Public Acceptability of Pavement Roughness on ...

Acceptability of Pavement Roughness on Urban Highways by the Driving Public

Kevan Shafizadeh Post-Doctoral Research Engineer Civil & Environmental Engineering University of California, Davis, CA 95616 Phone: (530) 752-8460; Fax: (530) 752-8924 E-mail: [email protected]

Fred Mannering Professor Civil & Environmental Engineering Purdue University, West Lafayette, IN 47907 Phone: (765) 494-2159; Fax: (765) 494-0395 E-mail: [email protected]

Submitted to the 2003 Transportation Research Board Annual Meeting August 1, 2002 Word Count: 7,312

TRB 2003 Annual Meeting CD-ROM

Paper revised from original submittal.

Shafizadeh & Mannering

1

ABSTRACT This paper explores the driving public’s attitude toward acceptable levels of road roughness using empirical data collected on urban highways. Individual driver acceptability levels are matched with International Roughness Index (IRI) levels to examine the existence of potential user acceptability thresholds. In particular, the observed trends are compared with the federal IRI guideline of 170 in/mi (2.68 m/km) for “acceptable ride quality,” recommended by the Federal Highway Administration (FHWA) in its 1998 National Strategic Plan for the National Highway System (NHS). This paper provides empirical support for the current recommended guideline. The research seems to provide the empirical support for the federal IRI guidelines that are already in existence. This study also found that IRI levels provided a very good indication driver acceptability, which agrees with past research based on antiquated present serviceability ratings (PSR). INTRODUCTION AND STUDY MOTIVATION In 1997, the Washington State legislature required that a series of internal audits be performed on its Department of Transportation. The audits, which were conducted for the Joint Legislative Audit and Review Committee (JLARC), led to a 1997 telephone survey of 508 Washington State residents that indicated that “poor road surface” ranked second only to “congestion/inadequate capacity issues” as the state’s biggest transportation problem (1). The telephone survey also revealed that the general public dissatisfaction due to pavement surface condition was perceived to be worse in Washington State than other states (p. 2-18, 2). The JLARC audit went on to conclude that while WSDOT has a pavement management system (PMS) that “has the analytic capabilities to help prioritize pavement projects, there are issues with how PMS applications and results are communicated” and that there is “a need for greater recognition of customer perceptions of pavement condition” (p. 2-20, 2). Finally, the JLARC report made the following recommendation: “The Washington State Department of Transportation should consider including pavement roughness… in its candidate pavement project thresholds” (p. 2-20, 2). Despite the report by Cambridge Systematics and the subsequent JLARC recommendation, WSDOT does, in fact, consider roughness as part of its pavement preservation assessment process. Currently, “WSDOT attempts to program rehabilitation for pavement segments when they are projected to reach an IRI [International Roughness Index] of 220 in/mi [3.5 m/km]” (p. 8, 3). International Roughness Index (IRI) is one extensively used quantifiable measure of roughness in Washington State and nationwide. This index is used by the Federal Highway Administration (FHWA) to assess changes in the condition of the nation’s highways and to forecast highway investment needs. Despite the widespread use of this index, work on linking the IRI with the motoring public’s perception of acceptable levels roughness has been limited. Little empirical data supports the establishment of a roughness threshold based on drivers’ perceptions of roughness. This research helps to fill this gap. While this study focused on issues in Washington State, it is important to note that any pavement-related findings could have national implications. In 2000, the Federal Highway Administration followed-up on a 1995 study by the National Partnership for Highway Quality (NPHQ), formerly known as the National Quality Initiative (NQI), to gauge the public’s satisfaction with the nation’s highway system. The results indicated that satisfaction with the pavement conditions of the national highway system remains low (4,5). Part of the improvement from 1995 was attributed to “the percentage of miles on the National Highway System (NHS) with an acceptable ride quality (based on an International Roughness Index [IRI] value of less than 170 in/mi) increased from 90.0% to 93.0% from 1995 to 1999” (p. 6, 4). When asked which highway characteristic should receive the most attention and resources for improvement, respondents chose pavement conditions (21%), only behind improvements to traffic flow (28%) and safety (26%). Some state and regional departments of transportation (DOTs) have also been discussing the merit of providing incentives to contractors who can construct a road with a high level of smoothness. One example of this “smoothness” incentive is demonstrated in Maricopa County, Arizona where “contractors, under this incentive program, can earn as much as an additional 10% of the total project paving costs in incentive bonuses by exceeding the preset standard for smoothness” (p. 17, 6): • • • •

IRI less than 51 in/mi (0.80 m/km) yields a 10% incentive. IRI of 51 – 60 in/mi (0.80 – 0.95 m/km) yields a 5% incentive, IRI of 61 – 80 in/mi (0.96 – 1.26 m/km) yields no incentive. IRI of 81 – 110 in/mi(1.27 – 1.58 m/km) yields a 5% penalty.




