feasibility, and safety in run-off-the-road accidents (via no lamp post to hit). There are ongoing developments in headlight technology that may enhance the visibility provided by vehicle headlights such as polarized headlights, ultraviolet headlights, high-intensity discharge lamps, focused beams, and hybrid devices that employ combinations of these technologies. Until these visibility enhancing technologies are implemented, analyses using the CIE analytical model suggest that fixed roadway lighting systems provide between 7x (low contrast conditions in presence of opposing vehicle) and 350x (high contrast in absence of opposing vehicle) more contrast under different target conditions (Staplin, Janoff and Decina, 1985). In terms of providing contrast to drivers at safe detection distances (198 m [650 ft] based on dry pavement conditions, a vehicle speed of 100 km/h [62 mi/h], and reaction time of 2.0 s), the analysis suggests that visibility provided by highway lighting is at least an order of magnitude higher than vehicle headlights alone.

While the benefits of highway lighting are unequivocal, the accident studies to date do not provide compelling evidence that there is an added benefit to older drivers. An analysis of light/dark twilight lighting conditions for different age groups and accident types is needed before any predictions can be made about the extent to which roadway lighting practices impact older driver safety. The fact that highway lights provide better target contrasts at safe detection distances is justification enough for using highway lighting. However, particular attention must be given to the increased susceptibility of older individuals to glare and reduced ocular transmittance. While age-related changes in glare susceptibility and contrast threshold are currently accounted for in lighting design criteria, there are other visual aging effects that are currently excluded from visibility criteria. These changes and the implications for lighting design are described in the following section.

Intersection Lighting. Rockwell, Hungerford, and Balasubramanuan (1976) studied the performance of drivers approaching four intersection treatments of special reflectorized delineators and signs, or illumination. A significant finding from observing 168 test approaches was that the use of roadway lighting significantly improved driving performance and earlier detection of the intersection, whereas signing, delineation and new pavement markings showed marginal changes in performance. Roadway lighting systems reveal objects in several different ways. Changes in the location of luminaires (such as reducing or increasing the spacing between poles, changing the mounting height, etc.) change the pattern of brightness as seen by the driver. All surfaces will differentially reflect the incident light depending upon the relative angles of incidence and observation. Therefore, as the driver proceeds down a roadway, the road itself and objects on it are observed under a lighting system that appears to move relative to the driver. The luminance of the object and the background it is seen against will change as the driver moves through the lighting system and the angle of observation changes. The primary luminance of vertical surfaces of objects as seen from the driver's view point are normally produced by different luminaires than the background road surface against which they are seen. Thus, changing the luminaire arrangement geometry and/or the light distribution of one or more of the luminaires may affect the contrast between objects and their background significantly while leaving the amount of light falling on the road unchanged.

Freeway Interchange Lighting. Janoff, Freedman, and Decina (1982) conducted a study to determine the effectiveness of partial lighting of interchanges, where partial interchange lighting (PIL) was defined as lighting that consists of a few luminaires located in the general areas where entrance and exit ramps connect with the through traffic lanes of the freeway (between the gore and the end of the acceleration ramp/beginning of the deceleration ramp). A complete interchange lighting (CIL) system includes lighting on both the acceleration and deceleration areas plus the ramps through to the terminus. In their survey of approximately 50 agencies which supplied information on over 14,000 interchanges and over 7,500 interchange lighting systems, it was found that 37 percent of the interchange lighting was complete interchange lighting, and 63 percent is partial interchange lighting. An observational field study was conducted to determine the effects of lighting level (various levels of PIL, CIL, no lighting, and daylight), geometry of the interchange (straight versus curved ramps), and presence of weaving area versus no weaving area on driver behavior and traffic operations. Partial interchange lighting was stratified by the number of lights at each ramp, and included three levels: PIL 1 (one light), PIL 2 (two lights), and PIL 4 (four lights). CIL test sites included a full cloverleaf in suburban Baltimore, MD and a three-leg interchange in suburban Philadelphia, PA with luminaire mounting heights of 12 and 9.4 m (40 and 31 ft), respectively. The dependent measures included: speed and acceleration of individual vehicles traversing the interchanges; merge and diverge points of individual vehicles entering the main road or leaving it; and erratic maneuvers such as brake activations, use of high beams, and gore or shoulder encroachments. Generally, both field studies indicated that CIL provided a better traffic operating environment than did PIL, and that any interchange lighting performed better than no lighting (although the differences were not always as great as between CIL and PIL). Results are presented below for each dependent measure.

