Appendix B. CMFs for Solid-State Roadway Lighting Applications

CHAPTER 1

CMFs for Solid-State Roadway Lighting Applications

Lighting can affect multiple aspects of safety. For example, well designed lighting extends the range of drivers’ visual detection and provides additional time for drivers to prepare for potential upcoming hazards. Lighting may also help reduce driving workload and therefore alleviate driver fatigue. On the other hand, roadway lighting may cause glare and introduce increased fixed-object crash risks. Lighting effects on safety during adverse weather conditions are also a subject of further research. The development of high-quality crash modification factors (CMFs) for lighting therefore needs to take into careful consideration several factors, such as study design, data quality and availability, sample size, and control of potential bias. In general, different CMFs can be developed for different crash types, crash severity levels, site conditions, and environmental conditions.

The CMF Clearinghouse currently uses a number of criteria to assess the quality of a CMF, including the study design, sample size, standard error, potential bias, and data source. Although most of these criteria can be relatively objectively controlled, the evaluation process is subject to opinions (CMFClearinghouse.org). Currently, NCHRP 17-72 is developing a new CMF rating procedure that revises the 5-star rating scale to a 150-point score system where CMFs scored 100 and above may be considered high-quality CMFs worthy of inclusion in the second edition of the Highway Safety Manual (HSM). The new scoring system proposed consists of three broad quality evaluation areas, including data (55 points), statistical analysis design (75 points), and statistical significance (20 points).

For before-after studies, the new scoring system requires studies to properly address issues such as regression to the mean bias, changes in traffic volume over time, suitability of reference/comparison groups, and quality of safety performance functions. For cross-sectional studies, the new system emphasizes aspects such as the similarity of sites with and without treatment, proper consideration of correlation between variables, and addressing of spatial and temporal correlations. Although on a different score scale, the new CMF scoring system is considered mostly consistent with the current star rating system when being applied to exiting CMFs in the current CMF Clearinghouse (CMFClearinghouse.org). Note that neither scoring system takes into consideration geographic regions where data is collected and therefore transferability of the results. For this reason, the rating/score of a particular CMF may not directly indicate its applicability for a specific location. For example, a 5-star fatal crash CMF developed based on European data may poorly reflect U.S. experiences.

This portion of the project provides guidelines on how to develop high-quality CMFs for roadway and street lighting measures based on this research, relevant literature, and the research team’s experience. While the results of this project are not immediately implementable, the results highlight data and informational needs to develop high-quality CMFs for local roads and residential areas.

Literature Review

Driving is a visual task, and drivers rely on the timely detection and identification of objects while driving to ensure safety. Roadway lighting intends to improve roadway and object visibility in dark driving

Roadway Lighting CMFs

The current CMF Clearinghouse database (Federal Highway Administration, 2020) included 95 CMFs for roadway/street lighting based on 14 studies (see Table 2). The database lists 55 CMFs for roadway

CMF Development

A number of methods have been commonly used for CMF development (Carter, Srinivasan, Gross, and Council, 2012; Carter, Srinivasan, Gross, Himes, et al., 2012).

• Before-after with comparison group approach. This method considers crash and traffic changes irrelevant to the subject treatment by including an untreated control group for comparison. An important aspect of this method is the selection of a suitable comparison group. Hauer (1997) previously proposed a statistical method using a sequence of odds ratios to quantitatively measure the suitability of the control group. The before-after with comparison group method is considered more accurate compared to naïve before-after studies. However, it often cannot address regression to the mean as well as those using empirical Bayes.
• Empirical Bayes before-after approach. This method is similar to the before-after with comparison group method in that both draw information on the expected crash frequencies from a group of untreated sites. However, the empirical Bayes before-after method utilizes safety performance functions (SPFs) to account for crash changes due to traffic and potentially roadway changes over the study period. Using SPFs in this method, however, may not account for changes in traffic control or factors other than those considered in the SPFs over time that may affect safety and therefore the estimated CMFs.
• Full Bayes study. The full Bayes method is another approach to derive CMF information based on before- and after-treatment data. Compared to the empirical Bayes method, the full Bayes method allows the development of complex models that are more flexible and accommodative for smaller sample sizes. If developed properly, full Bayes studies can consider spatial correlation between sites. The method in some cases can result in more reliable CMFs compared to the empirical Bayes before-after method but in general requires more statistical skills.
• Cross-sectional approach. Cross-sectional studies are based on comparisons between safety data from sites with and without the subject treatment. This approach is particularly suitable for studies where before-treatment data is not available. In cross-sectional studies, different regression methods may be used based on the data to derive the crash ratios (i.e., CMFs) between sites with and without the subject treatment. Cross-sectional analyses can be difficult to control in some cases and require careful treatment to avoid bias caused by multivariate regression issues such as collinearity. Therefore, CMFs based on cross-sectional studies in some cases can be less reliable compared to before-after studies.
• Case-control approach. The case-control method is similar to the cross-sectional method in that it also uses data with and without the treatment. However, case-control studies model data around the outcomes (e.g., severity levels or crash versus non-crash). Case-control studies rarely develop CMFs that compare crash frequencies.
• Cohort studies. Cohort studies compare risks of a certain outcome between two well-controlled groups with and without the subject treatment. This approach is less common in safety studies since it is difficult to conduct such controlled experiments on real roadways to enable sufficient crash data.