• • •

2

IRI of 101 – 110 in/mi (1.59 – 1.74 m/km) yields a 10% penalty. IRI of 111 – 120 in/mi (1.75 – 1.89 m/km) yields a 25% penalty. IRI of 120 in/mi (1.89 m/km) or more requires replacement.

While this incentive program may provide beneficial quality control, which, in turn, may have an impact on the structural integrity of the road, there is little empirical data to support the different incentive levels. This research helps to fill this gap. This paper has four primary sections. The first section is a brief literature review. The second section discusses aspects of the data collection of road roughness measurements along with the public’s general perception of pavement acceptability. The comparison and analysis of the public’s perceptions with the roughness guidelines recommended by the federal government along with other research is done in the penultimate section. Finally, conclusions and recommendations are made. LITERATURE REVIEW In 1982, the World Bank commissioned an experiment in Brazil to establish a roughness measurement standard, and the result was the International Roughness Index (IRI) (7). IRI is now considered the international standard for comparing roughness measurements. Since 1990, the Federal Highway Administration has required states to report road roughness on the IRI scale, which was later incorporated into the Highway Performance Monitoring System (HPMS). (Prior to 1990, all pavement conditions were evaluated for the federal government using present serviceability rating PSR values (8), which where developed in the 1960 American Association of State Highway Officials (AASHO) Road Test (9).) Typical IRI values range from 0 to 5 m/km (317 in/mi), with higher values indicating rougher pavement surface. TABLE 1 contains a qualitative pavement condition term and the approximate corresponding quantitative PSR or IRI values. TABLE 1 also indicates the FHWA descriptive term for pavement condition, “acceptable ride quality,” introduced in the 1998 FHWA National Strategic Plan. This plan stated that by 2008, 93% of the National Highway System (NHS) mileage should meet pavement recommended guidelines for “acceptable ride quality.” In order to be rated “acceptable,” pavement performance must have an IRI value of less than or equal to 170 in/mi (2.68 m/km). The term “less than acceptable” is currently used to describe lane miles that do not meet the “acceptable” threshold on the federal interstate system. While the threshold of 170 in/mi (2.68 m/km) is clear, it is not clear why this threshold value was set – or if there are data to support it. As will be shown, this report fills this gap in the literature by providing the empirical data to identify driver acceptability (and unacceptability) thresholds. Traditionally, a PSR of 2.0 to 3.0 has been used to define “failure” in the AASHTO pavement structural design method (10). (Also, see 11.) One can only guess that the mid-range value of 2.5 (corresponding to an IRI of 170 in/mi, 2.68 m/km) was used to set the IRI guideline. TABLE 1 Federal Pavement Roughness Thresholds for Interstate Facilities Condition Term Very Good Good Fair Mediocre Poor

PSR Rating

IRI

NHS Ride Quality

< 60 in/mi (< 0.95 m/km) 60 – 94 in/mi 3.5 - 3.9 (0.95 – 1.48 m/km) 95 – 119 in/mi 3.1 - 3.4 (1.50 – 1.88 m/km) 120 – 170 in/mi 2.6 - 3.0 (1.89 – 2.68 m/km) > 170 in/mi ≤ 2.5 (> 2.68 m/km) Source: Federal Highway Administration (8).

≥ 4.0

Acceptable: 0 – 170

Less than Acceptable: > 170

While there is a small list of researchers who many studies attempted to compare results of the AASHO Road Test using slightly modified experimental procedures and/or different or updated road roughness measuring devices in various states (e.g., 12, 13,14,15,16,17). The goal of most of these studies, however, was to update the




