Monast, Zorio, and Cook 1

1

2

3

Geographic and Demographic Methodology for Peer 4

Group Classification of Rural Demand Responsive 5

Transportation 6 7 8 9

Kai Monast 10 Institute for Transportation Research and Education 11 North Carolina State University 12 Centennial Campus Box 8601 13 Raleigh NC 27695 14 Phone: 919-515-8768 15 Fax: 919-515-8898 16 Email: kcmonast@ncsu.edu 17

18 Darcy Zorio 19 Institute for Transportation Research and Education 20 North Carolina State University 21 Centennial Campus Box 8601 22 Raleigh NC 27695 23 Phone: 919-515-8624 24 Fax: 919-515-8898 25 Email: dbdowns@ncsu.edu 26

27 Thomas J. Cook 28 Institute for Transportation Research and Education 29 North Carolina State University 30 Centennial Campus Box 8601 31 Raleigh NC 27695 32 Phone: 919-515-8622 33 Fax: 919-515-8898 34 Email: tjc@ncsu.edu 35

36 37

Submission Date: 07/30/10 38 39 40 Word Count: 5,163 41

42 43

44

TRB 2011 Annual Meeting Original paper submittal - not revised by author.

Monast, Zorio, and Cook 2

1 ABSTRACT 2

This research proposes the formation of rural demand response transportation peer groups based 3 on the challenges transportation systems face from uncontrollable geographic and demographic 4 factors. Using geographic and demographic factors that are outside of the control of the 5 transportation systems establishes peer groups whose members share similar challenges in 6 providing transportation services. By accounting for uncontrollable factors, the differences in 7 performance among transportation systems are more directly due to variances in controllable 8 factors. 9

The study area includes all 80 rural demand response transportation systems in North 10 Carolina. Four factors are used to sort transportation systems into peer groups: range of service 11 area elevation, highway density, population density, and the ratio of rural population to the total 12 service area population. Scores of 1-5 are assigned to transportation systems for each factor to 13 cluster transportation systems with similar characteristics. These scores are added to create a 14 single score reflecting the geographic and demographic profile of each transportation system. 15 Transportation systems are then sorted into peer groups based on their profile scores. 16

This methodology classifies transportation systems into peer groups where every 17 transportation system in the group has a similar opportunity to perform as well as the highest 18 performing member of its group. By accounting for uncontrollable factors, the differences 19 between the transportation systems are primarily due to controllable factors, allowing more 20 effective performance comparisons among transportation systems. 21

22

TRB 2011 Annual Meeting Original paper submittal - not revised by author.

Monast, Zorio, and Cook 3

INTRODUCTION 1 Given the wide range of operating characteristics, many demand responsive transportation 2 systems claim they are unique and cannot be effectively compared to other systems. This claim 3 has resulted in problems when attempting to create effective peer groups. 4

Most efforts at developing peer groups have been based on factors that are under the 5 control of or can be influenced by a transportation system. Purchasing new vehicles or 6 implementing a new technology, for instance, may correlate to improved ridership, fares, and 7 funding assistance. However, significant geographic and demographic factors are not 8 controllable by the transportation system. 9

This research proposes to create demand responsive transportation peer groups based on 10 uncontrollable (exogenous) geographic and demographic factors. Using uncontrollable 11 geographic and demographic factors establishes peer groups whose members share similar 12 opportunities to succeed. The degree to which a transportation system has succeeded in relation 13 to its peers is then due to controllable factors, including fleet size, presence of technology, local 14 funding, skills and experience of staff, etc. 15

Peer groups are determined using four factors for the 80 North Carolina demand response 16 public transportation systems. The two geographic factors are range of elevation and highway 17 miles per square mile. The two demographic factors are population density and rural population 18 ratio. Transportation systems are assigned a score for each of the four factors. 19

