Identification and Evaluation of the Cost-Effectiveness of Highway Design Features to Reduce Nonrecurrent Congestion (2014)

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.

4Background The Reliability area of the second Strategic Highway Research Program (SHRP 2) has focused on the need to improve travel time reliability on freeways and major arterials. The objec- tives of the present research were to (1) identify the full range of possible roadway design features used by transportation agencies to improve travel time reliability and reduce delays from key causes of nonrecurrent congestion, (2) assess their costs and operational and safety effectiveness, and (3) provide recommendations for their use and eventual incorporation into appropriate design guides. The research focused on geo- metric design treatments that can be used to reduce delays due to nonrecurrent congestion. Recurrent Congestion Versus Nonrecurrent Congestion Congestion and consequent delay to motorists result from both recurrent and nonrecurrent congestion. Recurrent Congestion Recurrent congestion is regularly occurring, predictable con- gestion that is generally experienced on a daily basis. On free- ways and major arterials, recurrent congestion is generally caused by traffic demand on a facility nearing or exceed- ing a facility’s capacity, and it is most frequently associated with commuter travel during the morning and evening peak periods. On local roads and at intersections, recur- rent congestion can also be caused by daily recurring events such as afternoon school dismissals or shift breaks at large employment sites. Recurrent congestion has traditionally been addressed through the design or redesign of highways, bridges, and intersections on which it has occurred or is expected to occur. Nonrecurrent Congestion Nonrecurrent congestion arises from random events that are generally unpredictable to the facility user, vary in degree from day to day and from one incident to the next, and create unreliable travel times that frustrate motorists. Sources of nonrecurrent congestion include the following: • Traffic incidents. Traffic incidents are events that disrupt the normal flow of traffic and often involve a blockage of one or more travel lanes. Incidents include such events as vehicle crashes, disabled vehicles, and debris in the travel lane. • Weather. Reduced visibility and roadway surface friction can affect driver behavior and, as a result, traffic flow. Drivers will usually lower their speeds and increase their headways when poor weather conditions are present. • Demand fluctuations. Demand fluctuation refers to the day-to-day variability in traffic demand that leads to higher traffic volumes on some days than on others. Fluctuating traffic demand volumes also result in variable travel times. • Work zones. Work zones are sections of the roadway, or roadside, on which construction, maintenance, or utility work activities take place. Work zones may involve a reduc- tion in the number or width of travel lanes, lane shifts, lane diversions, reduction or elimination of shoulders, or tem- porary roadway closures. • Special events. Special events include such occasions as major sporting events, festivals, concerts, and even seasonal shop- ping. Such events cause the traffic flow in the vicinity of the event to differ radically from typical patterns. Special events may cause surges in traffic demand that overwhelm the system. • Traffic control devices. Intermittent disruption of traffic flow by malfunctioning or poorly timed signals or by rail- road grade crossings contributes to congestion and travel time variability. C h a p t e r 1 Introduction

5 Nonrecurrent congestion has not traditionally been addressed through highway design. In recent decades, opera- tional solutions such as intelligent transportation systems and incident management techniques have been the chief means of combating nonrecurrent congestion. However, highway designers are more frequently considering infra- structure that directly addresses nonrecurrent congestion and that supports or facilitates operational strategies for addressing nonrecurrent congestion during roadway design and redesign projects. Nonrecurrent congestion is the cause of unpredictable delay; reliability is the measurement of its effects. As the fre- quency and severity of nonrecurrent congestion events on a facility increase, the reliability of that facility decreases. Definition of Reliability and Key Terms Reliability, which is shorthand for travel time reliability, is an important component of roadway performance and, per- haps more importantly, of motorists’ perceptions of roadway performance. Having accurate information about roadway performance significantly improves motorists’ perceptions of a trip because such information allows motorists to make decisions that give them more control over their trip. Reli- ability has not been widely used to describe performance, but increasingly agencies are recognizing its value in assessing their own performance and in communicating performance to the