The Value of Better Vehicle Occupancy Estimates for Project Prioritization

Abstract

Transportation projects are increasingly evaluated based on how they improve person throughput, which requires knowing the passenger car occupancy. Yet obtaining occupancies is labor intensive, leading agencies to rely on statewide values. How useful are locality-specific vehicle occupancies? A case study of 38 candidate Virginia highway investments showed an occupancy change of .15 randomly applied to one-half of the projects affects 24% to 42% of project rankings if locality-specific occupancies are used. Occupancy is not the only data element affecting project prioritization, but it is an underlooked one. The study showed that an occupancy uncertainty of .15 is equivalent to an uncertainty of 3.90% in the discount rate or 13.2% in assigned traffic volume. As these investments had a mean value of $17 M, this study demonstrated that when project prioritization metrics are based on person throughput, city-or county-specific occupancies are an integral component of a data-driven prioritization process.

Keywords

infrastructure: planning transportation: highways decision-making equity and efficiency issues infrastructure: capital budgeting

Introduction

Vehicle occupancy data is essential for transportation agencies in multiple criteria to prioritize transportation investments (Association of Central Oklahoma Governments [ACOG], 2019; Commonwealth Transportation Board, 2021d; FHWA and FTA, 2009; Turner, 2017). ACOG (2019) uses 10 criteria to evaluate projects, one of which is the extent to which the project is associated with a strategy to reduce travel delay, accounting for 25% of the value of an intersection project. For California, the combined objectives of delay reduction and increased travel time reliability account for about 17% of a project’s value (Turner, 2017). Virginia uses delay reduction and person throughput, which account for 10% to 45% of a project’s score depending on the region (Commonwealth Transportation Board, 2021d). These agencies illustrate a fairly common theme: delay reduction and increased person throughput comprise one but not the sole factor in project prioritization. Person-based, rather than vehicle-based measures are receiving increased attention. Buckeye (Buckeye, 2012) noted that higher passenger vehicle occupancies allow agencies to serve more customers “within the existing highway footprint,” thereby reducing costs and environmental impacts such as noise and emissions. Such person-based metrics necessitate a method for estimating passenger vehicle occupancy.

The Importance of Person Throughput

Successive federal reauthorizations since 1964 have increasingly encouraged the use of person movement, rather than vehicle movement, when measuring transportation system performance. In this vein, several authors have advocated the use of person throughput to select candidate transportation investments because that metric directly describes transportation efficiency for end users. Person throughput has classically been used to compare the performance of general purpose versus high occupancy vehicle lanes (Wellander & Leotta, 2000) and, unlike vehicle throughput, can indicate if a given HOV policy is materially reducing congestion (Kwon & Varaiya, 2008). Shbaklo and Krueger (1998) showed how consideration of person throughput can be used to assess different corridor management strategies, including transit and HOV options, especially for smaller communities. When multiple modes are compared, a key person-based measure is productive capacity—the maximum number passengers per hour served multiplied by the speed of travel, where bus rapid transit and light rail transit typically have several times the productive capacity of the auto mode (Vuchic, 2007). Person throughput is an essential element for determining VMT per household, which in turn can be used for an accessibility analysis and thus may provide a more accurate assessment of the number of people who are forecast to benefit from a proposed project (Maria Kockelman, 1997). Performance measures that rely solely on vehicle throughput prioritize access for persons who own vehicles at the expense of users who cannot drive (Litman, 2023). Although some have found that the choice to carpool is not correlated with income (Park et al., 2018), others have shown that for some trip types, individuals with a lower income tend to carpool more than individuals with a higher income (Benita, 2020), meaning that person throughput will better support equity analyses than vehicle throughput.

Although a conceptual case has been made for focusing on the movement of people rather than vehicles, the literature has not often cited specific examples of how a focus on person throughput would alter investment choices. One possible reason is that person movement requires some estimate of vehicle occupancy, and a common data source for such occupancies—the National Household Travel Survey (NHTS)—does not indicate how occupancy varies by location (Mitra & Saphores, 2018) unless agencies purchase additional samples. A survey conducted by the research team of 22 state DOTs showed that respondents mostly used the NHTS and the American Community Survey for estimating occupancy (Xu et al., 2022). The NHTS, which is updated every 5 to 7 years, generally can provide only a statewide figure unless the state purchases additional samples. Although the ACS is more frequent (e.g., annually if a city or county has at least 65,000 people), the focus is based only on respondents’ reporting of the work trip; more detailed occupancies are not available. Consequently, it is to be expected that since the data are not widely available, researchers have generally not examined how location-specific occupancies, as opposed to a default value, affect the evaluation of performance.

Use of Passenger Vehicle Occupancy in Project Prioritization

Three approaches exist for estimating person-based throughput for highway projects. First, one may ignore occupancy entirely, using measures such as vehicle delay (Lu & Wang, 2005); vehicle hours of travel as done in the Flagstaff Metropolitan Planning Organization and the Southwestern Pennsylvania Commission (Gunasekera, 2014); vehicle miles traveled as used in many rural areas (National Association of Development Organizations and FHWA, 2011); or the ratio of vehicle volume to capacity (Smith, 2018). This approach is understandable given the effort required to obtain occupancies. The Oregon Department of Transportation (DOT) (2019) pointed out that “occupancy factors may not be available.”

A second approach differentiates occupancy by vehicle type: buses, carpools, and passenger cars (Oregon DOT, 2019). Guler and Menendez (2014) evaluated how transit priority strategies, such as giving a bus a “pre-signal” that allows it to enter an intersection ahead of a car, based on different ratios of bus occupancy to vehicle occupancy. Assumptions of passenger car occupancy have included 1.0 (a “conservative” estimate [Guler & Menendez, 2014]); 1.4 (based on a survey conducted roughly 8 to 10 years earlier [Oregon DOT, 2019]); and 1.7 [FHWA, 2019; Margiotta et al., 2018]).

