**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

*Managing Speed: Review of Current Practices for Setting and Enforcing Speed Limits -- Special Report 254*. Washington, DC: The National Academies Press. doi: 10.17226/11387.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

*Managing Speed: Review of Current Practices for Setting and Enforcing Speed Limits -- Special Report 254*. Washington, DC: The National Academies Press. doi: 10.17226/11387.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

*Managing Speed: Review of Current Practices for Setting and Enforcing Speed Limits -- Special Report 254*. Washington, DC: The National Academies Press. doi: 10.17226/11387.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

**Suggested Citation:**"Appendix C: Effect of Speed Limits on Speed Distributions and Highway Safety." Transportation Research Board. 1998.

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.

Appendix C Effect of Speed Limits on Speed Distributions and Highway Safety: A Survey of the Literature Patrick McCarthy Department of Economics, Purdue University The purpose of this review is to examine recent work on the effect of motor vehicle speed limits on highway speeds and highway safety. The review is empirical and concentrates on identifying the quanti- tative effects of changes in regulatory speed limit policies on the dis- tribution of speeds and traffic safety. In general, the proposition that changes in speed limits induce observed changes in the distribution of speeds on a road network and have implications for the network's highway safety is straightforward. However, testing this proposition 277

MANAGING SPEED 278 in a scientific inquiry to reliably answer such questions as "How will a 5-mph (8-km/h) increase in the speed limit alter the average speed of travel on a road? Will the number of fatal crashes increase if driv- ers, on average, travel 5 mph faster? What effect will a 5-mph increase in the speed limit have on the fatality rate?" is not straight- forward. Through local, state, and federal provision, the country's trans- portation system comprises a network of roads that differ by size, quality, and location. The attributes of each road's users (e.g., socio- economic characteristics of the drivers, proportion of truck traffic) over some time period will generally differ by type, quality, and loca- tion of road. In combination with differential speed limits, traffic enforcement, and other government interventions, the nation's high- ways produce trips and, as a by-product, highway safety outcomes. If all roads and all road users were homogeneous, then determining the effect of alternative speed limit policies on speed distributions and highway safety would be relatively easy. However, the fact that both the individual components of a road system and its users are hetero- geneous complicates the task of identifying the effects of changes in speed limit policies on highway speed distributions and safety. Over the past 24 years, the U.S. Congress has passed three major pieces of federal legislation related to speed limits. First, responding to the oil crisis in the early 1970s, Congress passed the Emergency Highway Energy Conservation (EHEC) Act in 1974. Among its provisions, the act mandated a 55-mph (89-km/h) national maxi- mum speed limit (NMSL) on all U.S. highways and threatened a loss of highway funds if states did not sufficiently enforce the limit. A Transportation Research Board (TRB) study (1984) concluded that implementation of the 55-mph national speed limit saved 2,000 to 4,000 lives annually. In 1987, Congress passed the Surface Transportation Uniform Relocation Assistance (STURA) Act, which gave each state the right to increase speed limits on portions of the Interstate system lying within the least-populated areas of its boundaries. Thirty-eight states immediately responded to the legislation by raising speed limits on their rural Interstate highways, followed in 1988 by 2 additional states. Since passage of the 1987 legislation, there have been numer-

279 Effect of Speed Limits on Speed Distributions and Highway Safety ous national, regional, and statewide studies analyzing the effects of relaxed rural Interstate speed limits on highway safety. Most recently, Congress passed the National Highway System Designation Act of 1995, which gave states complete freedom to set speed limits within their jurisdictional boundaries. To date, little sci- entific information is available on the effect this has had on speed distributions and highway safety. By examining the empirical relationships among speed limits, speed distributions, and highway safety on nonlimited- and limited- access roads, this paper complements recent international reviews of speed limits and highway safety (Fildes and Lee 1993; Knowles et al. 1997). In general, this review covers domestic and international speed limit studies that have been conducted since the extensive 1984 TRB review. For a subset of studies, this review critically analyzes both the studies' findings and the strengths and weaknesses of their method- ological approaches. This review focuses on recent research that has analyzed the impli- cations for highway safety as a direct result of the relaxed (rural) Interstate 65-mph (105-km/h) speed limit embodied in the STURA Act. There are three primary reasons for this focus. First, passage of the STURA Act was national in scope, affecting all roads, Interstate and non-Interstate, posted with speed limits above 55 mph (89 km/h). Of the rural and urban Interstate highways in 1990 with eli- gible mileage, 98 and 97 percent, respectively, had posted speed lim- its of 65 mph (NHTSA 1992). Moreover, although accounting for only 7.3 percent of total lane miles in the national transportation net- work, Interstate highways account for 26.1 percent of vehicle miles traveled (FHWA 1995). Factors that affect travel on limited-access high-speed roads have the potential for significantly affecting high- way safety. Second, although it was national in scope, the 1987 STURA Act did not roll back the 1974 legislation. Rather, the act modified the structure of speed limits on limited-access roads by per- mitting higher speeds on sections of the system that past research had identified as the safest. This raises interesting and contentious issues concerning the law's effect on highway safety. Concentrating higher speed limits on the safest parts of the transportation network makes more plausible, for example, the controversial possibility that

MANAGING SPEED 280 "fine-tuning" Interstate speed limits may have actually raised overall safety on the nation's highways. Third, a sufficient amount of time since the law's 1987 passage has elapsed for researchers to conduct longer-term studies on the law's varied effects. In contrast to short- term "impact" studies that, at times, may provide misleading infor- mation to policy makers, studies based on a longer sample of postenactment data are likely to more fully capture the traveling pop- ulation's myriad adjustments to the new environment. This paper is organized as follows. First, an overall framework within which to place existing empirical studies on the effects of speed limit laws is discussed. The methodological constructs most frequently used to empirically evaluate the effect of speed limit laws on speed distributions and highway safety are summarized. The known information on the effects of changing posted speed limits on the distribution of speeds is discussed. The effect of speed distribu- tions on highway safety is examined. The findings of various studies examining the direct and indirect effects that relaxed speed limits on rural Interstate highways have had on the motoring public's safety are reported. International experience with speed limits is reviewed. Finally, a number of areas for further research are identified, and con- cluding comments are given. SPEED LIMITHIGHWAY SAFETY FRAMEWORK Figure C-1 shows the relationship between speed limits and highway safety on a given road. The basic mechanism between speed limits and highway safety is shown in the middle of the figure. Put simply, speed limits, among other factors, influence drivers' choices of opti- mal speeds. Solid arrows indicate a direction of direct causality. As shown by the figure, the posted speed limit is one of many objective factors that directly feed into a driver's speed decision. Other impor- tant determinants include highway and vehicle design, traffic enforcement and other highway government interventions (e.g., safety belt use laws), environmental attributes (e.g., weather condi- tions, topography), and characteristics of the driving population (e.g., proportion of younger drivers). At the same time, a driver has under- lying preferences for risk and, in any specific driving situation, a sub-

281 Effect of Speed Limits on Speed Distributions and Highway Safety Figure C-1 Relationship between speed limit and highway safety. jective view of traffic safety. By combining to determine a driver's optimal speed in each travel environment, these objective and subjec- tive factors produce a distribution of speeds on the roadway and a set of safety outcomes (e.g., number and type of crashes, number and type of injuries). Notice also that a driver's optimal speed is a latent variable. Although analysts can observe posted speed limits, measure the dis- tribution of speeds on roadways, and observe safety outcomes, analysts cannot directly observe a driver's optimal speed. However, analysts can potentially (i.e., under certain conditions) infer the empirical effect of highway safety policy on a typical driver's optimal speed by examining the effect of policy on the road's distribution of speeds. In addition to direct links between causal or determining factors, denoted by dotted-line boxes, and determined factors, denoted by

MANAGING SPEED 282 solid-line boxes, there are indirect links that represent feedback effects. The figure shows the feedback roles that speed distributions and highway safety play on roads. Not only does speed distribution affect highway safety, but also a road's safety record has a feedback effect on the road's distribution of speeds (through its effect on indi- vidual drivers' optimal speed decisions). Further, a road's safety record will ultimately have long-term effects on highway and vehi- cle design, governmental policy, and the topic of this review--posted speed limits. Two other points about Figure C-1 are worth mentioning. First, if, as hypothesized, the indicated linkages represent a given type of road (e.g., rural Interstate highway) in the system, then similar link- ages will characterize other roads in the system (e.g., urban Interstate highways, arterials). Each road does not operate in isolation but is linked with other roads in the network, which implies that policy changes specific to one type of road are not likely to produce effects specific to that road but will influence travel behavior on other roads in the network. Traffic diversion, reflecting changes in route choice, traffic generation, reflecting latent travel demands, and spillover effects, whereby travel behavior on the affected road carries over onto other roads, are indicators of these secondary or indirect effects of a tar- geted highway policy. Second, most empirical studies of highway safety, particularly as they relate to the 65-mph (105-km/h) speed limit, fall into one of three general categories. A relatively small group of studies examines how posted speed limits are set--in practice, what factors traffic engineers take into account when posting speed limits on alternative roadways. A related question concerns the effect of changes in the posted speed limit on speed distributions and driver compliance with the posted speed limit. A second group of empirical studies examines the relationship between attributes of the speed distribution, specifi- cally average speed and speed dispersion, and highway safety. An important issue in this research is the role that average speed plays in highway safety after controlling for the statistical effect of speed dispersion. The last, and largest, set of empirical studies focuses on the effect of changes in the posted speed limit on highway safety. In Figure C-1, this is the direct link between the Speed Limit

283 Effect of Speed Limits on Speed Distributions and Highway Safety box and the Highway Safety box (indicated by the dashes). Although a large body of research on this issue is extant, a consensus has yet to form on whether increasing speed limits, particularly on limited- access roads, deteriorates safety. Part of the current debate in this area concerns how the law affected alternative highway safety mea- sures. METHODOLOGICAL CONSIDERATIONS Empirical studies examining the effect of changes in posted speed limits on speed distributions or highway safety generally identify the primary null hypothesis as either (or at times both) of the following: H01: Increased speed limits have no effect on highway speed dis- tributions. H02: Increased speed limits have no effect on highway safety out- comes. The respective alternative hypotheses are as follows: HA1: Increased speed limits have a nonzero effect on highway speed distributions. HA2: Increased speed limits have a nonzero effect on highway safety outcomes. Depending on the analyses' objectives, there are numerous variations of these hypotheses. Highway speed limits, for example, may be referred to as having "direct" or "indirect" effects. A direct effect refers to the highway directly affected by the speed limit change. Indirect effects, on the other hand, correspond to speeds on highways whose speed limits have not changed. The direct effect of the STURA Act in 1987, for example, is the effect of a 65-mph (105- km/h) speed limit on rural Interstate speed distributions; the effect on speed distributions of all other roads is the indirect effect. If, all else being constant, relaxed speed limits on one road produce changes in the distribution of speeds on other road networks whose speed limits have not changed, there is said to be a "spillover" or "tainting"

MANAGING SPEED 284 effect associated with the speed limit change. Often, various null hypotheses are tested on characteristics of the distribution of speeds (on the affected and unaffected roads), including average speed, speed dispersion, 85th percentile speed, and the proportion of drivers traveling above a given speed. Also of interest are hypotheses related to a speed limit's effect on highway speeds by type of vehicle, road type, time of day, and a variety of other factors that differentiate travel in the highway system. Similarly, numerous hypotheses concerning the effect of relaxed speed limits on highway safety outcomes are tested. As with highway speed distributions, these hypotheses often focus on direct and indi- rect effects, as well as the effects by vehicle type, road type, time of travel, location, alcohol consumption, and socioeconomic factors (e.g., age and gender). Although most of the attention is on fatal crashes and fatalities, some studies also identify the effect of relaxed speed limits on the severity of crashes. In the empirical literature on speed limits and their effects, three methodological approaches have typically been used to test hypothe- ses concerning the effects of changes in posted speed limits: paired comparisons, regression analysis, and time series analysis. Paired Comparisons Ideally, testing null hypotheses on the effects of altered posted speed limits would require an experimental design whereby the analyst ran- domly selects a set of homogeneous roads (i.e., roads that are physi- cally identical and that have an identical user profile) for analysis. The analyst randomly divides the set of roads into two subsets, a con- trol group and an experimental group. The analyst then alters the speed limit for those roads in the experimental group, observes speed distributions and safety outcomes, and tests whether these outcomes are statistically different from similar measures for the control group. A benefit of this methodology is that the analyst can then draw infer- ences for the population of roads under study [e.g., a 10-mph (16- km/h) increase in rural Interstate speed limits increases average rural Interstate speeds by 3.2 mph (5.1 km/h), all else held constant]. Depending on whether the null hypothesis is accepted (rejected), the

285 Effect of Speed Limits on Speed Distributions and Highway Safety analyst concludes that the results are not (are) consistent with an altered speed limit having an effect on speed distributions and high- way safety. Often, to obtain additional insights on the extent to which the observed effect is (is not) close to 0, the analyst reports confidence intervals. Unfortunately, use of an experimental design approach to analyze the effect of speed limit changes is not generally feasible. Even assuming that one could sufficiently control for differences in a set of randomly selected roads to isolate the effect of changing speed limits, transportation agencies are reluctant to participate in such experiments for a variety of reasons. The alternative to an experimental design approach is a quasi-experimental methodology, which recognizes that speed limit changes do occur on some roads while not on others. In quasi-experimental procedures, there is a set of nonrandomly chosen roads (e.g., rural Interstate highways) on which the speed limit has changed--the experimental group--and a set of nonrandomly chosen roads on which the speed limit has not changed (e.g., rural non-Interstate highways)--the comparison group. After controlling for confounding factors that reflect heterogeneity across roads and road users, the analyst tests whether there are statistically significant differences between speed distributions and highway safety outcomes on the affected and unaffected roads. Depending on whether the null hypothesis is accepted, the analyst concludes that the change in speed limit has or has not had an effect. Paired comparison approaches come in many forms. In some cases, the analyst draws conclusions from simple comparisons of average speeds and safety outcomes for the affected and unaffected roads. Other applications use more formal testing procedures to determine whether there is a statistically significant difference in speeds or safety, or both. Further, the approach is used across roads at a given point in time (e.g., a comparison between states that did and did not increase rural Interstate speeds) or across time (a comparison of affected states before and after the speed limit change). A common approach is the use of odds ratios. Consider the fol- lowing table:

MANAGING SPEED 286 After Before Speed Limit Speed Limit Increase Increase 65-mph (105-km/h) rural Interstate states n11 n12 55-mph (89-km/h) rural Interstate states n21 n22 where nij (i, j = 1, 2) is the number of fatal crashes in state type i and speed limit environment j. Relative to 55-mph (89-km/h) states, the odds of a fatal crash in a 65-mph (105-km/h) state prior to the speed limit increase are (n12/n22). After the speed limit increase, the odds of a fatal crash in a 65-mph state relative to 55-mph states are (n11/n21). Thus, the odds ratio is (n11/n21)/(n12/n22). If the speed limit increase in 65-mph states has no effect on fatal crashes, then the odds ratio will be 1. Alternatively, if the law significantly increased (decreased) fatal crashes, then the ratio would be greater (less) than 1. Analysts typically calculate chi-square tests and confi- dence intervals to test null hypotheses using these methods. A primary advantage of quasi-experimental methodologies based on paired comparison approaches is that very little information is needed to conduct the test. In the preceding example, four bits of information are sufficient to test the hypothesis. However, the example, as with all paired comparison analyses, has a maintained hypothesis that relating outcomes in the experimental group [i.e., fatal crashes in 65-mph (105-km/h) states] to those in the compar- ison group [i.e., 55-mph (89-km/h) states] controls for all differ- ences between the two groups in all other determining factors. Whether these techniques provide sufficient control for confound- ing factors that may influence the variable of interest and, accord- ingly, affect the test results and the associated policy implications is an important empirical issue. Analysts often stratify the sample by other variables (e.g., highway exposure, socioeconomic character- istics) to explicitly control for other determining factors. In most cases, however, the analyst uses univariate stratification, that is, stratification of the sample one variable at a time. Simultaneously stratifying by multiple variables would enable an analyst to better isolate the effect of altered speed limits, although a practical limita-