public. Reliability and key terms related to reliability are defined below. Definition of Reliability Travel time reliability is a relatively new concept. Although various definitions of reliability have been proposed in the literature, no single definition has been universally accepted among traffic operations researchers and practitioners. Project L07 adopted the working definition for reliability that was developed by the research team for SHRP 2 Project L03, Analytical Procedures for Determining the Impacts of Reliability Mitigation Strategies (1): Reliability: The level of consistency in travel conditions over time, measured by describing the distribution of travel times that occur over a substantial period of time. This definition of reliability has two key parts: • Consistency in travel conditions, which refers to consis- tency in travel times and is mathematically represented by a statistical distribution of travel times. • Substantial period of time, which for convenience and prac- ticality has been defined in Projects L03 and L07 as 1 year. A period of 1 year also ensures a substantial enough data set on which to draw conclusions about how a facility generally operates. The measurement and prediction of reliability are mathe- matically rigorous. Therefore, several terms and concepts are presented here to set the foundation for analyzing travel time reliability later in this report. Time-Slice Because the reliability of a roadway may change throughout the day with changing traffic patterns and changing prob- ability of nonrecurrent congestion events, it is evaluated for specific time-slices. A time-slice is a single- or multi hour por- tion of a 24-h day, considered over an entire year (excluding weekends and holidays). For example, “the hour from 6:00 to 7:00 a.m. for every nonholiday weekday between Janu- ary 1 and December 31 of this year” is a single-hour time- slice. Single-hour time-slices are the simplest to work with because they are consistent with the way in which highway traffic volume data are typically collected and analyzed. One way to think of a single-hour time-slice is as an hour-year. Multihour time-slices defined and evaluated by Project L03 included the following: • Peak period: a continuous time period of at least 75 min during which the space–mean speed is less than 45 mph • Midday: 11:00 a.m. to 2:00 p.m. • Weekday: all 24 h aggregated In this research, only a single-hour time-slice was used for evaluation. Travel Time Index Although expected and actual travel times for a given highway segment or trip are intuitive measures for most drivers (“it should take me 15 minutes to travel from X to Y, but it actu- ally took me 17 minutes”), they are not necessarily convenient universal measures because analysis segments vary in length. Longer segments naturally require longer times to traverse, and comparison of travel times among segments of varying lengths would not be very meaningful. Thus, a numerical travel time measure exhibiting consistency across facilities of varying lengths is desirable. In reliability research, the travel time index (TTI) has emerged as such a measure. TTI is defined as the ratio of the actual time spent traversing a given distance to the free-flow travel time for that same distance. TTI can be measured at the scale of individual vehicles. For example, if the free-flow speed of a 2-mi freeway segment

6is 60 mph (meaning a vehicle could traverse the segment in 2.0 min), and a vehicle traverses the segment in 2.4 min, then the TTI for that vehicle is the ratio of 2.4 to 2.0 min (i.e., TTI = 1.2; note that as a ratio of two quantities mea- sured in consistent units of time, TTI is a unitless index). For reliability analysis, however, it is useful to aggregate TTI to larger scales, rather than at the scale of individual vehicles, such as all vehicles traversing a segment during a time-slice. At least two other measures, travel speed and travel rate, could be considered as fundamental measures of reliabil- ity, as each “normalizes” for both travel time and segment length: • Travel speed is expressed in the familiar units of miles per hour (the ratio of time to distance). • Travel rate is essentially the inverse of travel speed, expressed as a ratio of distance to time (e.g., seconds per mile). However, as a unitless measure, TTI is a preferable stan- dard because it can compare across different facilities regard- less of the speed for which they are designed. For example, an average travel speed of 55 mph (travel rate of 65 s/mi) would be quite acceptable on a facility with a free-flow speed of 55 mph (65 s/mi), but it would be less acceptable on a facil- ity with a free-flow speed of 70 mph (51 s/mi). In each case, the analyst would need two numbers to judge the reliability of the facility: the actual and free-flow speeds (or travel rates). In contrast, a single TTI value (a reliable 1.0 in the first case and a less reliable 1.27 in the second) would be sufficient to make this judgment. Because TTI is defined relative to the free-flow speed of the facility, motorists traveling