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.
125 5 Summary, Conclusions and Recommendations 5.1 Safety and Operational Impacts of Speed Limit Changes 5.1.1 Summary Results This project carried out a considerable number of analyses of the effects of speed limits and other factors on speed choice, crash incidence and crash severity. The analyses drew on a variety data types including loop detector measurements, stated preference surveys, revealed choices, and crash records containing information about crash counts and severities, vehicles and their occupants, and roadways and their environments. The project made extensive use of data obtained from Washington State because of its quality and the effort required to assemble a useful dataset from disparate original sources. However, data from a national driver safety survey, vehicle speed data from Southern California and Austin, Texas, and a national sample of crash records were also used. The analyses applied state-of-the-art statistical methods to address a number of data features that complicate traffic safety analyses. The projectâs datasets and analyses are thoroughly described in Chapter 4 of this report. As explained in Chapter 4, the projectâs research strategy was based on a high-level framework encompassing the relationships between driver speed choice behavior, crash occurrence and injury or crash severity. The project analyses led to the development of various quantitative models of these relationships, and the various datasets mentioned above were then used to estimate them. The data generally dictated the most appropriate methods for their analysis; these methods ranged from ARIMA models (of speed data over time) to weighted least squares regression (of crash counts for road segments clustered on the basis of design attributes), and from random-effects negative binomial models (of crash counts on short roadway segments each year) to heteroscedastic ordered logit models (for crash and injury severity). Although the data did not permit estimation of models that directly link crash counts and severity to actual speed choices39, the results do suggest how speed limits and roadway design and use, along with other control factors, may affect speed choice, crash frequency, and crash severity. Based on these analyses, a number of relatively simple conclusions can be drawn: ⢠A speed limit increase tends to be associated with higher average vehicle speeds. The average vehicle speed rises by less than half of the amount of the speed limit increase itself. ⢠Our predictions of the average speed increase associated with a speed limit increase may be overestimated because our models were based on cross-sectional data rather than actual before-after observations. 39 As discussed later in section 5.4, relatively few network sites offer reliable dual-loop detector stations for accurate recording of speed information. Most sites are urban in nature, and many are missing data. In addition, all aggregate their speed data temporally (e.g., to 5-minute intervals), losing valuable information on instantaneous variations in speed choice. The models of speed choice that were developed from the small subset of Washington State sites having such detectors performed very poorly in prediction. Ideally, one would like to have actual speed data for all sites used to calibrate models of crash frequency (and severity). Such a dataset is not yet available for sufficient sites.
126 ⢠If a roadwayâs design speed (or âsafe speedâ) is also increased, simulations of hypothetical driver behavior suggest that driver optimal speed choices rise by roughly the full amount of the design speed change (assuming that speed limits and design speeds rise together). ⢠Average speed and speed variability tend to reflect highway design and use characteristics more than they reflect speed limits. In fact, after controlling for these other characteristics, variations in observed speed choices seem largely unaffected by speed limits (and by changes in speed limits). ⢠Total crash counts tend to increase with speed limit increases, but not dramatically: other things equal, a 10 mi/h speed limit increase on a typical high-speed roadway can be expected to lead to a roughly 3% increase in total crash counts. ⢠Roadway features other than speed limit also affect the total crash rate. Sharper horizontal curves and steeper vertical grades are associated with higher rates. Roadways with medians tend to exhibit lower crash rates, as do those with wider shoulders. Other things equal, roads with 4-5 lanes tend to have higher crash rates than those with other numbers of lanes. ⢠Using flexible models of crash severity, estimated using occupant-level data, the project found an unambiguous association between speed limit increases and the distribution of crash severity. Models developed from Washington State and national databases both suggest that higher speed limits lead to a significant increase in the proportions of severe injuries and fatalities. However, the models differ (by a factor of approximately 2) in their predictions of the magnitude of these increases. ⢠Because the HSIS-based injury severity model was estimated using only data on high-speed roads, it is felt to be more applicable. For a speed limit increase from 55 to 65 mi/h, the increases in fatal injuries predicted by this model are on the order of 28%; the specific magnitude depends on occupant and roadway characteristics. ⢠The value of this predicted increase in fatalities is roughly consistent with the increases in fatality rates found by several researchers on rural interstates following the 1987 NMSL relaxation. However, this approximate correspondence does not constitute a validation of the HSIS-based model, and it must also be noted that other researchers have found smaller (or no) changes in fatality rates following the NMSL relaxation and repeal. Typically, such work has relied on more aggregate data, using statistical methods less able to exploit data characteristics, and a following a less comprehensive approach than were used here. ⢠Since injury severity and related predictions derive here from cross-sectional models, they will be tend to be overestimated if actual speed differences are higher than those that are estimated following changes in the speed limit. Strictly speaking, these conclusions apply only to the specific datasets that were analyzed and the particular geographic areas from which their data were collected. (The analyses of Washington State HSIS data were especially important in developing many of the conclusions regarding crash incidence.) However, insofar as there is nothing particularly distinctive or unusual regarding the highway, environmental, vehicle or driver characteristics in the geographic areas that provided the data for these analyses, one may consider these conclusions to be more broadly applicable. In any case, considering the effort and time that were required to assemble and analyze these datasets, it was not feasible for the project to attempt a comparably detailed research effort over a wider geographic scope. (This point is discussed further below.) National data were used for the severity model inferences, and these are consistent with the Washington State severity model results.