The scores are then added to determine the degree to which geographic and demographic 20 factors influence a transportation system’s potential operating performance. Five peer groups are 21 established by clustering the total scores. Since the peer groups are based on uncontrollable 22 geographic and demographic characteristics, transportation systems can be measured against 23 their peers using performance statistics and other controllable variables. 24 25 LITERATURE REVIEW 26 Ripplinger (2010) uses a hierarchical cluster analysis to classify rural and small urban transit 27 agencies into peer groupings (1). The author states that data from the Rural National Transit 28 Database (NTD) is useful because of its consistency and uniformity, but notes that some of the 29 data are not reliable. The study classified agencies into four groups based on service type. 30 Systems in each group were clustered using annual vehicle-miles, annual vehicle-hours and fleet 31 size and evaluated using operating statistics for each of these clusters. Regarding the use of 32 operational variables to create peer groups, the study concludes that collecting service area data 33 may result in “the greatest improvement in terms of constructing rural transit agency peer 34 groups.” The author continues by stating that service area data will allow peer groups to be 35 created based on uncontrollable factors. 36

Vaziri and Deacon (1984) provide an in depth methodology for the creation of peer 37 groups and their potential use in rating transit performance (2). The study asserts that exogenous 38 variables (those that cannot be controlled by the transit agency) “have no lesser impact on 39 performance” than endogenous variables (those that can be controlled by the transit agency). The 40 authors concluded that “peer groups should be based solely on those uncontrollable market and 41 environmental variables that significantly influence transit performance.” Vaziri and Deacon 42 selected six exogenous variables to be used in their study by conducting a factor analysis and 43 using their professional judgment. This method resulted in transit systems grouped by poverty 44 levels, size, youth population, education, automobile availability and density of development. 45 Performance was compared within peer groups based on the assumption that “the subject system 46

TRB 2011 Annual Meeting Original paper submittal - not revised by author.

Monast, Zorio, and Cook 4

could achieve performance levels demonstrated by others in its peer group if the proper policy 1 decisions were made.” The authors concluded that it is appropriate to compare transit systems 2 using peer groups and that these peer groups should be based on uncontrollable variables. 3

TCRP 136: Guidebook for Rural Demand Response Transportation: Measuring, 4 Assessing and Improving Performance (2009) discussed the formation of peer groups and the use 5 of performance measures specific to rural demand response transportation (DRT) (3). The 6 research acknowledged that rural DRT is diverse because of the variety of organizational 7 structures, service area sizes, fleet sizes and funding streams. The report describes 28 factors that 8 influence DRT performance and classifies each one as controllable, partially controllable or 9 uncontrollable based on the system’s ability to influence each factor. Many different 10 methodologies were discussed to assess DRT performance, including peer group comparisons. 11 The report created a typology of rural DRT systems and discussed various factors that affect 12 performance such as ridership market served, service area or operating environment and use of 13 advanced technology. For the research portion of the report, three service area-based criteria 14 (municipal DRT systems, county DRT systems and multi-county DRT systems) were used to 15 present performance data of the systems that participated in the study. 16 17 METHODOLOGY 18 The purpose of this research was to develop an adaptive methodology for determining peer 19 groups based on uncontrollable geographic and demographic variables. Grouping rural 20 transportation systems using uncontrollable geographic and demographic factors will result in 21 performance differences between the systems that are due to factors under the transportation 22 system’s control. Once systems are grouped based on these uncontrollable factors, differences in 23 performance among peers are explained by policies, scheduling practices, technology, and many 24 other such controllable factors. Transportation systems that desire to improve can compare 25 themselves to high performing peers and identify differences in controllable factors. 26

A four step process is used to determine the peer groups, as listed below: 27 1. Identify Study Area; 28 2. Calculate Factors; 29 3. Analyze Data; 30 4. Create Peer Groups. 31

32 1. Identify Study Area 33 The first step in determining new peer groups using the method is to determine the study area. 34 The study area used in this research includes all 80 rural demand response transportation systems 35 in North Carolina. 36

The rural demand response transportation systems in North Carolina are spread across a 37 large geographic area including mountainous regions with peaks over 6,000 feet above sea level 38 and flat coastal areas with island communities only accessible by boat. Some service areas are 39 completely rural, while others have dense urban population centers. Fixed route transportation is 40 available in some service areas, while in others the primary source of transportation is human 41 service agency-centered demand response transportation. 42

Table 1 shows some of the variations of the transportation systems in the study area. 43 44

45

TRB 2011 Annual Meeting Original paper submittal - not revised by author.