faster than the free-flow speed have a TTI value less than zero. For the purposes of this research, the 90th percentile speed (corresponding to the 10th per- centile TTI) was used as a surrogate for free-flow speed, and any TTI values less than 1.0 were set equal to 1.0. Scope and Scale of Reliability Travel time and reliability can generally be considered from two perspectives: • Facility based. At the smallest scale, travel time can be con- sidered over a short, uninterrupted, homogenous highway segment for all vehicles that travel the segment over a time- slice. Such facility-based measures could be extended to a highway corridor and ultimately to an entire metropolitan highway system. • Trip based. As experienced by the individual traveler, trip-based travel times are ultimately what truly matters. For example, an individual commute typically traverses numerous facility types and segment lengths, and the reli- ability of each contributes to the reliability of the entire commute. As the most microscopic measure, segment-based travel times can be aggregated to derive any other scale. For exam- ple, as described above, an individual trip is composed of a series of segments. And certainly, smaller segments (or at most, corridors) are the scale at which design decisions and investments are made. Reliability statistics can be disaggregated by travel mode (automobile, truck, transit), travel purpose (freight movement, commute to work, business travel, personal errands, leisure travel), or by both mode and purpose. Such categorizations are especially useful for economic evaluations in which reli- ability may be valued differently for different trip purposes. Fundamental Diagram of Reliability Reliability is described by a distribution of travel times. Graph A in Figure 1.1 illustrates a typical travel time proba- bility distribution function (TT-PDF) for travel times on a freeway segment. Such distributions can have many shapes and are not always unimodal (single-peaked). Graph B shows the same distribution presented as a travel time cumulative distribution function (TT-CDF). The result- ing S-curve shape, with a standardized vertical axis, allows easy visual extraction of cumulative percentiles, including the median (50th percentile). By incorporating the concept of TTI, Graph C creates a unitless horizontal axis. The resulting curve is normalized along both dimensions and can serve as the fundamental diagram of reliability, referred to throughout this report as a cumulative TTI curve. The cumulative TTI curve is a cumu- lative distribution function of the TTI for a given time-slice (TTI-CDF). This curve has a series of properties that are use- ful indicators of reliability. Perhaps the most fundamental is that the closer the curve’s shape is to a vertical line at TTI = 1.0 (the minimum x-value), the more reliable is the facility it describes. This reliability indicates that there is little differ- ence in the travel times between trips of the shortest dura- tion and trips of the longest duration, and that the travel time index for even longer trips is close to 1.0. Graph D illustrates this principle. Evaluating Reliability: Indicators Although the travel time distribution serves as a defining dia- gram for reliability, simpler quantitative measures are usually the backbone of analysis. Figure 1.2 shows a sample TTI-CDF

7 and extends the discussion to other measures that have been derived from the travel time distribution. Mean-Based Measures Certain measures, such as mean TTI and the lateness index, relate to the mean of the travel time distribution. Mean Travel TiMe index The mean of the TTI distribution (TTImean) can hint at a facili- ty’s reliability. A facility with a TTImean of 1.1 would probably be considered “reliable,” and a facility with a TTImean of 2.0 would probably be considered “unreliable.” Strictly speaking, the term undesirable is more appropriate than unreliable when refer- ring to a “bad” TTImean, because the mean generally conveys no information regarding the shape (variability) of the distri- bution. However, research has shown that reliability decreases with increasing congestion, to the extent that at least one report (1) has concluded that “reliability is a feature or attribute of congestion.” One could imagine a distribution such as the one in Figure 1.3, in which travel times are “reliably” clustered around an undesirable TTI (in this case, 2.0). However, such distributions are not common in reality, because when a facil- ity nears its capacity, delay values are very volatile, and so the of 1-year hourly time-slices from an actual highway segment; analysts desire numerical measures to distinguish among these curves. To be useful, such measures must describe an aspect of the travel time distribution (most often, its shape). The following discussion begins with the two fundamental descriptors of any statistical distribution (mean and variance) Figure 1.1. Cumulative TTI curve. Figure 1.2. Sample 24-h cumulative TTI curves. Figure 1.3. Reliable but undesirable distribution: a theoretical construct.