Brake activations (the key measure of effectiveness) are dependent on the visual quality in the interchange and are directly related to both the safety and smoothness of traffic flow in the interchange area. There were more brake activations under the no lighting condition than either the PIL or CIL condition, and higher brake activation frequencies were observed for PIL than for CIL. At the exit ramp at the Maryland test site, there was a definite increase in the frequency of brake activations as the lighting was reduced from CIL to PIL 2 to PIL 1 to no lighting. For the entrance, the frequencies were 2 to 2.5 times higher under PIL and no lighting than CIL. Additionally, the location of the brake activations was shifted later in the travel path when the lighting was reduced from CIL to PIL.

Driver uncertainty due to inadequate visibility was measured by the frequency of use of high beams. High-beam use increased as the lighting was reduced from CIL to PIL to no lighting on three of the four ramps.

Merging and diverging patterns directly relate to the smoothness of traffic flow, and were observed to worsen as the lighting was reduced from CIL to PIL. At the exit ramps, drivers diverged later as the lighting decreased from CIL to PIL to no lighting. At the main site, where weaving was required, the distance downstream of the gore at which diverging occurred under CIL was about 20 percent less [8.5 m (28 ft)] than under no lighting, and 10 percent less (4.8 m [16 ft]) than under the average of all PIL conditions. At entrance ramps, the mean point of merging also increased as lighting decreased. Under CIL the merging point was 11 percent less ( 9.75 m [32 ft]) than under no lighting and 10 percent less (8.8 m [29 ft]) than under PIL. These distances are equivalent to approximately 0.5 s at highway speeds. At this site, merging was required within 152 m (500 ft), thus a decrease in merge distance was considered better from an operational standpoint. At the Pennsylvania site, exit drivers diverged later under PIL conditions than under CIL conditions and merging drivers merged earlier under PIL conditions than under CIL conditions. Late divergence was 30 percent higher for PIL than for CIL. The visibility of the beginning of the deceleration ramp was 11 times higher under CIL than under PIL conditions. At this site, merging was relatively unconstrained as the entrance ramp/acceleration ramp continued for a few thousand feet, allowing drivers to merge at their leisure. The changes in distance at which diverging and merging occurred, according to the researchers, was probably indicative of better visibility under CIL, which allowed drivers to see the exit ramp or main lines more easily and thus diverge or merge sooner. Average velocities and accelerations, dependent measures which are also directly related to the smoothness of traffic flow at an interchange, were not affected by the changes in visual quality from CIL to PIL.

Gore and shoulder encroachments (erratic maneuvers) occurred more often at the Pennsylvania site as lighting was reduced from CIL to PIL. At the Maryland site, there were no gore encroachments and there was no consistent pattern of shoulder encroachments as a function of lighting.

It was concluded that if traffic flow and safety are important issues, then existing CIL systems should not be reduced to PIL systems. When installing new lighting and economics are not an overriding issue, then a CIL system is preferred over a PIL system. However, a PIL system with one or two luminaires per ramp will normally perform better than no lighting at far lower cost than a CIL system. PIL systems with fewer luminaires (one or two) frequently performed better than PIL systems with greater numbers of luminaires (four). This was explained by the fact that drivers may experience transitional visibility problems under the PIL conditions when they are forced to drive from dark to light to dark areas and at the same time perform complex maneuvers such as diverging, merging, and tracking a 90° curve.

Hostetter, Crowley, Dauber, and Seguin (1989) note that because luminaires are not usually placed downstream of the physical gore of a partially lighted exit ramp, a driver proceeds from a lighted area to a nonlighted area on the ramps. Citing evidence from various researchers (Boynton and Miller, 1963; Boynton, 1967; Boynton, Rinalducci, and Sternheim, 1969; Boynton, Corwin, and Sternheim, 1970; Rinalducci and Beare, 1974; and Frederickson and Rotne, 1978), they report that the effect of going from higher to lower levels of luminance results in a reduction in visual sensitivity, which would explain the findings of Janoff et al. (1982) that performance under partial lighting was better with fewer luminaires.