287 Effect of Speed Limits on Speed Distributions and Highway Safety tion of multivariate stratification is that it generates many empty cells. Regression Models An alternative procedure for analyzing the effect of altered speed limits on highway speeds and safety is to develop a statistical model that not only includes the relevant policy variable but also controls for other confounding factors. This is the regression approach, whose aim is to estimate equations of the following general form: N k j xit,j Hit i it i=1 j=1 where Hit = highway outcome (e.g., fatal crashes, fatality rate, injury crashes, injury rate) for cross section i and time period t (i = 1,..., N; t = 1,..., T ), xit,j = jth explanatory variable for cross section i and time period t (i = 1,..., N; t = 1,..., T; j = 1,..., k), j = parameter reflecting the marginal effect of the jth explana- tory variable on the highway outcome ( j = 1,..., k), and = error term for cross section i and time period t (i = 1,..., N; it t = 1,...,T ). The data for this regression are cross sections over a period of time, called a pooled data set, an example of which is the number of annual fatal crashes in each state from 1970 through 1995. The formulation in the preceding equation is often referred to as a fixed effects specification because the model includes separate parameters ( i) to reflect each of the cross sections included in the analysis. This pooled data formulation is a general specification that, depending on one's data set, collapses to simpler econometric models. The two most common are time series regression models and cross-section models. In a time series regression model, N = 1 and there is a single cross section (e.g., annual nationwide fatal crashes from 1970 through 1997, or the monthly fatality rate from January

MANAGING SPEED 288 1980 through December 1992). Further, if the model included only a constant term and a time trend as the only explanatory variable, then the regression equation would model historical trends. Alternatively, a cross-section model is based on a cross section of observations at a single point in time (e.g., fatal crashes in each county for 1994, or total crashes for each state in the nation in 1996). To the extent that the analyst can obtain accurate information on other confounding factors that affect the dependent variable of inter- est, a well-specified regression model controls for the statistical influ- ence of the confounding factors and better isolates the independent effect of the policy. In general, there are two difficulties with this approach. First, estimation of regression models is subject to several statistical pitfalls. In time series analysis, for example, error terms may be serially correlated, which, if not corrected, invalidates hypoth- esis tests. Alternatively, regression models based on highly collinear data are generally unable to isolate the independent effects of the collinear variables. Thus, the potential advantages of regression model approaches will be realized to the extent that the analyst tests and, if necessary, corrects for statistical and other problems encoun- tered in regression techniques. Second, the ability of the regression model approach to control for other determining factors implies that regression analysis is generally more data intensive. Unavailable, inappropriate, or unreliable data as well as time or resource con- straints on data collection may cause researchers to develop and esti- mate relatively simple models that fail to adequately control for a larger set of relevant determining factors. Similar to a paired comparison methodology, regression models may also reflect a quasi-experimental approach. Consider, for example, a regression analysis of fatalities on 50 observations, where each obser- vation represents the number of fatalities in a state during 1990. Because not all states with eligible mileage increased their rural Interstate speed limits when Congress passed STURA, the sample of observations includes states with 65-mph (105-km/h) highways and states with 55-mph (89-km/h) highways. Typically, a variable is included in the model to reflect a state's maximum speed limit status; the variable is equal to 1 if the state increased its speed limit on rural Interstate highways and 0 otherwise. Because the sample includes

289 Effect of Speed Limits on Speed Distributions and Highway Safety experimental states along with comparison states, the approach is quasi-experimental. If the analyst cannot reject the null hypothesis that the coefficient on the 0-1 speed limit variable equals 0, the inference is that the enactment of STURA is consistent with a null hypothesis that the act had no effect on fatalities, holding all else constant. Another example of a quasi-experimental approach within a regression framework occurs when there are no explanatory variables in the model. In this case, the model includes a constant, which reflects the overall mean of the dependent variable, and a set of 0-1 variables that reflect treatments. The analyst estimates this model, called analysis of variance (ANOVA), to examine whether the treat- ments (e.g., relaxing speed limits on rural Interstates, type of road) in the experimental states have a significant effect on the overall mean relative to the comparison states. Interrupted Time Series Analysis A third methodology often used to analyze the effect of altered speed limits on speed distributions and highway safety is time series inter- vention models. If a time series of the monthly fatalities from 1976 to 1990 were examined, two features would probably be immediately apparent. First, the series would exhibit a declining trend. Second, a repeating cyclical pattern that reflects seasonal variations in fatalities would appear. An analyst's objective is to develop an autoregressive integrated moving average (ARIMA) model that accounts for the trend of the series, seasonal patterns, and any serial dependencies that exist in the series itself or in the error term. In effect, ARIMA mod- els decompose the behavior of the series into three components: trend, which is captured by an integrated component of the model; autoregressive and moving average components that explain the cur- rent observation in terms of past observations and random shocks; and seasonal terms that capture the regularities in the series. The model is initially estimated for the preintervention period. Assuming that the process, in the absence of the intervention, would continue in accordance with the preintervention model, the model is then esti- mated with an additional function to identify the effect of the inter- vention (e.g., relaxed speed limit). There are several possibilities for

MANAGING SPEED 290 modeling the intervention. An "abrupt permanent" function reflects an intervention that has an immediate and permanent effect on the series, whereas an "abrupt temporary" function is one that immedi- ately affects the series but whose effect decays over time. Or there could be no initial effect but a gradual buildup to a permanent effect. A significant advantage of ARIMA time series models is data economy. In contrast to regression time series models that require data on the dependent variable and each of the explanatory variables, ARIMA models only require data on the dependent variable series and knowledge of when the intervention occurred. This can yield considerable savings on resources expended to collect the necessary data for analysis. However, the implications that ARIMA models have for data col- lection are not free. In time series models, a maintained hypothesis is that the effect of other determining factors is captured and that there are no disruptions in these series over the relevant period of analysis. Essentially, the assumption is made that the disruption occurring in the series is due only to the policy under study. For example, consider the effect that relaxed rural Interstate speed limits have on fatalities. An analyst estimates an ARIMA model of the process and finds that relaxed speed limits significantly reduced fatalities. This assumes that all other determining factors, including, for example, vehicle miles traveled, evolved during the postintervention period as it had during the preintervention period. Suppose, however, that gasoline prices sig- nificantly rose in the third quarter of 1987, shortly after relaxed speed limits were implemented. The assumption that preintervention vehi- cle miles traveled evolved in a manner consistent with postinterven- tion miles traveled is no longer tenable and the drop in fatalities would be inappropriately attributed to the relaxed speed limits. In recent work, researchers have included additional variables in ARIMA mod- els, referred to as ARIMAX models, to explicitly test the hypothesis that other determining factors have no effect on the series. POSTED SPEED LIMITS AND SPEEDING BEHAVIOR Figure C-1 indicates that changes in posted speed limits alter the distribution of speeds on a road to the extent that changed limits

291 Effect of Speed Limits on Speed Distributions and Highway Safety alter drivers' optimal speed choices. Holding all else constant, if altered speed limits have no effect on optimal speeds, speed distribu- tions will be unaffected. Although conceptually this is straightfor- ward, empirically the issue revolves around the analyst's ability to isolate the specific effect of posted speed limits from the effects of other confounding factors that influence optimal speed decisions, including traffic enforcement, environmental attributes, and public safety campaigns. Because roads are linked in a network, a related issue is whether a change in posted speed limits on one road alters drivers' optimal speeds on other roads, the spillover effect. Studies that have addressed the effects of changing posted speed limits on speed distributions can be usefully divided by two road types, nonlimited-access and limited-access roads. Nonlimited-Access Roads Table C-1 identifies recent studies that have examined the speed dis- tribution effects of posted speed limits on nonlimited-access roads. A study by Parker (1997) included 100 experimental [172 mi (277 km)] and 83 comparison [132 mi (212 km)] nonlimited-access sites in 22 states between June 1986 and July 1989. The primary objective in selecting a comparison site was to match as closely as possible the design, volume, and speed characteristics of the associated experi- mental site. In general, posted speed limits on the comparison sites included in the study were set at the 45th percentile speed. The following information is given for the experimental sites: · Sixty-three of the 100 experimental sites and 80 percent of the total mileage for the study were located in rural areas with popula- tions less than 5,000. Fifteen sites were located in urban areas with 50,000 persons or more. · Posted speed limits were lowered at 59 sites and raised at 41 sites. The most frequent speed limit decrease was 10 mph (16 km/h) (35 sites), whereas the prevailing increase was 5 mph (8 km/h) (26 sites). The maximum decrease in the posted limit was 20 mph (32 km/h) (three sites), and the maximum increase was 15 mph (24 km/h) (three sites). Before the speed limit change, the typical

Table C-1 U.S. Research on Speed Limits and Speeds--Nonlimited-Access Roads Study Database for Speeds Methodology Major Findings Comments Casey and Experimental/compar- ANOVA Supports speed adaptation, Tested three road configu- Lund ison site data, 55- Multiple regression 0 to 4 percent increase rations 1987a mphb freeways to mitigated by environmen- Lower speeds for commer- connecting roads, tal factors cial vehicles CA Differences in the effects 1985 of age and gender across field studies What is the geographical extent of adaptation? Casey and Experimental/compar- ANOVA Increase in average speeds Retested road configura- Lund ison site data Multiple regression with no change in speed tions from 1987 study 1992 Reanalysis of 1985 CA limits Lower speeds for commer- study Continued support for speed cial vehicles 1988 adaptation Sronger evidence that No increase in adapted younger drivers and speeds female drivers travel faster What is the geographical extent of adaptation?

Ullman Six urban fringe 55- Before/after analysis Lowering speed limits from No control for confound- mphb sites in Texas 55 b to 45 mphb had little and ing factors Dudek effect on speed distribu- 1987a tions Parker Experimental/compar- Quasi-experimental Increased speed limits have 100 experimental sites; 83 1997a ison site data, 22 significant but small comparison sites states absolute effect on speeds Nonrandom site selection June 1986July 1988 Change in speed alone has No control for cross-site Aug. 1987July 1989 little effect on driver and cross-state differ- behavior ences (e.g. enforcement, No apparent speed adapta- vehicles, drivers, educa- tion tion) Many results statistically significant but inter- preted as "not practically meaningful" a The text covers this study in more detail. b 55 mph = 89 km/h; 45 mph = 72 km/h.

MANAGING SPEED 294 posted speed limit for the experimental sites was set at the 20th per- centile speed; after the speed limit change, this increased to the 43rd percentile speed. · Sites whose speed limits were lowered by 15 mph (24 km/h) or more had the highest "before change" posted speed limits. Sites whose speed limits were increased by 10 mph (16 km/h) or more had the lowest "before change" speed limits. · Average 24-h volume at the experimental sites was 4,500 vehi- cles. For the comparison sites, the average 24-h volume was 3,400. In general, Parker's study found little evidence of a relationship between posted speed limits and speed distributions. The study's pri- mary findings with respect to speed distributions are as follows: · There was generally less than a 2-mph (3-km/h) difference in average speeds, speed standard deviation, and 85th percentile speed between the before and after speeds. These changes were statistically significant but were interpreted as "not sufficiently large to be of practical significance" (Parker 1997, 86). · There was little evidence of spillover effects. · Changes in posted speed limits led to changes in driver compli- ance, but this reflects the definition of compliance as driving at or below the posted speed limit rather than changes in driver behavior. In sum, Parker's study found that changing posted speed limits on nonlimited-access roads had little effect on speed distributions. It was further concluded that changes in posted speed limits had little effect on highway safety. The latter result is expected given the find- ing that altered limits had little effect on speed distributions. To the extent possible, Parker's study matches experimental with comparison sites to isolate the effect that posted speed limit changes have on driver behavior and speed distributions. At the start, Parker proposed an experimental design methodology for studying the effects of posted speed limit changes. However, because of legal, safety, and other concerns, state departments of transportation would only agree to participate if sites were nonrandomly selected. In particular, experimental sites were not randomly chosen from the

295 Effect of Speed Limits on Speed Distributions and Highway Safety population of nonlimited-access roads but from a set of sites whose posted speed limits were to be changed. Thus, the methodology actually used in the study was quasi-experimental. This has two implications: · Nonrandom selection of sites implies that the results emanating from a study of these sites cannot be generalized to a population of nonlimited-access roads. Inferences can only be drawn for the 172 mi (277 km) of experimental site roads included in the analysis. · Since, for all experimental sites included in the study, speed lim- its were to be changed, the posted speed limit changes may have sim- ply rationalized observed behavior. The first point is explicitly recognized in the study, which states: "The findings may apply to similar sites where the speed limits are changed for similar reasons. Generalizations to other roadways are not appropriate" (Parker 1997, 5). There is an allusion to but rela- tively little discussion of the second point--the study recognizes that "speed limit changes at the study sites were not made for the purpose of experimentation" (Parker 1997, 6). Reasons cited in the study for changing speed limits included requests from the pub- lic, leaders, or enforcement personnel; consistency of speed limits with traffic conditions; high incidence of crashes; compliance with local ordinances; and changing traffic volume and land use patterns. According to the study, the researchers were not aware of any major differences in enforcement or public information campaigns between the pre- and postchange environments. However, there was no attempt to explicitly evaluate whether traffic enforcement at the experimental sites differed between the pre- and postchange periods. This is potentially important, since differences in enforce- ment affect drivers' optimal speeds and, accordingly, speed distribu- tions. Higher (lower) posted speed limits combined with greater (less) enforcement could produce little difference in observed speeds. Further, if changes in the posted speed limit simply legalized exist- ing behavior, the findings would be significantly biased toward accepting the null hypothesis that altered speed limits have no effect

MANAGING SPEED 296 on driver behavior, and little insight would be offered into the inde- pendent effect of changing posted speed limits on speed distributions or highway safety. The study also notes that the modest changes in the speed distri- butions, on the order of 2 mph (3 km/h) or less, are statistically sig- nificant but not practically meaningful. An implication is that a statistically significant 2-mph increase in average speeds will have no effect on highway safety regardless of whether the road is posted at 20 or 40 mph (32 or 64 km/h), representing a 10 percent and 5 per- cent change, respectively, in average speed. The researchers do not comment on how large the distribution change must be to be "of practical significance." Ullman and Dudek (1987) examined the effect of lowering the speed limit from 55 to 45 mph (89 to 72 km/h) at six urban fringe highway sites in Texas where rapid urban development was occur- ring. On the basis of 1 year pre and 1 year postspeed limit change data, the authors generally find little change in average speed, 85th percentile speed, the proportion of drivers traveling above 60 mph (97 km/h), acceleration, or skewness. However, there was no control for other confounding factors, including population changes, traffic congestion, and traffic enforcement, which weakens the authors' con- clusions that lowering speed limits below the 85th percentile speed had no "conclusive effect on absolute speeds, speed distributions, or speed-changing activities" (Ullman and Dudek 1987, 48). In related studies, Casey and Lund (1987, 1992) analyzed the extent to which drivers who are exposed to and have adapted to higher-speed environments drive faster when entering lower-speed environments compared with drivers who have not been exposed to higher-speed environments. Holding all else constant, the null hypothesis is that driver speeds in low-speed environments will be no different whether or not these drivers were previously exposed to a higher-speed environment. Rejection of the null hypothesis is con- sistent with speed adaptation. In their 1987 paper, Casey and Lund tested three California field locations (representing six sites), reflect- ing rural and urban settings and alternative connecting road config- urations and speed limits. Casey and Lund reached the following conclusions:

297 Effect of Speed Limits on Speed Distributions and Highway Safety · Drivers traveled more slowly on the connecting roads. However, drivers exiting an expressway generally traveled faster on the connecting road than those not exiting an expressway. At one site there was no difference between adapted and nonadapted speeds; at the other five sites, the difference ranged between 1.8 and 4.7 percent. · At two of the three field locales, close to 100 percent of the drivers were required to stop before entering the connecting road. This provides stronger evidence of speed adaptation behavior, since the observed speed behavior on the connecting road was not simply an uninterrupted continuation from the higher-speed road, a phe- nomenon often referred to as speed perpetuation. · In a 1992 study, the authors retested these sites to assess the effect of the 65-mph (105-km/h) speed limit that California imple- mented on a portion of its eligible roads. Although none of the sites included in the study were eligible for the higher speed limit, the authors found that average speed increased at two of the three free- way sites and three of the four connecting roads in the study. Speed adaptation continued to be observed but did not worsen in the post- 65-mph environment. The results reported in these studies are consistent with previous research [e.g., Matthews (1978)], although the extent of speed adap- tation is less in Casey and Lund's work, which may reflect better control of confounding factors. This suggests the need for similar studies in different areas and with different road configurations to determine whether Casey and Lund's results are representative or, for some reason, unique to the location and set of roads studied. Further, a potentially important question that this study raises is the extent to which the speed adaptation effect continues. In the study, drivers are either "adapted" or "nonadapted," reflecting the freeway exposure of the "adapted" drivers. If speed adaptation is a general phenomenon, will expressway drivers adapt to the lower speeds on the connecting road? If so, we might expect to initially see higher speeds for the adapted drivers on exiting the expressway but little difference in speeds between the adapted and nonadapted drivers after a time.