Compared to many other safety treatments, roadway and street lighting systems are somewhat unique in nature because it is difficult to identify locations with both before-and-after data. In state and local jurisdictions that have a lighting policy, roadway and street lighting systems are typically installed during roadway construction or at a time historically beyond current institutional records, especially in urban areas. In states and localities that do not have a lighting policy, lighting is rarely used as a safety treatment either at the project level or systemically. For this reason, a cross-sectional approach is frequently more suitable for the development of CMF-based systemic lighting-related treatments (Gross and Donnell, 2011).

The CMF Clearinghouse currently uses a number of criteria to assess the quality of a CMF, including the study design, sample size, standard error, potential bias, and data source. Although most of these criteria can be relatively objectively controlled, the evaluation process is more or less subject to opinions (CMFClearinghouse.org). Currently, NCHRP 17-72 is developing a new CMF rating procedure that revises the 5-star rating scale to a 150-point score system where CMFs scored 100 and above may be considered high-quality CMFs worthy of inclusion in the second edition of the Highway Safety Manual (HSM) (Srinivasan, 2018). The new scoring system proposed consists of three broad quality evaluation areas including data (55 points), statistical analysis design (75 points), and statistical significance (20 points).

For before-after studies, the new scoring system requires studies to properly address issues such as regression to the mean bias, changes in traffic volume over time, suitability of reference/comparison groups, and quality of safety performance functions. For cross-sectional studies, the new system emphasizes aspects such as the similarity of sites with and without treatment, proper consideration of correlation between variables, and addressing of spatial and temporal correlations. Although on a different score scale, the new CMF scoring system is considered mostly consistent with the current star rating system when being applied to exiting CMFs in the current CMF Clearinghouse (Srinivasan, 2018). Note that neither scoring system takes into consideration geographic regions where data is collected and therefore does not account for transferability of the results. Accordingly, the rating/score of a particular CMF may not directly indicate its applicability for a specific location. For example, a 5-star fatal crash CMF developed based on European data may poorly reflect U.S. experience.

Roadway Lighting Safety and CMF Gaps

Based on the literature review, roadway lighting safety studies have shown conflicting results. While most studies revealed a positive safety effect for lighting, some showed insignificant or adverse effects of lighting on safety. The lighting CMFs included in the CMF Clearinghouse therefore also vary significantly within each treatment category. Many lighting CMFs were developed using cross-sectional analyses based on negative binomial modeling. Most such studies involved daytime crashes as an exposure measure in the form of night-to-day crash ratio, night-to-day crash rate ratio, or nighttime crash frequency weighted by daytime crashes. However, some focused on only nighttime crash frequencies most likely following the HSM methods for other safety infrastructure treatments. The project team also found the following gaps relevant to lighting CMFs:

The project team could not identify CMFs developed specifically for SSL systems. In addition, none of the CMFs compared different CCTs and/or different lighting sources.

• A number of studies used measured lighting data and therefore were able to analyze a number of detailed lighting variables. However, many studies only analyzed roadway lighting as a binary variable (i.e., with versus without), therefore ignoring the effects of different lighting performance metrics (e.g., illuminance/luminance levels and uniformity) on safety.
• The CMFs in the current CMF Clearinghouse covered intersection lighting, freeway interchange lighting, and surface roadway segment lighting. None of the CMFs however were for lighting systems on freeway continuous segments.
• The CMFs developed in domestic studies were rated between 2 and 3 stars. Most of the CMFs were based on data from single states.

CHAPTER 2

CMF Development

General Methods

A number of methods have been commonly used for CMF development: (Carter, Srinivasan, Gross, and Council, 2012; Carter, Srinivasan, Gross, Himes, et al., 2012; Gross and Donnell, 2011).

Before-After Comparison Group Approach

This method takes into account crash and traffic changes irrelevant to the subject treatment by including an untreated control group for comparison. An important aspect of this method is the selection of a suitable comparison group. Hauer (Hauer, 1997) previously proposed a statistical method using a sequence of odds ratios to quantitatively measure the suitability of the control group. The before-after with comparison group method is considered more accurate compared to naïve before-after studies. It, however, often cannot address regression to the mean as well as those using an empirical Bayes approach.

Empirical Bayes Before-After Approach

This method is similar to the before-after with comparison group method in that both draw information on the expected crash frequencies from a group of untreated sites. However, the empirical Bayes before-after method utilizes safety performance function (SPFs) to account for crash changes due to traffic and potential roadway changes over the study period. Using SPFs in this method, however, may not account for changes in traffic control or factors other than those considered in the SPFs over time that may affect safety and therefore the estimated CMFs.

