Influence of Infrastructure Design on Distracted Driving (2022)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

70 Chapter 7. Safety Framework for the Use of Crash Data to Assess the Impact of Infrastructure Design on Distraction 7.1 Introduction 7.1.1 Background Police crash reports are a rich source of data about the locations, characteristics, and contributing circumstances of roadway crashes, especially when crash locations are geospatially coded. A key advantage using crash data for safety analyses is that crash data is an actual indicator of safety. While the use of safety surrogates plays an important role in evaluating and ultimately improving safety, reducing crash frequency and severity is the end goal of safety activities. Crash evaluations can take a number of forms, such as evaluating crashes before and after installation of various roadway elements, cross-sectional analyses (which compare crashes on similar roadway sections with and without the feature of interest), geospatial analyses (which attempt to derive some relationship between crashes and the spatial proximity of an infrastructure element), and so on. Crash analyses can also target a particular type of crash that may be an indicator of distraction, such as run-off-road or rear end. As noted in Section 6.3, crash reporting forms used in most states now include a place to indicate OVD. The crash narrative analysis presented in Chapter 6 did suggest that crashes coded as involving OVDs are unlikely to be a good source for identifying IRD crashes in general. However, clusters of crashes noted as involving OVDs may suggest that some factor is at play, even if officers are not able to capture the source of the distraction. The impact of IRD can be evaluated a number of different ways using crash data. As described in this chapter, a Safety Framework was developed that agencies can use to assess the impact of an infrastructure element on distraction using crash data and then test the efficacy of the approach. 7.1.2 Scope of Framework Road safety decision-makers at all levels of government rely heavily on data from motor vehicle crash reports to drive strategic and tactical decisions about policy priorities and expenditures. The purpose of this analysis was to determine whether existing motor vehicle crash reports are a viable source of information about the role of IRDs in crash causation. Examples of research questions that could potentially be addressed through an analysis of crash reports include the following: • How often are external distractors identified as causal factors in police crash reports? • Are OVD crashes more prevalent in certain environments? • What are the physical characteristics of locations where OVD crashes are reported to have occurred (e.g., roadway type, driving environment [urban, suburban, or rural], distractor type, distance from the traveled way to the distractor)? • What attributes or features of roadside distractors contribute to elevated crash rates (e.g., physical size, luminance, moveability, or changeable elements)?

71 • Do specific images or content result in a spike in crashes near billboards or variable message signs? • What is the relationship between a particular infrastructure element (e.g., wind turbine, billboard, bridge structure) and crashes? • Can relationships between a particular infrastructure element and crashes be used to infer distraction? • Are certain types of crashes (e.g., run-off-road, rear-end) more likely to be related/correlated to a particular infrastructure element (e.g., overpass, wind turbine)? To evaluate the efficacy of using crash data to identify the impact on crashes of distraction resulting from infrastructure elements, the following research question was proposed and used to develop a Safety Framework: What is the spatial relationship between the presence of wind turbines and crashes? This research question was selected because wind turbines are an easily identifiable infrastructure element. Additionally, since wind turbines are highly visible and not necessarily an expected part of the landscape, it was felt that it would be more feasible to characterize the relationship between wind turbines and crashes than between crashes and other infrastructure elements, given the available project resources. According to the AWEA, Iowa currently ranks second in the nation in total installed wind generating capacity, with 5,590 wind turbines capable of generating 10.7 gigawatts annually (equivalent to the power consumed by 2.4 million homes). Iowa leads the nation in the number of wind turbines per square mile. Some are located within the viewshed of major freeways and expressways, and many are visible from two-lane highways or county secondary roads (Figure 16). Source: Justin C. Hilts, Shutterstock Figure 16. Wind turbines along a two-lane roadway in Iowa. Another reason for selecting wind turbines for this analysis is the fact that a correlation between increased crash risk and the presence of a wind turbine could rather easily be attributed to drivers