3

AASHO Road test to create a unique set of linear regression equations for predicting panel ratings (PSRs) in a different location or based on different roughness measurement systems. (For a more thorough review of this literature, please refer to Shafizadeh et al. (18).) In 1985, Nair et al. produced interesting findings in terms of user acceptability. The only problem is that all results were in terms of PSR, “Sections with high PSRs (greater than 3.5) were rated to be acceptable unanimously by the panel” (p. 81, 19). The study also reveled that “Up to 88% of the variation in PSR can be explained by the roughness variables” (p. 81, 19). This study was among the first to establish that a quantitative roughness measurement alone accounted for the majority of the variability in the subjective evaluations of roughness. Most recently, a series of reports published in 2000, titled “Public Perceptions of the Midwest’s Pavements,” was considered by its authors to be “the largest survey of public perceptions of satisfaction and improvement policies on rural, two-lane highways ever conducted in the 20th Century in the USA” and was a fiveyear, three-phased effort supported by pooled funds from Wisconsin, Iowa, and Minnesota by Kuemmel et al. (20,21,22). As one part of this project, participants were recruited by phone and asked to drive using their own vehicle over selected rural highway segments (“within 10 minutes drive time of a city of 500 population or more”). Participants were called back at a later time and asked general questions about their satisfaction with the segment. Participants received $10 for their time and any related expenses – if they completed the follow-up phone survey within approximately one week. In six months, over 450 highway segments were selected, and 2,300 surveys were completed in the three states. While the details on the data collection were limited, it appears that the data were collected under unrestricted conditions, insofar as no effort was made to control for conditions during which drivers evaluated the segment. No aspect of this phase of the data collection appears to have been controlled other than the selection of the test section. Using these data, researchers tried to determine the roughness and distress levels that are tolerated by the public. TABLE 2 shows the cumulative percentage of participants who “agreed” or “strongly agreed” with the statement: “I am satisfied by with the pavement on this section of highway” (p. 12, 20). The authors of this study wanted to find the IRI value at which 70% of drivers, collectively, would be satisfied with a given section of highway. This study is one of the other few that identified IRI acceptability levels. TABLE 2 IRI Acceptability Levels in Kuemmel et al. Cumulative Percent of Participants 10% 20% 30% 40% 50% 60% 70%

PCC (N = 240) 203 in/mi (3.2 m/km) 158 in/mi (2.8 m/km) 158 in/mi (2.5 m/km) 146 in/mi (2.3 m/km) 120 in/mi (1.9 m/km) 114 in/mi (1.8 m/km) 70 in/mi (1.1 m/km)

Pavement Type Asphalt Composite (N = 171) (N = 203) 158 in/mi 171 in/mi (2.5 m/km) (2.7 m/km) 120 in/mi 120 in/mi (1.9 m/km) (1.9 m/km) 108 in/mi 89 in/mi (1.7 m/km) (1.4 m/km) 76 in/mi 76 in/mi (1.2 m/km) (1.2 m/km) 63 in/mi 70 in/mi (1.0 m/km) (1.1 m/km) 51 in/mi 57 in/mi (0.8 m/km) (0.9 m/km) 44 in/mi – (0.7 m/km) Source: Kuemmel et al. (20).

Total (N = 614) 184 in/mi (2.9 m/km) 158 in/mi (2.5 m/km) 127 in/mi (2.0 m/km) 108 in/mi (1.7 m/km) 76 in/mi (1.2 m/km) 63 in/mi (1.0 m/km) –

EMPIRICAL SETTING Data in this study originated from two primary sources: 1) an in-vehicle study, and 2) the Washington State Department of Transportation (WSDOT) and its Pavement Management System (WSPMS). There were 56 participants who provided in-vehicle data. Each individual evaluated 40 highway segments and produced nearly 2,240 unique “observations.” (Actually, there were 2,180 valid “observations” because of missing or incomplete




4

data for some participants.) Participants for the in-vehicle study were selected after a series of screening processes. First, a simple mailout/mailback survey was sent to over 2,500 registered vehicle owners in Seattle area whose license plate was collected at random from vehicles entering and exiting SR 520 near the University of Washington. Over 2,800 license plates were gathered during various hours of the day and random days of the week in the fall of 2000. Addresses corresponding to the collected license plates were obtained through the Department of Licensing, and a two-page survey with a postage-paid business reply was sent to the registered owner of the vehicle. Respondents who were willing to participate in an in-vehicle study were also asked to provide contact information (name, address, and phone number) so they could be contacted at a later time. The response rate of the preliminary survey was over 21.4%, which is comparable to other mailout/mailback surveys with response rates of 33% (23) and 15% (24), among others. Drivers, who agreed to participate in a driving experiment, were contacted in the late spring of 2001 with a simple one-page letter and instructed to telephone the University of Washington if they were still interested in participating. Notable descriptive statistics characterizing the pool of driving subjects are detailed in a report by Shafizadeh et al. (18), including: • • •