8SeMivariance Although calculating the variance about the mean (as shown by Equation 1.2) is fairly common statistical practice, describ- ing how travel times differ from the mean is potentially not as useful as describing how they differ from the ideal. Therefore, the concept of the semivariance (sr) has been used. Statisti- cally, semivariance can be described as the second moment of the travel time distribution about the minimum, as shown by Equation 1.3: ∫∫ ( ) ( ) ( )σ = − = − == 1 TTI TTI 1 1.32 2 0 100% 0n r di dir i i ii n In this case, r, the reference value from which deviation is calculated, is set to 1.0, the minimum possible (or ideal) TTI. The value of n is set to 100%, echoing the upper limits of the cumulative TTI distribution. Figure 1.5 illustrates the difference in the values used to build the variance (TTI - TTImean)2 and semivariance (TTI - 1)2. The variance curve is constructed from the cumulative TTI curve by calculating (TTI - c)2 for each percentile p, the dif- ference between TTIp and the vertical line y = TTImean, and then squaring that difference. The semivariance curve is constructed the same way, except using the vertical line y = 1. Thus, s and sr can be computed by taking the area to the left of the appropriate cumulative curve generally leans forward like the outer curves in Figure 1.2. Thus, more “unreliable” curves will have higher TTImean values. laTeneSS index A slight enhancement to TTImean acknowledges the distinction between travel time and delay. The difference between a user’s actual travel time and desired free-flow travel time across a segment (or for an entire trip) can be said to be equivalent to that user’s delay. Since a TTI of 1.0 equates to free-flow con- ditions, delay can be thought of as proportional to TTI - 1. This quantity, although trivial to calculate once a TTI has been calculated, has a physical analog, as illustrated in Figure 1.4. Because the cumulative TTI curve is unitless, the shaded area in Figure 1.4 is equal to TTImean - 1. A suggested name for this quantity is the lateness index (LI). If the LI is multiplied by the total number of vehicles in the time-slice (V) and the free-flow travel time of the segment (TTFF), then the result is the total delay experienced by all vehicles, as shown by Equation 1.1: ( )= × ×Total delay LI TT 1.1FFV Variance-Based Measures Certain measures relate to the variance of the travel time dis- tribution, as described below. variance and STandard deviaTion A distribution’s variance and standard deviation are indicators of how far the distribution spreads out. As such, these measures are more powerful descriptors of reliability than the mean, because reliability is primarily concerned with variability. Variance about the mean (s) is calculated as shown by Equa- tion 1.2 (assuming a continuous distribution), with TTIi rep- resenting the ith percentile TTI and n representing 100% (the maximum y-value on the cumulative TTI curve). The standard deviation is given by σ . 1 TTI TTI 1.2mean 2 0n dii i n∫ ( ) ( )σ = − = Shaded area = Lateness Index TTI Figure 1.4. Lateness index. Figure 1.5. (Top) variance and (bottom) semivariance buildups.