127 5.1.2 Detailed Results The high-level conclusions listed above are derived from and supported by detailed results from the many analyses of speed and crash data that the project carried out. The main analyses and their results are briefly summarized in the following paragraphs. Again, Chapter 4 describes these in detail, while some of the projectâs secondary analyses are presented in the various Appendices to this report. 5.1.2.1 Speed Choice Models Highway Driving Speeds Reported in the MVOSS The project analyzed results from the 2000 Motor Vehicle Occupant Survey (MVOSS), a nationwide telephone survey of roughly 6,000 persons aged 16 or over. The survey questions emphasized traffic safety issues, including crash exposure, travel choices (such as usual driving speed, driving frequency and seatbelt use), and attitudes towards driving and current speed limits. Basic demographic information about the respondent, and about the type of vehicle usually driven, was also obtained. The projectâs analysis highlighted the intrinsic variability in speed choices across drivers, particularly under relatively uncongested conditions. For example, other things equal, males tend to report driving highways at speeds that are almost 1 mi/h faster than females, college-educated persons 1.6 mi/h faster (than those without college educations), and persons living in central cities/urban areas about 1.3 mi/h faster than rural residents. For every $50,000 rise in household income, drivers report driving about 2 mi/h faster, while those typically driving SUVs report driving 1.6 mi/h slower on average. Speed Choice on Orange County Freeways Variation in speed choices also reflects highway design attributes and environmental conditions. Using 30-sec loop detector data from freeways in Orange County, California in 1998, within- lane average speeds (Table 4-6) were estimated to rise almost 4 mi/h as the number of lanes (in one direction) went from 3 to 5 or more, and by 6.2 mi/h as design speeds rose 10 mi/h. Average section speeds (Table 4-9) were predicted to fall almost 5 mi/h under wet conditions, 4.8 mi/h at nighttime when lighting is provided, and another 3.5 mi/h when lighting is not provided. There is also great variation in average lane speeds across lanes (Table 4-6), with inside lanes averaging 6.2 to 7.3 mi/h more than right-side lanes, and 1.7 mi/h more than the next-to-inside lane. Of course, traffic intensity also has a significant effect, suggesting a reduction of 0.62 mi/h in average lane speed for every added vehicle per lane mile (Table 4-6), or 0.71 mi/h in average section speed (Table 4-9). Estimates of the standard deviation of between-vehicle speeds in the Orange County dataset (based on variations in successive average speed values) suggested that increases in design speed and average speed contribute in minor ways to speed variability. For example, within-lane speed standard deviation (Table 4-5) is predicted to rise 1.39 mi/h with a 10 mi/h increase in design speed, and another 0.46 mi/h with a 10 mi/h increase in the average (within-lane) travel speed. Darkness, wet pavements, obstructions and construction also raise within-lane speed variability.