A third (less frequent) approach is to include vehicle occupancy explicitly. For a corridor study conducted by the Wasatch Front Regional Council, Vyas et al. (2018) noted the importance of considering person throughput, which for passenger cars was based on a statewide household travel survey conducted 6 years previously. In an evaluation of candidate project level performance measures prepared for consideration by the Maricopa Association of Governments, Cambridge Systematics (2020) listed using person throughput as a best practice, noting that data included household travel surveys, occupancies for commuters based on Census data, and self-reported data.

The Challenge of Obtaining Vehicle Occupancies

FHWA and FTA reported that data essential for congestion management processes, i.e., “routine vehicle occupancy counts” (FHWA and FTA, 2009), are not widely available. Guo et al. (2020) reviewed various metrics used by transportation agencies for ranking projects, and although noting that hours of delay was “somewhat common,” the inclusion of passenger counts was “rare.” A 2021 survey conducted by the (Virginia) research team supported the earlier work reported by FHWA and FTA (2009) and Guo et al. (2020), finding that of 21 responding state DOTs, only two collected vehicle occupancy data through field measurements; three others used household questionnaires.

Obtaining occupancy count data on a widespread basis has historically been a labor-intensive effort (D’Ambrosio, 2011; Greene et al., 2019): either a stationary observer counts the number of occupants by looking through the windshield of passing vehicles or an observer performs this count while being driven in a test vehicle at a speed 10 to 15 mph slower than the average traffic speed. Although some approaches offer promise, such as image-based detection algorithms (Chan et al., 2011) and the use of crash record systems (Asante et al., 1996; Krile et al., 2019), such efforts are not yet widely deployed and require the use of field data for calibration. Occupancy estimates are usually obtained only through special studies, as indicated by one 2021 survey respondent who noted the existence of a special 2016 occupancy study (Colorado State University, 2016).

In short, project prioritization processes can be based on a wide range of data elements covering socioeconomic factors, traffic engineering, cost, performance, environmental impacts, modal split, and land development. Thus, because each data element has a cost, work is needed to understand how to compare the benefits of more accurate occupancies to an improvement in accuracy of other data elements, such as traffic volume or the discount rate, so that agencies may allocate resources accordingly.

Research Question and Scope

With a high data collection cost and no compelling benefit, an agency’s decision not to collect occupancy data is a rational one. Yet given the increased emphasis on person throughput rather than vehicle throughput when project investments are made, do locality-specific vehicle occupancies materially influence the project prioritization process? This study quantified the value of detailed passenger vehicle occupancy estimates on project prioritization using a case study in the Hampton Roads region of southeastern Virginia, an area with roughly 1.1 million people.

Figure 1 shows three lines of inquiry. First, in a controlled experiment, the study measured how deviations in occupancy from a presumed value could affect the selection of candidate transportation projects. Second, occupancy deviations were compared to changes in other variables. Third, based on three methods for estimating occupancy, the study identified a range of uncertainty for these estimates.

Figure 1.

Process for evaluating the importance of occupancy data.

The scope did not include determination of a best method for determining occupancy; rather, the focus was the relative importance of such estimates. The scope was also limited to uncertainty in passenger vehicle occupancy as opposed to occupancy for commercial or transit vehicles.

Case Study Approach

Virginia’s SMART SCALE prioritization process evaluates candidate investments through seven benefits: safety, congestion mitigation, accessibility, environment, economic development, air quality, and land use and transportation coordination. The case study focused on the 38 projects in VDOT’s Hampton Roads District (Figure 2).

Figure 2.

38 SMART SCALE projects in VDOT’s Hampton Rroads district.

In the case study area, congestion mitigation was weighted to account for 45% of these benefits and was based on how a project improves two performance measures: (1) additional peak period person throughput, determined by multiplying the number of vehicles by the “average vehicle occupancy rate,” and (2) reduction in peak period person delay, which is the product of reduction in vehicle delay and average vehicle occupancy rate. The research team based the case study on Virginia’s process (Commonwealth Transportation Board, 2021d). That process may be summarized as follows for the purposes of this paper:

1. Identify from all Virginia projects—not just those in the case study area—the single project with the best performance measure for person throughput and the single project with the best performance measure for delay (yellow part of Figure 3).

2. Obtain a congestion weighted score. Figure 3 shows person throughput is the increase in users served in the peak period (e.g., number of vehicles served multiplied by the vehicle occupancy). This step is done for each of the 38 projects in this case study.

3. Repeat a similar process to obtain the scores for the six non-congestion categories: safety, accessibility, environment, economic development, air quality, and land use and transportation coordination. (The metrics differ for each category. For example, whereas person-hours of delay saved was used for part of the congestion score, crash reduction is used for the safety score.)

4. Sum the seven scores, divide by the money requested by the project sponsor in units of tens of millions of dollars (Figure 4), and then rank the projects based on the score/cost ratio.

Figure 3.

Example to obtain total congestion weighted score.

Methodology

The research team investigated how variations in occupancy estimates could affect project rankings by first ensuring that the project prioritization steps could be replicated and then second by examining the sensitivity of the rankings to changes in occupancy.

Figure 4.

Obtain the score/cost ratio for each candidate project.

Replicate Elements of the Prioritization Process

The research team replicated this process such that the results matched those used in Virginia (Commonwealth Transportation Board, 2021b, 2021c, 2021d, 2021e). For instance, considering project 6690 (an improvement on Holland Road in Virginia Beach), Table 1 (columns 1 and 2) forecast that this project would increase person throughput by 348.27 and reduce person-hours of delay by 121.5 (Commonwealth Transportation Board, 2021a, 2021c) during the peak period. The best project for these two performance measures in all of Virginia was project 6641 in Loudoun County, forecast to increase person throughput by 1862.33 and to reduce delay by 610.1 person-hours. Steps 1 through 3 showed that the congestion score for project 6690 (equation (1)) was 8.69 (column 3). For the performance measures of safety, accessibility, environment, economic development, and coordination of land use and transportation, scores were determined in a similar manner (albeit with different factors, weights, and single best projects in those respective measures), and case study data (Jackson, 2022) showed a total of 10.37 (column 4). The total benefit, based on summing all six scores, was 8.69 + 10.37= 19.06 (column 5). The funds sought for this project by the sponsor (Virginia Beach City) were $16.8 million (column 6). The final score (calculated by 19.06/1.68) was 11.35. This Holland Road project received a higher final score than the next highest project, which was project 6692 (11.22).