Full Bayes Study

This method is another approach to derive CMF information based on before- and after-treatment data. Compared to the empirical Bayes method, the full Bayes method allows the development of complex models that are more flexible and accommodative for smaller sample sizes. If developed properly, full Bayes studies have the ability to consider spatial correlation between sites. The method in some cases can result in more reliable CMFs compared to the empirical Bayes before-after method but in general requires more statistical skills.

Cross-Sectional Approach

Cross-sectional studies are based on comparisons between safety data from sites with and without the subject treatment. This approach is particularly suitable for studies where before-treatment data is not available. In cross-sectional studies, different regression methods may be used based on the data to derive the crash ratios (i.e., CMFs) between sites with and sites without the subject treatment. Cross-sectional analyses can be difficult to control in some cases and require careful treatment to avoid bias caused by multivariate regression issues such as multi-collinearity. Therefore, CMFs based on cross-sectional studies in some cases can be less reliable compared to before-after studies. Due to the nature of crash data (i.e.,

Safety Influence Areas of Sites

When conducting safety studies of point features (e.g., intersections and ramp junctions), it may be necessary to clearly understand the influence areas of such point features. For example, many studies use a 250-ft influence area for intersections. Depending on the location, size, and nearby access points, the area of influence may be adjusted to better reflect the site conditions.

Data Sample Size

Analysis areas that are small may result in fewer crashes identified within each analysis area. During safety analysis and CMF development, researchers need to design the study approach carefully to address all these factors. For example, to understand the safety effect of lighting at intersections, many researchers divide the intersections into the intersection box (bounded by stop bars on each approach or consistent with the intersection lighting design area) and intersection approaches.

Isolating Safety Impacts

Isolating the safety impacts of lighting during safety studies can be particularly challenging because, as previously stated, lighting is frequently installed at locations where it is most likely to benefit safety and it can be difficult to identify unlighted sites with similar conditions in the same region. Many agencies conduct warranting analyses when determining whether lighting should be installed at certain locations. Most such analyses take into consideration nighttime crash data as a primary factor. Without properly addressing the location selection issue, this practice results in an endogeneity issue (due to simultaneity of higher nighttime crash frequencies and use of lighting) for safety researchers analyzing lighting effects (Banks et al., 2014; Kim and Washington, 2006). If the lighting system cannot reduce nighttime crash frequencies to the level below that at locations without a nighttime safety challenge, improper cross-sectional analyses may lead to the problematic conclusion that lighting correlates with higher nighttime crash frequencies.

Statistical modeling techniques have been developed to address endogeneity issues. In the case of nighttime crash analysis, it is a fairly common practice for researchers to use daytime crash frequencies as a controlling method for evaluating safety treatment effectiveness during nighttime (Donnell et al., 2010). Such an approach assumes that lighting systems do not have any safety effects during daytime and therefore, separate daytime models involving the lighting variable would enable an estimate of the crash risks at subject sites in cross-section studies. This assumption, if not carefully controlled for, it does not take into consideration factors such as different safety issues during nighttime and daytime (i.e., shown by a relatively low correlation between nighttime and daytime frequencies at certain locations) and other roadway features functioning differently during reduced visibility conditions, such as, retroreflective delineation treatments.

To isolate the safety impacts attributable to lighting, studies should include variables describing the complexity of the study site as it pertains to crash risks. Examples of such variables include number of lanes, geometric alignment, presence of hazardous roadway features (e.g., turning lanes, merging lanes, and raised channelization), pavement marking conditions, and potential visual distractions in the background. In addition, studies should carefully consider key factors that help isolate lighting safety impacts, such as: average annual daily traffic and hourly volume, crash types, weather conditions, and traffic conditions.

Average Annual Daily Traffic and Hourly Volume

Depending on traffic condition, the lighting can potentially affect safety significantly in different ways. To correctly account for traffic conditions, average annual daily traffic (AADT) may be less significant in

Field Measurement

Note that while current lighting systems, when designed to standard (some roadway/street lighting applications at certain agencies do not necessarily go through the design process), mostly follow the same recommendations outlined in Illuminating Engineering Society (IES) RP- 8 (various years) or the AASHTO Roadway Lighting Design Guide. The implemented lighting, however, can be significantly different from the computer-generated designs, and agencies rarely inspect lighting after the systems are installed. To accurately analyze lighting effects, researchers may need to conduct field lighting measurements to clearly understand lighting levels at the study sites. While luminance is a more useful measure from a design standpoint, this study targeted illuminance because it is a more reliable measure with which to conduct analysis.

CMF Types

Different types of CMFs may be developed for lighting based on various factors, such as:

Crash Types

Crash type examples include separate CMFs for single-vehicle crashes and multi-vehicle crashes.

Location

Location examples include CMFs for intersections, freeway ramp locations, and freeway segments. For intersections, researchers may develop separate CMFs for approaches and intersection boxes.