72 having been distracted by the turbine, even if the turbine is not noted as an external distraction in the corresponding crash reports. That is, because wind turbines are located a significant distance from the roadway, their presence cannot create an on-road scenario that increases crash risk. In contrast, crashes could be correlated to the presence of an overpass, but it would be more difficult to determine that distraction was the likely cause of those crashes (if the crash records do not sufficiently document an OVD) because other factors may be present that contributed to the increased number of crashes. During the winter, for example, ice or snow could linger under overhead structures due to less sun exposure. Bridges may also attract some types of animals, which could also be a contributing factor in an elevated crash rate. As a result, the presence of wind turbines offered the best scenario for inferring distraction as a cause of crashes due to infrastructure elements. 7.2 Data Sets Utilized 7.2.1 Crash Data Iowa crash data from 2018 and 2019 were selected for use in this Safety Framework. Iowa data were selected because Iowa is one of the national leaders in wind energy production and a large number of wind turbines are located across the state. Iowa crash data are also spatially located, and the research team had ready access to both the crash data (except crash narratives) and roadway data. Iowa leads a 15-state coalition that has developed a centralized electronic reporting portal called TraCS. The Iowa instance of TraCS is jointly maintained by the Iowa DOT and the Iowa State Patrol. As of March 2019, the system was used by 368 state, county, and municipal enforcement agencies in Iowa, with 99.2% of crash reports generated electronically. The centralized recordkeeping, searchability, and greater uniformity made possible by TraCS facilitate roadway safety research. Importantly for the purposes of this project, the Iowa TraCS obtains the location of each crash from the GPS data in the responding officer’s mobile data terminal, eliminating the need for manually coding crash locations when linking crash reports with physical features of the roadway or roadside environment. TraCS data—stripped of crash narratives and personal identifying information—is freely available through the online Iowa Crash Analysis Tool. The following general categories of variables can be extracted from the Iowa DOT crash data set: • Crash ID. • Date. • Manner of crash. • Crash severity. • Surface condition. • Contributing factors. • Vehicle characteristics. • Driver characteristics (age, gender). • Drug or alcohol involvement.

73 • Driver contributing circumstances, including distraction (the distraction element of the Iowa TraCS data model became operational on January 1, 2015). • Sequence of events. • Roadway type. 7.2.2 Roadway Data The research team also had access to the Iowa DOT’s GIS-based roadway database. The database includes roadway characteristics such as number of lanes, shoulder width, roadway surface type, and so on for freeways, expressways, and primary two-lane highways. Some verification of these features is required. The database also includes traffic characteristics (e.g., volume, percent trucks). A GIS-based record of the location of Iowa municipalities is also available through the database, which can be used to identify crashes that occur in rural versus urban areas. 7.2.3 U.S. Wind Turbine Database The U.S. Wind Turbine Database is a database and online map containing location data and specification information on both land-based and offshore wind turbines in the U.S. Data are compiled from the FAA, Lawrence Berkley National Laboratory, AWEA, USGS, and other online sources. This database is maintained through funding from the U.S. DOE WETO, USGS, and AWEA. The location for each wind turbine is not field verified; instead, locations are visually verified using high-resolution aerial imagery. For this reason, a confidence rating of the location for each turbine is included with the location data. Additional information provided for each turbine includes GPS coordinates, the installation year, the height of the turbine, and the length of various components of the turbine. Source data can be downloaded and used in GIS software to correlate wind turbine information to geocoded crash data (USGS 2020). The database is available in both a comma delimited text file as well as a shapefile, the latter of which can be imported into ArcGIS or another GIS software. For this project, the shapefile was used in ArcGIS to select wind turbines in the vicinity of selected crashes. Figure 17 shows the locations of wind turbines and 2019 police-reported crashes in Iowa.

74 Figure 17. Wind turbines (red dots) and 2019 crashes (blue dots) in Iowa. The following relevant variables are available in the U.S. Wind Turbine Database: • GPS location. • Confidence rating of the location. • Installation year. • Turbine height. • Length of various components of the turbine. 7.3 Data Reduction Once the targeted crashes for 2018 and 2019 were extracted, the Iowa DOT’s GIS-based roadway database was used to identify two sets of road segments. The first set included segments along I-80 in Iowa that are outside incorporated boundaries (rural). I-80 is the major east-west Interstate that traverses the entire state and includes both segments with high concentrations of wind turbines and segments with none. This set included 894 segments. The second set included paved roadway segments throughout Iowa near wind turbines. These segments represented a subset of roadways near wind turbines in Iowa; project resources were not sufficient to identify all segments near wind turbines in the state. Segments near wind turbines (within 2 miles) were manually identified in ArcGIS and were considered treatment segments. Roadway segments that were in proximity to treatment segments but were not within 4 miles of a wind turbine were selected as control segments. A total of 499 segments were selected, including both control and treatment segments. A large number of wind turbines were installed in 2018, though the specific month and day could not be determined. As a result, crashes for 2018 and 2019 were aggregated separately for each segment, and the number of operational turbines up to the preceding year was calculated. Spatial queries were conducted in ArcGIS for each segment, and the number of wind turbines within a 2-mile and 4-mile radius was calculated along with the number of crashes that occurred on the