C o n g e s t i o n S c o r e = (50 % * 348.27 / 1862.33 * 100 + 50 % * 121.5 / 610.1 * 100) * . 45

(1)

Table 1.

Impact of Reducing Holland Road Project Occupancy on Project Rankings.

Situation	Project	Increase in Person Through-Put (1)	Delay Savings in Person-Hours (2)	Congestion Score (3)	Other PM Scores (4)	Total Benefit Score (5)	Requested Cost in $10 Million (6)	Final Score (7)
Holland occupancy unchanged^a	Holland (6690)	348.27	121.50	8.688	10.37	19.06	$1.68	11.35
Holland occupancy unchanged^a	6692	110.76	24.55	2.244	8.75	11.00	$.98	11.22
Holland occupancy drops by .05^b	Holland (6690)	333.76	116.43	8.326	10.37	18.70	$1.68	11.13

PM = performance measure.

^aThe data in rows one and two match those available from Commonwealth Transportation Board (2021a).

^bThe data in row three were determined by the research team based on a presumed occupancy of 1.2 that then drops to 1.15, with the benefits computed as the ratio of 1.15/1.2.

Examine the Sensitivity of Rankings to Changes in Occupancy

The research team then examined how changes in occupancy affected the rankings. The default occupancy used for the Holland Road project was 1.2, and a spreadsheet was created that based on this occupancy and other project data elements computed a final score of 11.35. Then, in order to make the Holland Road project ranking score lower than that for project 6692, a reduction in occupancy of at least .05 was needed, which, in turn, reduced the person throughput and person-hours of delay as shown in the last row of Table 1.

A more systematic way to ascertain the importance of occupancy is to determine how many of the rankings for the 38 projects in the Hampton Roads District would change if for each project the occupancy for the next highest or next lowest project were to change by a particular amount. Accordingly, three separate experiments were undertaken.

Experiment 1 was performed twice—once with a baseline occupancy of 1.20 and once with a baseline occupancy of 1.14 (the mean vehicle occupancy for American Community Survey (ACS) Census data for commuting to work in the region). In experiment 1, the researchers defined each of the 38 projects based on its initial rank as even (ranked 2, 4, … 38) or odd (ranked 1, 3, … 37). Then, as shown in the light blue part of Figure 5, the occupancies for the odd projects were reduced from a presumed baseline of 1.20 to 1.15 (a deviation of −.05). Then, the entire process was continued for other occupancy deviations (e.g., +.05, −.10, +.10, stopping at +0.40 and −.20). Occupancies ranged from a low of 1.00 (one driver per vehicle) to 1.60 (chosen based on NHTS data [Buchanan, 2022]). The same steps were repeated by changing occupancies for only the even projects. The reason for doing odd projects first and then even projects was to avoid project selection bias. The number of projects for which the rankings changed was then determined.

Figure 5.

Experiment 1 example for changing odd ranking projects.

In experiment 2, since all project factors, not just those related to occupancy, are subject to some uncertainty, experiment 1 was repeated such that for each project, the congestion score (which is influenced by occupancy) was left unchanged but the remaining six non-congestion factors were allowed to vary. The impact of this variation on project ranking was then compared to that in experiment 1.

In experiment 3, the new scores in experiment 2 for the six non-congestion factors were used to develop new rankings. Then, the procedure for changing occupancy in experiment 1 was repeated.

Results

Replicate Elements of the Prioritization Process

A spreadsheet was devised that enabled the research team to examine how changing the occupancy for individual projects could affect the final score. For instance, to surpass the next higher ranking project, Holland Road needed an occupancy increase of 0.18, increasing its score to 12.12 (Table 2). Although similar, Tables 1 and 2 illustrate that the importance of occupancy is relative: for Holland Road, either an occupancy decrease of .05 (Table 1) or an occupancy increase of .18 (Table 2) was needed to change how the project was ranked.

Table 2.

Impact of Increasing Holland Road Project Occupancy on Project Rankings.

Situation	Project	Increase in Person Through-Put (1)	Delay Savings in Person-Hours (2)	Congestion Score (3)	Other PM Scores (4)	Total Benefit Score (5)	Requested Cost in $10 million (6)	Final Score (7)
Holland occupancy unchanged^a	7005	34.34	.00	.415	2.70	3.12	$.26	12.09
Holland occupancy unchanged^a	Holland (6690)	348.27	121.50	8.688	10.37	19.06	$1.68	11.35
Holland occupancy rises by .18^b	Holland (6690)	400.51	139.72	9.992	10.37	20.36	$1.68	12.12

PM = performance measure.

^aThe data in rows one and two match those available from Commonwealth Transportation Board (2021a).

^bThe data in the last row were determined by the research team based on a presumed occupancy increase of .18, from 1.20 to 1.38, with the benefits computed as the ratio of 1.38/1.20.

Simulation Where Only Occupancy Varies

Holland Road demonstrates that the importance of occupancy varies by project: a large change in occupancy will not affect two projects with very different scores but may affect two similar scoring projects. A simulation was performed to determine, systematically, how changes in occupancy could affect rankings, with different results obtained for even and odd projects, as shown in Table 3.

Table 3.

Change in Project Rankings as a Result of Changing Occupancy.