MANAGING SPEED 298 Limited-Access Roads Table C-2 summarizes recent research on the effect of increased speed limits, and specifically the 65-mph (105-km/h) speed limit embodied in the STURA Act, on limited-access roads. Significant aspects of these studies are as follows: · Regardless of research methodology, unit of observation, or time period, the 65-mph (105-km/h) speed limit generally led to an increase in average speeds on the rural Interstate systems. In various studies, the National Highway Traffic Safety Administration (NHTSA) and others consistently measured significant speed increases in the post-65-mph environment. For example, between the fourth quarter of 1986 and the fourth quarter of 1990, NHTSA (1992) estimated a 3.4-mph (5.5-km/h) increase in average speeds, from 60.6 mph (97.5 km/h) to 64 mph (103 km/h). · Consistent with the increase in average speeds has been a gen- eral increase in speed dispersion, 85th percentile speed, and the pro- portion of drivers traveling over 65 mph (105 km/h). · Virtually no study explicitly controlled for other confounding fac- tors across either time or unit of observation, although some studies used 55-mph (89-km/h) environments to normalize for other factors. · Although average speeds increased after states implemented 65- mph (105-km/h) speed limits on eligible parts of their Interstate sys- tems, average speeds did not increase by the amount of the speed limit increase. · There is mixed evidence on the effect that relaxed rural Interstate speed limits had on speed distributions in 55-mph (89- km/h) states or on 55-mph highways in 65-mph (105-km/h) states. Similar to most studies in the highway safety literature, the stud- ies cited in Table C-2 use a quasi-experimental design methodology, and none sufficiently controls for other factors that may influence speed distributions. Although one cannot conclude from the existing evidence that, all else constant, an increase in rural Interstate speed limits caused an increase in mean and 85th percentile speeds, the immediate and persistent increase in speeds identified in this work is

Table C-2 U.S. Research on Speed Limits and Speeds--Limited-Access Roads Database for Major Findings Methodology Study Speeds Comments 65-mph b Increased average speed in 65- Before/after analy- NHTSA Thirteen Time series, based on averages mphb states 1989a sis states across states Eight 55-mph b Increased 85th percentile speed in Limited sample Regression trend 65-mphb states analysis states 3rd qtr. 19821987 4th qtr. 19851st qtr. 1988 Increased speeds by 0.25 mphb Eighteen 65-mph b Before/after analy- NHTSA Update of 1989 study per year sis 1990 states continuing No control found differences Increased 85th percentile speed Regression to monitor across states by 0.30 mphb per year speeds Significant (but unreported) effect 4th qtr. 19851st of speed limit on average speed qtr. 1990 and 85th percentile speed Eighteen 65-mphb Increased average speed from Before/after analy- NHTSA Update of 1990 study 60.6b to 64 mphb sis 1992 states monitoring No control found differences Increased 85th percentile speed speeds, two across states from 66.6b to 70.7 mphb periods Increased speed dispersion by 0.7 4th qtr. 1986, 4th mphb qtr. 1990 (continued on next page)

Table C-2 (continued) Database for Major Findings Methodology Study Comments Speeds mph,b Reports 1993 statis- Average speed = 56.9 range FHWA No statistical analysis State-monitored = (49.4, 59.6)b 55-mphb roads tics 1995 85th percentile speed = 64.0,b 1993 range = (56.4, 68.3)b 6.5 million citations, range = (2,081, 980,258) 65-mphb states, percent driving Aggregates over all 55-mph b Nine 65-mphb Quasi-experimental McKnight over 65 mphb and 65-mph b states ARIMA models et al. states 1989a Seven 55-mphb 48.2% increase on rural Analyzed percentage of drivers traveling above 65 mph b Interstates states 9.1% increase on 55-mphb roads Data could not answer why Quarterly data, 55-mphb states, percent driving speeds in the 55-mph b states 19821988 over 65 mphb increased 18% increase on rural No control for differences across Interstates states or systems 37% increase on other 55- mphb roads

Freedman Rural Interstates Quasi-experimental Average and 85th percentile speed No data on estimated trends increase in 65-mphb rural and (MD, NM, VA) No statistical tests Esterlitz Urban Interstate Interstate states (VA, NM) No control for differences across Little change in 55-mphb rural 1990 (NM) states or systems April 1987July Interstate state (MD) Differentiates speeds by vehicle 1989 Little change on urban Interstate (NM) type Similar increase in tractor-trailer speeds (MD, VA) 3.9-mph b increase in rural Mace and 51 rural Interstate Before/after analy- No control for confounding Heckard speed study sites sis Interstate speeds effects 4.3-mph b increase in rural 1991 in AL, AZ, CA, For spillover analysis, FL, IL, OH, TN, Interstate 85th percentile speed "toward/away" approach used 0.65-mph b increase in rural TX for some states 1986, 1988/1989 Interstate speed dispersion Small number of spillover sites Little change in speeds from 1988 Detailed analysis of Illinois data to 1989 Dual speed limits inhibit car Lttle local spillover effect observed speeds and no evidence of spillover onto urban Interstates Average car speed in 65-mphb states Freedman Northeastern states Quasi-experimental Increased rural Interstate speeds in 65-mph b states and (CT, MD, MA, less than in dual speed limit states Williams NJ, NY, PA, NH, No effect on rural Interstate speeds No statistical tests in 55-mph b states 1992 OH, VT, VA, No control for differences across WV) Lower truck speeds in dual limit states or systems Oct. 1989Jan. 1990 states Differentiates speeds by vehicle type (continued on next page)

302 Table C-2 (continued) Database for MANAGING SPEED Study Methodology Major Findings Comments Speeds Parker Quasi-experimental Increased average speed at experi- Nonrandom site selection Experimental/com- mental sites, range = (0.2, 2.3)b Small sample size 1997 parison site data, Decrease in speed standard devia- No control for differences across 10 sites, 4 states tion at three of four experimen- states or systems (CA, MD, MI, tal sites, range = ( 0.9, 0.2)b VA) Changes in average speed and April 1989August 85th percentile speed less than 1989 0.5 mphb at comparison sites a The text covers this study in more detail. b Measurements given in miles per hour in this table are converted to kilometers per hour as follows: mph km/h mph km/h mph km/h -0.1 -0.2 2.3 3.7 59.6 95.9 0.2 0.3 3.9 6.3 60.6 97.5 0.25 0.40 4.3 6.9 64 103 0.3 0.5 49.4 79.5 65 105 0.5 0.8 55 89 66.6 107.2 0.65 1.05 56.4 90.8 68.3 109.9 0.7 1.1 56.9 91.6 70.7 113.8

303 Effect of Speed Limits on Speed Distributions and Highway Safety certainly consistent with this hypothesis. The evidence strongly sug- gests that drivers' optimal speeds on rural Interstate systems in states whose speed limits did rise lies above 55 mph (89 km/h) and for some drivers may lie above 65 mph (105 km/h). This further implies that an enforced 55-mph NMSL did constrain rural Interstate driv- ers' speed decisions. On the other hand, the mixed evidence on speed distributions in 55-mph (89-km/h) environments, combined with a lack of control for other confounding factors, precludes the drawing of any firm con- clusions from these studies concerning the spillover effects of relaxed rural Interstate speed limits. As indicated in Table C-2, NHTSA produced a series of studies on the speed distribution effects of the 65-mph (105-km/h) speed limit. Relative to speeds expected from historical trends, average and 85th percentile speeds on rural Interstate systems that raised the speed limits consistently rose. Average and 85th percentile rural Interstate speeds in thirteen 65-mph states increased 1.9 and 1.3 mph (3.1 and 2.1 km/h), respectively, in 1987 over 1986 (NHTSA 1989a, 5657). Average speeds in eight states that did not raise rural Interstate speed limits also increased, but the increase was much less. Rural Interstate average speeds in the 13 states, which had been trending upwards by about 0.2 mph (0.3 km/h) per year, appeared to have gotten a boost from the STURA Act in 1987. But there are inconsistencies even in these data. Of the thirteen 65-mph (105-km/h) states in the study, average speeds increased in 8 (Arizona, Arkansas, California, Colorado, Illinois, Iowa, Nevada, and Washington) but actually fell in 4 (Indiana, Mississippi, South Dakota, and Tennessee) and did not change in 1 (Wisconsin) between the first quarter of 1987 and the first quarter of 1988. Eighty-fifth percentile speeds increased in nine but fell in four states (Indiana, Mississippi, South Dakota, and Tennessee). A similar phe- nomenon occurred in the eight 55-mph (89-km/h) states in the study. Between the first quarter of 1987 and the first quarter of 1988, average speeds rose in five states and fell in four (Connecticut, New Jersey, New York, and Rhode Island). Eighty-fifth percentile speeds fell in the same four states. On the basis of individual state data, the apparently strong national evidence that relaxed rural Interstate

MANAGING SPEED 304 speed limits led to a uniform rise in speeds weakens. These findings emphasize the need to develop models that sufficiently control for confounding factors in order to accurately identify the isolated effect of the relaxed speed limits on speed distributions. They also imply that conclusions based on aggregates may be oversimplified and that aggregation masks the underlying mechanisms through which speed limit changes affect speed distributions. The analysis by McKnight et al. (1989) of 55- and 65-mph (89- and 105-km/h) states gives limited speed distribution data, but the information provided is consistent with NHTSA's research. There is no information on differences in average speeds or 85th percentile speeds between the 65-mph and 55-mph states. However, McKnight et al. provide information on the proportion of drivers traveling over 65 mph. On the basis of 16 states providing quarterly speed infor- mation between 1982 and 1989, McKnight et al. found that, between 1986 and 1988, the ratio of rural Interstate drivers to 55-mph road drivers traveling over 65 mph significantly increased in 65-mph states but fell in 55-mph states This is not a surprising result for the 65-mph states, given the observed increase in average speeds in the 65-mph environments. An intriguing finding of the study by McKnight et al. is that, in the post-65-mph (105-km/h) environment, the percentage of drivers exceeding 65 mph in 55-mph (89-km/h) states increased 18 percent on rural Interstate systems and 37 percent on other 55-mph highways. By comparison, in 65-mph states, there were 48 and 9 percent increases. The authors suggest that this may reflect a tainting or spillover effect but provide no evidence of this. Traffic diversion may explain the large difference in 65-mph states but does not explain the findings in the 55-mph environments. These findings again demonstrate the need to develop models that sufficiently control for relevant confounding fac- tors in order to fully understand the mechanisms at work. Summary From the work cited, several conclusions can be drawn concerning the effect on speed distributions of increasing speed limits from 55 mph (89 km/h) to 65 mph (105 km/h).

305 Effect of Speed Limits on Speed Distributions and Highway Safety · On nonlimited-access roads, evidence suggests that changes in speed limits affect average speeds and speed dispersions, but the magnitude of these changes appears to be small, and possibly much smaller than the change in speed limit. · Drivers exhibit speed adaptation, but the difference between adapted and nonadapted speeds is less than 5 percent. The extent of speed adaptation appears not to worsen with increasing speed limits on limited-access high-speed roads. · On limited-access high-speed roads, work indicates that increased speed limits lead to higher average speeds [on the order of 4 mph (6 km/h) or less for a 10-mph (16-km/h) increase in the speed limit] and increased 85th percentile speeds (of a similar magnitude) with small increases in speed dispersion [by less than 1 mph (2 km/h)], but these findings are based on limited control for other confounding factors. The effects that increased speed limits on limited-access roads have on speed distributions of nonlimited-access roads are considerably more ambiguous. · Many of these studies comment on the importance of speed enforcement in determining the effects of speed limit changes on speed distributions, but there is little empirical evidence on the explicit role that enforcement plays. SPEED DISTRIBUTION AND HIGHWAY SAFETY Studies The preceding section summarized a set of empirical studies examin- ing the effect of changes in speed limits on speed distributions on nonlimited-access and limited-access roads. Particularly for relaxed rural Interstate speed limits passed in 1987, there is evidence, albeit not consistent, that relaxed speed limits on rural Interstate highways increased mean and 85th percentile speeds as well as speed dispersion on these roads. The effect that speed limits have on roads' speed dis- tributions, however, is primarily relevant because of its implications for highway safety. A critical question then is, What effect do speed distributions have on highway safety? Since speed distributions can be characterized by average speed, speed dispersion, 85th percentile

MANAGING SPEED 306 speed, and the proportion of traffic traveling above some speed, it is important to understand whether there is an identifiable relationship between a road's speed distribution and highway safety. Unfortunately, relatively little research has focused on the rela- tionship between a road's speed distribution and measures of the road's safety. Recent work, identified in Table C-3, has typically cen- tered on the highway safety effects of average speeds versus speed dispersion. In contrast, as will be seen in the next section, a signifi- cant amount of research exists on the effect of speed limits on high- way safety. For much of this research, the hypothesis of interest is that an increase in speed limits leads to a deterioration in highway safety. The implication is that increases in speed limits induce drivers to travel faster, which leads to crashes involving more serious injuries and more fatalities. That is, "speed kills." Solomon (1964) initially investigated the relationship between crash involvement rate and speed dispersion, finding that the involvement rate is lowest very near average speeds. At speeds signif- icantly lower or higher than average speed, involvement rates increase. According to Solomon's findings, slower drivers can be just as dangerous as faster drivers. Cerrelli (1981) also found a U-shaped curve between crash rates and deviations from average speeds. In related work, Garber and Gadiraju (1988) found a U-shaped rela- tionship between speed dispersion and the difference between a road's design speed and the posted speed limit. Minimum speed dis- persion tends to occur when the design speed is 5 to 10 mph (8 to 16 km/h) higher than the posted speed limit. The authors also estimated a positive but not U-shaped relationship between speed dispersion and the crash rate. There are relatively few empirical studies that examine the direct relationship between speed distribution and highway safety after controlling for other relevant determinants. Lave (1985) explored the effects of average speed and speed dispersion on rural and urban fatality rates for Interstate systems, arterial roads, and collector roads. Each of the 6 models was estimated for 1981 and 1982, for a total of 12 models. Lave's basic hypothesis is Fatality rate = f (crash severity, crash involvement)