23 (41%)of whom were women, and 33 (59%) of whom were men, 17 (30%)of whom were age 21 to 35, 18 (32%)of whom were age 36 to 50, and 21 (38%) of whom were age 51 or over, 13 (23%) of whom had annual household incomes of less than $45,000, 16 (29%) of whom had incomes between $45,000 and $75,000, and 22 (39%) of whom had incomes of $75,000 and up

In-Vehicle Data Collection At the randomized starting location, each of the 56 participants were simply told that the Department of Transportation wanted their opinions on road roughness for highways in the area. They were shown a small area on a highway map and told that they would be driving over 40 predetermined highway test “segments.” As they drove over the test segments, they were asked, among other things, “Is this level of roughness acceptable to you?” Participants were not provided any more instructions. They were told that during the driving experiment they would be notified when each test segment started and ended and could provide their response at any point during the test segment. They were told that their response to the should be a simple “yes” or “no” answer. They were not given any other explicit instructions – except where to drive. If they asked how “acceptable” were defined, they were told to use their own judgment and to rank the roughness of the road in comparison to other roads in the state. In other words, participants were not given preset evaluation criteria and were forced to develop their own acceptability criteria. They were also instructed to drive in a way that was “consistent with their regular driving behavior.” They were instructed to drive at a speed that they felt most comfortable driving. The purpose of this stipulation was to capture as much of the participant’s normal driving behavior as possible with the hopes that the “research setting” would not affect their responses. Study Area and Route Selection A 25-mile circular loop on I-5, I-90, I-405 and SR 520 around the Seattle-Bellevue area was selected. This loop was chosen, because it was easily accessible by many participants because it was close to the Seattle and Bellevue areas and sampled four major facilities in the with unique attributes. To maximize the amount of data collected, the loop was driven twice – once in a clockwise direction and once in a counterclockwise direction. There were 20 test segments on each loop (40 total segments for both loops). A summary of the route selection criteria along with detailed in IRI measurements on the four sampled facilities is available in Shafizadeh et al. (18). The decision to repeat the loop in the opposite direction was deliberately done for both logistical and experimental reasons. It allowed drivers (and the researcher) the ability to return to the starting location halfway through the experiment if any unforeseen problems prevented the experiment from being completed in less than two hours (e.g., a major traffic accident). A loop was also selected because it allowed the participant to drive on each facility twice (albeit in opposite directions) to increase driver comfort with the general route. Randomization of Starting Locations As part of the experimental design, the starting location was randomized for each participant to minimize the effects of driver fatigue. One start location was University Village, near the junction of I-5 and SR 520, and the second start location was the Eastgate Park and Ride (P&R), near the junction of I-90 and I-405 in Bellevue. These




5

start locations were selected because they essentially allowed the experiment to begin on different facilities, depending on the start direction. For example, when starting at the Eastgate P&R and driving in the clockwise direction, the participant experienced (in order) I-90, I-5, SR 520, and I-405; if instructed to drive in the counterclockwise direction from the Eastgate P&R, the participant experienced (in order) I-405, SR 520, I-5, and I90. Selection of Test Vehicles Given that vehicle type could have an impact on the perception of road roughness, drivers were also assigned to different vehicle types. Vehicle type was thought to be important, not only because of the characteristics of the vehicle, but also because of the way the vehicle interacted with the roadway. Four types of vehicles were used in the study: • • • •

2001 Kia Optima sedan (18 participants – 32%) 2001 Jeep Grand Cherokee (16 participants – 29%) 2001 Ford Ranger pickup (12 participants – 21%), and 2001 Ford Windstar minivan (10 participants – 18%).

WSDOT IRI Data The last data source was the Washington State Department of Transportation (WSDOT) and its pavement management system. The Washington State Pavement Management System (WSPMS) provided the measured IRI on each segment. This database also contained information that was instrumental in the segment selection process by providing valuable geometric information (e.g. terrain, shoulder widths, the number of lanes, and roadway width), all of which were arbitrarily required to be homogeneous within each segment as part of the selection criteria. A WSDOT data collection crew measured the roughness profile and updated IRI values for the test segments to ensure accuracy of the data at the time of the study. Potential Acceptable/Unacceptable IRI Thresholds The goal of this research was to identify potential roughness thresholds at which the majority of drivers found the roughness levels to be distinctly acceptable or unacceptable. These threshold values are important, because an unacceptable threshold could be an indicator that the section is in need of rehabilitation. If the proportion of acceptable and unacceptable ratings for each segment is plotted against the IRI of the segment, FIGURE 1 is obtained. (There is an acceptable portion and an unacceptable portion for each segment, noted by its IRI on the x-axis.) For segments with a low IRI measurement (e.g., less than 100 in/mi or 1.6 m/km), nearly all ratings were “acceptable” and only a fraction was unacceptable. For test segments that exceeded the federal IRI guideline, there was little agreement about was “acceptable” and what was “unacceptable.” This figure is also elucidating, because it indicates that a pavement section at the federal IRI guideline of 170 in/mi (2.7 m/km) would be expected to have around a 65% acceptability rating (a 35% unacceptable rating).