Row	Deviation	Occupancy	Number of Projects Whose Rank Changed (Cost of Those Projects)
Row	Deviation	Occupancy	Add Deviation to All Even-Ranked Projects	Add Deviation to All Odd-Ranked Projects
1	.00	1.20 (baseline)	0 ($0 M)	0 ($0 M)
2	−.05	1.15	2 ($50 M)	2 ($27 M)
3	.05	1.25	4 ($34 M)	2 ($50 M)
4	−.10	1.10	6 ($101 M)	11 ($127 M)
5	.10	1.30	12 ($130 M)	4 ($105 M)
6	−.15	1.05	10 ($171 M)	12 ($134 M)
7	.15	1.35	16 ($339 M)	9 ($175 M)
8	−.20	1.00	10 ($171 M)	12 ($134 M)
9	.20	1.40	16 ($339 M)	11 ($195 M)
10	.25	1.45	16 ($339 M)	11 ($195 M)
11	.30	1.50	16 ($339 M)	11 ($195 M)
12	.35	1.55	17 ($340 M)	12 ($195 M)
13	.40	1.60	18 ($342 M)	13 ($188 M)

Baseline Occupancy of 1.20

Starting with a default baseline occupancy of 1.20, row two in Table 3 shows the number of projects whose ranking changes if the initially even-ranked projects (e.g., projects 6, 8…38) saw their occupancy decrease by −.05 to 1.15: only two of the 38 projects saw their rankings change. These projects entailed a total cost of $50 M. Row one also shows, however, that if one were instead to decrease the occupancy by −.05 for the projects initially ranked odd (e.g., project 5, 7…37), then the other two of the 38 projects would see their rank shift with a lower investment of $27 M.

Table 3 suggests that based on the Hampton Roads District, a change in occupancy of .10 from a particular baseline could alter the rankings for 4 to 12 projects of the 38 total; i.e., it could affect between 11% and 32% of all project rankings. If the occupancy were to deviate by .20, then 26% to 42% of projects could see their project ranking change. A deviation of .05 might affect at most (roughly) 10% of rankings.

Baseline Occupancy of 1.14

Virginia has examined locality-specific occupancies based on the 2016 ACS, which although focused on only the work trip (as opposed to all trip purposes) was viewed as advantageous by agency leadership because such an occupancy was more likely applicable for the peak period (Buchanan, 2022). In the case study area, occupancies are given for seven jurisdictions and range from 1.12 to 1.20 with a mean of 1.14 (if occupancies from those seven jurisdictions are weighted equally).

Repetition of the simulations of Table 3 with the Hampton Roads mean baseline occupancy of 1.14 (rather than a baseline occupancy of 1.20) gave results that were similar but with occupancy having a slightly greater impact due to the nature of the simulation: a change in occupancy of .05 had a larger impact on a smaller baseline occupancy than on a larger baseline occupancy. With a deviation of .05, Table 3 showed that two to four projects could see their ranks shift, whereas a new analysis with a lower occupancy showed that from two and six projects could see their ranks shift. A deviation of .10 had shifted ranks for 4 to 11 projects in Table 3; the new analysis caused a shift of 4 to 12 projects. With a mean occupancy of 1.14, the impact of a negative deviation of .15 could not be examined. However, a positive deviation of .15 could shift from 7 to 14 projects (Table 3), and the new analysis could shift rankings for 9 to 16 projects. No change was observed for a positive deviation of .20: from 11 to 16 projects could see their rank shift. In short, the results shown in Table 3 appear generally to convey the sensitivity of project prioritization to changes in occupancy for this particular case study.

Simulation Where Other Factors Vary

All project factors—not just those related to occupancy—are subject to some uncertainty. The question arises: to what extent does uncertainty in occupancy compare with the uncertainty of these other factors? The researchers are not aware of any work that has directly quantified uncertainty in these project scores although earlier work examined contingencies in Virginia’s process from previous years (P. Singla and J. S. Miller, unpublished data). That work computed for certain lane widening projects a total “alternative contingency” percentage, which was the stated contingency cost divided by an adjusted base construction cost, the latter of which included miscellaneous items but excluded inflation, contingency, and construction engineering and inspection. This alternative contingency percentage was larger than the stated contingency and was designed to represent fully any factors that could affect costs. Such contingencies ranged from a low of 0% to a high of almost 57%, with a mean contingency of 18% and a standard deviation of 15%.

A three-part experiment was devised where for each project 45% of the score based on congestion mitigation (which is influenced by occupancy) was left unchanged but the remaining six non-congestion factors were allowed to vary based on multiplying the sum of the original six scores by a simulated normal distribution with a mean of 1.0 and a standard deviation of 15% (Figure 6). For instance, column 4 of Table 2 shows that for project 6690 (Holland Road), the non-congestion scores summed to 10.37. The random simulation increased this score from 10.37 to 10.66—a modest change of about 3%. This modest change, coupled with changes of scores for other projects, did not affect the ranking for project 6690. Figure 7 demonstrates three sets of simulations and Table 4 shows these three sets of simulation results.

Figure 6.

Example of simulating contingencies in non-congestion scores.

Figure 7.

Three sets of simulation results.

Table 4.

Change in Project Rankings as a Result of Changing Occupancy and Other Factors.

Row (1)	Deviation (2)	Occupancy (3)	Change in Project Rankings (Ground Truth is the Original Rankings)		Change in Project Rankings (Ground Truth is the Revised Rankings Based on Random Alterations to the Non-Congestion Factors)
Row (1)	Deviation (2)	Occupancy (3)	Add Deviation to All Even-Ranked Projects (4)	Add Deviation to All Odd-Ranked Projects (5)	Add Deviation to All Even Ranked Projects (6)	Add Deviation to All Odd- Ranked Projects (7)
1	.00	1.20	22	17	—	—
2	−.05	1.15	23	16	4	2
3	.05	1.25	23	14	4	4
4	−.10	1.10	23	19	4	6
5	.10	1.30	22	19	8	9
6	−.15	1.05	20	22	7	10
7	.15	1.35	22	21	10	13
8	−.20	1.00	20	24	7	13
9	.20	1.40	22	21	10	13
10	.25	1.45	22	21	17	13
11	.30	1.50	21	21	18	14
12	.35	1.55	24	20	19	13
13	.40	1.60	23	19	20	14

1. Change non-congestion factors only. Row 1, columns 4 and 5, shows that 22 and 17 projects had a change in ranking if occupancy was held constant but other factors were allowed to vary for even-ranked projects and odd-ranked projects, respectively. Previously, when only occupancy was changed and other factors were held constant, fewer projects had a change in rankings unless occupancy changed dramatically. This result was not surprising: the impact of altering factors that account for 55% of a project’s prioritization score should be greater than that of altering a multiplier that affects part of the remaining 45% of the score.