128 Inside lanes exhibit substantially less variation than their right-side counterparts, and congestion or traffic intensity also is a key factor. Total variations in average speeds across and within lanes oppose many of these effects (Table 4-8), resulting in net effects on total-segment speed variation of 1.8 mi/h following a 10 mi/h increase in design speed, and 0.78 mi/h following a 10 mi/h rise in average section speed. As was expected, predicting traffic speeds is easier than predicting variations in these speeds. The R2 goodness-of-fit values of the Orange County regression models of average speeds were 0.59 or higher (Tables 4-6 and 4-9), while those for models of speed standard deviation (within and across lanes) ranged from 0.12 to 0.41 (Tables 4-5, 4-7, and 4-8). Individual Vehicle Speed Choice in Austin, Texas To complement the analyses based on datasets of aggregate vehicle speeds, a limited set of observations of individual vehicle speeds was collected using a radar gun on a variety of high- speed highways in Austin, Texas. Weighted least squares models were developed to assess the effects of flow, number of lanes and other variables on the average and standard error of individual vehicle speeds. This was the only dataset containing individual vehicle speed measurements that was available to the project. The final model of average vehicle speed included five statistically significant explanatory variables: pavement dry/wet condition, presence of a downstream intersection within 0.25 mile, equivalent hourly lane flow volume, speed limit, and facility access control (Table 4-11). Other things equal, a 5 mi/h increase in speed limit is estimated to result in a roughly 3.4 mi/h increase in average vehicle speed. Facility access control results in a 4 mi/h increase, while wet pavement conditions result in a 3 mi/h decrease in average speeds. Higher flow rates and the presence of a nearby downstream intersection both reduce the average vehicle speed. The adjusted R2 was 0.64, which appears quite satisfactory but may be biased upwards since the least squares assumption of independent error terms is violated by the repeated observations taken with the radar gun. The final model of speed standard error had a minimal goodness of fit (adjusted R2 of 0.01). Estimation results suggest that a 5 mi/h increase in the speed limit reduces vehicle speed standard error by 0.1 mi/h. Other things equal, vehicle speed variations are higher on access controlled freeways, rural facilities, those with more lanes, and near a downstream intersection. Speed Limit Change Intervention Analysis in Washington State WSDOTâs system of permanent traffic recorders (PTRs) provided hourly vehicle counts by speed for four sites from 1995 through 1997. These sites reflect combinations of urban and rural locations, with and without a speed limit increase (of 5 mi/h). ARIMA intervention analysis (Table 4-15) suggests that the 5 mi/h speed limit increase at two of the sites was associated with an average speed increase in the range of 1.2-1.6 mi/h. Speed variance at the rural site increased by roughly 5 mi2/h2, but there was no statistically significant speed variance change at the urban site. Urban congestion may be one reason for these different results. The sites with no speed limit change exhibited practically no change in speed or speed variance.
129 Analysis of Rational Speed Choice Using Simulated Data The project also developed a theoretical model of how rational drivers choose their driving speeds, in order to minimize a generalized cost of travel, based on a highly non-linear formulation. Optimal speeds were found for a wide range of plausible parameter values, and linear regression models were developed to relate the optimal speeds that were developed from this procedure to the key explanatory variables (Table 4-17). It was found that a 10 mi/h speed limit increase results in an increase in chosen speed of between 3.7 and 4.4 mi/h, which is very consistent with actual, observed speed changes. Moreover, estimation results suggested that âsafeâ speeds (for which design speeds may be a reasonable proxy) are more important in determining actual speed choice than are speed limits. However, these two variables do appear to complement each other: a simultaneous increase of 1 mi/h in both speeds results in an almost identical increase in the chosen speed. 