75 segment for each of 2018 and 2019. For the 2018 and 2019 crash data, a segment was classified according to whether a wind turbine was operational by 2017 or 2018, respectively. Appropriate roadway characteristics were also included for each roadway segment. The roadway elements were sourced from the Iowa GIMS data for 2016, the last year in which these data were available before the Iowa DOT switched to an LRS, which complicates the extraction of roadway features. The corresponding roadway elements for each crash were extracted by completing a spatial join between the GIMS data and the crash data. It was assumed that each crash occurred on the nearest roadway. The following data were then extracted from the roadway data and included as independent variables: • NUMLANES: Number of lanes on the roadway. • SURFWIDTH: Surface width in feet. • SURFTYPE: Surface type. • SHDTYPER: Type of the right shoulder. • SHDWIDTHR: Width of the right shoulder to the nearest foot. • RUMBLER: Rumble strip presence on the right shoulder. • SHDTYPEL: Type of the left shoulder. • SHDWIDTHL: Width of the left shoulder to the nearest foot. • RUMBLEL: Rumble strip presence on the left shoulder. • LIMITMPH: Roadway speed limit. • AADT: Annual average daily traffic. • MEDTYPE: Type of median (e.g., grass, flush). • MEDWIDTH: Width of the median between the edges of the traffic lanes recorded to the nearest foot. • LANELENG: Length of the road segment in miles. 7.4 Analysis of Segments in the Vicinity of Wind Turbines This analysis explored whether the presence of a wind turbine influences roadway crash rates. To explore this hypothesis, the analysis combined statewide crash data with data on wind turbine locations. 7.4.1 Variables Included Crash frequency was the dependent variable included in the analysis. Traffic volume and segment length are measures of exposure and are expected to impact the number of crashes on a segment. Both were included as independent variables. The complete set of independent variables included the following: • Roadway characteristics (see list of data in Section 7.3). • Segment length (miles). • Annual average daily traffic (AADT). • Presence of a wind turbine within 2 miles. • Presence of a wind turbine with 4 miles.

76 • Number of wind turbines within 2 miles. • Number of wind turbines within 4 miles. The average height of the wind turbines that fell within a 2-mile buffer of a crash was also calculated, but in many cases a value was missing from the database field describing turbine height, resulting in an insufficient amount of information about this variable to include in the analyses. 7.4.2 Modeling Approach The first modeling approach that was attempted was to assess crashes on rural roadways (nonfreeway). A total of 299 segments within 4 miles of a wind turbine were manually extracted, representing 256 miles. Crash data were extracted for one year (2018), but only five roadway departure crashes were found within this time period. Two hundred segments that were more than 4 miles from a wind turbine were also extracted, representing 171 miles. A total of 11 roadway departure crashes were found, resulting in 0.055 crashes per mile. The data for 2019 also included a low number of crashes. Due to the low numbers of crashes, no analysis could be completed for rural roadway (nonfreeway) segments. The second modeling approach attempted was to assess crashes on roadway segments on I-80. A total of 894 segments were identified (241 miles total), with around 7% and 11% of those being near a wind turbine for the 2018 and 2019 crash data, respectively. Models using segments within 2 miles and within 4 miles of a wind turbine were developed for the I-80 data. The models were reasonably similar, so results are only presented for those segments within 4 miles. A cross-sectional model was created that compared crashes on different segments of I-80 and that included the presence of a wind turbine as an explanatory variable. A Poisson-lognormal model was selected as the best fit for the data, with crash frequency as the dependent variable. The crash counts for each year were simultaneously modeled by a Poisson-lognormal regression as follows: 𝑦 ~𝑃𝑜𝑖𝑠𝑠𝑜𝑛 𝜃( ) , 𝑦( )~𝑃𝑜𝑖𝑠𝑠𝑜𝑛 𝜃( ) , 𝑙𝑜𝑔𝜃( )~𝑁𝑜𝑟𝑚𝑎𝑙(𝑥 𝛽 + 𝑤( ) 𝛾 ,𝜎 ), 𝑙𝑜𝑔𝜃( )~𝑁𝑜𝑟𝑚𝑎𝑙(𝑥 𝛽 + 𝑤( ) 𝛾 ,𝜎 ). In the model above, the data are as follows: • 𝑦( ): Crashes on the 𝑖-th segment during year 𝑗, 𝑗 = 1 or 2. • 𝑥 : Vector of common features for the 𝑖-th segment, such as AADT, number of lanes, and segment length.