Lighting Variables

Many previous safety studies analyzed lighting as a binary variable (i.e., present versus absent). While this type of CMFs provides valuable information on the safety effects of lighting, it may not accurately quantify lighting effects due to the different safety performance of lighting systems at different illuminance, luminance, veiling luminance, surround ratio, and uniformity levels. As such, it may be of best interest to practitioners in some cases to use detailed lighting variables, such as minimum illuminance levels, average illuminance levels, and lighting uniformity levels. The increased implementation of LED systems (versus legacy HPS systems) currently also provides the safety community an opportunity to study the effect of correlated color temperature (CCT) of lighting systems on safety.

CHAPTER 3

CMF Development Examples

During this project, the team attempted to develop the following example lighting CMFs as examples for future development:

• Adaptive LED system: to replace traditional static HPS system in an urban setting for major and local roadways.
• Static LED system: to replace traditional static HPS system in an urban setting for major and local roadways.
• Increasing the horizontal illuminance of an LED system in an urban setting for major and local roadways.

The purpose of the CMF example development was primarily to illustrate the effects of different methods and consideration on the results of CMFs.

Lighting Systems Overview

The example CMFs were developed using crash and lighting data from the Cities of Cambridge and Somerville in Massachusetts. The City of Cambridge replaced the approximately 7,000 traditional HPS lights citywide on streets and parks during late 2014 and early 2015 with an adaptive 4000 K LED lighting system. With the new system, the light levels are dimmed to 50% of their normal brightness following a dimming schedule configured according to roadway types. This SSL system was one of the early adaptive systems across the nation, providing an opportunity to study the safety effects of adaptive LED lighting systems. During late 2015 and early 2016, the City of Somerville replaced citywide HPS roadway luminaires with a new LED system (U.S. Department of Energy, 2016). The Cities of Cambridge and Somerville are located side by side, sharing relatively similar roadway and traffic characteristics, providing an opportunity to evaluate both adaptive LED systems in the case of Cambridge, and a static LED system in Somerville.

Roadway Segments Selection

The selection of roadway segments involved field lighting data collection for the purpose of identifying the potential safety effects of different lighting levels. The project team selected a number of roadways in both cities (Figure 3) that were representative of different types of roadways by land use (i.e., commercial versus residential) and AADT, referring to major surface roadways or classified by the Highway Performance Monitoring System (HPMS) roadways and local roadways. The Cambridge lighting system follows a dimming schedule based on three types of roadways: C1 roadways or major HPMS roads, C2 roadways consisting of smaller HPMS roads and major local roads, and C3 roadways that are minor local roads. The project team made an effort to ensure the selected roadways in Cambridge were representative

Crash and Roadway Data Collection

The project team obtained 2010–2019 crash data from the Massachusetts Department of Transportation (MassDOT). Based on crash time and location, the project team determined nighttime (defined when the sun angles are below -6° in this study) crashes based on the National Oceanic and Atmospheric Administration (NOAA) solar angle calculation method (Global Monitoring Laboratory).

For roadway and traffic variables, the project team collected the following information:

• AADT: collected based on 2010–2019 yearly HPMS submittals by MassDOT.
• Number of lanes: collected based on satellite imagery on ArcGIS^TM and Google® Maps.
• Speed limit: collected based on satellite imagery on ArcGIS and Google Maps. Note that most major roadways analyzed had a speed limit between 20 mph and 35 mph. In addition, many local roadways did not have speed limit signs posted. As a result, the project team did not include this variable in the modeling process.
• Sidewalk presence: collected based on satellite imagery on ArcGIS and Google Maps.
• Street parking presence: collected based on satellite imagery on ArcGIS and Google Maps.
• Bike lane presence: collected based on satellite imagery on ArcGIS and Google Maps.
• One way street indicator: collected based on satellite imagery on ArcGIS and Google Maps.
• Median presence: collected based on satellite imagery on ArcGIS and Google Maps.
• Land use type: included commercial, mixed, and residential; determined using GIS shapefiles published by the municipalities.

Methodological Considerations

For methodological considerations, the team collected information on both local and HPMS roadways in the two cities. For the demonstration of the CMF development process, however, the analysis mostly focused on HPMS roadways only. HPMS roadways are maintained by the state and had relatively complete

Where:

• E (N_LED) = expected nighttime frequency for the LED period;
• E (D_LED) = expected daytime frequency for the LED period;
• E (N_HPS) = expected nighttime frequency for the HPS period before LED conversion; and
• E (D_HPS) = expected daytime frequency for the HPS period before LED conversion.

CMF Analysis Scenarios and Results

Replacing Traditional HPS Systems with an Adaptive LED System

Description of Treatment and Settings

This comparison analysis attempted to develop a CMF for the conversion of traditional HPS lighting systems to an adaptive LED system on major surface roadways in an urban area. The current lighting system is an adaptive LED system with a CCT of 4000 K that dims to half of its normal levels at midnight for major arterials (i.e., C1 roadways in Cambridge, MA) and 10:00 p.m. for minor arterials (i.e., C2 roadways in Cambridge MA). Prior to the LED conversion, the city had a legacy HPS system for citywide roadways. The CMF was developed based on mostly HPMS roadways in the City of Cambridge, MA whose lighting policy is that it illuminates all roadways in the city limit. This provides an opportunity for an ideal sample that does not include potential bias due to lighting used as a safety treatment following crash-based warranting analyses for many state-maintained roadways. The city also was one of the pioneers implementing an LED adaptive system in the early years, allowing sufficient before-and-after crash data for analysis.