5.1.2.2 Crash Occurrence Models The analysis used HSIS data for Washington State, covering the years 1993-1996 and 1999- 2002. Segments were manually clustered on the basis of design details, resulting in a panel of relatively homogeneous clusters. Linear random effects models of total crash counts (all severities combined) were statistically preferred to fixed-effects models (Table 4-20). Crash counts were predicted to rise in a concave, quadratic fashion with speed limits. It was predicted that 3.29% more crashes would occur if speed limits were to increase from 55 mi/h to 65 mi/h on an âaverageâ roadway section. This number fell slightly (to 2.90%) when the data were examined using a before-after regression. This approach reduced the dataset size (per cluster) significantly, but was pursued in order to remove the potential for omitted-variables biases in the speed limit coefficients. In both models it was found that speed limits, right shoulder width, degree of horizontal curvature and the presence of a median were the most important factors affecting crash frequency. 5.1.2.3 Crash Severity Models Crash severity models are concerned with predicting the distribution of injuries by severity, given that a crash has already occurred. Heteroscedastic ordered logit models were estimated, using regional and national datasets. These offered similar results and conclusions. The first model used Washington State HSIS data from 1993 through 1996 for high-speed roadways. The second pair of models relied on a national database (the NASS CDS) of vehicle and crash data from 1998 through 2001, covering all roadway types (not just high-speed roads). This pair of models distinguished single-vehicle from multi-vehicle crashes. All models were based on five injury severity levels for occupants. While the Washington model offered more information on crash site characteristics, the national database contained information on vehicle weight, a potentially key variable that is changing over time. Results from both models suggest that roadways with higher speed limits experience significantly higher fatality rates (Tables Table 4-26 and Table 4-30) â everything else constant
130 (including design and use attributes). This occurs primarily through increases in the (predicted) variance of the modelâs latent error term, but also through increases in the general, latent severity level â particularly in the case of multi-vehicle crashes (as examined with the national database). However, the models differ considerably in their estimates of the magnitude of these impacts; the NASS CDS-based model estimates fatality increases roughly twice as large as those of the HSIS- based model. The statistical association between the probability of fatal injury and speed limit has a concave form. Since actual, practical speed limits lie below the speed limit associated with the highest fatal injury probability, the increase in fatal injury probability tapers with increases in speed limits. This may occur due to a compensation effect, where drivers drive more carefully at higher speeds. It also may be due to a threshold or saturation effect (where drivers and their vehicles are not willing or able to travel any faster), or to latent variable effects (such as the highest-speed roadways occurring only in the safest driving environments.) The HSIS-based model is felt to be more applicable to the analysis of the impacts of speed limit changes on high-speed roads, because it was estimated using only data from such roads. It is likely that the predictions of the NASS CDS-based model for high-speed roads are unduly affected by its observations for lower-speed facilities. Focusing on the HSIS-based model, the probability of sustaining a fatal injury following a crash is estimated to rise by 24% if the speed limit increases from 55 to 65 mi/h, and by 12% if the increase is from 65 to 75 mi/h. Other predictions can be found in Table 4-27. However, to the extent that actual speed choices do not rise as much following a speed limit change as they appear to do in cross-sectional databases, these severity results may be overestimated. The predicted increase is roughly consistent with the increase in fatalities on rural interstates found by several researchers (Rock, 1995; Ledolter and Chan, 1996; Brownstone, 2002) in their analyses of the 1987 NMSL relaxation. However, this cannot be considered a âvalidationâ of the HSIS-based model because of the many difficulties associated with prior worksâ analysis of aggregate crash statistics, due to the effects of multiple confounding changes (such as route changes). Moreover, other researchers conducting aggregate analyses of the NMSL relaxation and repeal have found small or no fatality rate impacts. 5.2 Non-Safety Impacts of Speed Limit Changes The project also examined non-safety impacts associated with speed limit changes; this was, however, a lower priority activity than investigating their safety impacts. The examination was based in part on a review of the relevant technical literature, as well as on survey responses received from state DOTs. Some state DOTs carried out studies of impacts of the NMSL repeal, and some of these studies included qualitative consideration of such non-safety factors. Available reports from these studies were obtained and reviewed as well for information on non- safety impacts.