6

Measured IRI (m/km) 0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

4.5

100% Acceptable Ratings

90% Percent of Observations Per Segment

Unacceptable Ratings

80% 70% Note: For test segments with low IRI measurements, nearly all ratings were "acceptable" and only a few were "unacceptable."

60% 50% 40% 30%

FHWA Guideline

20%

Acceptable

10%

Unacceptable

Note: For test segments that exceeded the federal IRI guideline, there was little agreement about what was "acceptable" and what was "unacceptable."

0% 0

50

100

150

200

250

300

Measured IRI (in/mi)

FIGURE 1 Proportion of Acceptable and Unacceptable Ratings for Each Test Segment (Sorted by IRI). It was also helpful to sort all observations by IRI and group them into acceptable and unacceptable groups, shown in TABLE 3. The fraction of acceptability ratings drops most dramatically when IRI increases from the range between 1.0 and 1.9, where 95% of the ratings were considered to be acceptable, and between 2.0 and 2.9, where only 68% of the ratings were considered to be acceptable. TABLE 3 Distribution of Acceptable/Unacceptable Observations, Grouped by IRI IRI Grouping Total Observations Acceptable [Cumulative] [Cumulative] [Cumulative] < 63 in/mi(< 1.0 m/km) 111 (5.1%) 110 (99.1%) [< 63 in/mi (< 1.0 m/km)] [111 (5.1%)] [110 (99.1%)] 63 – 126 in/mi (1.0 – 1.9 m/km) 1296 (59.4%) 1226 (94.6%) [< 127 in/mi (< 2.0 m/km)] [1,407 (64.5%)] [1,336 (95.0%)] 127 – 189 in/mi (2.0 – 2.9 m/km) 386 (17.7%) 263 (68.1%) [< 190 in/mi (< 3.0 m/km)] [1,793 (82.2%)] [1,599 (89.2%)] 190 – 252 in/mi (3.0 – 3.9 m/km) 331 (15.2%) 189 (57.1%) [< 253 in/mi (< 4.0 m/km)] [2,124 (97.4%)] [1,788 (84.2%)] ≥ 253 in/mi (≥ 4.0 m/km) 56 (2.6%) 21 (37.5%) [< 317 in/mi (< 5.0 m/km)] [2,180 (100%)] [1,809 (83.0%)] Note: The acceptable and unacceptable ratings for each IRI grouping total 100%.

Unacceptable [Cumulative] 1 (0.9%) [1 (0.9%)] 70 (5.4%) [71 (5.0%)] 123 (31.9%) [194 (10.8%)] 142 (42.9%) [336 (15.8%)] 35 (62.5%) [371 (17.0%)]

When the observations are sorted into cumulative IRI groupings, the differences are more subtle and illustrates that drivers, on a whole, seemed relatively pleased with the roads in the study area. Over the entire sample, only 17% of the observations were still deemed to be acceptable as shown in the last row. It is important to recognize that the sample of test segments was not uniformly distributed and that it consisted primarily (82%) of test segments with an IRI of less than 190 in/mi (3.0 m/km). The fact that the sample was more heavily weighted with smooth roads could also have contributed, in part, to the high number of acceptable evaluations.