Table C-3 U.S. Research on Speed, Speed Dispersion, and Highway Safety Study Database for Speeds Methodology Major Findings Comments 1985a Lave Statewide, by func- Regression analysis After controlling for speed dis- Controlled for enforcement tional persion, average speed had lit- and hospital access class tle effect on fatality rates Limited control for other fac- 1981, 1982 More emphasis on speed disper- tors sion as a coordinating device Garber and Interstate, arterial, Regression analysis Crash rates increase with Speed dispersion is lowest Gadiraju and major rural ANOVA increasing speed when difference is between 1988a 5 and 10 mphb collector test sites dispersion on all roads in Virginia Relationship between speed dis- Crash rate is not necessarily persion and (design speed positively related to average posted speed limit) is U speed shaped McCarthy Countywide Regression analysis After controlling for speed dis- Controls for a variety of 1988 State of Indiana, persion, average speed had lit- socioeconomic factors 1987 tle effect on safety Limited sample (continued on next page)

Table C-3 (continued) Study Database for Speeds Methodology Major Findings Comments Levy and Statewide data for Regression analysis Average speed, speed dispersion, Controls for a variety of Asch 1985 and their interaction have socioeconomic factors 1989 important effects on the fatal- Speed dispersion significantly ity rate increased Interstate fatality rates when average speed 63.8 mphb Average speed significantly increased Interstate fatality rates when speed dispersion 8.2 mphb Similar results for total fatal- ity rates Significant speed effect results from weighted average speed measures (Lave 1989)

Fowles and Statewide data for Regression analysis Speed and speed dispersion con- Results robust to model speci- Loeb 1979 sistently significant determi- fication 1989 nants of fatalities Controls for a variety of socioeconomic factors Uses Interstate speeds to explain total statewide fatalities (Lave 1989) Snyder Primary federal-aid Regression analysis Average speed and speed disper- Lack of control for confound- 1989 rural highways sion for the fastest vehicles are ing effects offset, to some Annual data: significant determinants of degree, by fixed effects 19721974 for fatalities model 26 states Speed dispersion for the slowest Rural speed data reflect vehicles is unimportant Interstates, rural primary and secondary roads (Lave 1989) Use of 1974 data is a compli- cating factor (Lave 1989) aThe text covers this study in more detail. b5 mph = 8 km/h; 10 mph = 16 km/h; 8.2 mph = 13.2 km/h; 63.8 mph = 102.7 km/h.

MANAGING SPEED 310 arguing that crash involvement depends on whether traffic is flowing smoothly (smoothness varies inversely with speed dispersion) and that crash severity depends on how fast traffic is moving--which is a positive function of average speed. Since smoother traffic flows imply lower speed dispersions, Lave posited that speed dispersion reflects a coordinating mechanism whereas average speed reflects a limiting mechanism on highway traffic. Lave's empirical formulation of the above model is Fatality rate = f (average speed, speed dispersion, other factors) In defining the fatality rate as the dependent variable, Lave con- trolled for exposure. He included two other variables in the empiri- cal equation to control for other determining influences of highway fatality rates, enforcement and access to medical care. Average speed and speed dispersion (measured as the difference between 85th per- centile speed and average speed) are the two variables of most inter- est, and both are expected to have a nonnegative coefficient. From this research, Lave reached the following conclusions: · After controlling for speed dispersion and other determining fac- tors of the fatality rate, "there is no discernible effect of speed on the fatality rate" (Lave 1985, 1162). In all 12 equations, average speed is not statistically significant, and in 10 of 12 equations, the coefficient of average speed is negative. · Speed law enforcement should focus on reducing speed disper- sion to maintain a smoother traffic flow rather than on speed per se. Slow drivers as well as fast drivers are hazardous to public safety. Lave did not use these findings to support higher posted speed limits because there is little information on the relationship between speed dispersion and average speed. For his data, there was "generally a negative correlation" between the two, but he took this only as sug- gestive of a negative relationship. Supporting Lave, Garber and Gadiraju (1988) also found an inverse relationship between average speed and speed dispersion, but their model fails to control for other confounding factors.

311 Effect of Speed Limits on Speed Distributions and Highway Safety Lave's general conclusion that average speed has little effect after controlling for speed dispersion is not fully supported by the estima- tion results. Of the 12 final regression equations reported, speed dis- persion is consistently insignificant in half, corresponding to rural collectors, urban freeways, and urban Interstate roads. For these sys- tems, the result is counterintuitive--neither speed dispersion nor average speed has any effect on the fatality rate. High correlation among explanatory variables will drive down t-statistics and could explain the insignificance in some of Lave's equations. However, arguing against this is the nonuniform correlation between speed and variance (negative correlation in eight cases, nonnegative in four) as well as the generally low reported values of R2 (ranging between .019 and .269 when neither speed measure is significant). In only two equations, those for rural Interstate highways in each year, do the included explanatory variables explain more than half of the variation in the fatality rate. An alternative explanation for the poor performance of some models may be the omission of other factors that explain the fatality rate. Levy and Asch (1989) and Fowles and Loeb (1989) estimated regression models with a larger set of explanatory variables that cap- tured additional influences including gender, age distribution of driv- ers, income, alcohol, and population. Enforcement, however, was absent in both studies. Both of these studies focused on the role of average speed in highway safety. Snyder (1989), on the other hand, estimated a fixed effects model (to account for the absence of con- trolling factors) that differed from the others in explicitly consider- ing whether the effect of speed dispersion was symmetric for slower and faster drivers. Among the findings from these studies are the fol- lowing: · After controlling for other influences on safety, average speed and speed dispersion are both important determinants of highway safety. · Levy and Asch (1989) found that the interaction between aver- age speed and speed dispersion has a positive effect on fatality rates. Speed dispersion significantly increases the fatality rate when average speed reaches 63.8 mph (102.7 km/h), considerably higher than the sample average speed of 55.2 mph (88.8 km/h).

MANAGING SPEED 312 · Speed dispersion has an asymmetric effect. Snyder (1989) found that the effect of speed dispersion for the slower drivers (measured as median speed minus 15th percentile speed) had no effect on fatali- ties, whereas speed dispersion for faster drivers (measured as 85th percentile speed minus median speed) had a positive and significant effect on fatalities. These results are both consistent and inconsistent with those of Lave. In that speed dispersion is found to be an important influence on highway safety, these studies support Lave's findings. However, in contrast to Lave, each study concludes that average speed is sepa- rately important. Further, Snyder's results call into question Lave's conclusion that speed dispersion is symmetric for slow- as well as fast-moving traffic. In a response, Lave (1989) argued that the con- tradictory findings in each of these analyses could be attributed to aggregation problems in that either the fatality measures or speed measures were inappropriately aggregated across road types. He reported the empirical results of alternative specifications of the authors' models that demonstrated the implications of alternative aggregation schemes and produced results generally consistent with his 1985 study. All of these studies produce strong evidence that speed dispersion is an important influence on highway safety. Although the outcome of a crash is more severe the higher the speed, all else being constant, there is greater uncertainty as to the effects of average speed on crash probability. As noted by Lave, there is a need to obtain improved data on a road's speed characteristics as well as associated data that con- trol for other determining factors of highway safety. This would reduce potential aggregation problems and enable researchers to bet- ter isolate the effects of average speed and speed dispersion on high- way safety. Additional work is needed on the appropriate level of aggregation. In some cases, aggregation may be more beneficial than harmful. Does the speed distribution on a collector road affect speed distribu- tions on Interstate highways? Probably not. However, much of the current debate on rural Interstate speed limits concerns speed adap- tation, whereby drivers adapted to higher speeds on roads posted

313 Effect of Speed Limits on Speed Distributions and Highway Safety with high speed limits will also drive at higher speeds on roads posted with lower speed limits. In other words, the speed distribution on higher-speed rural Interstate roads will affect speed distributions on other road systems with lower posted speed limits. Empirically, this implies that rural Interstate speed measures may be appropriate variables in models that analyze total fatality rates. And the finding that rural Interstate speeds affect the total fatality measures, rather than reflecting spurious correlation, may be capturing relationships between driving activities on road networks. Summary Among the implications from this discussion are the following: · There is a positive relationship between crash severity and speed dispersion, particularly for rural Interstate roads. Also, evidence sug- gests that minimum speed dispersion occurs when the difference between a road's design speed and the posted limit lies between 5 and 10 mph (8 and 16 km/h). · There is some evidence for the notion that the marginal effect of average speed (speed dispersion) on highway safety depends on the level of speed dispersion (average speed). · The safety effect of speed dispersion appears to be most impor- tant for the fastest rather than the slowest drivers, thus suggesting the need for maximum speed limits. Minimum speed limits could also reduce speed dispersion, but more research is needed to evaluate the effects of minimum speed limits on speed distribution and highway safety. · More detailed and informative data should be collected to better understand the relationship between average speed, speed dispersion, and highway safety. Speed distribution measures that characterize the environment of the safety outcome (e.g., late night crashes on arter- ial roads) would be ideal. Existing measures, however, are typically highly aggregated proxies, and the extent to which these proxies bias the results and lead to inappropriate inferences is not known. · There is a need for research on the aggregation issue and specif- ically what level of aggregation is appropriate for accurately charac-

MANAGING SPEED 314 terizing the relationship between speed distribution and highway safety. · Literature on average speed and speed dispersion focuses pri- marily on fatality measures. With the exception of McCarthy (1988), who found that speed dispersion is a significant determinant of total and injury as well as fatality measures, current literature is silent on the effect of speed distribution on the spectrum of crashes. RELAXED RURAL INTERSTATE SPEED LIMITS AND HIGHWAY SAFETY Preceding sections have reviewed the empirical literature on the effects of posted speed limits on speed distributions and work on how changes in speed distributions affect highway safety. As shown in Figure C-1, the extent to which changes in posted speed limits affect highway safety depends on their effects on drivers' optimal speeds and, therefore, on the distribution of speeds. The two preced- ing sections have identified a body of research that addresses these questions, but the literature on these issues is not extensive. In con- trast, a large body of work on the relationship between posted speed limits and highway safety exists. Much of the difference reflects the availability of data. Speed distribution information is not widely available and, when available, may not be appropriate to the study's objectives. Highway safety data and information on changes in speed limits, particularly those initiated or facilitated at the federal level that are likely to have a broad influence on highway safety, are readily avail- able and amenable to time series, cross section, panel data, or simply before-and-after analyses. For the 55-mph (89-km/h) NMSL and the relaxed 65-mph (105-km/h) speed limit on rural Interstate roads embodied in the EHEC and STURA Acts, respectively, the avail- ability of data has had the positive effect of generating many studies on the relationship between changes in speed limits and highway safety. However, empirically establishing a relationship (or the absence of one) presents a difficult task. In addition to the hypothe- sized effects of changes in posted speed limits on speed distributions and, therefore, highway safety, account must be taken of a multitude

315 Effect of Speed Limits on Speed Distributions and Highway Safety of confounding factors, identified in Figure C-1, that influence high- way safety. Recent literature analyzing the highway safety effects of an increase to 65 mph (105 km/h) in the posted speed limit on rural Interstate highways that Congress permitted in the STURA Act is reviewed in this section. The review is divided into two subsections depending on whether the study is national in scope or concentrates on a single state. Further, to more reliably evaluate the effects of relaxed rural Interstate speed limits, the review concentrates, with some exceptions, on studies based on a postenactment environment of at least 2 years. Rural Interstate Speed Limits of 65 mph (105 km/h) and Highway Safety--National Perspective Table C-4 identifies 14 studies analyzing the overall effects of relaxed rural Interstate (RI) speed limits on highway safety. Immediately after passage of the STURA Act in 1987, 38 states raised speed lim- its on some portion of, if not all, eligible RI mileage, and 2 other states increased their limits in 1988. Among the salient points in Table C-4 are the following: · Compared with 55-mph (89-km/h) states, most studies con- cluded that highway safety on RI highways deteriorated with the higher speed limits. · Most of the studies fail to adequately control for other factors that determine RI fatalities and fatality rates. · The primary focus of many studies is the direct effect of the speed limit law to the exclusion of the law's systemwide effect. · There is relatively little attention given to nonfatal crashes, non- fatality measures, or the distribution effect of the speed limit on crash and injury severity. · The bulk of the studies occurred in the late 1980s and early 1990s, with relatively few studies in the past 2 to 3 years. Since passage of the law, NHTSA has completed a number of studies on the law's effects. In the appendices of its first assessment

Table C-4 Research on 65-mpha Speed Limit and Highway Safety--U.S. National Perspective Study Database for Study Methodology Major Findings Comments 65-mpha states NHTSA Thirty-eight 65- Before/after Inconclusive results on selective 1989ab mpha states Regression analysis 19% increase in RI fatalities, speed limit increases and Ten 55-mpha states with comparison 19861987 dual speed limits Annual data: series 7% decrease in UI fatalities, Limited control for confound- 19751987 19861987 ing factors 55-mpha states Small numbers problem for 7% increase in RI fatalities, some state analysis 55-mpha states--East Coast 19861987 10% increase in UI fatalities, Limited evidence of increases 19861987 in nonfatal injuries Before/after analysis 65-mpha states NHTSA Thirty-eight 65- Updates 1989 study mpha states 1990 Regression analysis 13% increase in RI fatalities, Limited control for confound- Ten 55-mpha states with comparison 19871988; 2% decrease, ing factors Annual data: series 19881989 Variability across states 19751988 7% increase in RI fatality rate, 20% increase in RI vehicle 19871988; 7% decrease, miles traveled, 19861989, 19881989 and accounts for one-third of 55-mpha states increase in fatalities 12% decrease in RI fatalities, Small numbers problem for 19861989 individual states 13% increase in UI fatalities, Weak information on total 19861989 crashes and nonfatal injuries Monthly changes mirror annual changes

Before/after analysis 65-mpha states NHTSA Thirty-eight 65- Updates 1990 study mpha states 1992 Regression analysis Limited control for confound- 4% decrease in RI fatalities, Ten 55-mpha states with comparison ing factors 19891990; 27% increase, Annual data: series Variability across states 19861990 19751990 19% decrease in RI fatality rate, 19891990; 0% change, 19861990 55-mpha states 17% increase in RI fatalities, 19891990; 3% increase, 19861990 Reports 1993 statis- 65-mpha states FHWA All states, 1993 No analysis 1995 tics 2.4% increase in RI fatalities, 19921993 55-mpha states 4.5% decrease in RI fatalities, 19921993 Garber and Forty 65-mpha states Regression analysis Controlled for seasonal effects, Estimated median effect of law on Graham Monthly data: Jan. safety belt law, economy, and RI fatalities--15% increase 1989b 1976Nov. 1988 exposure (i.e., trend term) Estimated median effect of law on Large differences across indi- non-RI fatalities--5% increase vidual states Only considered fatalities (continued on next page)

Table C-4 (continued) Study Database for Study Methodology Major Findings Comments 65-mpha 65-mpha McKnight Twenty Before/after analysis states Aggregate model over states et al. states Quasi-experimental 21.8% increase in RI fatal crashes Identified nonuniform changes 1989b Eight 55-mpha states ARIMA models 1.2% increase in fatal crashes on over states other 55-mpha roads Six dual limit states Before/after analysis identified 55-mpha states Six experimental regional differences states 10.4% increase in RI fatal crashes Safety belt laws and traffic Monthly data: Jan. 12.7% increase in fatal crashes on density not important in other 55-mpha roads 1982July 1989 explaining law's effects Limited positive effect on injury Abrupt intervention crashes Dual limit and experimental states, significant fatality increase on RI and other 55- mpha roads 65-mpha states Baum et al. Thirty-eight 65- Before/after odds Odds ratio assumes indepen- a states mph ratio 15% increase in RI fatalities rela- dent series 1989 Eight 55-mpha states 19821986 versus tive to other rural roads (odds Similar results for uniform versus Annual data: 1987 ratio = 1.15) dual speed limit laws, safety 55-mpha states 19821987 belt/no safety belt law, day- No effect on RI odds ratio (odds time/nighttime crashes, and ratio = .94) (4% increase in all single/multiple vehicle crashes RI fatalities; 12% increase in Odds ratio increased in 24 of fatalities on other rural roads) 38 states Analyzed fatalities only