2. Change all factors simultaneously. After the non-congestion factors in row one were changed, altering occupancy further (see rows 2–13 and columns 4-5) had only a modest impact on project rankings: the number of affected project rankings was between 14 and 24. In short, once the non-congestion factors have been altered, changes to occupancy have a modest effect.

3. Redo the experiment shown in Table 3 with new non-congestion factors. This was to presume that changing non-congestion factors would yield new rankings and to rerun the experiment shown in Table 3 where only the occupancy was changed. These results are shown in columns 6 and 7 of Table 4. The results were close to those shown in Table 3 which revealed given the change in other factors, the same change in occupancy (.10) may have a similar impact on project rankings. For instance, a shift in occupancy of .10 altered project rankings for four to nine of the 38 projects (similar to 4 to 12 of the projects in Table 3).

Discussion

Occupancy is just one of several data elements in the project prioritization process subject to uncertainty. How important is occupancy compared to these other data elements? One way to make this determination is to compare how a change in occupancy influences project prioritization with respect to two other data elements common to most prioritization processes: vehicle volume and the discount rate.

Importance of Occupancy Versus Traffic Volume

The assigned traffic volume for the link where a future project will be built is also a forecast and subject to uncertainty. The fact that occupancy is larger than 1.0 means that the impact of a change in occupancy alone will be somewhat less than the impact of a change in the link volume. If occupancy is exactly 1.00, then an increase in occupancy of .05 persons corresponds to a 5% change in link vehicle volume. With this case study’s default occupancy of 1.2, an occupancy increase of .05 corresponds to a change in link volume of 4.17%. If the FHWA average occupancy of 1.7 (FHWA, 2019) is used, then a change in occupancy of .05 corresponds to a change in link volume of 2.9%.

Importance of Occupancy Versus the Discount Rate

The discount rate is also uncertain (Davies, 2012). With regard to the Holland Road project, how might alternative discount rates influence project prioritization relative to different occupancy rates?

Figure 8 shows that project 6690 has a total person throughput of 348.27 and a total person delay savings of 121.5 hours when occupancy is 1.20 (upper left). With a 5-year discount rate of 5%, the person throughput and delay savings are 80.44 persons and 28.06 hours (upper middle). Changing the discount rate from 5% to 6.5% yields a person throughput of 334.29 and a delay savings of 116.62 hours (upper right), which corresponds to a new SMART SCALE score of 11.13 (last row of Table 1). Thus, the 1.5% net change in the discount rate (5% to 6.5%) equals a .05 change in occupancy (1.20 to 1.15). This can be standardized to say that a .01 change in the occupancy rate is equivalent to a .30% change in the discount rate (middle of Figure 8).

Figure 8.

Comparison of a change in discount rate with a change in occupancy.

The same approach with a different project—that shown in Table 2 rather than Table 1—yielded a similar standardized answer: a .01 change in occupancy equates to a .26% change in the discount rate.

Potential Variation in Feasible Occupancy Estimation Methods

Table 5 summarizes three methods for estimating auto occupancy: (1) use of values from the 2009 NHTS provided by VDOT staff; (2) use of values from the ACS, which are based on the work trip as extracted from the 2016 ACS by VDOT staff; and (3) extraction of occupancies from crash data where potential bias associated with crash data at the jurisdiction level has been addressed (Xu et al., 2022). (Virginia records all occupants, not just those who are injured.) All three methods target the morning peak period. Columns 2 through five show that by city, the mean range of these occupancies is .22, with the NHTS values being the highest and the crash or ACS values being the lowest. The mean difference in each possible pair of methods for all cities (e.g., NHTS vs. ACS, NHTS vs. crash, and ACS vs. crash) was .15.

Table 5.

Vehicle Occupancies by Select Jurisdictions in Hampton Roads.

City (1)	Original				Scale to NHTS				Scale to ACS
City (1)	NHTS (2)	ACS (3)	Crash (4)	Range (5)	NHTS (6)	ACS (7)	Crash (8)	Range (9)	NHTS (10)	ACS (11)	Crash (12)	Range (13)
Chesapeake	1.36	1.13	1.15	.22^a	1.36	1.28	1.39	.11	1.15	1.13	1.15	.02
Hampton	1.32	1.13	1.20	.19	1.32	1.27	1.56	.29	1.14	1.13	1.18	.05
Newport^b	1.36	1.20	1.08	.28	1.36	1.62	1.16	.46	1.15	1.20	1.11	.09
Norfolk	1.20	1.14	1.14	.06	1.20	1.33	1.36	.16	1.12	1.14	1.15	.03
Portsmouth	1.60	1.15	1.12	.48	1.60	1.36	1.29	.30	1.19	1.15	1.13	.06
Suffolk	1.24	1.12	1.11	.13	1.24	1.23	1.26	.03	1.12	1.12	1.13	.01
Va. Beach^b	1.31	1.13	1.14	.17	1.31	1.29	1.36	.07	1.14	1.13	1.15	.01
Mean	1.34	1.14	1.13	.22				.20				.04

NHTS = national household travel survey; ACS = American community survey; Crash = crash data with bias correction.

^aAll values are rounded to two decimal places.

^bProper names are Newport news and Virginia Beach, respectively.

Given a value of .15, the occupancy uncertainty is equivalent to around .26%*.15/0.01 = 3.90% for the discount rate (left side of Figure 1) or around 4.4% *.15/.05 = 13.2% in terms of traffic volume for a particular project. With regard to Table 3, this uncertainty could change the rankings for 9 to 16 projects (of 38) in the case study area, where such projects have a capital investment of $171 M to $339 M. The question then arises: would such changes in rankings affect transportation as perceived by users? If all projects can be built, then the answer is no—and arguably, because projects tended to shift only one ranking, a more likely impact is that with an average cost of $17 M, just one project would not be built that otherwise would be built and vice versa. Although Virginia’s SMART SCALE process ultimately determines which projects are built, other agencies might use a project prioritization process to determine when projects are built. For example, the Missouri DOT (Metropolitan Planning Council, 2007) commits to building projects in a “High” category first and then to build projects in a “Medium” category as resources permit.