7

Using this approach, it was also found that nearly half of all “acceptable” ratings were made on segments with an IRI of less than around 90 in/mi (1.42 m/km), and that almost 85% of all “acceptable” evaluations were made on test segments below the federal guideline with an IRI of 170 in/mi (2.7 m/km) or less. About 50% of unacceptable evaluations still occurred at the federal roughness guideline, which indicates that all unacceptable ratings are not made exclusively at high IRI levels and that other factors may influence the fewer number of unacceptability ratings. About 10% of all acceptable observations had a corresponding IRI above 190 in/mi (3.0 m/km). Almost as many unacceptable ratings occurred where the IRI was less than 95.0 in/mi (1.5 m/km). ROC Curves A robust way of interpreting these data can be borrowed from epidemiological research through the use of receiver operating characteristic (ROC) curves, which are typically used for the prediction of disease based on the presence of some indicator variable. In this case, this methodology can be employed to test for the prediction of roadway acceptability based on an IRI threshold. An ROC curve is a plot of the true positive rate (i.e., the rate of individuals who were predicted to find pavement “agreeable” based on an IRI less than the threshold value and actually found the segment to be acceptable) against the false positive rate (i.e., the rate of individuals who were predicted to find pavement “agreeable” based on the threshold value but actually found the segment to be unacceptable) for different thresholds. In this case, a true positive (TP) response is defined as an observation that had an IRI less than or equal to 170 in/mi and was found to be acceptable by the participant, as one would expect. There were 1,531 “true positive” ratings, which represented 70.2% of all responses. A true negative (TN) is defined as an observation that had an IRI greater than 170 in/mi (2.68 km) and was found to be unacceptable by the participant, as one would also expect. There were 220 “true negative” ratings, which represented 10.0% of all responses. A false positive (FP) test is defined as an observation that had an IRI less than or equal to 170 in/mi (2.68 m/km) but was found to be unacceptable by the participant. There were 151 false positive ratings, which represented 7.0% of all responses. A false negative (FN) test is defined as an observation that had an IRI greater than 170 in/mi but was found to be acceptable by the participant. There were 278 false negative ratings, which represented 12.8% of all responses. In short, with 170 in/mi as a threshold, approximately 80.2% (1,809 of 2,180) of all responses held “true” to their expectation. These response types (TP, TN, FP, and FN) are illustrated in FIGURE 2. The acceptability data were sorted by IRI to generate two histograms, along with their accompanying normal curves. (Because there a uniformly distributed number of segments at all IRI values, the data is more discrete than continuous and the result is plots that may not look very normally distributed.) Regardless, this analysis is useful illustrates the areas under the histograms that correspond to true positive, false positive, true negative, and false negative evaluations. FIGURE 2 is important because it helps to illustrate how the positive and negative evaluations would be altered if the IRI threshold were different. In other words, if the IRI threshold were increased, it would result in greater true negative evaluations, which is desirable, but it would also result in fewer true positive evaluations, which is undesirable. (It would also result in a smaller number of false positive evaluations, which is desirable but also result in a greater number of false negatives, which is undesirable.) In short, this figure helps to illustrate the fact that any adjustment to the IRI threshold would be a tradeoff.




8

IRI (m/km) 0.00

0.79

1.58

2.37

3.16

3.95

4.73

300

Federal Guideline 250

Acceptable

Frequency

200

Unacceptable

Acceptable Histogram Unacceptable Histogram

150

False Negative True Negative

100

True Positive 50

False Positive 0 0

50

100

150

200

250

300

IRI (in/mi)

FIGURE 2 Acceptability and Unacceptablility Histograms with Fitted Normal Curves. With this data, the sensitivity and specificity of the federal threshold can also be calculated to be 0.846 and 0.593, respectively. The sensitivity is defined as the power of this threshold to correctly identify acceptable roughness (positive) responses, and the specificity is defined as the power of the test to correctly identify unacceptable roughness (negative) responses – both of which can be expressed mathematically as: Sensitivity = TP / (TP + FN) Specificity = TN / (TN + FP) Calculations of sensitivity and specificity at various threshold values can also be used to generate an ROC curve for this data, shown in FIGURE 3. An ROC curve visually demonstrates this tradeoff by plotting a curve that shows all of the possible thresholds and how they would capture the desirable true positives (sensitivity) versus the undesirable false positives (1 – specificity). The ROC curve also provides some indication as to the discrimination, or accuracy, of the indicator variable as a predictor, that is, the ability of IRI to correctly classify both acceptable and unacceptable ratings. The closer the curve follows the left-hand border and the top border of the ROC space, the more accurate the test. Conversely, the closer the curve comes to the 45-degree diagonal of the ROC space, the less accurate the test. The area underneath the ROC curve also provides an indication to the accurate the test. An area of 1.0 represents a perfect test, while an area of 0.5 represents a worthless test. FIGURE 3 has an area of 0.822, indicating that IRI is a good test criterion for roughness acceptability.




9

True Positive Rate (Sensitivity)

1.00

.75

.50

.25

0.00 0.00

.25

.50

.75

1.00

False Positive Rate (1 - Specificity) Diagonal segments are produced by ties.

FIGURE 3 ROC Curve for Acceptability Ratings.