Hostetter, Carter, and Dauber (1993) also recommended that to minimize the effects of transient adaptation, the number of refractor type luminaires used for partial lighting be kept to a maximum of three. In this study, 25 subjects between the ages of 18 and 59+ drove an instrumented vehicle on entrance and exit ramps where the following variables were manipulated: number of luminaires (no lighting, one luminaire, two luminaires, three luminaires, or four luminaires), luminaire type (refractor versus sharp cut-off), lateral target placement (left shoulder or right shoulder) and longitudinal target placement (near and far). The sharp cut-off type luminaires are designed to reduce glare. The near target placement was 61 m (350 ft) downstream of the last luminaire (which was 45.8 m [150 ft] past the influence of the luminaire and approximates the design of partial lighting on sharply curved ramps) and the far target placement was 144.9 m (475 ft) from the last luminaire. The targets were placed 0.61 m (2 ft) from the left or right edgeline on the shoulder. The objective was to determine the extent to which transient visual adaptation (TVA) affects drivers' detection of targets along partially lighted freeway interchanges. TVA occurs when the eye moves from one luminance level to another and is continuously adapting to higher and lower levels, thereby temporarily reducing contrast sensitivity A reduction in contrast sensitivity can reduce the probability of detecting a target on the road or roadside. No analyses were performed to determine the effects of age. This study was concerned with evaluating the effects of luminaire type and number on TVA, and not with the same driving performance measures studied by Janoff et al. (1982).

Hostetter et al. (1993) found that any number of luminaires (both refractor and sharp cut-off) produced a transient effect, reflected in shorter target detection distances associated with lighting for the near targets. In other words, the target detection distance was greater when there was no lighting. Similarly, for the refractor luminaires and the far targets, the difference in mean target detection between any number of luminaires and no luminaires was significant at the p=0.01 level, where no lighting produced the farthest target detection distances. Drivers traveling a partially lighted section detected a roadside target 0.6 to 1.3 s later than was the case without lighting. There were also significant differences between four luminaires and one, two and three luminaires, where the mean detection distance associated with four refractor type luminaires (112.8 m [370 ft]) was substantially shorter than any of the other far target detection distances. For the sharp cut-off luminaires and far targets, there was very little difference between the target detection means for one through four luminaires; the maximum mean difference was 3.3 m (11 ft). The effect of target location was consistent for both the refractor and the sharp cut-off type luminaires; the far target location resulted in longer detection distances than those for the near targets, regardless of the number of luminaires. The fact that the detection distances associated with the far targets were closer to the detection distance observed under nonilluminated conditions, indicates that the TVA effect is partially dissipated as the driver approached the target located 144.7 m (475 ft) from the last luminaire. Since the difference between the near and far targets was 38 m (125 ft), and the subjects were driving at 72.4 km/h (45 mi/h), the TVA effect began to dissipate in less than 2 s. With the exception of the four luminaire/refractor type condition, the near-far differences in mean detection distance were statistically significant. The authors explain that this exception indicates that with the longer exposure to light (four luminaires) and the higher levels of glare associated with the refractor-type luminaires, the TVA dissipation effect noted for most conditions does not occur. In an earlier study conducted by Hostetter, Crowley, Dauber, and Seguin (1989) at the same interchange, the TVA effect associated with four luminaires of the refractor type was totally eliminated by the time the driver approached a point 182.9 m (600 ft), approximately 9 s, beyond the last luminaire. Finally, with regard to type of luminaire, an analysis found no statistical difference in target detection distance, even though subjects reported improved visibility and comfort associated with the sharp cut-off luminaires. The authors caution that this finding may have resulted from the requirement to manually adjust the sharp cut-off luminaires due to erroneous information provided about the luminaire mounting height at the test site, to bring the illuminance performance closer to the legal requirements set by FHWA and AASHTO. It was noted that veiling glare from the sharp cut-off installation was considerably lower than that found with the refractor type, and system glare (light emitted at high angles) is virtually nonexistent with the sharp cut-off units. Hostetter et al. (1993) recommended that if more than three luminaires are required to meet the visibility need at a particular location, then they should be of the sharp cut-off type, since this type of luminaire produced a less severe transient effect.