Potential Variation in Occupancy by City

If a consistent method of determining occupancy is used for all projects, then the mean range of .22 (Table 5, column 5) overstates the importance of occupancy in project prioritization. Rather, of interest is how a given method affects the relative occupancies by city. How would a Chesapeake project’s occupancy compare to that of a Portsmouth project if NHTS values were used for both versus if ACS values were used for both? The span of occupancies differs by method: ACS values show a span of slightly less than .08 (e.g., 1.12 in Suffolk to 1.20 in Newport News) whereas NHTS values show a span of about .40 (e.g., 1.20 in Norfolk to 1.60 in Portsmouth). A prioritization process based on a very tight span of occupancies may be less affected by a deviation in occupancy than a process based on a wide range of occupancies.

To aid in understanding how these three methods affect occupancy differences by location, columns 6 to 13 scale all three of these occupancies to an NHTS baseline and then to an ACS baseline. For instance, in the example of Norfolk (with an NHTS occupancy of 1.20, which is .14 below the mean NHTS occupancy of 1.34), the NHTS span (.40) is much larger than the ACS span (.075). Converting Norfolk’s original NHTS occupancy to an ACS baseline places this occupancy at 1.12 (equations (2) and (3)).

{N o r f o l k}_{N H T S}^{A C S b a s e l i n e} = \bar{A C S} + (\frac{{N o r f o l k}_{N H T S} - \bar{N H T S}}{{S p a n}_{N H T S}}) {S p a n}_{A C S}

(2)

1.12 = 1.14 + (\frac{1.20 - 1.34}{0.40}) 0.075

(3)

The ranges in columns 9 and 13 account for differences in the means for each method. For instance, ACS values give Newport News projects the highest occupancy, NHTS values place them near the middle, and crash data give them the lowest occupancy. Column 13 shows that this variation when scaled to ACS data means that Newport News sees a variation of .09 in occupancy depending on which method is used. Overall, if the NHTS occupancy is used as a baseline and then ACS and crash occupancies are scaled to the NHTS, then the mean range by city for all three methods is .20; if the ACS occupancy is used as the baseline, then the mean range is much smaller, at .04. These mean ranges may be compared to the results in Table 3: with an ACS baseline, for instance, a deviation of .04 (close to the .05 in Table 3) suggests that two to four projects could see their rankings affected (of 38 total).

Applicability to Other Agencies

The credibility of the approach presented here is influenced by three factors: repeatability, scalability, and bias. First, there simply is no perfect ground truth data set for vehicle occupancy at every location in a state. Thus, one can collect data manually (through field observations) at a small number of sites and compare observations and estimates, but this comparison becomes infeasible as the number of sites grows. For example, Virginia’s prioritization process considers almost 400 different projects. Thus, repeatability is an issue. Second, the estimate is affected by consistency in terms of choosing a data collection method. For instance, one may consider occupancies reported from two cities: Chesapeake and Norfolk. The difference in occupancy is .16 if NHTS is the data source, which is much larger than the difference of .01 if based on crash data. Thus, scalability is an issue. Third, the crash-based approach discussed herein appears to benefit from Virginia’s practice of recording occupancies for all crashes—even if no injury results. For other states that may collect occupancy data for injury crashes only, additional bias correction measures will be needed. Thus, bias is an issue. That said, if the three factors can be addressed, then these results suggest that prioritization systems can benefit from the use of local data rather than default values.

Conclusions

Data-driven prioritization methods that link candidate investments to desired outcomes have the potential to improve the transparency of project prioritization processes. Zhao and Manaugh (2023) showed how the use of various data elements (e.g., socioeconomic information, infrastructure condition, and travel demand) could help decision makers choose among worthy but different goals such as projects that improve the connectivity of the transportation network versus projects that reduce greenhouse gas emissions. Quantitative methods are not limited to point-based systems; they can also inform expert assessments. The U.S. Government Accountability Office (2019) showed that quantitative prioritization methods that connect candidate projects to particular outcomes can increase the “accountability” of an investment program. A well-structured systematic approach for prioritization can increase “stakeholder participation” and therefore support transparency—but data quality is an integral component of these methods (Li et al., 2016). In this vein, this study evaluated the impact of one particular data element—vehicle occupancy—through a sensitivity analysis based on a case study in Virginia’s Hampton Roads region. This sensitivity analysis showed how the use of location-specific data, rather than a statewide default value, can affect investment decisions.

This study showed that a location-specific occupancy value has a modest potential to influence project rankings. A minor variation of .05 in occupancy could affect 5% to 11% of project rankings, whereas larger fluctuations in occupancy, such as .10 or .20, can lead to a more substantial impact on project rankings, ranging from 11% to 32% and 26% to 42%, respectively. Generally, such modifications affected no more than a one-position shift in the 38-project case study. With an average project cost of $17 million, locally specific occupancies could redirect at most $34 million in investments (due to not building one project and instead constructing another).

This exercise enables one to compare the relative importance of data elements. Because person throughput is the product of vehicle volume and occupancy, an occupancy change of .05 (compared to a default value of 1.20) corresponds to a volume change of 4.2%. Alternatively, based on Figure 4, an uncertainty in occupancy of .05 was roughly equivalent to a change in the discount rate of 1.5% in the case study. Although this paper focused on occupancy, the general principle of replacing a default value with location-specific values may be explored with other data elements used in project prioritization. For instance, a prioritization system could be considered that assigns a set additional number of points for any project that benefits two modes of transportation (ACOG, 2019). One could examine how prioritization of projects is affected if one replaces this default premium with a sliding scale where more points are assigned as the mode shares become more similar.

The relative importance of occupancy will vary based on the project prioritization system. For instance, in the case study, congestion benefits (which are based on occupancy) affect 45% of a project’s score. Thus, as shown in Table 4, changing the other factors (which account for 55% of a project’s score) will have a greater impact than occupancy. Because other agencies’ prioritization processes (ACOG, 2019; FHWA and FTA, 2009; Turner, 2017) also do not have delay as the dominant factor in project prioritization, this finding may extend to other locations.