131 In broad terms, non-safety impacts of speed limit changes may include effects on economic, environmental and/or commercial conditions. Unfortunately, generally applicable conclusions regarding such effects are mostly lacking. Speed limit increases translate into less-than-equivalent increases in average travel speed. The project found, for example, that a 10 mi/h speed limit increase would result in average travel speeds roughly 4 mi/h higher, provided that other factors such as congestion did not constrain travel speeds. The reduced travel times made possible by higher travel speeds have an economic value. US DOT (1997, 2003) guidelines, for example, suggest that the travel time of intercity passengers on surface modes should be valued at approximately $15/h on average. When considering the system-wide impacts of a speed limit change, it must be remembered that not all trips will be fully affected by the change. For example, trips for which the average speed is significantly constrained by congestion will not experience the full effect of a speed limit change whereas those less affected by congestion are likely to experience greater impacts. Moreover, to the extent that higher speeds translate into a slightly higher crash rate (the project found that a 10 mi/h speed limit increase would result in a crash rate increase of roughly 3%), the travel delays resulting from crashes (known as non-recurrent congestion) will also increase, offsetting somewhat the reduction in travel times made possible by higher average speeds. Finally, travel time reliability also has an intrinsic economic value (Small et al., 1999), and the reduced time reliability resulting from slightly higher crash rates at higher speeds would also offset to some extent the economic value of the lower travel times. Changes in average travel speed also affect vehicle operating costs. Of the various cost components that contribute to overall operating costs, running costs (those that directly result from vehicle operation) are most significantly impacted by speed; and of running cost components, fuel consumption costs are the largest portion. For a medium or large car, fuel consumption at 55 and 65 mi/h calculated using the FHWAâs HERS-ST model (FHWA, 2002) with an economic fuel cost of $1.50/gallon40 shows that the 4 mi/h average speed increase noted above would lead to an operating cost increase that is roughly half the estimated value of travel time savings. Thus, the net (time and cost-related) benefit resulting from higher average speed is further reduced. The project reviewed the two main approaches used to quantifying the economic costs of injuries and fatalities: the human capital approach, and the willingness-to-pay approach. With respect to the environmental impacts of speed limit changes, the little evidence available suggests that these are small to negligible. The noise impacts of post-NMSL speed limit increases were modeled in New Jersey using actual before-after speeds and traffic volumes on affected facilities. It was concluded that the change in noise level would be imperceptible in the noise environment on and surrounding the roadways. Air quality issues have been evoked in some locations as a reason for lowering speed limits. (These are sometimes called environmental speed limits.) In the Houston-Galveston (Texas) area, for example, preliminary modeling analysis of a proposal to reduce speed limits on high- 40 $2/gallon pump price minus $0.50/gallon transfer payments such as Federal and state taxes.
132 speed roads suggested that such limits would improve compliance with EPA air quality standards. Opposition to this suggestion led to an analysis using a more up-to-date air quality model that predicted a much smaller change in air quality from a speed limit reduction, and the proposal was suspended. The New Jersey study mentioned in the preceding paragraph also investigated air quality changes resulting from the higher speed limits, and found these to be ânominalâ (i.e. insignificant). The project was unable to find any empirical or documentary evidence regarding possible commercial impacts of speed limit increases. The resulting (smaller) increases in average speeds of commercial vehicles should, in the medium to long term, result in opportunities for more efficient transportation and business operations. However, such speed changes are typically small, and the productivity of a commercial vehicle (and of the operations that it serves) depends only partly on its travel speed since it may spend significant time in loading/unloading operations or waiting for cargo. Thus, the impacts on business and commerce of speed limit changes are likely to be marginal. 5.3 Enforcement Policy Responses to the NMSL and its Repeal Institutional memory concerning the specific decisions that were taken by DOTs and state police agencies at the time of the large-scale speed limit changes is beginning to be lost. Many of the personnel who were involved in developing high level policies and strategies to respond to the NMSL imposition, relaxation and repeal were, at that time, in relatively senior positions. Many of these officials are no longer accessible. Consequently, much the detailed information about the policy and strategy responses of DOTs and State Police, and about how these responses were developed, is now either unavailable or only anecdotal in nature. It is sometimes claimed that the NMSL imposition and related Federal mandates led to a systematic concentration of speed limit enforcement efforts on high-speed roads, to the detriment of potentially more beneficial traffic enforcement efforts of other kinds or on other facility types. Available data from DOTs and state police agencies do not allow a rigorous investigation of this assertion. Similarly, available data do not allow a rigorous investigation of the converse hypothesis, namely that the NMSL relaxation and repeal were accompanied by a widespread redeployment of enforcement resources away from speed enforcement on high-speed roads and towards other activities with potentially higher traffic safety benefits. Nonetheless, anecdotal evidence collected by the project through surveys of state DOT and police officials across the country does suggest that neither of these things happened systematically or on a large scale. Some respondents acknowledged that there was a concern to demonstrate compliance with the NMSL in order to avoid Federal sanctions. However, respondents were adamant that no enforcement actions taken during the period of the NMSL were of a nature to compromise traffic safety. Similarly, respondents cited no examples of systematic changes in enforcement practices away from speed limit enforcement on high-speed roads following the NMSL relaxation and repeal. Indeed, several respondents and DOT reports noted that speed limit enforcement activities actually became more intensive on high-speed roads in the period following the repeal, out of concern that drivers who formerly ignored the 55 mi/h limit might continue their scofflaw habits at the higher speed limit, with potentially more dangerous consequences.