Baum et al. Forty 65-mpha states Before/after odds Updates of 1988 study 65-mpha states (1991 study) b, 1990 ratio Odds ratio assumes indepen- a states Eight 55-mph 29% increase (19% after adjust- 1991b 19821986 versus dent series Annual data: ments for VMT and passenger 1988 Adjusted for VMT increases 19821989 vehicle occupancy rates) in RI 19821986 versus and changes in vehicle occu- fatalities relative to other rural Effect of Speed Limits on Speed Distributions and Highway Safety 1989 pancy roads (odds ratio = 1.29) No adjustment for other con- 55-mpha states (1991 study) founding factors No effect on RI odds ratio Analyzed fatalities only Mace and Before/after analysis General increase in total and injury No control for confounding Twenty-eight rural Heckard effects Interstate speed crashes 1991 study sites in IL, Insufficient data on fatal crashes OH, TX Annual data: 1986, 1988 Chang et al. Thirty-two 65-mpha Before/after analysis No effect of the speed limit on Aggregate model over states 1991b ARIMA models Interstate fatalities for larger states Investigated alternative inter- states vention functional forms Six 55-mpha states Increase in Interstate fatalities, with some decaying effects, for smaller Significant change in fatalities Monthly data compounded by change in Jan. 1975Dec. 1989 states unknown exogenous factors An initial increase with decaying effects occurred in the 55-mpha states (continued on next page) 319

Table C-4 (continued) Study Database for Study Methodology Major Findings Comments Lave 1992 Thirty-eight 65- Before/after analysis Systemwide, the law decreased fatali- Comment on Godwin's (1992) mpha states (pub- ties literature review Eight 55-mpha states lished Decrease due to traffic diver- with Annual data: 1986, sion and more efficient allo- Godwin 1988 cation of policing resources 1992) Other than VMT, no control for confounding factors How comparable is the com- parison group? Godwin Thirty-eight 65- Before/after analysis Decreased fatalities systemwide Comment on Lave's (1992) 1992b mpha states implies unreasonably high VMT finding Eight 55-mpha states shift from non-RI to RI roads Other than VMT, no control Annual data: 1986, for confounding factors 1988 How comparable is the com- parison group?

McCoy et 19 pairs of state Quasi-experimental Speed zones with "reasonable" posted Focused on urban areas al. 1993 highway urban Poisson regression speed limits have lower crash rates Reasonable and unreasonable speed zones than zones with lower "unreason- speeds based on zone's pre- 19851988 able" limits vailing speed and test run speed Comparison of reasonable and unreasonable zones with same proposed speeds Regression controlled for AADT, length, and number of businesses Lave and Thirty-eight 65- Before/after analysis Systemwide, 3.4% to 5.1% decrease Updates and extends Lave's mpha states Elias Regression analysis in the fatality rate (1992) comment 1994b Eight 55-mpha states Systemwide approach Annual data: 1986, State analysis controls for sea- 1988 sonal effects, safety belt law, Monthly data: Jan. and the economy 1976Dec. 1990 Note: AADT = average annual daily traffic; RI = rural Interstate; UI = urban Interstate; VMT = vehicle miles traveled. a55 mph = 89 km/h; 65 mph = 105 km/h. bThe text covers this study in more detail.

MANAGING SPEED 322 of the law, NHTSA (1989a) discusses the analytical methodology used in the first and subsequent updates to the 1989 study (NHTSA 1989b; NHTSA 1990; NHTSA 1992). In a before-and-after com- parison, NHTSA (1989a) found that RI fatalities in 65-mph (105- km/h) states were 18 percent higher between 1986 and 1987, which compared with a 7 percent increase in 55-mph (89-km/h) states. Similarly, in its 1992 study, 65-mph states experienced a 27 percent increase in fatalities between 1986 and 1990 compared with a much smaller 3 percent increase in the 55-mph states. However, between 1989 and 1990, RI fatalities in 65-mph states decreased 4 percent compared with a 17 percent increase in 55-mph states. By not adequately controlling for other influences on highway safety, before-and-after comparisons can produce a distorted picture of the law's effects. To control for the effects of confounding factors, NHTSA relates changes in RI fatalities to changes in a companion series. An ideal companion series would have the following attri- butes: (a) the companion series is conceptually related to the RI series, (b) it is statistically related to the RI series over time, and (c) it is not contaminated by the intervention in the RI fatality series. According to the study (p. II-7), the basic assumption is that the rural Interstate series and the companion series move well enough together historically that a deviation in the historical pattern can be interpreted as the result of a higher speed limit in 1987. This assumption is justified as long as no other changes affected the relationship between the two series. This assumption is not a statistical concern but one requiring knowledge about the highway safety environment. [Emphasis added.] NHTSA experimented with various companion series and found that "fatalities on other roads" worked best. An ideal companion series enables the analyst to control for other confounding factors that influence both series. However, underlying this methodology are two important assumptions. First, no other changes occurred that affected one series but not the other. If this does not hold, the model does not sufficiently account for the influence of other confounding

323 Effect of Speed Limits on Speed Distributions and Highway Safety factors. Second, there are no spillover effects. As Attribute c indi- cates, the ideal companion series is not contaminated by the inter- vention, an assumption that is unlikely to be satisfied. There is sufficient evidence in the literature [e.g., Garber and Graham (1989)] to seriously question the assumption of no spillover effects. The fact that the implications of these assumptions are only dis- cussed in the appendix is unfortunate, since readers will likely place greater importance on the reported 18 percent increase in fatalities between 1986 and 1987 than may be justified by the analysis. NHTSA (1989a) also explores the effects of the speed limit by state and finds that most states, although not all, experienced an increase in fatalities. The change in the ratio of RI fatalities to all other fatalities between 1986 and 1987 ranged from a low of .053 for Montana to a high of .1483 for Wyoming. Of the 21 states included in this analysis, 9 states had ratios less than .01. Whether these changes are statistically different from 0 is not reported. Subsequent studies by NHTSA (1989b; 1990; 1992) used similar methodologies and newly available data to update the estimates reported in the 1989 study. One difference in the 1992 study is the explicit control for vehicles miles traveled (VMT). According to NHTSA, the 20 percent increase in RI VMT between 1986 and 1990 accounted for one-third of the increase in fatalities on these roads, once again signifying the importance of appropriately control- ling for other factors. Baum et al. (1989; 1990; 1991) used a before-and-after analysis based on odds ratios to assess the effect of higher RI speed limits on highway safety. The studies defined two groups --38 (or 40, depend- ing on the study) 65-mph (105-km/h) states that raised the speed limit in the post-STURA environment and eight 55-mph (89-km/h) states--and two time periods, 1982 through 1986 versus the year of study (1987, 1988, or 1989). Similar to NHTSA's regression strategy, the odds ratio identifies a companion series and statistically investi- gates whether the change in the odds ratio of RI to other rural fatal- ities was significantly different between 65-mph and 55-mph states. Overall, Baum et al. (1989) found that relaxed speed limits signifi- cantly increased the odds of an RI fatality in 65-mph states, whereas in the 55-mph states, no significant effect was found. In 65-mph

MANAGING SPEED 324 states, fatalities increased 19 and 4 percent on RI and other rural roads, respectively, yielding a net 15 percent increase in fatalities, a result similar to the 18 percent initially estimated by NHTSA (1989a). Similar results were found for comparisons of 55-mph states with and without dual speed limits, states with and without safety belt laws, and daytime versus nighttime crashes. However, in a state- by-state analysis, the study observed that 24 of the 38 states had an odds ratio greater than 1. This implies a decline in the odds ratio in 14 of the 38 states, suggesting that safety improved in those locales. This reflects, at least in part, the lack of control for other factors that influence highway safety. In contrast to the 1989 study, Baum et al. (1991) controlled for changes in VMT and vehicle occupancy. With no adjustment, the study found that, relative to 19821986, the odds of an RI fatality in 65-mph (105-km/h) states in 1989 increased 29 percent. After adjustment for changes in VMT and vehicle occupancy, the percent- age increase was 19 percent. The qualitative importance of this find- ing is twofold. First, the 1991 study again stresses the importance of controlling for other determining factors and implies that, by not controlling for them, the initial estimates of the law's general effect were biased upward. Second, it raises the question, How would these estimated effects change if the study included controls for changes in other known influences on highway safety such as alcohol consump- tion and traffic enforcement? In a widely cited paper, Garber and Graham (1989) estimated separate regression models based on monthly data for the 40 states that raised RI speed limits. In the study, the authors explicitly con- trolled for a subset of determining factors for which monthly data were available, including economic performance, seasonal effects, weekend travel, and safety belt laws. The models also included a time trend to capture the influence of VMT. Consistent with NHTSA's estimates and those of Baum et al. (1991), Garber and Graham estimated that the median effect of the higher speed limit law was a 15 percent increase in fatalities on RI highways. The median effect on rural non-Interstate roads was a 5 percent increase in fatalities. From their work, the authors reached the following con- clusions:

325 Effect of Speed Limits on Speed Distributions and Highway Safety · The 65-mph (105-km/h) speed limit did not have uniform effects across the 40 states. All else being constant, fatalities increased in 28 of the 40 states and either decreased or were unchanged in 12 states. · The higher speed limit generally increased rural non-Interstate fatalities, implying that the spillover effects from the law dominated the law's traffic diversion effects. · There is a need for additional research on identifying factors that explain cross-state heterogeneity, which implies the need to col- lect more detailed state-level data and to develop more reliable mod- els of highway safety. Although they controlled for a subset of determining factors, the authors expressed concern about the extent to which the variables included sufficiently control for other determinants of highway fatalities. Garber and Graham's finding of positive and negative effects of the speed limit on statewide fatalities is consistent with the notion that existing studies have not sufficiently identified or controlled for the influence of other factors on state fatalities. If a model adequately accounted for these differences, the speed limit law could be expected to have the same effect on highway safety in different states. Whether it is possible to control for the various effects is an empiri- cal issue, but future research should move in this direction. McKnight et al. (1989) and Chang et al. (1991) estimated ARIMA intervention time series models to assess the effect of the 65-mph (105-km/h) speed limit. McKnight et al. (1989) aggregated over twenty 65-mph states from January 1982 through July 1989. They eliminated observations prior to 1982 due to "major shifts in crash trends occurring in the years prior to 1982" (McKnight et al. 1989, 10), although there was no discussion of the nature of these shifts or why they could not be included in the empirical models. As indicated in Table C-4, McKnight et al. estimated a significant increase in RI fatal crashes in 65-mph states but no effect on non- rural Interstate roads in these states, a finding that suggests the pos- sibility of traffic diversion toward 65-mph RI highways in the 65-mph states. Dual limit states and experimental states exhibited similar effects. Further, as confounding factors, neither safety belt use

MANAGING SPEED 326 nor traffic density affected the results. In 55-mph (89-km/h) states, on the other hand, there was an inexplicably significant increase in both RI and nonrural Interstate fatal crashes. These findings, and particularly the significant fatal crash increase in 55-mph (89-km/h) states, raise more questions than they answer. How would the results change if a longer time series were used? What role does aggregation of 65-mph (105-km/h) and 55-mph states, respectively, have on the results? Would the results change if factors in addition to safety belt use and traffic density were accounted for in the analysis? Are the results for the 55-mph states the result of spillover effects and speed adaptation or do they reflect uncontrolled-for heterogeneity? In a related analysis, Chang et al. (1991) used ARIMA time series intervention methodology that included a longer monthly time series, January 1975 to December 1989, to estimate fatality models for 32 states with a 65-mph (105-km/h) limit and 6 states with a 55- mph (89-km/h) limit. In contrast to McKnight et al.'s abrupt perma- nent intervention, Chang et al. tested alternative formulations of the intervention, including abrupt permanent, abrupt temporary, an increasing effect up to a permanent level, and so forth. Further, the authors tested the sensitivity of the interventions to alternative start- ing and ending periods (if appropriate). Overall, the authors con- cluded that the 65-mph speed limit had a statistically significant effect on fatalities, but after a year's "learning period" the effect decayed over time. In separate analyses based on state clusters, simi- lar effects were present in smaller states; large states, on the other hand, were "virtually insensitive to the speed limit change" (Chang et al. 1991, 68). On the basis of their sensitivity analyses, which identified signifi- cant effects on fatalities prior to implementation of the speed limit, Chang et al. (p. 68) concluded that "the level of impacts exhibited in the `after' fatality records . . . represented the compounded effects of the increased speed limit and some other exogenous factors. . . . Those unknown factors have consistently caused an increase in fatality numbers since 1986. . . ." (emphasis added). In contrast, one could argue that driver anticipation of higher speeds, rather than changes in exogenous factors, generated the fatal-

327 Effect of Speed Limits on Speed Distributions and Highway Safety ity increases in the months preceding the 65-mph (105-km/h) lim- its. The proximity between 1986 and the speed limit law in 1987 raises the question of what other exogenous factors could have resulted in the observed increase in fatalities. Without further and more explicit information concerning the "unknown" exogenous fac- tors and an explanation of the mechanism through which changes in these factors generated the increase in fatalities, the evidence does not support the authors' conclusions that the exogenous factors have consistently increased fatalities since 1986. This again underscores the need to appropriately account for determinants of highway safety not related to speed limits. Also, Chang et al.'s analysis finds that speed limit effects are not uniform across relatively homogeneous groups of 65-mph states, which, similar to other studies, raises important issues of the appropriate level of aggregation and hetero- geneity. RI Speed Limits of 65 mph (105 km/h) and Highway Safety-- Statewide Perspective Table C-5 summarizes a number of state-specific studies on the effects of the 65-mph (105-km/h) speed limit, most but not all of which are for the more populous states. Similar to the national stud- ies, the state-specific analyses use a variety of methodological approaches, but the state-specific studies are typically more general in that many examine the effects of the speed limit on fatal as well as nonfatal measures of highway safety. Some relevant points from Table C-5 are the following: · With the exception of a study of Illinois by Pfefer et al. (1991), the speed limit significantly increased RI fatality measures. Fatal crashes on 65-mph (105-km/h) roads increased by as much as 45 percent. · Fatal crashes on 65-mph (105-km/h) roads increased by 45 per- cent in Iowa (Ledolter and Chan 1994), and fatalities increased by as much as 40 percent in Illinois (Rock 1995). · On 55-mph (89-km/h) roads, effects of the speed limit law were mixed. Streff and Schultz (1990) found no effect on 55-mph roads in

Table C-5 U.S. Research on Speed Limits and Highway Safety--State-Specific Studies Study Database for Study Methodology Major Findings Comments Wagenaar Michigan ARIMAX inter- 19% increase in fatalities on 65- Controls for various con- mpha roads et al. Jan. 1978Dec. vention analysis founding effects but some 1989 1988 40% increase in serious injuries on are significant with perverse 65-mpha roads Monthly signs 38% increase in fatalities on 55- Models variable levels only mpha roads Large confidence intervals Fatality equation: R2 = .03 Limited postenactment sample Streff and Michigan ARIMAX inter- 28% increase in fatalities Extends Wagenaar et al. Schultz Jan. 1978Dec. vention 39% increase in serious injuries (1989) study No effect on 55-mpha roads or 1990 1989 analysis No effect on number of vehi- Monthly other roads cles involved, implying that main effect is on severity Wide confidence intervals, often including 0 Why include insignificant covariates? Small number problem?