CONCLUSIONS AND LIMITATIONS In this study, drivers were placed in real world driving scenarios and asked to reveal their opinion about pavement roughness. This study provided empirical data that can be used to support an International Roughness Index (IRI) acceptability threshold. While the data are subject to interpretation, approximately 85% of all “acceptable” evaluations were made on test segments with an IRI of 170 in/mi (2.7 m/km) or less, which is the federal IRI guideline recommended by the 1998 National Strategic Plan set forth by FHWA. The findings of this research seem to provide empirical support for the federal IRI guidelines that are already in existence. This study also found that IRI levels provided a very good indication of rankings and acceptability, which agree with Nair et al. (19) who used the PSR ratings. These findings, however, do not seem to coincide with Kuemmel et al. (20), who indicated that less than 20% of the participants found an IRI of 170 in/mi to be acceptable, as was shown in TABLE 2. This study does not find any substantial evidence to suggest that the federal IRI guideline be moved. However, this study does beg the question of state and federal decision-makers in deciding where the threshold (or thresholds) should be placed: what fraction of the driving population do you want to “satisfy?” It is unrealistic to satisfy all drivers’ tastes and preferences, but this study provides the empirical data identify the IRI values that satisfy an optimal portion of the driving population, given other constraints. The main limitations of this study include the fact that results were based on pavements with a limited range of IRI values. A wider range of “good” and “poor” pavements may have provided more elucidating results. Unfortunately, this project was limited by the accessibility of facilities around the Seattle area to individuals who were willing to participate. The fact that a limited number of participants who may not be representative of the typical highway user in Washington State contributed to this study was also a limitation that has been present in past research studies. ACKNOWLEDGEMENTS This research was done entirely at the University of Washington (UW) and funded by the Washington State Department of Transportation (WSDOT). This paper has benefited from comments by our colleagues in the UW




10

Department of Civil and Environmental Engineering, particularly Dr. Joe Mahoney, Dr. Venkataraman (Venky) Shankar, and Dr. G. Scott Rutherford, and from comments by our colleagues at WSDOT, specifically Keith Anderson, Tom Baker, Kevin Dayton, Linda Pierce, Marty Peitz, Dr. Nadara (Siva) Sivaneswaran, Jim Spaid, Jeff Uhlmeyer, and Kim Willoughby. Dr. Sivaneswaran and John Livingston also deserve special thanks for leading collecting IRI data on the test segments. Dr. Linda Boyle provided input and feedback throughout the project, and Patrick Vu provided invaluable assistance in the preparation and collection of in-vehicle data. REFERENCES 1.

Elway Research Inc. Transportation Agencies Performance Audit: Citizen Input Survey. Prepared for the Washington State Joint Legislative Audit and Review Committees (JLARC), 1997. September.

2.

Cambridge Systematics. Department of Transportation Highways and Rail Programs Performance Audit. Prepared for the Washington State Joint Legislative Audit and Review Committee (JLARC), Report 98-2, 1998. Available on-line at: http://jlarc.leg.wa.gov/Reports/98-2HwyRails.PDF. Accessed August 1, 2002.

3.

Washington State Department of Transportation. Washington State Highway Pavements: Trends, Conditions, and Strategic Plan. Olympia, Washington. May 1999. Available on-line at: http://www.wsdot.wa.gov/fossc/mats/pavement/Pavement%20Plan.pdf. Accessed August 1, 2002.

4.

Keever, D.B., K.E. Weiss, and R.C. Quarles. Moving Ahead: The American Public Speaks on Roadways and Transportation in Communities. Federal Highway Administration (FHWA) Report: FHWA-OP-01-017. Washington, D.C. 40 pp. February 2001. Available on-line at: http://www.fhwa.dot.gov/reports/movingahead.pdf. Accessed August 1, 2002.

5.

Coopers and Lybrand, L.L.P.. National Quality Initiative (NQI) National Highway User Survey (NHUS). Federal Highway Administration, Washington, D.C., 1995. Available on-line at: http://ntl.bts.gov/data/nqinhus.pdf. Accessed August 1, 2002.

6.

Boeger, R. and R.J. Crowe. “It’s the Ride that Counts.” Public Road, 65 (4). January/February 2002. pp. 1719. Available on-line at: http://www.tfhrc.gov/pubrds/janfeb02/maricopa.htm. Accessed August 1, 2002.

7.

Sayers, M.W., T.D. Gillespie, and C.A.V. Queiroz. The International Road Roughness Experiment: A Basis for Establishing a Standard Scale for Road Roughness Measurements. I Transportation Research Record 1084, TRB, National Research Council, Washington, D.C., 1986, pp. 76-85.

8.