Freeway Mainline Lighting. The effect of reducing lighting on the freeway mainline has been studied by several researchers. When lighting on one side of a Texas Interstate highway was eliminated as an energy savings measure, night accidents on an unlit side increased by 47 percent, whereas accidents on the lit side and daytime accidents decreased during the study period (Richards, 1979). A controlled field study (Staplin, Janoff, and Decina, 1985) was performed to determine the effect on driver behavior and safety of extinguishing mainline freeway lighting during nighttime periods of low traffic volume (part-night lighting), in which 24 subjects between the ages of 20 and 67 encountered and responded to an 18-percent-reflective, 15.24-cm (6-in) "bullet" target at 9 sites while driving on inservice urban and suburban freeways. Six lighting conditions were selected for evaluation: full lighting (200-W high-pressure sodium lamps, mounted 9.1 m [30 ft] at spacings of between 20.7 and 26.8 m [68 and 88 ft], in a staggered arrangement); dimming to 30 percent light output; dimming to 50 percent light output; extinguishing every other luminaire; extinguishing luminaires on one side of the roadway where luminaires are located on both sides of the roadway; and no lighting. An experimenter accompanied each partially-alerted subject on the road course. The subject indicated target detection by pressing a steering wheel-mounted switch, which recorded the distance to the target. The MOE was the distance at which drivers could first detect a simulated hazard in the roadway ahead of their vehicles. It was hypothesized that as the lighting was reduced, the distance at which the driver detected the simulated hazard would decrease, and this change would serve as a surrogate measure of the relative changes in safety with the various reduced lighting techniques.

The best detection performance in the Staplin et al. study was achieved under the full lighting condition (mean detection distance = 87.8 m [288 ft]), with orderly decrements in performance noted for the uniform-dimming-to-75-percent-power (71 m [233 ft]), uniform-dimming-to-50-percent-power (68.3 m [224 ft]), and every-other-luminaire-extinguished conditions (62.5 m [205 ft]), respectively. The one-side-extinguished and the no-lighting conditions showed the poorest performance, with a mean detection distance in each condition of 49.7 m (163 ft). The surrogate detection measure for reduction in safety was statistically significant for the all-off and one-side-only lighting techniques, but not for the dimmed tactics and the every-other-off tactic. The highway safety implications of this research may be gauged in terms of AASHTO minimum braking distance requirements, in which design braking distance is as follows:

[7] 

Under the existing dry and level conditions with a worn portland cement surface and using average new tires, the detection distances obtained under full lighting and both uniformly dimmed lighting conditions, all exceed the current design requirement, and the distance-to-target results associated with the every-other luminaire condition provided for 98 percent of the minimum requirement. However, when worst-case conditions are considered (e.g., wet pavements and worn tires), none of the detection distances obtained in the study were large enough to ensure a safe stop. Janoff and Staplin (1987) note that because design criteria within this context cannot reasonably be fixed in terms of the absolute worst-case scenario, and because a motorist's response to a perceived hazard while driving on a freeway is typically a combination of braking and lateral movement (and not braking alone), a generalized distance-to-target criterion was calculated in this study to express the relative impact of the alternative reduced lighting tactics on highway safety. This generalized criterion was the average of the existing and worst-case braking requirements, based on the presented braking distance formula. A value of 76.8 m (252 ft) was obtained. Measured against this criterion, only the full lighting condition evaluated in this experiment provided a detection distance which was, on average, adequate for safe and effective vehicle control for all drivers.

Prior to 1983, the Illuminating Engineering Society (IES) recommended design of roadway lighting using pavement illuminance. The design was based on the amount and uniformity of light reaching the pavement surface from the luminaire. The level of illuminance (in terms of average horizontal footcandles) and uniformity ratio (the ratio of average to minimum footcandles) were outputs from this method. This method was straightforward, easy to calculate, and simple to measure and verify. However, it had little relationship to what the driver saw.

The 1983 Edition of the American National Standard Practice for Roadway Lighting, ANSI/IES RP8 (IES, 1983) recognized that roadway lighting could also be designed on the basis of pavement luminance, and the standard included parameters, formulae, and techniques for designing roadway lighting on the basis of pavement luminance as well as the conventional pavement illuminance. A proposed 1990 Edition (ANSI/IES RP8) has been under review for some time. It deletes the traditional illuminance method and proposes instead the use of Small Target Visibility (STV) as a design method for roadways that are lighted for the primary purpose of improving safety. The STV method requires the calculation of the visibility of a standard target with diffuse reflectance as it is positioned at various locations along the roadway. The proposed draft standard, which considers a simplified object or target on a roadway, and a standard target and driver geometry, is currently undergoing refinement by the Illuminating Engineering Society of America (Lewin, 1996).