133 The evidence suggests instead that the response of most police agencies to the NMSL relaxation and repeal generally took more measured forms: for example, reduced tolerance for speeds higher than the new limits together with, in some cases, a new speeding fine structure and/or an aggressive information campaign to notify the public of the tougher post-repeal policy. 5.4 Data Recommendations The methods used in this work were guided, and limited, by the extent and quality of existing datasets. For example, Washington Stateâs HSIS dataset is felt to be the best that the U.S. presently offers, but its panel datasets are missing key years (1997 and 1998). The dual-loop detectors in Washington Stateâs northwest region were originally thought to provide speed averages at 30-second intervals, but it was found that the original detailed data had been lost through aggregation to 5-minute intervals. Although the characteristics of the available data frequently constrained the types of analyses that the project could perform, the datasets that were assembled and used by the project were typically of a quality higher than (and at least comparable to) those that are generally available elsewhere in the U.S. and abroad. Thus, the data limitations present in the project datasets are likely also to be present in all but very specialized and focused traffic and crash datasets available elsewhere. Broadly speaking, datasets covering extensive geographic areas are likely to be less detailed, while those that include very detailed data are likely to focus on relatively limited geographic areas, highway facilities and/or time periods. The ideal dataset for traffic safety research purposes would offer true counts and speeds, fully integrated data on design, operations and crashes for a wide range of sites (on the order of at least 500 centerline miles, rural and urban), over several years, both before and after speed limit changes. Exposure (VMT) would be accurately estimated, rather than derived from very imprecise estimates of AADT based on a sparse set of periodic (i.e. occasional) short-term traffic counts, as is frequently the case. Towards this goal, the project has a number of recommendations regarding future data collection efforts to support fundamental research into crash causality and characteristics, but these recommendations are conditioned by the considerations expressed above. Research-oriented data collection efforts should, as much as possible, complement and build on the crash, traffic, and highway inventory data collection efforts routinely carried out. Given these sources of currently available data, it is worthwhile to focus research-oriented data collection in a few specific ways. These recommendations echo and parallel those of a recent government review of the NHTSA grant program that helps states improve their safety data systems (GAO 2004). First, traffic safety research would benefit from the collection and assembly of additional types of information on the characteristics of roadways and their environments. This could include information on pavement and weather conditions; the presence and nature of embankments, barriers and culverts; driveway and cross-road frequencies; clear zone width; and sight distances. None of the datasets that the project analyzed contained such data. As explained in Chapter 4, one of the analytical difficulties that had to be confronted was the potential for correlations
134 between speed limits and unobserved roadway and environmental characteristics such as these. As discussed in the report, such correlations can bias speed limit impact estimates by attributing to speed limits some of the effects that are actually due to the unobserved characteristics. A dataset containing such data could considerably reduce this difficulty by allowing the effects of these characteristics to be estimated explicitly. However, this work's analysis of crash rate changes resulted in estimates similar to those arising from an analysis of counts (as described in section 4.3), suggesting that this issue may not lead to practically different conclusions. Second, as a practical matter it would be more efficient to concentrate near term research- oriented data collection efforts on a subsystem of the overall highway system. This would ideally be a subset for which some of the required research-related traffic safety data already exist in some form, and for which the remainder can be expeditiously collected and processed. The high-speed roadway subsystem would seem to be a good initial candidate in this regard. Over the longer term, it would be desirable to extend such data collection efforts to other components of the overall system. It should be noted that the number of urban areas deploying high-performance traffic sensor systems continues to increase. Such instrumentation and the associated data processing systems can be used to support freeway and/or arterial management systems, incident response systems, and advanced traveler information systems (ATIS), among other uses. The data generated by these traffic measurement systems is frequently preserved and stored; indeed, the on-going Federally-sponsored Archived Data User Service (ADUS) represents a national significant effort to standardize and make available