Federal Highway Administration. 1999 Status of the Nation's Highways, Bridges, and Transit: Conditions and Performance, Report to Congress. Federal Highway Administration. Report FHWA-PL-08-017. Washington D.C., May 2001. Available on-line at: http://www.fhwa.dot.gov////policy/1999cpr/index.htm. Accessed August 1, 2002.

9.

Carey, W.N., Jr. and P.E. Irick. The Pavement-Serviceability Concept. Highway Research Board Bulletin 250. National Research Council. Washington, D.C., 1960, pp 40-58.

10. American Association of State Highway and Transportation Officials. AASHTO Guide for Design of Pavement Structures. Washington D.C. June 1993. 11. Washington State Department of Transportation. WSDOT Pavement Guide, Volume 2: Pavement Notes for Design, Evaluation, and Rehabilitation. Olympia, Washington. February 1995. Available on-line at: http://www.wsdot.wa.gov/fasc/engineeringpublications/Manuals/Volume2.pdf . Accessed August 1, 2002. 12. Nakamura V.F. and H.L. Michael. Serviceability Ratings of Highway Pavements. In Highway Research Record 40, Highway Research Board, National Research Council, Washington, D.C., 1963, pp. 21-36.




11

13. Scrivner, F.H. and W.R. Hudson. A Modification of the AASHO Road Test Serviceability Index Formula. In Highway Research Record 46, Highway Research Board, National Research Council, Washington, D.C., 1964, pp. 71-87. 14. Karan, M.A., D.H. Kobi, C.B. Bauman, R.C. Hass. Subjective and Mechanical Estimations of Pavement Serviceability for Rural-Urban Road Networks, Transportation Research Record 715, TRB, National Research Council, Washington, D.C., 1979, pp. 31-36. 15. Moore, R.K., G.N. Clark, and G.N. Plumb. Present Serviceability–Roughness Correlations Using Rating Panel Data. In Transportation Research Record 1117, TRB, National Research Council, Washington, D.C., 1985, pp. 152-158. 16. Arterburn, S. and B. Suprenant. Public Perception of Pavement Rideability, Colorado Department of Highways. Report CDOH-UCB-R-90-10. Denver, Colorado. July 1990. 17. Ward D.R., K.J. Kercher, and S. Gulen. Correlation of IRI to Public Perception of Pavement Roughness: Final Report. Indiana Department of Transportation, Division of Research. West Lafayette, Indiana. May 1993. 18. Shafizadeh, K., F. Mannering, and L. Pierce (forthcoming). A Statistical Analysis of Factors Associated with Driver-Perceived Road Roughness on Urban Highways. Washington State Department of Transportation Research Report, WA-RD 583.1. 19. Nair, S.K., W.R. Hudson, and C.E. Lee. Realistic Pavement Serviceability Equations Using the 690D Surface Dynamics Profilometer. Research Report 354-1F, Center for Transportation Research, University of Texas, Austin. Austin, Texas. August 1985. 20. Kuemmel, D.A., R.K. Robinson, R.J. Griffin, R.C. Sonntag, and J.K. Giese. Public Perceptions of the Midwest's Pavements –Executive Summary – Iowa. Report CHTE 2001-02. WisDOT Highway Research Study #94-07. January 2001. Available on-line at: http://www.trc.marquette.edu/public_perception/. Accessed August 1, 2002. 21. Kuemmel, D.A., R.K. Robinson, R.J. Griffin, R.C. Sonntag, and J.K. Giese. Public Perceptions of the Midwest's Pavements –Executive Summary – Minnesota. Report CHTE 2001-02. WisDOT Highway Research Study #94-07. February 2001. Available on-line at: http://www.trc.marquette.edu/public_perception/. Accessed August 1, 2002. 22. Kuemmel, D.A., R.K. Robinson, R.J. Griffin, R.C. Sonntag, and J.K. Giese. Public Perceptions of the Midwest's Pavements –Executive Summary – Wisconsin. Report WI/SPR-01-01. WisDOT Highway Research Study #9407. January 2001. Available on-line at: http://www.trc.marquette.edu/public_perception/. Accessed August 1, 2002. 23. Khattak, A., J. Schofer, and F. Koppelman. Commuters’ Enroute Diversion and Return Decisions: Analysis and Implications for Advanced Traveler Information Systems. Transportation Research 27A(2), 1993, pp. 101-111. 24. Ng, L., W. Barfield, and F. Mannering. A Survey-Based Methodology To Determine Information Requirements For Advanced Traveler Information Systems. Transportation Research 3C(2), 1995, pp. 113-127.