The following discussion deals with the effects of lighting geometry on visibility. Although the studies cited below do not address age effects of lighting geometry per se, the implications are that reductions in glare and/or increases in target contrast will have added benefits for older drivers. For instance, changes in lighting distribution (i.e., unidirectional upstream or downstream) can create intolerable glare. To minimize the glare for unidirectional lighting systems, glare cutoffs can be used to increase visibility (Ketvirtis and Cooper, 1975). Based on what is known about the effects of glare on older drivers, glare reduction should have an added benefit.

There has been some suggestion that luminaire spacing can be reduced to increase VL via an increase in adaptation level and contrast. However, Keck (1989) believes this lighting strategy can lead to an overall reduction in STV. The reason is that reduced spacing produces large zones over which targets undergo contrast reversal. These contrast reversal zones are a result of the fact that the light from the luminaires overlap. With high overlap, there are large regions over which target contrast is determined by lighting from behind and in front of the target. In spite of an overall increase in adaptation level under these lighting conditions, overall contrast is reduced because the two different lighting directions produce opposite contrast polarity. As noted earlier, contrast is negative when luminance comes primarily from behind the target and positive when illuminated from the front.

Based on Keck's analysis, it is recommended that lighting systems be developed with low overlap. This can be accomplished by either changing the angular coverage of the luminaire, reducing the height and/or increasing spacing. In any case, minimizing overlap will result in a decrease in uniformity. What must be assessed is if minimal overlap can be accomplished while meeting the maximum non-uniformity specification and maintaining a minimum visibility level (or ratio of actual contrast to contrast required for detection). For example, Shelby (1983) has shown that increased luminaire spacing has complex, unpredictable effects. Some work is needed to determine a minimum non-uniformity specification based on dynamic visual performance measures. Until then, Keck's analysis suggests that a non-uniformity of 1 (average luminance divided by minimum luminance) should never be considered the ultimate design goal. Rather, measured non-uniformities should lie within a specified range of values.

A second consideration is the directionality of the lighting system. Two different approaches are typically discussed: upstream and downstream lighting. In upstream lighting distributions, the light source is directed toward the stream of traffic. The advantage of this approach is that it is the most energy-efficient in terms of producing a high adaptation level with the lowest power watts/ft². Targets are seen in negative contrast. The disadvantages of this approach are (1) it increases the amount of glare and (2) it tends to cancel the positive contrast produced by vehicle headlights. The glare problem can be reduced with the proper design of glare shields. However, the contrast cancellation problem is difficult to resolve.

Downstream directional lighting produces positive contrast targets by aiming most of the light in the same direction as traffic flow. The advantages of this approach are that it is more efficient than vehicle headlights at increasing pavement luminance (although less efficient than upstream distributions) and the effects of vehicle headlights on target contrast are additive. One disadvantage of this geometry is that it produces a strong illumination within the vehicle (Odle, 1989). This can produce a veiling luminance over the forward driving scene via reflections off the dashboard, which can be particularly dangerous for older drivers.

Odle (1989) discussed a hybrid approach that takes advantages of the three light distributions discussed here. Counterbeam distribution is a bilaterally asymmetric lighting distribution that directs most of the light in the same direction as driver travel but some light is also directed in the opposite direction. Ketvirtis and Moonah (1995) have designed such a system into a high-mast lighting system that employs a new technique called "composite photometry." The counterbeam arrangement takes advantage of the efficiency with which pavement luminance can be increased with the upstream light and the positive contrast additivity with vehicle headlights in the downstream light.

Shelby (1983) also assessed the effects of changing mounting height. It was shown that reducing the mounting height from 13.7 to 7.9 m (45 to 26 ft) produced small increases in pavement luminance (0.9 to 1.1 cd/m²) but doubled the average "visibility index," a relative measure that is independent of observer age and task factors. This is an important finding, since it suggests visibility can be improved more efficiently by reducing mounting height than by increasing lamp lumens. However, the question that remains unanswered is the extent to which decreased uniformity that results from decreasing mounting height cancels any benefits due to increased visibility.