77 • 𝑤( ): Vector of features for the 𝑗-th year. This variable consists of an intercept (first entry) and an indicator of wind turbines in a 4-mile radius (yes/no) during year 𝑗. The parameters are as follows: • 𝛽: Vector of common coefficients. If the 𝑘-th entry is binary, then it represents the expected log ratio of crashes under the same conditions (e.g., same segment length, AADT, number of lanes, etc.) in the presence/absence of that covariate. • 𝛾 : Intercept coefficient and log ratio of expected crashes under the same conditions in the presence/absence of at least one wind turbine within four miles. • 𝜎 : Standard deviation of the 𝑗-th year. 7.4.3 Results The estimates for the best fit model are presented in Table 8. Each parameter is accompanied by a 95% credible interval (i.e., the probability that a parameter falls within that interval is 0.95). If the interval includes zero, then the parameter is not significantly different from zero (i.e., the null effect of that variable). Moreover, because data for both years were highly correlated, the results in Table 7.1 correspond to a model where both years have a common intercept rather than separate intercepts. AADT was divided by 10,000 for modeling efficiency. Table 8. Parameter estimates of wind turbines crash model. Parameter Mean SD Lower (95 CI) Upper (95 CI) Intercept -2.5961 0.1571 -2.9042 -2.2905 AADT (per 10,000) 0.3917 0.0441 0.3051 0.4774 5 lanes (base: 4) -0.4367 0.1347 -0.7063 -0.1766 6+ lanes (base: 4) -0.317 0.1391 -0.5863 -0.0459 Lane length 2.2909 0.1103 2.0771 2.5094 Nearby wm-year 1 0.0441 0.1454 -0.2492 0.3297 Nearby wm-year2 0.1377 0.1208 -0.1033 0.3683 Std. error yr. 1 0.6031 0.0673 0.4716 0.7374 Std. error yr. 2 0.6795 0.0649 0.5523 0.8079 As expected, crash frequency is most correlated to segment length and traffic volume (AADT). The two variables describing the number of lanes were the only variables that were statistically significant and included in the best fit model. Fewer crashes are expected when a segment has five lanes or six or more lanes. Crashes for 2018 were modeled for segments where a wind turbine was present as of 2017, and crashes for 2019 were modeled for segments where a wind turbine was present as of 2018, and the data for these years are therefore presented separately. Presence of a turbine for both the 2018 and 2019 data was associated with a modest increase in crashes. For 2018 (variable = Nearby wm-year 1), presence of a turbine was associated with an increase of 1.045 (exp(0.0441)), with a credible set of 0.080 to 1.39. The credible set includes zero, which, as noted above, indicates lack of significance. The results for 2019 (variable =