The impact of occupancy also depends on how repeatability is addressed. If a consistent method for determining occupancy is applied, the relative uncertainty in occupancy after adjustment to a common baseline is .04 if ACS is the method and .20 if NHTS is the method. The ACS baseline difference, as it is close to the simulated difference of .05, would suggest that location-specific occupancies could affect roughly 5% to 11% of project rankings. Because the NHTS method shows a greater range in occupancies, it suggests that location-specific occupancies could affect about 26% to 42% of project rankings.

Footnotes

Acknowledgements

This research benefitted from insights provided by David Caudill, Donna Chen, Ben Cottrell, Linda Evans, Mike Fontaine, Jungwook Jun, Sanhita Lahiri, Andrew Mondschein, Brian Park, Ivan Rucker, Eric Stringfield, Rahul Travedi, Peng Xiao, and Mo Zhao and data provided by Jared Buchanan, Shan Di, and Tina Simmons. The authors are responsible for any errors herein.

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: all authors; data collection: all authors; analysis and interpretation of results: all authors; draft manuscript preparation: all authors. All authors approved the final version of the manuscript.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received financial support for the research, authorship, and/or publication of this article: This work was supported by Virginia Department of Transportation under grant 118055.

ORCID iD

Yiqing Xu

Author Biographies

Yiqing Xu is a third-year transportation PhD student at the University of Virginia. She received a bachelor's degree from Zhejiang University of Technology and a master's degree in Urban and Environmental Planning from the University of Virginia. She is interested in shared mobility infrastructure, demand modeling, and transportation equity.

Lance Dougald is an associate principal research scientist at VTRC where his research emphasis areas are in traffic engineering and transportation planning. He received his master’s degree in civil engineering at the University of Virginia and is a member of ASCE’s Transportation and Development Institute’s Safety Committee.

John Miller is a professional engineer and the Associate Director for Environment, Planning, and Economics at the Virginia Transportation Research Council. His recent research has been focused primarily within the applied transportation planning area.

References

Asante

S. A.

Adams

L. H.

Shufon

J. J.

McClean

J. P.

(1996). Estimating average automobile occupancy from accident data in New York state. Transportation Research Record: Journal of the Transportation Research Board, 1553(1), 115–123. https://doi.org/10.1177/0361198196155300117

Association of Central Oklahoma Governments (2019). STBG-UZA application guidebook. Association of Central Oklahoma Governments.

Benita

(2020). Carpool to work: Determinants at the county-level in the United States. Journal of Transport Geography, 87, 102791. https://doi.org/10.1016/j.jtrangeo.2020.102791

Buchanan

(2022). Email to J. S. Miller.

Buckeye

K. R.

(2012). Innovations on managed lanes in Minnesota, Public Works Management & Policy. 17(2), pp. 152–169.

Cambridge Systematics, Inc (2020). Project Evaluation Best Practices. Cambridge Systematics, Inc. https://azmag.gov/Portals/0/Documents/MagContent/2020-Project-Evaluation-Best-Practices-Report.pdf?ver=2021-01-27-090203-857 (Accessed 13 June 2020).

Chan

C.-Y.

Singa

Wang

(2011). Implementation and evaluation of automated vehicle occupancy verification California PATH research report No. UCB-ITS-PRR-2011-04. University of California.

Colorado State University (2016). Average vehicle occupancy study for the Colorado department of transportation, Division of Transportation Development. Unpublished manuscript.

Commonwealth Transportation Board (2021a). Consensus scenario approved may 19, 2021 (excel). Commonwealth Transportation Board. https://smartscale.org/documents/round_4_consensus_scenario_selected_projects_for_posting.xlsx (Accessed 4 Oct 2021).

10.

Commonwealth Transportation Board (2021b). Project scores. Commonwealth Transportation Board. https://smartscale.org/documents/fy2022-resource-documents/project-scores-fy2022.xlsx (Accessed 4 Oct 2021).

11.

Commonwealth Transportation Board (2021c). Project scoring calculations (excel). Commonwealth Transportation Board. https://smartscale.org/documents/fy2022-resource-documents/r4_detailed_project_scoring_calculations.xlsx (Accessed 4 Oct 2021).

12.

Commonwealth Transportation Board (2021d). SMART SCALE technical guide. Commonwealth Transportation Board. https://smartscale.org/documents/2020documents/technical-guide-2022.pdf (Accessed 4 Oct 2021).

13.

Commonwealth Transportation Board (2021e). Viewing SMART SCALE application: Holland road phase I. Commonwealth Transportation Board. https://smartportal.virginiahb2.org/#/public/applications/2022/smartScale/view/F30-0000007344-R01 (Accessed 4 Oct 2021).

14.

D’Ambrosio

K. T.

(2011). Methodology for collecting vehicle occupancy data on multi-lane interstate highways: A GA 400 case study. M.S. Thesis, Georgia Institute of Technology.

15.

Davies

(2012). Proposed modifications to the cost-benefit analysis decision criteria for road project evaluation to improve decisionmaking. Transportation Journal, 51(4), 473–487. https://doi.org/10.5325/transportationj.51.4.0473

16.

FHWA (2019). National household travel survey. FHWA. https://nhts.ornl.gov/ (Accessed 11 July 2022).

17.

FHWA and FTA (2009). Placing the congestion management process in the context of metropolitan transportation planning goals and objectives. FHWA-HOP-09-043 Capital District Transportation Committee Albany. https://ops.fhwa.dot.gov/publications/fhwahop09043/fhwahop09043.pdf (Accessed 15 June 2022).

18.

Greene

MacDonald

Arena

Srini

Gourevitch

Ezike

Stern

(2019). Technology and equity in cities: Emerging challenges and opportunities. The Urban Institute. https://www.urban.org/sites/default/files/publication/101360/technology_and_equity_in_cities_1.pdf (Accessed 17 Feb 2022).