9 PercenTage of TriPS on TiMe Percentage of trips on time (labeled “PTOT” in Figure 1.6) essentially works in the reverse direction of the PTI and MI, in effect specifying a target TTI and then extracting the corre- sponding percentile from the cumulative TTI curve. Percentage of trips on time represents the percentage of trips completed within a certain speed or time range, such as the percentage of trips that arrived on time with a speed of 45 mph or greater or the percentage of trips that arrived on time with a TTI of 1.5 or less. Overall, no single point (or small region) in the travel time distribution is a comprehensive descriptor of reliability. For example, Figure 1.7 illustrates two curves with identical PTI values (TTI95%) but very different behavior in the upper tails. Nevertheless, values in the upper percentiles can certainly convey a sense of how much the cumulative distribution leans forward. Curvature Indices Several reliability measures are built on ratios that describe aspects of the curvature of the cumulative curve. Figure 1.8a illustrates these measures in relation to the cumulative curve. Buffer index The buffer index (BI) describes how much the cumulative TTI curve “leans forward” beyond the mean or median. The term buffer indicates the extra time that travelers should add to their average travel times to ensure on-time arrival (buffer time equals planning time minus average time). Like the PTI, BI hinges on the 95th percentile TTI, but it uses a ratio involving either the mean (Equation 1.4) or the median (Equation 1.5): ( ) ( ) ( )=BI TTI – TTI TTI 1.4mean 95% mean mean ( ) ( ) ( )=BI TTI – TTI TTI 1.550% 95% 50% 50% Recent research has raised doubts about the use of the BI as a primary reliability metric for tracking trends in reliability curves (shaded in the figure). With curves that lean forward, sr will always be much larger than s. A “reliable but undesirable” TTI distribution like the one shown in Figure 1.3 would have a very low s (indicating low variability with respect to the mean), but a higher sr (indicating high variability from the ideal). LI (Figure 1.4) and the semivariance provide roughly the same information about the cumulative TTI curve: the former is a summation of TTI - 1, and the latter is a summation of (TTI - 1)2. The semivariance places disproportionate emphasis on larger deviations, and therefore may better gauge reliability. Single-Point and Regime Indices Several measures used in reliability analysis relate to points or regions on the cumulative TTI curve. Figure 1.6 illustrates these measures in the context of the cumulative curve. Gener- ally, such measures have been developed for values well above the median TTI, because the upper portion of the cumulative curve yields the most information about reliability. Planning TiMe index The planning time index (PTI) is equal to the 95th percen- tile TTI. Its name derives from the idea that it represents the total time travelers should allow to ensure on-time arrival 95% of the time. MiSery index The misery index (MI) represents the average of the highest 5% of travel times (“the worst day of the month”). On the cumulative TTI curve, it is equal to the average x-coordinate in the circled area in Figure 1.6. MI may be especially useful in characterizing rural reliability, for which even a relatively small number of very delayed trips can be a source of major frustration for motorists. One approximation for the MI is TTI97.5%; this approximation assumes roughly linear behavior of the cumulative TTI curve above the 95th percentile. Figure 1.6. Single-point and regime indices. Figure 1.7. Identical PTIs.

10 that very different distributions can have identical BIs and skew statistics, respectively. Summary of Reliability Indicators As discussed above, it is not merely unreliability, but unde- sirable unreliability that must be quantified. One can analo- gize to capacity-based analyses, in which an index (level of service) gets worse as an undesirable quantity (delay) gets larger. Similarly, it is logical for a reliability-based index to increase as undesirable variability increases. Measures of area around the cumulative TTI distribution such as LI and semivariance best exhibit this behavior. Curvature indices (BI, skew statistics) do not do so reliably. Point measures cannot always tell the full story. The cumulative curve itself is the best metric of reliability. By studying its shape in a given situation, the analyst can determine which supple- mental measures are appropriate. No universal standard has yet been developed for accept- able values of any reliability index. When standards are devel- oped, they will likely vary for different physical environments (e.g., large metropolitan area, smaller metropolitan area, rural area) and differing facility types (e.g., freeway or arterial). Comparing Reliability The cumulative travel time distribution and its properties can be used to compare reliability conditions on a facility before and after the implementation of a proposed improvement. For example, the cumulative TTI graph in Figure 1.9 shows data from an actual freeway segment before and after a reli- ability improvement (a ramp-metering implementation). The shaded area is equal to the differences in the LI and can be termed the “lateness reduction,” which, when multi- plied by the segment’s volume (V) and free-flow travel time (TTFF), translates to an overall delay reduction. Thus, the area between the TTI curves before and after improvement is pro- portional to the overall delay reduction. due to its erratic and unstable nature (1). Treatments that tend to uniformly decrease travel times (rather than affecting only the extremes) can result in counterintuitive BIs (falsely indi- cating reliability degradations when conditions are actually improving). However, BI remains useful as a secondary metric. Skew STaTiSTic The skew statistic (SS) is a measure of symmetry in the travel time distribution, calculated as a ratio of 40th percentile TTI ranges on either side of the median TTI, as shown by Equation 1.6: ( ) ( ) ( )=SS TTI – TTI TTI – TTI 1.690% 50% 50% 10% Measures such as the BI and skew statistic, although pro- viding information about the shape of the travel time dis- tribution, do not provide sufficient information about the desirability of the distribution. Figure 1.8, b and c, illustrates Figure 1.8. Curvature indices. (a) (b) (c) Figure 1.9. Delay reduction.