Description of Data

The CMF was developed based on 246 roadway segments totaling 40.7 centerline miles in Cambridge, MA. The analysis involved a total of 13,820 daytime crashes and 4,080 nighttime crashes between 2010 and 2019 (Table 3). Based on the LED conversion schedule, crashes between 2010 and 2014 were used for the before (i.e., HPS) period and 2015–2019 crash data were used for the after (i.e., LED) period.

Table 3. Crash counts by year for the adaptive LED analysis.

CMF Results for Adaptive LED System

Table 4 lists the mixed-effect negative binomial regression model for the Cambridge adaptive LED system based on nighttime crashes. Table 5 further lists the model developed based on daytime crashes for exposure measurement. Note that, while the LED indicator variable was significant (at a 0.05 level of significance) in the nighttime model, the same variable was not significant in the daytime model. Based on only the nighttime model, the CMF for converting a traditional HPS system to an adaptive LED system was determined as 1.14 (i.e., Exp (0.133)), suggesting a 14% increase in nighttime crashes after the LED conversion. Because the LED factor was not significant for the daytime crash model, it was not necessary to use the daytime crashes as an exposure measurement.

Replacing Traditional HPS Systems with a Static LED System

Description of Treatment and Settings

This comparison analysis attempted to develop a CMF for the conversion of traditional HPS lighting systems to a static LED system on major surface roadways in an urban area. The current lighting system is a static LED system that maintains the full lighting characteristics throughout the entire nighttime. The City of Somerville converted the citywide HPS system to the current LED system during late 2015 and early 2016. The city lights all roadways within the city limits, providing a good opportunity to understand the safety effects of lighting.

Description of Data

The CMF was developed based on 137 roadway segments in Somerville, MA with a total length of 28.4 miles. The analysis involved a total of 5,721 daytime crashes and 2,041 nighttime crashes between 2010 and 2019 (see Table 6). Based on the LED conversion schedule, crashes between 2010 and 2015 were used for the before (i.e., HPS) period and 2016–2019 crash data were used for the after (i.e., LED) period.

Table 6. Crash counts by year for the Static LED analysis.

CMF Results for Static LED System

Table 7 lists the mixed-effects model for the nighttime crashes for Somerville comparing the nighttime crash frequencies before and after the LED conversion for major roadways (i.e., HPMS roads). Table 8 further lists the mixed-effects model for daytime crashes for comparison. As the models indicate, the LED system was a significant factor in the nighttime model but not significant in the daytime model. Based on the nighttime model, conversion to an LED system was found to have a CMF of 1.22 (i.e., Exp (0.197)), suggesting a 22% increase in nighttime crashes after the LED conversion.

*indicates statistical significance (α=0.05)

Increasing Horizontal Illuminance of an LED System

The adaptive lighting system at Cambridge provided an opportunity to analyze the correlations between different luminance levels and nighttime crash experience. Using the lighting measurement data, the project team attempted to understand this correlation for different types of roadways in the two cities during different time periods based on the diming schedule of the Cambridge system. For this analysis, the project team analyzed four lighting measures for three time period-roadway type combinations. The four lighting measures included:

• Average illuminance
• Minimum illuminance
• Maximum illuminance
• Standard deviation of illuminance (or variance)

The three time-roadway type combinations were:

• C1 roadways in Cambridge and HPMS roadways in Somerville for nighttime after midnight.
• C2 roadways in Cambridge and non-HPMS roadways in Somerville for nighttime after 10:00 p.m. and before midnight.
• All HPMS roadways in both cities for nighttime before 10:00 p.m..

The analyses used only crash data for the after-LED periods of the two cities.

During the analyses, the project team first attempted to develop mixed-effect negative binomial regression models. For those scenarios where SAS could not find a global solution for the random-effect models, the team developed fixed-effect negative binomial models instead. The analysis results did not show any significant correlations between the four lighting measures and nighttime crashes for the before 10:00 p.m. (HPMS roads) and the after 10:00 p.m. (C2 roadways) scenarios. The analysis of the C1 roadways for the after-midnight period did show that both mean horizontal illuminance and standard deviation of horizontal illuminance were significant for nighttime crashes. The results for the C1 roadway analysis, however, were based on regular negative binomial regression models after SAS failed to find a global solution for all four lighting variables for this scenario.

Table 9 and Table 10 list the negative binomial regression models for mean illuminance and standard deviation of illuminance, respectively, for the C1 roadways after midnight. As the models suggested, on the major roadways in both cities and after midnight, higher horizontal illuminance levels corresponded with lower nighttime crash frequency. In addition, more uniform lighting (i.e., lower illuminance variance) correlated with higher nighttime crash frequency. Holding all other variables constant, a 1 lux increase in mean illuminance correlated with a crash modification of 0.983 [i.e., exp (-0.017)], indicating a reduction of 1.7% in nighttime crash frequency. A 1 lux increase in illuminance standard deviation correlated with a 1.8% decrease in crash frequency (a CMF of 0.982 [i.e., exp (-0.018)]). It should be noted that these decreases in crash frequency are valid in the range of the mean illuminances and the standard deviations measured in the data. Crash frequency cannot be expected to decrease consistently with increases in illuminance and standard deviation of illuminance; this effect will plateau at a certain level of the illuminance variables.