78 Nearby wm-year 2) were slightly higher, with an expected increase of 1.148 (CI = 0.902, 1.445). Although the credible set includes zero, the values were further from zero than for the 2018 data. The lack of significance may be due to sample size. Only two years of crash data were included in the analysis. A large number of wind turbines were installed between 2017 and 2019, so including additional years of previous data would not have yielded a larger sample. The analysis does show some indication that the presence of wind turbines increases crash risk, but this finding would need to be explored further. 7.5 Analysis of OVD Crashes A simple geospatial analysis was conducted that assessed how OVD crashes were coded in the vicinity of wind turbines. As noted in the Safety Framework for crash narratives (Chapter 6), officers did not appear to consider infrastructure elements when noting OVDs on their crash forms. However, since wind turbines are a unique feature, officers may have been more likely to have coded them as distractions. All crashes for 2019 were further examined to determine whether a pattern could be detected for OVD crashes and the presence of wind turbines. Proximity matching was conducted in ArcMap, and the presence of a wind turbine within 2 miles was noted for each crash. As Table 9 shows, 3,307 crashes were located within 2 miles of a wind turbine (17.7% of the 58,554 total crashes in 2019). Of the 58,554 crashes in 2019, 851 were coded as involving an OVD. Of the 3,307 crashes located within 2 miles of a wind turbine, 62 (1.87%) were coded as involving an OVD, while of the 55,247 crashes located more than 2 miles from a wind turbine, 789 (1.43%) were coded as involving an OVD. As a result, an OVD crash was 1.3 times more likely to occur (CI = 1.01, 1.70) in locations within 2 miles of a wind turbine than in locations not within 2 miles of a wind turbine, and the results are statistically significant. Although this was a simple analysis, it suggests that officers may be more likely to code an OVD near an unusual infrastructure feature. It also suggests that some distraction potential exists near wind turbines. Table 9. Comparison of OVD crashes near wind turbines. Total Crashes Coded as OVD Percent OVD Total crashes 58,554 851 1.45% < 2 miles from wind turbine 3,307 62 1.87% > 2 miles from wind turbine 55,247 789 1.43% 7.6 Outcomes and Discussion 7.6.1 Summary Crash evaluations can take a number of forms, such as an evaluation of crashes before and after the installation of various roadway elements, a cross-sectional analysis that compares similar sections with and without a feature of interest, a geospatial analysis that attempts to derive some relationship between crashes and the spatial proximity of an infrastructure element, and so on. The objective of the work described in this chapter was to develop a Safety Framework that

79 agencies could use to assess the impact of an infrastructure element on distraction and then test the efficacy of the approach. This Safety Framework used data from a wind turbine data set and Iowa crash and roadway data sets. The use of data from Iowa provided a unique opportunity for this analysis because nationally Iowa ranks among the top states in wind energy production and the team already had access to the Iowa crash and roadway data. Three different analyses were attempted. First, a set of rural roadways (nonfreeway) in Iowa was selected within 4 miles of a wind turbine, along with a control set of similar roadways. Crashes for 2018 were extracted, but only a total of 16 crashes over 550 miles was found. As a result, a crash analysis could not be conducted with these data. The second analysis compared segments of I-80 across Iowa that were within 4 miles of a wind turbine to segments that were not near a wind turbine. A Poisson-lognormal model was developed. The results showed that crashes were higher on segments near wind turbines, but the results were not conclusive. Only 2 years of crash data were included in the analysis. A large number of wind turbines were installed between 2017 and 2019, so obtaining additional years of previous data would not have yielded a larger sample. The analysis does show some indication that the presence of wind turbines increases crash risk, but this finding would need to be explored further. Using crash data from 2019, the third analysis geospatially linked crashes in Iowa that were coded as involving an OVD to the presence of wind turbines. Proximity matching was conducted in ArcMap, and the presence of a wind turbine within 2 miles was noted for each crash. An OVD crash was 1.3 times more likely to occur (CI = 1.01, 1.70) in locations within 2 miles of a wind turbine than in locations not within 2 miles of a wind turbine, and the results are statistically significant. Although this was a simple analysis, it suggests that officers may be more likely to code an OVD near an unusual infrastructure feature. 7.6.2 Performance Metrics Performance metrics suitable for measuring the success of the use of crash data to assess IRDs were identified, as described below. Feasibility of the Data Sets Utilized The analysis utilized the Iowa crash database, Iowa roadway database, and U.S. Wind Turbine Database. Crash reports are a rich source of data about the locations, characteristics, and contributing circumstances of roadway crashes, especially if the crash locations have been geospatially coded. A key advantage to the use of crash data for safety analyses is that they are an actual indicator of safety. While the use of safety surrogates plays an important role in evaluating and ultimately improving safety, reducing crash severity is usually the end goal of safety activities. Many states capture geocoded crash data, which facilitates the correlation of crash occurrence to infrastructure elements of interest. Since a large number of crashes occur each year, a reasonable sample size can be found for most roadway features of interest. Even for more unusual elements,