Monast, Zorio, and Cook 5

TABLE 1. Study Area Characteristics 1

Variable (n=80) Mean Standard Deviation Minimum Maximum

Service Area Size (miles2) 609 380 199 2,314 Fleet Size 19 11 2 69 Annual Passenger Trips 89,450 145,960 9,068 1,203,377 Annual Service Miles 626,795 630,565 93,084 4,091,492 Annual Service Hours 33,633 40,691 3,448 316,092 Passengers per Service Mile 0.16 0.20 0.04 1.84 Passengers per Service Hour 2.96 2.78 1.17 24.76

Source: 2009 NCDOT Operating Statistics 2 3 2. Calculate Factors 4 Many geographic and demographic factors were considered. Geographic factors address the 5 relative difficulties of providing service posed by the service area. Variables used to create these 6 factors may include service area size, speed, trip length, elevation, intersections, highway miles, 7 road miles, or a combination of two or more variables. These data are available from the US 8 Census, USGS, private companies, and state and local sources. 9

Demographic factors address difficulties in providing service related to the dispersion of 10 trip origins and destinations. Variables used to create these factors may include population, urban 11 population, rural population, elderly population, youth population, disabled population, 12 households without a vehicle, population below the poverty level or median income, or a 13 combination of two or more variables. Many demographic variables are available from the US 14 Census, other Federal sources including NTD, and state and local sources. 15

The geographic and demographic data listed above and their combinations were considered 16 as possible factors. Proposed factors were tested using the following criteria to determine their 17 appropriateness for creating peer groups: 18

1. Are the data readily available, accurate, and complete? Previous peer group studies have 19 pointed out potential issues with NTD statistics (1). Issues with data reliability exist in 20 other data sources as well. For instance, geographic information such as road miles may 21 not include privately owned neighborhood roads. 22

2. Do the factors address significant differences among the transportation systems in the 23 study area? Factors that do not reflect differences are ill-suited for establishing peer 24 groups. 25

3. Does a single geographic factor and single demographic factor adequately capture the 26 diversity of the study area? If not, it is necessary to select additional factors. 27 28

After applying the selection criteria to potential factors, four factors are used in this analysis. 29 Range of Elevation is calculated by subtracting the lowest elevation point from the highest 30

elevation point in the service area (data source: NCDOT). Range of Elevation is a geographic 31 factor that indicates the potential difficulty in operating transportation service due to lower 32 operating speeds resulting from long driveways, steep inclines, curvy roads, etc. Transportation 33 systems with a large range of elevation encounter the greatest degree of challenges in service 34 provision. 35

However, the elevation factor alone does not adequately account for potential geographic 36 difficulties in providing service experienced by coastal communities and isolated areas with 37 small ranges in elevation. Therefore, it was necessary to select an additional geographic measure. 38

TRB 2011 Annual Meeting Original paper submittal - not revised by author.

Monast, Zorio, and Cook 6

Highway Density is calculated by dividing the total length in miles of State and Federal 1 highways by the service area size (data sources: NCDOT, Census 2000). Highway Density is a 2 geographic factor that indicates the potential mobility network constraints, as highways tend to 3 increase mobility options by offering a greater number of routes. It would be preferable to use all 4 road miles in the service area for this calculation. However, including local roads in the analysis 5 normally requires collection of local data whose quality and completeness can be questionable. 6 Highway miles are readily available from state and Federal sources, and are accurate and 7 complete. 8

The Highway Density factor accounts for coastal areas with expansive water bodies to 9 circumnavigate and for rural areas that have few transportation network options due to lack of 10 highways. Transportation systems with lower Highway Density encounter greater challenges in 11 service provision. 12

Together, Range of Elevation and Highway Density provide an effective measure of the 13 difficulty of providing service posed by the geographic constraints of the service area. 14

Population Density is calculated by dividing the total population by the service area size 15 (data source: Census 2000). Population Density indicates the relative proximity of trip origins. 16 Transportation systems with lower Population Density will be more likely to have longer trip 17 lengths, which will be more difficult to serve. 18