CHAPTER 4

Discussion

During this analysis, the project team illustrated the development of CMFs for converting HPS systems to an LED system and for changes in LED lighting metrics. The first group of CMFs illustrated the safety implications of converting a traditional HPS system to an adaptive LED system and a static system in the Cities of Cambridge and Somerville, MA. For the changes in LED lighting metrics, the project team analyzed the correlations between four lighting metrics (average horizontal illuminance, maximum horizontal illuminance, minimum horizontal illuminance, and standard deviation of horizontal illuminance) and nighttime crashes.

When looking at the analyses for the individual CMFs, the results showed:

• A CMF of 1.14 was determined for converting a traditional HPS system to an adaptive LED system. This CMF was determined based on the crash and roadway data from the City of Cambridge, MA. For the analysis, the team used 2010–2014 crash data for the HPS period and 2015–2019 crash data for the LED period. This CMF seemed to suggest that the conversion of a traditional HPS system to the adaptive LED system in Cambridge correlated to a 14% increase in nighttime crashes.
• A CMF of 1.22 was determined for converting a traditional HPS system to a static LED system. This CMF was determined based on crash and roadway data in Somerville, MA. The analysis used 2010–2015 crash data for the HPS period and 2016–2019 crash data for the LED period. This CMF seemed to suggest that the conversion to the LED system coincided with a 22% increase in nighttime crashes.
• The LED conversion did not significantly correlate with an increase in daytime crashes during the same periods in either city.
• A CMF of 0.983 was determined for average horizontal illuminance for major HPMS roadways after midnight. This CMF seemed to suggest that a 1.7% of crash reduction correlated with each unit (i.e., lux) increase in average horizontal illuminance.
• A CMF of 0.982 was found for standard deviation of horizontal illuminance for major HPMS roadways after midnight, indicating a 1.8% decrease in nighttime crashes for each unit increase of horizontal illuminance standard deviation.
• Different lighting levels did not correlate with significant differences in nighttime crashes for HPMS roadways before 10:00 p.m. and for C2 roadways after 10:00 p.m.

Interpreting CMF Results for Converting HPS to LED

The previous analyses comparing the current LED systems with the previous HPS lighting showed that both cities with LED systems experienced an increase in nighttime crashes. The analyses, therefore attributed the increase in nighttime crashes to the LED conversion as the CMFs would imply. Note that the Cities of Cambridge and Somerville are located next to each other with comparable traffic patterns and driver populations.

LED systems are used increasingly for roadway and street lighting for several reasons, such as increased energy efficiency, more manageable performance variables (e.g., uniformity, distribution, and CCT), longer service life, and less maintenance. By looking at either system independently, a CMF that is greater than one (1.0) may appear plausible at first when considering the following factors:

• Lighting levels for the Cambridge system were dimmed too low later in the night for the adaptive system; and/or
• System-wide lighting design/installation issues are causing problematic uniformity or luminance levels.

Both factors may result in significant visibility challenges, causing risky driver behaviors. However, in this study, both LED systems that were implemented individually in the two cities were found with a CMF greater than 1 (1.0), which appeared to be highly unlikely. Both LED systems provide luminance levels comparable to that of the previous HPS systems at full level. Therefore, merely using LED instead of HPS should not result in significant changes in nighttime crashes. The crash changes, therefore, could be due to factors not considered in the models, such as increase of nighttime traffic in the region or change in nighttime crashes due to other factors.

The researchers plotted the average night-to-day (ND) crash ratios for the analyzed segments (Figure 5) to understand the crash trends over the analysis period. As the figure illustrates, the major HPMS roadways in Cambridge (i.e., C1 roadways) generally experienced increasing trends in ND crash ratios between 2012 and 2015, while the minor HPMS roadways in Cambridge (i.e., C2 roadways) experienced a dip between 2011 and 2015. The ND crash ratios for both types of roadways in general exhibited a declining trend between 2015 and 2018. Neither roadway types appeared to show an increasing ND crash ratio trend starting in 2015 after the LED system was implemented. Similarly, for the Somerville HPMS roadways, the ND crash ratios showed a decreasing trend starting in 2016, after the implementation of the new LED system. Prior to 2016, the ND ratio showed a dip between 2012 and 2016. These crash trends by no means suggest adverse effects of the LED systems on nighttime crashes. Note that the use of ND crash ratio in this discussion is appropriate, as the CMF analysis showed that the daytime crashes did not have a significant correlation with the implementation of the LED systems in either city.