80 such as wind turbines, as shown in Figure 18, a large number of roadways are still in some proximity to the feature of interest. Based on the availability of data and likely sufficient sample sizes, crash data sets were determined to be feasible for conducting an evaluation of IRDs. The U.S. Wind Turbine Database was a viable data set for identifying wind turbines. Although the accuracy of the location data was not field verified by the team, the location for each wind turbine is visually verified for the database using high-resolution aerial imagery, and a confidence rating of the location of each turbine is included in the data. Additionally, for each turbine, the information provided includes GPS coordinates, installation year, height of the turbine, and length of various components of the turbine. Source data can be downloaded and used in GIS software to correlate wind turbine information to geocoded crash data (USGS 2020). The wind turbine height was determined to be missing for a number of turbines, so that characteristic is not likely to be useful in further analyses. Because the database includes the installation year, it was possible to ensure that crashes for a particular year correlated to the presence of a wind turbine. Although beyond the scope of this project, this feature of the wind turbine database would facilitate a before-and-after analysis. The Iowa DOT roadway database and Google Street View were also utilized for the analysis and were sufficient for cataloging roadway characteristics as needed. Sample Size It was difficult to determine the relationship between the presence of a wind turbine and crashes. This was likely because only a subset of locations could be selected for inclusion in the analysis due to project resources. With sufficient resources, the sample size just within Iowa could have been increased fourfold. Additionally, the U.S. Wind Turbine Database has locations for over 60,000 wind turbines in 37 states. Even when disaggregated by roadway type, a large sample size could reasonably be obtained for this study. Ideally, a minimum of 50 segments for each independent variable is needed. Performance of Statistical Models A number of different models could be utilized to conduct an evaluation of the impact of wind turbines or other infrastructure elements. Common models include generalized linear models, empirical Bayes, ordered probit, and so on. The analyses described in this chapter were cross sectional in that they compared roadway segments that were near (treatment segments) or not near (control segments) wind turbines for the same time period. Cross-sectional analysis can be challenging. Hauer (2010) suggests that cross-sectional studies have not proven successful in identifying cause and effect in road safety because multivariable regression often does not produce consistent results between studies. Wood et al. (2015) noted that cross-sectional studies do not control for regression to the mean, changes in safety performance, or time trends. Observational before-and-after studies are considered the industry standard for developing crash modification factors due to specific interventions. As a result, the use of before-and-after studies is recommended when feasible.

81 7.6.3 Discussion Although the analyses conducted to demonstrate the efficacy of this Safety Framework were not strongly conclusive, the use of crash data to evaluate IRDs appears to be feasible. The results of the analyses using crash data suggest the following strengths, challenges, and recommendations regarding the use of crash data for the evaluation of IRDs. Strengths include the following: • Crash data provide an actual measure of safety. • Crash data are readily available in most states, and, as a result, significant effort for data collection is not needed. • When differences in data collection and reporting are properly accounted for, data can be combined for different geographic areas (e.g., states, jurisdictions), which increases the sample size available. This is particularly useful when evaluating the impact of rare infrastructure items. For instance, community entrance signs/art installations are not frequent, so their impact could be evaluated using occurrences over multiple states. Challenges include the following: • Except for unique infrastructure elements, crashes coded as involving an OVD do not appear to be a good indicator of IRD, making it difficult to determine whether a crash in the vicinity of a particular roadway element is related to that element. Without indication of an OVD, a crash can only be inferred to be related to the element. • If a correlation is shown to exist between crashes and a roadway element, it is challenging to determine whether distraction was the likely cause of those crashes (if the crash records do not sufficiently document an OVD) because other factors could be present that contributed to the increased number of crashes. For instance, a relationship between overpasses and crashes could be due to distraction or other factors, such as ice or snow lingering under overhead structures due to low sun exposure. Bridges may also attract some types of animals, which could also be a contributing factor in an elevated crash rate. Recommendations for using a crash analysis to assess the impact of infrastructure on distraction include the following: • Evaluation of IRDs using crash analyses is best suited to rural areas. Since it is difficult to attribute a crash to a particular infrastructure element, it is necessary to study areas where the number of overlapping roadway elements is minimized. • The use of crash data to assess IRDs is likely best suited for situations where an infrastructure element is either isolated or uncommon. For instance, elements such as overpasses, wind turbines, community entrance signs/art installations, and so on can be clearly isolated from other features that might also be distracting. • A before-and-after analysis is also a recommended method for evaluation of IRDs. Crashes before and after the infrastructure element is in place can be compared. In this manner, the main difference between the two sets of crashes is the presence of the feature. This method is much more likely to show a difference (if one exists) than a cross-sectional analysis.