11 from the National Climactic Data Center; a literature review; and interviews with state highway agency staff. Chapter 3 also lists all the nonrecurrent congestion design treatments con- sidered in this research. Chapter 4 explains in mathematical terms how the pre- dicted TTI distributions for a section of freeway during a specific time of day can be used to calculate operational ben- efits of design treatments in terms of reduced total delay and improved reliability. The mechanics of mapping the effects a given treatment has on operations into the reliability model variables (demand-to-capacity ratio, lane hours lost, rainfall, and snowfall) are presented in detail. Chapter 5 presents the methodology for estimating the direct and indirect safety benefits of design treatments for nonrecurrent congestion, so that they can be accounted for in the benefit–cost analysis. The direct benefits include the reduction in crash frequency or severity expected as a result of changes to lane width, shoulder width, or other geomet- ric features related to base capacity as indicated by Highway Capacity Manual procedures; and other roadway and road- side design features that may affect driver behavior, likeli- hood of a crash, or severity of a crash. The research team found a relationship between crash frequency and level of service, described in Chapter 4. This relationship predicts the indirect safety benefits expected as a result of an improve- ment in level of service. Chapter 6 describes the methodology for placing the oper- ational and safety benefits estimated in Chapter 4 and Chap- ter 5 in economic terms to compute the net present benefit of a design treatment. In addition, a procedure is described for determining a treatment’s net present cost and computing the benefit–cost ratio. The research team developed a “reasonableness test” to evaluate the outputs provided by the procedures described in this report and implemented in the Analysis Tool described in Chapter 2. The test was used to initiate an iterative quality control process of implementing changes based on test results and then retesting. This effort is described in Chapter 7. Major findings from all phases of the research are sum- marized in Chapter 8. These conclusions came not only from the literature and meetings with highway agencies, but also from the development of the various models and procedures presented in this report. They include insights gained by the research team through careful study of previously developed reliability measures and visual presentation of those measures. Chapter 8 concludes with recommendations for how the results of this research might be implemented by highway agencies. The delay reduction illustrated in Figure 1.9 can further be translated into economic terms by using the monetary value of time. Research has shown that motorists directly value reduc- tions in travel time variability, leading to the idea that a simi- lar graph could be constructed for some measure of variance and translated into economic terms by using a monetary value of reliability. Predicting Reliability Essential to reliability’s application as a measure of highway system performance is the ability to forecast the effect of an improvement strategy (or even a “do-nothing” strategy) on a facility’s near- and long-term reliability. Recent research has broken new ground in correlating reliability measures to pre- dictable attributes or events, proposing a series of equations for predicting reliability based on three highway and environ- ment attributes (1): • A general measure of highway congestion (ratio of demand to capacity) • A measure of temporal–spatial impacts of incidents and work zones (lane hours lost) • A measure of precipitation amount (rain and snow) As explained in Chapter 4, these predictive formulas are the foundation for most of the operational analysis work of Project L07. Organization of the report The remainder of this report is organized as follows. Chapter 2 describes the original research objective and scope and how they grew over the life of the project to address additional research needs. It explains the evolution of the research approach based on the reliability models developed by another SHRP 2 project that preceded this effort. Chapter 2 also briefly summarizes the three research products in addi- tion to this final report: the Design Guide, Analysis Tool, and Dissemination Plan. Chapter 3 describes the various sources of data used to develop the methods, models, and default values found in the products of this research. Data sources include the reli- ability models developed by SHRP 2 Project L03; the traf- fic operational databases available from Seattle, Washington, and Minneapolis–St. Paul, Minnesota; crash data from the same cities; weather data for stations around the United States