The CMF modeling process for the LED conversion CMFs was a multivariate process involving, in particular, yearly ADT information. Taking into account the ADT development process at state DOTs, the yearly ADT information may be considered a particularly weak link in the CMF modeling process. States typically develop ADT values for roadways based on permanent and temporary traffic counts conducted at strategic locations. Such locations are frequently determined based on factors such as functional class and

CHAPTER 5

Conclusions

The purpose of this investigation was to define the methods and potential pitfalls in the development of CMFs for lighting. Three examples are used to highlight these difficulties. In a study of lighting in Cambridge and Somerville Massachusetts, three sample CMFs considering the application LED lighting, adaptive lighting, and overall lighting level were developed. The results indicated that crashes increase with the application of LED lighting but less so when the lighting is adapted. However, this effect was countered by crashes being reduced when using a higher lighting level. Upon consideration, these three results contradict each other for the light level CMF. The lower light levels in adaptive lighting would create the expectation of more crashes than the higher light levels in the non-adaptive case but this was not what the analysis revealed. This highlights the difficulties in creating the CMF.

The analysis indicated that:

• More sites considered over fewer years of data collection may provide more reliable results, as other factors such as traffic volume and composition are less likely to change over a shorter timeframe.
• Use of properly measured lighting data is beneficial to the analysis, as the impacts of the overall lighting condition are assessed based on the installed lighting as well as the natural variation in lighting design. This is true whether the lighting is designed and installed or if it is light from off-road sources. This poses an issue, as measured data is not typically available.
• Similar sites need to be considered, as the impact of one lighting design may not be the same under all application conditions.
• To consider the specific roadway conditions more accurately at the time period used in the analysis and to ensure that nighty traffic exposure data is accurate, it is best to use hourly, or at least ADT, data that considers time of day, weekly, and seasonal variations.

Limitations

While attempting to develop CMFs for converting HPS systems to LED systems, this study was not able to identify suitable sites to compare safety experiences between with-LED lighting and without lighting conditions. The study also used sites on surface streets in an urban setting, where lighting may not necessarily generate the most benefits in reducing motor vehicle crashes. Due to the limited severe crashes, this study also did not attempt to separately derive CMFs for severe crashes (i.e., injury and fatal crashes). Furthermore, the analysis included a significant number of local streets without accurate traffic volume data have fewer crashes, which affected CMF development. As smart cities and intelligent transportation systems are widely deployed, the hourly traffic volume data for local street and residential areas should be made easily accessible to enable more accurate CMF calculations. Finally, it should be noted that the results of this project are not immediately implementable; the results highlight data and informational needs to develop high-quality CMFs and lay the foundation for upcoming work in this area.

References

AASHTO. (2014). Highway Safety Manual (1st Edition) with Supplement. American Association of State Highway and Transportation Officials, Washington, DC.

AASHTO. (2018). Roadway Lighting Design Guide, 7th Edition. American Association of State Highway and Transportation Officials, Washington, DC.

Banks, D., Persaud, B., Lyon, C., Eccles, K., and Himes, S. (2014). Enhancing Statistical Methodologies for Highway Safety Research – Impetus from FHWA.

Bell, A., M. Fairbrother, and K. Jones. (2019). Fixed and Random Effects Models: Making an Informed Choice. Quality & Quantity, 53, 1051–1074.

Bhagavathula, R., and Gibbons, R. B. (2018). Effects of Mounting Height, Offset Distance, and Number of Light Towers on Drivers’ Visual Performance and Discomfort Glare in Work Zones. Transportation Research Record: Journal of the Transportation Research Board, No. 2672(16), 105–118.

Bhagavathula, R., Gibbons, R. B., and Edwards, C. J. (2015). Relationship between Roadway Illuminance Level and Nighttime Rural Intersection Safety. Transportation Research Record: Journal of the Transportation Research Board, No. 2485(1), 8–15. https://doi.org/10.3141/2485-02.

Bhagavathula, R., Gibbons, R. B., and Nussbaum, M. A. Effects of Intersection Lighting Design on Nighttime Visual Performance of Drivers. Leukos, 14 (1), 25–43.

Bhagavathula, R., and Gibbons., R. B. (2013). Role of Expectancy, Motion and Overhead Lighting on Nighttime Visibility. Human Factors and Ergonomics Society Annual Meeting.

Box, P. C. (1989). Major Road Accident Reduction by Illumination. Transportation Research Record, No. 1247, 32–38.

Boyce, P. R. Lighting for Driving: Roads, Vehicles, Signs, and Signals. CRC Press.

Carter, D., Srinivasan, R., Gross, F., and Council, F. (February 2012). Recommended Protocols for Developing Crash Modification Factors. American Association of State Highway and Transportation Officials. Retrieved 06/02/2020 from http://www.cmfclearinghouse.org/collateral/CMF_Protocols.pdf.

Carter, D., Srinivasan, R., Gross, F., Himes, S., Le, T., Persaud, B., Lyon, C., Bonneson, J., and Hauer., E. Appendix C: Guidance for Developing Crash Modification Functions; Guidance for the Development and Application of Crash Modification Factors.

CMFClearinghouse.org. About the Star Quality Rating. Retrieved 06/02/2020 from http://www.cmfclearinghouse.org/sqr.cfm.