Pant et al. Ohio Quasi-experimental Increase in injuries and PDO Interesting findings but no crashes on 65-mpha roads 1991 July 1984June Before/after analy- attempt to explain 1987 versus Aug. sis Increase in fatal, decrease in injury Levels only and PDO crashes on 55-mpha Little control for confounding 1987July 1990 Poisson analysis RI roads factors Decrease in injury and PDO crashes on non-RI 55-mpha roads Pfefer et al. Illinois ARIMA interven- No significant effect on passenger Only considered rural 1991 Jan. 1983July 1988 tion analysis car crash rate Interstate crashes Monthly Decrease in fatal-injury car-truck How sensitive are results to crash rate intervention month? McCarthy Indiana Regression analysis Increase in total, fatal/injury, and Fixed effects model 1993 Countywide data PDO alcohol-related crashes Controlled for exposure, age 19811989 Redistribution of alcohol-related distribution, population, crashes from higher-speed to economy, alcohol availabil- lower-speed environments ity, and enforcement Similar effects for most cate- gories of alcohol-related crashes (e.g., daytime, single-vehicle, non-truck- involved) (continued on next page)

Table C-5 (continued) Study Database for Study Methodology Major Findings Comments Jernigan et Virginia Before/after analy- Increase in RI fatalities Levels only al. 1994 19851987 versus sis Decrease in systemwide fatalities No control for confounding 19891992 ANOVA factors Effect on RI fatalities stabi- lized in 19901991 Dual speed limit had no effect on car-truck crashes Khorashadi California Before/after analy- Increase in fatal crashes on 65- Fixed versus random effects RI 65-mpha roads, mpha roads, RI and RNI 1994 sis ANOVA 1,155 mia ANOVA Increase in total, fatal, injury, and Examined various crash RI 55-mpha roads, fatal and injury crashes on 65- causes but no control for 343 mia mpha roads relative to 55-mpha VMT or other confounding RNI 65-mpha roads effects roads, 132 mia 19821986 versus 19881992

Ledolter Iowa Regression time Systemwide Implications for law's effect and Quarterly, series 18% increase in fatal crashes on severity not discussed Chan 19811991 Seemingly unre- 2.4% increase in major injury No control for confounding 1994 lated regression crashes factors Rural roads Possible small number prob- Effect of Speed Limits on Speed Distributions and Highway Safety 45% increase in RI fatal crashes lem for road type analysis 17% increase in rural primary road fatal crashes 12% increase in rural secondary road fatal crashes McCarthy California Regression analysis Systemwide Garber and Graham (1989) 1994b Jan. 1981Dec. No effect on total, fatal, injury, type specification 1989 PDO crashes Results not completely consis- Monthly Road type tent with panel data analy- Law had no effect on injury or sis (by county by year) over fatal crashes on Interstate the same period. McCarthy roads, U.S. highways, state (1994a) found redistribu- highways, or county roads tion away from counties with and toward counties without Interstate roads (continued on next page) 331

Table C-5 (continued) Study Database for Study Methodology Major Findings Comments Rock 1995 Illinois ARIMA interven- 40% increase in RI fatalities, 65- Modeled levels mpha roads May 1982April tion analysis Abrupt permanent interven- 1991 25% increase in RI fatalities, 55- tion mpha roads Monthly Considered only rural high- ways Note: PDO = property damage only; RI = rural Interstate; RNI = rural non-Interstate; VMT = vehicle miles traveled. a55 mph = 89 km/h; 65 mph = 105 km/h; 132 mi = 212 km; 343 mi = 552 km; 1,155 mi = 1859 km.

333 Effect of Speed Limits on Speed Distributions and Highway Safety Michigan, and Pant et al. (1991) obtained mixed results, increased fatal crashes but decreased injury and property damage crashes. · Most studies did not control for confounding factors. · Wagenaar et al. (1989) and Streff and Schultz (1990) estimated time series ARIMAX models (ARIMA models that include explana- tory variables), but the confidence intervals were often wide and the coefficients of other determinants often carried perverse signs. · Only one study, McCarthy (1993), investigated the effect of the law on a subset of crashes, those that were alcohol related. He found no systemwide effect but significant redistribution of crashes away from higher-speed environments and toward lower-speed environ- ments. · In a time series study of California, McCarthy (1994b) found that the law had a statistically insignificant effect on fatal and nonfa- tal injury crashes on Interstate roads, U.S. highways, state highways, and county roads. · For the smaller states that experience few fatal crashes and fatal- ities, a small number of crashes involving many fatalities could sig- nificantly affect observed percentage changes. Also, with a small number of incidents, classical regression models [e.g., Garber and Graham (1989)] are inferior to other modeling procedures. Poisson regression, for example, explicitly accounts for the fact that the dependent variable represents count data, which are likely to include a large number of zeroes. Since fatal crash or fatality data are often more consistent with Poisson rather than classical regression models, Poisson regression is a more appropriate framework for analysis. Local Versus Systemwide Effects An interesting twist in this literature occurred in 1992 when Lave, in a response published with a survey paper by Godwin (1992) on the effects of the 65-mph (105-km/h) speed limit, commented that the new law could have actually saved lives. Using data from an NHTSA (1989b) study, Godwin presented information that, between 1986 and 1988, RI and non-Interstate fatalities increased 35 and 0 percent, respectively, in 65-mph states; in 55-mph (89-km/h) states, the respective increases were 9 and 2 percent. These data are consistent

MANAGING SPEED 334 with much of the literature (as indicated in Table C-4) that 65-mph states experienced a disproportionate increase in RI fatalities. Lave's twist had two points, one theoretical and one empirical. The theoret- ical point was that the 65-mph law, by reducing the relative general- ized cost of driving on RI highways, would alter drivers' route choices away from the less safe non-Interstate roads to the safer RI highways. That is, traffic would be diverted to the RI roads. Further, by remov- ing the threat of cutting off funding if states did not enforce the 55- mph speed limit, the law enabled enforcement agencies to optimally reallocate traffic resources from catching speeders to more productive traffic activities such as targeting DUI drivers. Lave argued that these two effects would produce a systemwide decrease in fatalities. Empirically, Lave demonstrated this by reworking the numbers in Godwin's Table 2. Overall, fatalities in 1988 were 1.86 percent higher in 65-mph (105-km/h) states but 1.92 percent higher in 55-mph (89-km/h) states. Alternatively, to account for exposure, Lave con- sidered fatality rates and found that, for the 65-mph states, the 1988 fatality rate was 2.42 per 100 million VMT (1.50 per 100 million VKT) compared with 2.57 per 100 million VMT (1.60 per 100 mil- lion VKT) in 1986. There was no change in the fatality rate for the 55-mph states. Lave concluded that "if the 1986 fatality rate had still prevailed in 1988, there would have been 2206 more fatalities" (p. 12). Lave attributed the reduction to the 65-mph speed limit since fatality rates in the 55-mph states remained constant. In a reply, Godwin (1992) pointed out that if the fatality savings occurred because of traffic diversion, the required shift in traffic to generate the expected savings would be 88.5 million VMT (142.4 million VKT), much higher than the observed 14.8 percent increase in RI VMT. Lave and Elias (1994) refined Lave's 1992 argument. Using updated data, they estimated models of the Garber and Graham type using systemwide fatalities and fatality rates rather than RI fatality measures as the dependent variable. Separate models were fitted to each of 46 states. Consistent with Lave's earlier comment, the authors in this study concluded that systemwide fatality rates in 65- mph (105-km/h) states fell between 3.4 and 5.1 percent (significant in 14 of 40 states). Reasons cited for the decline included traffic

335 Effect of Speed Limits on Speed Distributions and Highway Safety diversion, reallocation of enforcement resources, and possible declines in speed dispersion. Griffith (1995) and Lave (in a response published with Griffith's comment) discussed issues concerning Lave and Elias's 1994 paper. Two of the more important concerns raised by Griffith, relative to the time series regression results, and Lave's responses are as follows: · Concern: Statewide fatality rates may be too broad to capture systemwide effects of the speed limit. Response: The error associated with too narrow a definition of systemwide effects will miss some of the law's effects, whereas too general a definition will reduce the model's explanatory power by adding noise to the model. If the additional error is systematic, that is, the model missed some strong influence that occurred indepen- dently of the speed limit law but had a negative effect on the fatality measure, then it is important to identify an alternative explanation for the observed results. · Concern: The use of fatality rates assumes that fatalities and VMT move proportionately, contrary to empirical findings that the fatality rate decreases with traffic density. Response: Fatality rates control for exposure, and if the speed limit law diverted traffic to the higher-density Interstate highways, then the law should be credited with the associated lower fatality rates. In general, both of these issues are empirical and can be subjected to hypothesis testing. The first point suggests that other excluded determinants of fatality rates may be responsible for the observed results. Garber and Graham (1989) found considerable heterogene- ity across states, as did Lave and Elias, and raise the specification issue in the concluding remarks of their initial state-by-state analy- sis. Although Lave and Elias (1994, 53) claim that Garber and Graham's empirical model is the "most sophisticated analysis" of the effects of the 65-mph (105-km/h) limit, an unnecessarily strong statement that even Garber and Graham may question, Lave's basic point is relevant. Lave and Elias have proposed a theory that identi- fies the expected systemwide effects of the 65-mph speed limit. As

MANAGING SPEED 336 with all empirical models, their model does not prove the theory but is consistent with it. To falsify their theory, it is not sufficient to argue that some other factors may be responsible. Indeed, this was also a point of criticism for the unwarranted conclusion of Chang et al. (1991) that some exogenous factors, not the speed limit change, drove the observed change in fatalities. Whether VMT is proportional to fatalities is also testable. Estimating, for example, a double-log model and testing the null hypothesis that the coefficient of VMT is 1 is a test of the propor- tionality hypothesis. Although concerns about the direction and size of Lave and Elias's quantitative results are likely to continue, their analysis is important because it tends to reorient one's perspective on the various effects of the speed limit change. There has been a significant amount of research on the direct effects of the higher speed limits. The results of the research suggest that the higher speed limits have increased fatali- ties on RI highways. As indicated in Table C-4, many studies have also found various effects on 55-mph (89-km/h) roads (or in 55-mph states) and interpret them as detrimental spillover effects of the law. There has been relatively little discussion of traffic diversion or, for that matter, traffic generation effects of the higher speed limits. Consider two sets of hypotheses, A and B. Hypothesis Set A is as follows: H0: The 65-mph (105-km/h) speed limit had no effect on sys- temwide highway safety. H1: The 65-mph speed limit decreased systemwide highway safety. Hypothesis Set B is as follows: H0: The 65-mph speed limit had no effect on systemwide highway safety. H1: The 65-mph speed limit increased or decreased systemwide highway safety. If the direct and spillover effects of the law are emphasized to the exclusion of the traffic diversion effects, highway safety is expected to

337 Effect of Speed Limits on Speed Distributions and Highway Safety fall, consistent with Hypothesis Set A. Contrary to most work in the area, Lave and Elias identify a mechanism, based less on speed adap- tation and more on traffic diversion, that supports Hypothesis Set B. In other words, their perspective focuses more on traffic diversion (complemented with a reallocation of enforcement resources). This allows a trade-off between the deteriorating and enhancing effects of the law and produces an alternative hypothesis that admits the pos- sibility of systemwide highway safety improvements as a result of the law. On the basis of time series and panel data for California, McCarthy (1994a, 1994b), consistent with Lave and Elias, found no systemwide effect from the 65-mph (105-km/h) speed limit law. Lave and Elias's paper and subsequent comments and replies gen- erate a number of research questions on the effects of speed limits: · Is it possible to validate Lave and Elias's underlying assumptions on traffic diversion and reallocation of resources? · What role do other confounding factors not included in Lave and Elias's models have on the systemwide highway safetyenhanc- ing effects of the 65-mph (105-km/h) speed limit law? · What is the appropriate geographic size for internalizing the systemwide effects of the law and what are its determinants? · To what extent do national or statewide speed limit laws pro- duce traffic generation effects as the generalized trip cost falls below a road user's reservation price (i.e., the price at which trip demand is zero) or as users shift from nonhighway to highway modes? · As with most work in this area, Lave and Elias's model examines only fatality measures. What effect would a systemwide approach have on the distribution of nonfatal injuries and property damage crashes? · Would multiequation frameworks that modeled highway safety on alternative road types improve our understanding of the mecha- nisms that link changes in speed limits on high-speed roads to sys- temwide highway safety? Summary To sum up, evidence on the effect of higher RI speed limits on high- way safety indicates the following:

MANAGING SPEED 338 · The higher RI speed limits initiated in mid-1987 have generally led to an increase in RI fatalities and fatal crashes. Because other determining factors were not appropriately controlled for, however, initial study estimates of fatality increases in the range of 15 to 30 percent were too high. · The effects of the speed limit exhibited considerable hetero- geneity across states. Part of this may reflect the fact that small states with few crashes will have large proportional changes relative to larger states. However, part also reflects differences across states that are not adequately accounted for in the empirical models. · Similar to the law's mixed effects on the speed distributions of nonlimited-access roads, the higher RI speed limits have produced mixed highway safety effects on nonrural Interstate roads. · There is sufficient evidence in the literature to seriously question whether the net effect of the law is unambiguously negative. More work is needed on the distribution effects and the overall net effects of higher speed limits on limited-access roads. INTERNATIONAL WORK ON SPEED LIMITS In addition to U.S. work on the linkages between speed limits and highway safety, a number of international studies examined this issue. Tables C-6 and C-7 summarize the findings of recent international analyses for lower- and higher-speed roads, respectively. Lower-Speed Roads Table C-6 indicates that, unlike the United States, substantial work has been done in international studies on the effects of speed limits on lower-speed roads [roads ranging from 19 mph (30 km/h) to 50 mph (80 km/h)]. In general, the analysis in these studies is very sim- ilar to that used in many U.S. studies, namely, quasi-experimental approaches dominated by a paired comparisons methodology. As such, these studies tend to generate similar effects and suffer the same drawbacks. On the positive side, the imposition of speed limits in lower-speed environments is typically associated with a decrease in crashes and crash severity. However, these analyses generally suffer

Table C-6 International Research on Speed Limits and Highway Safety--Lower-Speed Roads Study Database for Study Methodology Major Findings Comments Prior speed limits were 37 mpha Engel and Denmark Logit regression 9% decrease in crashes Thomsen Introduced a 31- 24% decrease in fatalities Limited sample size mpha speed limit, 1988 No control for confounding factors urban areas Assumes that the proportion of Quarterly data, Oct. urban to rural miles traveled is 1985Oct. 1987 the same before and after the law change General decrease in speeds and Prior speed limits were 31 mpha Schleicher- Germany Before/after Jester Implemented 19- analysis crash severity Speed limit decreases combined mpha speed zones 1990 with public information, traffic 19831986 control, speed control, and street design changes (continued on next page)

Table C-6 (continued) Study Database for Study Methodology Major Findings Comments Vis et al. Netherlands, 15 Before/after 20% speed reduction, generally No information on prior speed 1992 municipal analysis resulting in an 85th per- limits centile speed of 19 mpha areas Quasi-experi- Speed limit aimed to "integrate" Implemented 19- mental Traffic volume fell 5% to 30% road user categories, where the mpha speed zones motorist identifies 19 mpha as 5% trend-adjusted decrease in 1980s all crashes the appropriate speed 25% trend-adjusted decrease in Combined with engineering mea- injury crashes sures to slow traffic (e.g., humps, axis realignments, traffic islands) Where did the decreased traffic go? No experimental site; changed only the speed limit sign