traffic and operations data from traffic sensor systems and other ITS components around the country. The project examined most of the metropolitan areas with currently operational traffic sensor systems as possible sources of data for its analyses and model development activities. For a variety of reasons, the data from most of the examined systems were found to be unsuitable for project use. Some systems, for example, only covered a relatively small length of roadway, so that the number of crashes occurring on them would be too small to constitute a statistically valid sample. Others aggregated the archived traffic data into time intervals that were too long to be useful for the projectâs disaggregate analysis of traffic characteristics. In those cases where the data could potentially have been used by the project, the task of assembling and integrating the disparate sources of required data (highway inventory, traffic and crash data) exceeded the resources available to the project. However, it is likely that over time local agencies will find it advantageous to develop and maintain such integrated datasets themselves, and as this happens these will become an increasingly valuable and accessible source of data for traffic safety research. Towards this end, data producing agencies should be encouraged to adopt consistent geo- or linear referencing systems to facilitate the assembly of integrated sets of disparate data types. Furthermore, agencies should be encouraged to preserve collected data in the most disaggregate form feasible, rather than aggregating it in order to reduce its archiving costs. The declining costs of data storage should make this option more attractive to agenciesâ data services.
135 5.5 Overall Conclusions and Recommendations The NMSL was adopted in 1974 in response to the first energy crisis. Its adoption, together with its relaxation on rural interstate highways in 1987 and its complete repeal in 1995, created the conditions for a unique large-scale natural experiment on speed limits and their safety and other effects. It is not likely that our nation will have another occasion to experience speed limit changes on such a broad scale in the foreseeable future. It is clear that the more dire predictions that were made about the likely safety impacts of the NMSL relaxation and repeal have not come to pass. Although some researchers have found significant changes in the crash experience of roadways that underwent speed limit changes, others have not, and it is fair to say that a broad consensus as to the effects of the speed limit changes still has not emerged. This suggests that at an aggregate level the overall magnitude of such effects, if indeed they exist, is as small as or smaller than those of changes in a wide variety of other safety-related factors that were occurring at the same time as but (mostly) independently of the speed limit changes themselves. Such changes include, among others: ⢠variability in weather conditions; ⢠improvements in roadway design; ⢠changes in DUI and young driver laws; ⢠changes in traffic police practices and policies; ⢠changes in driversâ seatbelt usage habits; ⢠more effective driver education and public traffic safety awareness programs; ⢠demographic shifts in the driving population, including driver ages and gender distributions; ⢠changes in driving patterns, including the distribution of travel between day and night hours, urban and rural locations, and interstate and other facility types; ⢠improved safety features in vehicle designs; ⢠increases in VMT per lane mile of network capacity and increases in congestion; ⢠increased usage of in-vehicle communications devices (e.g. cellular telephones), leading to more rapid notification of and response to crash situations; and ⢠improved capabilities and effectiveness of emergency response services. The aggregate combined effects of these changes, together with whatever effects the speed limit increases themselves may have had, appear to have been small. This project carried out much more detailed disaggregate-level analyses, however, and the conclusions that emerge from these are somewhat clearer. The project found that small (roughly 3%) increase in total crash rates are associated with a speed limit increase from 55 to 65 mi/h on an âaverageâ high-speed roadway section. It found that a significant increase in the probability of fatalities and incapacitating injuries are associated with higher speed limits. For this particular 10 mi/h speed limit change, a 24% increase in the fatal injury probability would be expected. These predictions would of course be different for different roadway sections and speed limit changes.
136 Application of the cross-sectional models underlying these crash-severity predictions may tend to over-predict the impacts of speed limit changes because actual changes in average travel speeds following changes in speed limits may be lower than those observed across a set of existing roadways with different speed limits. Nonetheless, even if actual speed changes are expected to be 50% lower than those implied by the cross-sectional models, their impact on the crash fatality rates (and more generally on the injury severity distribution) would in many cases remain statistically and practically significant.