19.

Guler

S. I.

Menendez

(2014). Evaluation of presignals at oversaturated signalized intersections. Transportation Research Record: Journal of the Transportation Research Board, 2418(1), 11–19. https://doi.org/10.3141/2418-02

20.

Gunasekera

Hirschman

(2014). NCHRP 08‐36, task 112 cross mode project prioritization. Transportation Research Board. https://onlinepubs.trb.org/onlinepubs/nchrp/docs/NCHRP08-36(112)_FR.pdf (Accessed 13 June 2022).

21.

Guo

S. B.

Jha

Greeman

(2020). Technical memorandum: Project prioritization practices and methods: Task 4–synthesis development. Texas A&M Transportation Institute.

22.

Jackson

(2022). Email to J. S. Miller.

23.

Krile

Landgraf

Slone

(2019). Developing vehicle occupancy factors and percent of non-single occupancy vehicle travel. FHWA, U.S. Department of Transportation. Publication FHWA-PL-18-020.

24.

Kwon

Varaiya

(2008). Effectiveness of California’s high occupancy vehicle (HOV) system. Transportation Research Part C: Emerging Technologies, 16(1), 98–115. https://doi.org/10.1016/j.trc.2007.06.008

25.

T. H. Y.

Thomas Ng

Skitmore

(2016). Modeling multi-stakeholder multi-objective decisions during public participation in major infrastructure and construction projects: A decision rule approach. Journal of Construction Engineering and Management, 142(3). https://doi.org/10.1061/(asce)co.1943-7862.0001066

26.

Litman

(2023). Evaluating transportation equity: Guidance for incorporating distributional impacts in transportation planning. Victoria Transport Policy Institute. https://www.vtpi.org/equity.pdf (Accessed 19 April 2023).

27.

J. J.

Wang

(2005). Development of a procedure for prioritizing intersections for improvements considering safety and operational factors. University of South Florida.

28.

Margiotta

R. A.

Turner

Taylor

(2018). National performance measures for congestion, reliability, and freight, and CMAQ traffic congestion: General guidance and step-by-step metric calculation procedures. Cambridge Systematics, Inc. Publication FHWA-HIF-18-040.

29.

Maria Kockelman

(1997). Travel behavior as function of accessibility, land use mixing, and land use balance: Evidence from san francisco bay area. Transportation Research Record: Journal of the Transportation Research Board, 1607(1), 116–125. https://doi.org/10.3141/1607-16

30.

Metropolitan Planning Council (2007). States’ approaches to transportation project prioritization linking policy, planning and programming. Metropolitan Planning Council. https://www.metroplanning.org/uploads/cms/documents/nationalpractices.pdf (Accessed 30 Oct 2022).

31.

Mitra

S. K.

Saphores

J.-D.

(2018). An analysis of travel characteristics of carless households in California. University of California Irvine. https://escholarship.org/uc/item/4j54k2bv (Accessed 7 Jan 2022).

32.

National Association of Development Organizations and FHWA (2011). Transportation project prioritization and performance-based planning efforts in rural and small metropolitan regions. National Association of Development Organizations and FHWA. https://www.nado.org/wp-content/uploads/2011/11/RPOprioritization.pdf (Accessed 13 June 2022).

33.

Oregon Department of Transportation (2019). Transportation analysis performance measures In Analysis procedure manual version 2. Oregon Department of Transportation. https://www.oregon.gov/ODOT/Planning/Documents/APMv2_Ch9.pdf (Accessed 30 June 2022).

34.

Park

Chen

Akar

(2018). Who is interested in carpooling and why: The importance of individual characteristics, role preferences and carpool markets. Transportation Research Record: Journal of the Transportation Research Board, 2672(8), 708–718. https://doi.org/10.1177/0361198118756883

35.

Shbaklo

S. A.

Krueger

L. B.

(1998). A pilot application study of corridor performance indicators. Florida Department of Transportation. https://rosap.ntl.bts.gov/view/dot/12716 (Accessed 19 April 2023).

36.

Smith

P. N.

(2018). Full-cycle performance-based planning and programming. Texas Department of Transportation. https://transops.s3.amazonaws.com/uploaded_files/Texas_DOT_Slides-NOCoE_Webinar_Future-Proofing_Performance_Management.pdf Accessed June 13, 2022.

37.

Turner

L. T.

(2017). Application of a project prioritization framework to the 2016 SHOPP. California Department of Transportation. Publication CA17-2921 https://dot.ca.gov/-/media/dot-media/programs/research-innovation-system-information/documents/final-reports/ca16-2785-finalreport-a11y.pdf (Accessed 13 June 2022).

38.

U.S. Government Accountability Office (2019). Discretionary transportation grants—actions needed to improve consistency and transparency in DOT’s application evaluations. U.S. Government Accountability Office. https://www.gao.gov/assets/gao-19-541.pdf (Accessed 24 April 2023).

39.

Vuchic

V. R.

(2007). Urban transit systems and technology. John Wiley & Sons, Inc.

40.

Vyas

Harris

Knowlton

LaBonty

G. J.

Milam

R. T.

Seager

Williges

Zundel

(2018). Applying innovative performance metrics for corridor evaluation. Transportation Research Record: Journal of the Transportation Research Board, 2672(44), 93–102. https://doi.org/10.1177/0361198118788463

41.

Wellander

Leotta

(2000). Are high-occupancy vehicle lanes effective? Overview of high-occupancy vehicle facilities across north America. Transportation Research Record: Journal of the Transportation Research Board, 1711(1), 23–30. https://doi.org/10.3141/1711-04

42.

Dougald

L. E.

Miller

J. S.

(2022). Development of guidance for a vehicle occupancy rate data collection program. Virginia Transportation Research Council. https://www.virginiadot.org/vtrc/main/online_reports/pdf/23-R5.pdf (Accessed 22 Jan 2023).

43.

Zhao

Manaugh

(2023). Introducing a framework for cycling investment prioritization. Transportation Research Record: Journal of the Transportation Research Board, 0(0), 1–13. https://doi.org/10.1177/03611981231152241