Engel and Denmark, residential Quasi-experi- 18.4% decrease in control No information on prior speed Thomsen areas mental group adjusted crashes limits Introduced 19-mpha Before/after 1992 21.1% decrease in control 3 years of before data, 3 years of speed zones analysis group adjusted injuries after data 139 mi,a experimental group; 44 experimental Regression analy- 72% decrease in casualties per 11,766 mi,a control group areas, 53 control sis road user, experimental areas areas No change in crash risk per Status of streets changed from 1980s user in experimental areas "traffic streets" to "living areas" 96% increase in casualties per Speed-reducing measures also road user, just outside experi- implemented 0.4-in.a increase in height of hump mental areas decreased speed by 0.6 mpha Road narrowing decreased speed by 2.9 mpha No discussion of effect on casual- ties per road user in outer areas (continued on next page)

342 Table C-6 (continued) Study Database for Study Methodology Major Findings Comments MANAGING SPEED 25-mpha limit led to perma- Cairney City of Unley, Before/after No information on prior speed nent 3-mpha reduction in and Australia analysis limits 25-mpha speed zone 2-mia by 660-fta study area in Fackrell speed 1993 19911993 Initial temporary fall in traffic Unley volume Size of speed reduction at experi- Effect of increased enforce- mental sites varied ment ambiguous Examines effects of speed limit changes with and without speed camera enforcement Authors question whether changes in economic factors may have affected results Newstead Victoria, Australia Before/after No systemwide effect Speed limits increased on 1,196 31-,a 43-,a and 50- mia of roads, decreased on 342 and analysis 6.9% increase in injury crashes mpha speed limit mia of roads Mullan Quasi-experimen- for metropolitan Melbourne, 1996 zones tal marginally significant For Melbourne, 47% decrease in 19921993, 32.9% reduction in injury injury crashes when limit increased from 37 mpha to 50 mpha and a 19941995 crashes in the rest of Victoria, marginally signifi- 10.5% increase when limit increased from 47 mpha to 50 mpha cant No control for other confounding factors

aStudy used Standard International units, which were converted to U.S. equivalents in the table. Correspondences are as follows: 0.4 in. = 1 cm; 660 ft = 200 m. mi or mph km or km/h mi or mph km or km/h 0.6 1 43 70 2 4 47 75 2.9 4.7 50 80 3 5 139 223 19 30 342 550 25 40 1,196 1925 31 50 11,766 18935 37 60

Table C-7 International Research on Speed Limits and Highway Safety--Higher-Speed Roads Study Database for Study Methodology Major Findings Comments Fieldwick Europe and the Regression Decrease in urban speed limit from 37 Confidence interval for pre- mpha to 31 mpha would decrease fatal and United States cross-section dicted effects not given Brown 1984 analysis and nonfatal injuries 25% No control for cross-section 1987 Similar but smaller effect if rural speed heterogeneity limits decrease from 62 mpha to 56 Authors caution that other mpha excluded variables could reduce the beneficial effects found in their analysis Nilsson Sweden Before/after Decrease in average speed by less than Assumes that speed limit 56-mpha speed 1990 analysis speed limit decrease change had no effect on pre- Relative to other 56-mpha roads, 15% vious 56-mpha roads limit on 3,400 mia of roads (11%) decrease in injury crashes No control for accompanying 1988, 1989 (injuries), neither effect statistically changes in public informa- significant tion and enforcement or other confounding factors Controlled speed limit was 62 mpha Sliogeris Victoria, Australia Before/after Statistically significant 24% increase in 1992 Imposition and analysis injury crashes per mile after introduc- No control for other factors tion of 68-mpha speed limit removal of a Regression Control group is all other 62- 68-mpha speed mpha signed roads in analysis Statistically significant 19% decrease in limit injury crashes per mile after removal Victoria of 68-mpha speed limit 19851991 Similar results for rural and urban roads

Borsje Netherlands Before/after Differentiated speeds on motorways 75 mpha on 80% of motor- 1995 Introduction of analysis decreased average speed and had a ways, 62 mpha on 20% of general 75-mpha nonincreasing effect on speed disper- motorways sion for 62-mpha and 75-mpha roads speed limit Statistical significance of 19881992 Positive effect on crash incidence results not reported Accompanying policies included greater enforce- ment, media campaigns, infrastructure changes Johansson Sweden Poisson time No statistically significant effect on fatal Methodology accounts for 56-mpha speed 1996 series analy- or serious injury crashes overdispersion and serial limit sis Statistically significant decrease in minor correlation Monthly, injury and vehicle damage crashes Controls for exposure 19821991 (through economic vari- ables), seasonal effects, safety belt law aStudy used Standard International units, which were converted to U.S. equivalents in the table. Correspondences are as follows: mi or mph km or km/h 31 50 37 60 56 90 62 100 75 120 3,400 5500

MANAGING SPEED 346 from not appropriately accounting for confounding factors and using a comparison series that may also be affected by the speed limit change. There are interesting and unique aspects to some of these experi- ments. First, three European countries--Germany, the Netherlands, and Denmark--have analyzed the effects of a 19-mph (30-km/h) speed zone in urban areas. In each of these cases, the speed limit was part of an urban planning policy whereby traffic users shared the streets with other users. Complementing the reduced limit were other actions, including public information campaigns, increased enforcement, engineering speed measures, and so forth, intended to inform the public (directly or indirectly) that the appropriate speed on the affected roads was lower than in the surrounding areas. Moreover, in the Netherlands study, Vis et al. (1992) report that for all experimental sites a combination of actions was taken. In other words, in no case did the speed limit change simply involve a speed limit sign change. Thus, it is not possible in these studies to draw any conclusions concerning the effect of a speed limit sign change only. However, there were differential effects, depending on the specific combination of actions taken, which suggests that effective speed limit changes involve the implementation of reinforcing policies. Faure and de Neuville (1992) also make this point in a description of France's "Safer City, Accident-Free Districts" program. A second point of interest is that part of the decrease in crashes in some studies was due to a decrease in traffic volume, which raises the question of the traffic distribution effects of the speed limit. An illus- tration of this was Denmark's 19-mph (30-km/h) speed zone in res- idential areas. In areas just outside the speed limit effect area, Engel and Thomsen (1992) report (with no discussion) a 96 percent increase (significance level not reported) in casualties per road user. This leads to the question of whether the 30-km/h zone has suffi- ciently diverted traffic to outer areas that the net effect is a deterio- ration of safety. Local versus systemwide effects are also present in Newstead and Mullan's (1996) recent study of the speed limit policy of Victoria, Australia, which attempted to rationalize speed limits on more than 1,550 mi (2500 km) of its roads. The authors found that differenti-

347 Effect of Speed Limits on Speed Distributions and Highway Safety ated speed limits increased injury crashes in metropolitan Melbourne, decreased injury crashes in the rest of Victoria, and pro- duced no overall systemwide effect. Pedestrian Safety There has been some but not much work on the relationship between speed limits and pedestrian safety. The evidence leads to an expected positive relationship between vehicular speed and the incidence and severity of pedestrian crashes: · Ashton and Mackay (1979) report that 5, 45, and 85 percent of pedestrians hit by vehicles traveling 20, 30, and 40 mph (32, 48, and 64 km/h), respectively, end in a fatality. Pasanen (1992) makes a sim- ilar point in a Finnish study, finding that the risk of fatality when a pedestrian is hit by a vehicle traveling 31 mph (50 km/h) is nearly 8 times higher than when a pedestrian is hit by a vehicle traveling 19 mph (30 km/h). In a study of 118 speed-related pedestrian fatalities in Adelaide, Anderson et al. (1997) found that a small reduction in vehicle speed could produce a large decrease in pedestrian crash risk by decreasing vehicle impact speed. In a 37-mph (60-km/h) speed zone, for example, the authors determined that a 6-mph (10-km/h) fall in traveling speed would produce 48 percent fewer pedestrian fatalities, and that for 22 percent of the cases, the pedestrian-related crash would not have occurred at the lower speed limit. · Pasanen (1992) found that pedestrians were at fault in 84 per- cent of the crashes. In many of these cases, the pedestrian was not aware of an approaching vehicle. Higher-Speed Roads The relatively few recent international studies on higher speed lim- its, reported in Table C-7, indicate a positive correlation between relaxed speed limits and average speeds and crashes. Nilsson (1990) found, for example, that Sweden's speed limit reduction (imple- mented in June 1989) from 68 to 56 mph (110 to 90 km/h) had ben- eficial effects. On the basis of speed and crash information in 1988

MANAGING SPEED 348 (before the change) and 1989 (after the change), average speeds fell, but not by as much as the speed limit reduction. Total and injury crashes fell 15 and 11 percent, respectively. However, since there was not an attempt to control for changes in enforcement, public infor- mation, or other confounding factors, the extent to which the observed beneficial effects are specific to speed limit changes is unclear. Johansson's (1996) analysis of the same event sheds addi- tional light on the law's effects. On the basis of monthly data cover- ing the period January 1982 through December 1991, Johansson estimated a Poisson time series model, which controlled for serial correlation, seasonal effects, safety belt legislation, and exposure. Although citing alcohol and driver age as potential determinants of safety, Johansson excluded each from the analysis, the former because alcohol policy had not changed in Sweden during the analysis period and the latter because data were unavailable. Like Nilsson, Johansson found that the speed limit reduction was beneficial in that the num- ber of minor injuries and property damage only crashes fell. However, unlike Nilsson, Johansson found that the law had no effect on the number of fatal or serious injury crashes. In addition to the studies reported in Table C-7, there has been considerable work in Finland on the effects of speed limit changes on higher-speed roads. From the 1960s through the 1970s, Finland undertook a series of speed limit experiments, the results of which were broadly consistent with the findings reported in Table C-7 (Salusjarvi 1988). In addition, between 1987 and 1988, Finland ini- tiated a series of seasonal speed limit experiments (Finch et al. 1994) that included (a) reducing speed limits on 1,200 mi (2000 km) of roads from 62 to 50 mph (100 to 80 km/h) in the winter of 1987 [in the winter of 1988, speed limits on an additional 1,200 mi (2000 km) of roads were similarly reduced], (b) reducing speed limits on all 75- mph (120-km/h) motorways to 62 mph (100 km/h) during each win- ter period, and (c) increasing speed limits on 870 mi (1400 km) of roads from 50 to 62 mph (80 to 100 km/h) in the summers of 1988 and 1989, half each summer. Findings from these experiments indi- cated that reductions (increases) in speed limits were associated with decreases (increases) in average speed and speed dispersion. In addi- tion, there was a positive correlation between the direction of the

349 Effect of Speed Limits on Speed Distributions and Highway Safety speed limit change and the effect on crashes. The extent to which other factors may have affected these results, however, is not known. Summary This group of international studies yields the following insights in setting speed limits: · For nonlimited-access urban roads, local speed limit zones were successful in reducing speeds and crashes when implemented with complementary policies such as public information campaigns, greater enforcement, and engineering measures. However, part of the improvement in urban speed zones was due to reduced volumes, and there was little analysis of where the traffic went, that is, the policy's effects outside the zones. Limited evidence suggests that the sys- temwide effects may be zero. · For nonlimited-access roads, there is limited evidence that enforcement is an important determinant of safety. This is consistent with more recent work by Elvik (1997), who used meta-analysis to explore the effect of automated speed enforcement on traffic safety. On the basis of work in Norway, Germany, Sweden, England, the Netherlands, and Australia, Elvik estimated that automated speed enforcement reduced injury crashes by 17 percent. · On higher-speed roads, the international work is broadly con- sistent with that in the United States. With little control for other confounding factors, higher (lower) speed limits generally lead to higher (lower) speeds and more (fewer) crashes. The absolute change in speed is less than the absolute change in the speed limit. As a final point, there has been limited international work on the role of enforcement in highway safety. CONCLUSIONS AND AREAS FOR FUTURE RESEARCH The purpose of this review is to provide an overview of recent empir- ical research on the effect of changes in posted speed limits on speed distributions and highway safety. To provide a contextual basis for the

MANAGING SPEED 350 review, the conceptual framework and statistical methodologies com- mon to many of the reviewed studies were summarized. Recent domestic and international research on the effects of posted speed limits was reviewed. What implications does this work have for determining speed limit policies? This section focuses on two areas. First, notwithstanding the often significant differences in geographic and temporal scope as well as methodological approach, can anything specific be said about the effect of alternative speed limits? Second, although there has been a significant amount of work on speed limits, are there important gaps in existing research that future research should address? Conclusions With regard to specific implications for speed limit policy, existing studies provide support for the following conclusions: · Speed limits must be perceived by the traveling public as "rea- sonable," that is, as consistent with the enforced traffic conditions experienced by the typical driver. To the extent that this is not true, more drivers will be noncompliant, which could compromise high- way safety. · On nonlimited-access roads, speed limit changes of 5 to 10 mph (8 to 16 km/h) "for cause" (e.g., based on crash experience, increased pedestrian traffic, more businesses) will likely have little effect on speed distribution and highway safety. · For nonlimited-access roads, the greatest effect on speed distri- butions and highway safety occurs when a speed zone is implemented as part of an urban planning policy that simultaneously introduces complementary measures (e.g., greater enforcement, public informa- tion campaigns, engineering measures) to slow drivers down. · For limited-access roads, a 10-mph (16-km/h) increase in RI speed limits from 55 mph (89 km/h) to 65 mph (105 km/h) has gen- erally increased nationwide average speeds by less than 4 mph (6 km/h) and increased nationwide speed dispersion by less than 1 mph (2 km/h). But there is considerable cross-state variation in these effects.