Donnell, E., Shankar, V., Porter, R., and Larson, T. (2009). Analysis of Safety Effects for the Presence of Roadway Lighting.

Donnell, E. T., Porter, R. J., and Shankar, V. N. (2010). A framework for estimating the safety effects of roadway lighting at intersections. Safety Science, 48 (10), 1436–1444.

Edwards, C. J. (2015). Lighting levels for isolated intersections leading to safety improvements. https://conservancy.umn.edu/handle/11299/170427.

Federal Highway Administration. (2020). Crash Modification Factors Clearinghouse. Federal Highway Administration. Retrieved 10/15/2020. from http://www.cmfclearinghouse.org/.

Gibbons, R., Guo, F., Medina, A., Terry, T., Du, J., Lutkevich, P., and Li, Q. (2014). Design Criteria for Adaptive Roadway Lighting.

Gibbons, R. B., Cheung, J., and Lutkevich., P. The Future of Roadway Lighting. Public Roads 79 (3).

Gibbons, R. B., Meyer, J. E., Terry, T. N., Bhagavathula, R., Lewis, A., Flanagan, M., and Connell, C. (2015). Evaluation of the Impact of Spectral Power Distribution on Driver Performance [Tech Report]. https://rosap.ntl.bts.gov/view/dot/35871.

Global Monitoring Laboratory, NOAA Solar Calculation Details. Retrieved 09/21/2021 from https://gml.noaa.gov/grad/solcalc/calcdetails.html.

Gross, F., and Donnell, E. T. Case-Control and Cross-Sectional Methods for Estimating Crash Modification Factors: Comparisons from Roadway Lighting and Lane and Shoulder Width Safety Effect Studies. Journal of Safety Research 42, 117–129.

Gross, F., Persaud, B., and Lyon., C. A Guide to Developing Quality Crash Modification Factors.

Hauer, E. (1997). Observational Before-After Studies in Road Safety - Estimating the Effect of Highway and Traffic Engineering Measures on Road Safety. Pergamon Press.

Illuminating Engineering Society. (2022). Recommended Practice: Lighting Roadway and Parking Facilities. ANSI/IES RP-8-22. New York.

Isebrands, H. N., Hallmark, S. L., Li, W., McDonald, T., Storm, R., and Preston, H. (2010). Roadway lighting shows safety benefits at rural intersections. Journal of Transportation Engineering, 136 (11), 949–955.

Jin, H., Jin, S., Chen, L., Cen, S., and Yuan, K. Research on the Lighting Performance of LED Street Lights with Different Color Temperatures. IEEE Photonics Journal, 7 (6).

Kim, D., and Washington, S. (2006). The Significance of Endogeneity Problems in Crash Models: An Examination of Left-Turn Lanes in Intersection Crash Models. Accident Analysis & Prevention, 38 (6), 1094–1100.

Li, Y., Bhagavathula, R., Terry, T. N., Gibbons, R. B., and Medina, A. Safety Benefits and Best Practices for Intersection Lighting. Final Report.

Lord, D., and Mannering, F. (2010). The Statistical Analysis of Crash-Frequency Data: A Review and Assessment of Methodological Alternatives. Transportation Research Part A: Policy and Practice, 44 (4), 291–305.

Lutkevich, P., Gibbons, R., Bhagavathula, R., and McLean, D. (2020). NCHRP Research Report 920: Solid-State Roadway Lighting Design, Volume 2: Research Overview, Transportation Research Board, Washington, DC.

Lutkevich, P., McLean, D., and Cheung, J. FHWA Lighting Handbook.

Monsere, C. M., and Fischer, E. L. (2008). Safety Effects of Reducing Highway Illumination for Energy Conservation. Presented at 87th Annual Meeting of the Transportation Research Board, Washington, DC.

SAS Institute Inc. (2015). The NLMIXED Procedure. SAS/STAT® 14.1 User’s Guide.

Srinivasan, R. (n.d.) UNC Highway Safety Research Center Presentation: NCHRP 17-72: Update of Crash Modification Factors for the Highway Safety Manual. Presentation. The University of North Carolina Highway Safety Research Center. Retrieved 06/02/2020 from https://www.larson.psu.edu/assets/docs/TESC-presentations/6A-HSM-Safety-Future-Generations/Srinivasan-NCHRP-17-72-CMFs.pdf.

U.S. Department of Energy. (2016). Outdoor Lighting Challenges and Solution Pathways.

Van Schalkwyk, I., Venkataraman, N., Shankar, V., Milton, J., Bailey, T., and Calais, K. Evaluation of the Safety Performance of Continuous Mainline Roadway Lighting on Freeway Segments in Washington State.

Washington State Department of Transportation. Design Manual.

Yang, R., Wang, Z., Lin, P., Li, X., Chen, Y., Hsu, P. P., and Henry, A. Safety Effects of Street Lighting on Roadway Segments: Development of a Crash Modification Function. Traffic Injury Prevention. https://doi.org/10.1080/15389588.2019.1573317.