351 Effect of Speed Limits on Speed Distributions and Highway Safety · For limited-access roads, both average speed and speed disper- sion are inversely related to highway safety in general and fatalities in particular. The effect of speed dispersion is most important for RI roads. Drivers traveling in the top 15th percentile appear to compro- mise highway safety more than those traveling in the lowest 15th percentile. · The increase in RI speed limits from 55 to 65 mph (89 to 105 km/h) has generated mixed results with respect to its effect on non- rural Interstate roads whose speed limits remained the same. · Although increasing RI speed limits from 55 to 65 mph (89 to 105 km/h) produces highway fatality distribution effects, the evi- dence is consistent with (at least) a zero net systemwide effect. · Evidence indicates that speed adaptation occurs, but the effect appears to be small, suggesting that the highway safety effect will also be small. · The few studies that have analyzed the effects of alternative enforcement levels indicate that traffic enforcement is an important determinant of highway safety. Areas for Future Research There are a number of areas in which additional research could sig- nificantly contribute to an understanding of the effect of speed lim- its on speed distributions and highway safety. These areas are discussed in the remainder of this section. Data Collection Common among many, if not most, of the studies reviewed in this paper was either the absence of or insufficient control of confound- ing factors that affect speed distributions and highway safety. Studies that do not adequately control for these related factors will likely be biased and lead to improper inferences. Although in some cases it is not possible to obtain additional information, in most cases there is at least some possibility of improved specification to better isolate the true effect of speed limit changes. An important omission in most analyses is enforcement, yet the few studies that

MANAGING SPEED 352 include this variable find that enforcement is an important determi- nant of highway safety. Much of the data on speed distributions is highly aggregated. Future research should aim to generate speed distribution data that are better aligned with the environment in which the crash occurred (e.g., late-night crashes on rural two-lane roads). Further, free-flow data are often collected to generate measures of speed distribution, but how useful are these data in determining highway safety out- comes in non-free-flow periods (i.e., during congested periods)? This is an open question. Also, since travel occurs on a network of inter- connected roads, are existing aggregate measures of highway safety reasonable speed distribution proxies for different functional road types, or does use of these measures cause important biases? Methodology As noted earlier, existing research on speed limits generally uses uni- variate classification procedures, regression analysis, or ARIMA time series models. Multivariate classification models are rarely used to analyze the effects of highway safety. Among simple regression mod- els, there is often a surprising lack of diagnostics and, if necessary, correction for common statistical problems (e.g., serial correlation in time series analysis). The relationship between speed limits, speed distribution, and highway safety would likely be better understood if researchers experimented more with more general models of highway safety or alternative methodologies. The infrequency of fatal crashes and fatal- ities is amenable to Poisson regression techniques, yet these methods have been used less frequently than might be expected. In addition, there has been little work on developing and estimating simultaneous frameworks to capture the interaction between the demand for road space and highway safety. A third area is the use of various probabil- ity models. Ordered probit (Greene 1997), for example, is one methodology that could be used to examine the effect of speed lim- its on crash severity. To date, most work on speed limits applies a particular methodol- ogy to a particular data set (e.g., regression analysis on cross-section

353 Effect of Speed Limits on Speed Distributions and Highway Safety data from California or ARIMA models on monthly time series data from Michigan). Given the availability of state and national data and computing technology, developing multiple data sets from a given base of information is relatively easy and would enable researchers to make different passes at the same underlying information. Using data from 1980 through 1996 for a given state that raised its speed limit in 1987, for example, would an aggregate regression model based on annual data produce similar effects of the law as would an ARIMA model for that state based on monthly data over the same time period? Would the effects be similar to those generated from uni- variate classification models that compare crash rates between 1980 and 1986 with those between 1987 and 1996? Such analyses may produce dramatically different estimates of the law's effects and, in so doing, identify potentially fruitful areas for further research. Unresolved Research Issues Following are a number of issues on which some research exists but on which further research is warranted. · How robust is the empirical finding that a 10-mph (16-km/h) increase in RI speed limits leads to an increase in average speed and speed dispersion, respectively, of around 4 mph (6 km/h) and under 1 mph (2 km/h), particularly if there is greater control of other deter- mining factors? This is particularly important given the recent increase in speed limits to 70 mph (113 km/h) that some states implemented following passage of the National Highway System Designation Act in 1995. · How important is enforcement in determining speed distribu- tion properties and highway safety when speed limits change? · Although there is some information on the relationship between average speed, speed dispersion, and highway safety, much additional work is needed to improve the understanding of these relationships. Are average speed and speed dispersion negatively related? If so, does this relationship hold at all speeds or is there a nonlinear (or inde- pendent) relationship between these measures at other speeds? A related question is whether speed dispersion is only important on

MANAGING SPEED 354 high-speed RI highways, as suggested in some research, and less important on lower-speed nonlimited-access roads. · What level of aggregation is appropriate for accurately charac- terizing the relationship between speed distribution and highway safety? · More information is needed on how speed limit laws affect the spectrum of crashes, injury related as well as noninjury related. Although nonfatal crash and nonfatal injury data are less reliable, this does not justify ignoring all the available information. · Additional research is needed on the spillover effects associated with speed limit changes for both limited- and nonlimited-access roads. The literature on this issue is mixed, and there is no consen- sus on either the direction or the magnitude of the effect. Better models are needed to identify the linkages that produce spillover effects. · In addition to hypothesized spillover effects, speed limit changes, particularly on limited-access high-speed roads, are likely to produce traffic generation effects. It is not known whether these effects exist and, if so, how important they are in determining the highway safety effects of speed limit changes. · Sufficient evidence exists to question whether the net effect on highway safety of speed limit laws is to deteriorate highway safety. More research is needed on the distribution effects of speed limit laws to evaluate their net effects on highway safety. · Reducing the impact speed of vehicles can have a significant effect on pedestrian crashes and fatalities. For the United States there is little information on the extent to which pedestrian fatalities are an important consideration when speed limits change. REFERENCES ABBREVIATIONS FHWA Federal Highway Administration NHTSA National Highway Traffic Safety Administration TRB Transportation Research Board Anderson, R. W. G., et al. 1997. Vehicle Travel Speeds and the Incidence of Fatal Pedestrian Crashes. Accident Analysis and Prevention, Vol. 29, pp. 667674.

355 Effect of Speed Limits on Speed Distributions and Highway Safety Ashton, S. J., and G.M. Mackay. 1979. Some Characteristics of the Population Who Suffer Trauma as Pedestrians When Hit by Cars and Some Resulting Implications. Proceedings of the IRCOBI Conference, Gothenburg, Sweden. Baum, H., et al. 1989. The Mortality Consequences of Raising the Speed Limit to 65 mph on Rural Interstates. American Journal of Public Health, Vol. 79, pp. 13921395. Baum, H., et al. 1990. Motor Vehicle Crash Fatalities in the Second Year of 65 mph Speed Limits. Journal of Safety Research, Vol. 21, pp. 18. Baum, H., et al. 1991. The Fatality Consequences of the 65 mph Speed Limits, 1989. Journal of Safety Research, Vol. 22, pp. 171177. Borsje, J. F. 1995. The Effects of the 1988 Differentiation of Speed Limits in the Netherlands. Proceedings, Road Safety in Europe and Strategic Highway Research Program (SHRP), Swedish Road and Transport Research Institute Conference 2A, Part 4, Lille, France, Sept. 26-28, 1994, pp. 2536. Cairney, P. T., and H. L. Fackrell. 1993. The Effects of a 40 km/h Local Area Speed Limit on Traffic Behaviour and Community Perceptions and Opinion. Research Report ARR 243. Australian Road Research Board Ltd., July. Casey, S. M., and A. K. Lund. 1987. Three Field Studies of Driver Speed Adaptation. Human Factors, Vol. 23, pp. 541550. Casey, S. M., and A. K. Lund. 1992. Changes in Speed and Speed Adaptation Following Increase in National Maximum Speed Limit. Journal of Safety Research, Vol. 23, pp. 135146. Cerrelli, E. C. 1981. Safety Consequences of Raising the National Speed Limit from 55 mph to 60 mph. National Highway Traffic Safety Administration, U.S. Department of Transportation. Chang, G. L., et al. 1991. Safety Impacts of the 65-mph Speed Limit on Interstate Highways. AAA Foundation for Traffic Safety, Aug. Elvik, R. 1997. Effects on Accidents of Automatic Speed Enforcement in Norway. Presented at 76th Annual Meeting of Transportation Research Board, Washington, D.C. Engel, U., and L. Thomsen. 1988. Speeds, Speed Limits, and Accidents. The Road Trend in Built Up Areas after the Introduction of the 50 km/h Limit. Technical Report 3/1988. Danish Council of Road Safety Research, Oct. Engel, U., and L. K. Thomsen. 1992. Safety Effects of Speed Reducing Measures in Danish Residential Areas. Accident Analysis and Prevention, Vol. 24, pp. 1728. Faure, A., and A. de Neuville. 1992. Safety in Urban Areas: The French Program "Safer City, Accident-Free Districts." Accident Analysis and Prevention, Vol. 24, pp. 3944. FHWA. 1995. Travel Speeds, Enforcement Efforts, and Speed-Related Highway Statistics. FHWA-SA-95-051. U.S. Department of Transportation, Oct. Fieldwick, R., and R. J. Brown. 1987. The Effect of Speed Limits on Road Casualties. Traffic Engineering Control, Vol. 28, pp. 635640. Fildes, B. N., and S. J. Lee. 1993. The Speed Review: Road Environment, Behaviour, Speed Limits, Enforcement and Crashes. Roads and Traffic Authority of New South Wales and the Department of Transport and Communications, Australia, Sept.

MANAGING SPEED 356 Finch, D. J., et al. 1994. Speed, Speed Limits, and Accidents. Project Report 58. Safety Resource Centre, Transport Research Laboratory, United Kingdom. Fowles, R., and P. D. Loeb. 1989. Speeding, Coordination, and the 55 mph Limit: Comment. American Economic Review, Vol. 79, pp. 916921. Freedman, M., and A. F. Williams. 1992. Speeds Associated with 55-mph and 65- mph Speed Limits in Northeastern States. ITE Journal, Feb., pp. 1721. Freedman, M., and J. R. Esterlitz. 1990. Human Factors and Safety Research Related to Highway Design and Operation. In Transportation Research Record 1281, Transportation Research Board, National Research Council, Washington, D.C., pp. 5261. Garber, N., and R. Gadiraju. 1988. Speed Dispersion and Its Influence on Accidents. AAA Foundation for Traffic Safety, July. Garber, S., and J. D. Graham. 1989. The Effects of the New 65 Mile-per-Hour Speed Limit on Rural Highway Fatalities: A State-By-State Analysis. Final Grant Report DOT HS 807 452, July. Godwin, S. R. 1992. Effect of the 65 m.p.h. Speed Limit on Highway Safety in the U.S.A. With comments by C. Lave and reply to comments. Transport Reviews, Vol. 12, pp. 114. Greene, W. H. 1997. Econometric Analysis (third edition). Prentice-Hall, Inc., Upper Saddle River, N.J. Griffith, M. S. 1995. Discussion of "Did the 65 mph Speed Limit Save Lives?" With response by C. Lave. News, Queries & Comment Section, Accident Analysis and Prevention, Vol. 27, pp. 137140. Jernigan, J. D., et al. 1994. Impact of 65-mph Speed Limit on Virginia's Rural Interstate Highways: 19891992. Final Report VTRC 95-R7. Virginia Transportation Research Council, Nov. Johansson, P. 1996. Speed Limitation and Motorway Casualties: A Time Series Count Data Regression Approach. Accident Analysis and Prevention, Vol. 28, pp. 7387. Khorashadi, A. 1994. Accidents Before and After the 65 mph Speed Limit in California. Final Report CA-TO-94-1. California Department of Transportation, Sacramento, June. Knowles, V., B. Persaud, M. Parker, and G. Wilde. 1997. Safety, Speed and Speed Management: A Canadian Review. Final Report, Contract T8080-5-6858. Transport Canada, March. Lave, C. 1985. Speeding, Coordination, and the 55 mph Limit. American Economic Review, Vol. 75, pp. 11591164. Lave, C. 1989. Speeding, Coordination, and the 55 mph Limit: Reply. American Economic Review, Vol. 79, pp. 926931. Lave, C., and P. Elias. 1994. Did the 65 mph Speed Limit Save Lives? Accident Analysis and Prevention, Vol. 26, pp. 4962. Ledolter, J., and K. S. Chan. 1994. Safety Impact of the Increased 65 mph Speed Limit on Iowa Rural Interstates. Midwest Transportation Center, University of Iowa, Iowa City, July. Levy, D. T., and P. Asch. 1989. Speeding, Coordination, and the 55 mph Limit: Comment. American Economic Review, Vol. 79, pp. 913915.

357 Effect of Speed Limits on Speed Distributions and Highway Safety Mace, D. J., and R. Heckard. 1991. Effect of the 65 mph Speed Limit on Travel Speeds and Related Crashes. Final Report DOT HS 807 764, March. Matthews, M. L. 1978. A Field Study of the Effects of Drivers' Adaptation to Automobile Velocity. Human Factors, Vol. 20, No. 6, pp. 709716. McCarthy, P. S. 1988. Highway Safety and the 65 mph Maximum Speed Limit: An Empirical Study. AAA Foundation for Traffic Safety, Oct. McCarthy, P. S. 1993. The Effect of Higher Rural Interstate Speed Limits in Alcohol Related Accidents. Journal of Health Economics, Vol. 11, pp. 281299. McCarthy, P. S. 1994a. An Empirical Analysis of the Direct and Indirect Effects of Relaxed Interstate Speed Limits on Highway Safety. Journal of Urban Economics, Vol. 36, pp. 353364. McCarthy, P. S. 1994b. Relaxed Speed Limits and Highway Safety: New Evidence from California. Economics Letters, Vol. 46, pp. 173179. McCoy, P. T., et al. 1993. Evaluation of Lower Speed Limits on Urban Highways. Research Report TRP-02-26-92. Nebraska Department of Roads, May. McKnight, A. J., et al. 1989. The Effect of the 65 mph Limit on Speeds and Accidents. Final Grant Report DOT HS 807 463, Aug. NHTSA. 1989a. The Effects of the 65 mph Speed Limit During 1987: A Report to Congress. U.S. Department of Transportation, Jan. NHTSA. 1989b. The Effects of the 65 mph Speed Limit Through 1988: A Report to Congress. U.S. Department of Transportation, Oct. NHTSA. 1990. The Effects of the 65 mph Speed Limit Through 1989: A Report to Congress. U.S. Department of Transportation, Dec. NHTSA. 1992. The Effects of the 65 mph Speed Limit Through 1990: A Report to Congress. U.S. Department of Transportation, May. Newstead, S., and N. Mullan. 1996. Evaluation of the Crash Effects of the Changes in Speed Zones in Victoria During 1993-94 (Excluding 100 to 110 km/h). Report 93. VicRoads, Australia, June. Nilsson, G. 1990. Reduction in the Speed Limit from 110 km/h to 90 km/h During Summer 1989. VTI Report 358A. Swedish Road Safety Office. Pant, P. D., et al. 1991. Effects of the 65-mph Speed Limit on Traffic Accidents in Ohio. In Transportation Research Record 1375, Transportation Research Board, National Research Council, Washington, D.C., pp. 5360. Parker, M. R., Jr. 1997. Effects of Raising and Lowering Speed Limits on Selected Roadway Sections. Final Report FHWA-RD-92-084. Federal Highway Administration, U.S. Department of Transportation, Jan. Pasanen, E. 1992. Driving Speeds and Pedestrian Safety: A Mathematical Model. Transportation Engineering Publication 77. Helsinki University of Technology, Finland. Pfefer, R. C., W. W. Stenzel, and B. D. Lee. 1991. Safety Impact of the 65-mph Speed Limit: A Time Series Analysis. In Transportation Research Record 1318, Transportation Research Board, National Research Council, Washington, D.C., pp. 2233. Rock, S. M. 1995. Impact of the 65 mph Speed Limit on Accidents, Deaths, and Injuries in Illinois. Accident Analysis and Prevention, Vol. 27, pp. 207214.

MANAGING SPEED 358 Salusjarvi, M. 1988. The Speed Limit Experiments on Public Roads in Finland. In Proceedings of Roads and Traffic Safety on Two Continents, VTI Report 332A, Vagoch Trafik Institute, Linkoping, Sweden. Schleicher-Jester, F. 1990. TEMPO 30 in Towns--Results of a German Experiment. Living and Moving in Cities. Centre d'Études des Transports Urbans, Bagneux, France, pp. 100107. Sliogeris, J. 1992. 100 Kilometre per Hour Speed Limit--Evaluation of Road Safety Effects. Report GR 92-8. VicRoads, Australia, Aug. Snyder, D. 1989. Speeding, Coordination, and the 55 mph Limit: Comment. American Economic Review, Vol. 79, pp. 922-925. Solomon, D. 1964. Accidents on Main Rural Highways Related to Speed, Driver, and Vehicle. Federal Highway Administration, U.S. Department of Transportation, July. Streff, F. M., and R. H. Schultz. 1990. The 65-mph Speed Limit in Michigan: A Second Year Analysis of the Effects on Crashes and Crash Casualties. Final Report UMTRI- 90-37. The University of Michigan Transportation Research Institute, Sept. TRB. 1984. Special Report 204: 55: A Decade of Experience. National Research Council, Washington, D.C. Ullman, G. L., and C. L. Dudek. 1987. Effects of Reduced Speed Limits in Rapidly Developing Urban Fringe Areas. In Transportation Research Record 1114, Transportation Research Board, National Research Council, Washington, D.C., pp. 4553. Vis, A. A., et al. 1992. Safety Effects of 30 km/h Zones in the Netherlands. Accident Analysis and Prevention, Vol. 24, pp. 7586. Wagenaar, A. C., et al. 1989. The 65 mph Speed Limit in Michigan: Effects on Injury and Death. Final Report UMTRI-89-28. The University of Michigan Transportation Research Institute, Dec.