However, Population Density alone does not account for the potential dispersion of 19 destinations. Some rural transportation systems have urban clusters and/or urbanized areas within 20 their service area that are able to provide access to basic employment, medical, shopping, and 21 educational facilities. Other rural transportation systems may have similar overall population 22 densities, but do not include urban clusters or urbanized areas- forcing the transportation system 23 to travel outside their service area to provide access to these basic employment, medical, 24 shopping, and educational facilities. 25

Rural Population Ratio addresses this concern by dividing the rural population by the total 26 population of the service area using Census 2000 data and definitions. Rural Population Ratio 27 indicates the demand for trips outside of the service area, as rural areas will have less services 28 available within the area. Leaving the service area to provide trips can be costly, time 29 consuming, and inefficient for transportation systems. A service area with no urban population 30 (Rural Population Ratio of 1.0) does not include an urban cluster or urbanized area within its 31 boundaries. Transportation systems with higher Rural Population Ratios will have a higher 32 likelihood of costly and time consuming out of service area trips. 33

Together, Population Density and Rural Population Ratio measure the difficulties in 34 providing service posed by the demographic constraints of the service area. 35

Rural transportation systems in North Carolina are allocated funding for and expected to 36 provide general public service, which is why persons with disabilities, households without 37 vehicles, population below the poverty level, and other similar transportation disadvantaged 38 factors are not included. 39

The presence of a separate transportation system operating fixed-route within a rural 40 transportation system’s service area was also considered for use as a factor. These fixed-route 41 trips are not provided by the rural transportation system, but could impact a rural system’s 42 demand for service. After analysis, it was determined that the presence of a fixed route system 43 was inversely related to the Rural Population Ratio. Therefore, inclusion of a fixed route factor 44 was not necessary. 45

TRB 2011 Annual Meeting Original paper submittal - not revised by author.

Monast, Zorio, and Cook 7

Table 2 below shows the mean, mode, minimum, maximum, and standard deviation for 1 the four selected factors, as applied to the North Carolina study area. 2 3 TABLE 2. Descriptive Statistics of Factors 4

Variable (n=80) Mean Mode Standard Deviation Minimum Maximum

Range of Elevation 1451.0 560.5 1723.0 35.0 5556.0 Highway Density 1.7 1.8 0.6 0.5 3.1 Population Density 180.8 97.9 221.0 9.5 1321.5 Rural Population Percent 0.62 0.65 0.28 0.04 1.0

5 3. Analyze Data 6 Numerous methods are available to cluster transportation systems for each factor. Since this 7 research is intended to be easily replicated, it was necessary to select a common and easy to use 8 clustering method. The selected method uses Jenks Natural Breaks, also known as goodness of 9 variance fit (GVF). This method incorporates common statistical formulas that measure variance 10 in an iterative process. This process optimizes homogeneity within the peer groups (4, 5). 11

The Jenks process is incorporated into GIS software as a standard classification method. 12 Therefore, all transportation system data and factor data are linked to GIS software for analysis 13 and visual representation. 14

Two of the selected factors, range of elevation and population density, have broad ranges 15 of values and significant differences between the mean and the mode, indicating that there are 16 outliers in the datasets (see Table 2). It is difficult to classify datasets with many outliers into a 17 small number of categories that will adequately describe the members of the group. The 18 researchers determined through trial and error that those two factors require a minimum of five 19 categories to sufficiently create groups of transportation systems that are similar. For 20 consistency, five categories are used for all four factors. 21

The groups are assigned a score from 1-5 for each factor to cluster similar transportation 22 systems. For Range of Elevation, Population Density and Rural Population Ratio, the 23 transportation systems in the group with the highest values receive a score of five and the 24 transportation systems in the group with the lowest values receive a score of one. For Highway 25 Density, the transportation systems in the group with the highest values receive a score of one 26 and the transportation systems in the group with the lowest values receive a score of five. 27

Figures 1-4 display the clusters for each factor, along with the group size (n) and number 28 of transportation systems in each range. 29 30

31

TRB 2011 Annual Meeting Original paper submittal - not revised by author.

Monast, Zorio, and Cook 8

1

2 3 FIGURE 1. Range of Elevation 4 5 6

7 8 FIGURE 2. Highway Density 9 10

11

TRB 2011 Annual Meeting Original paper submittal - not revised by author.

Monast, Zorio, and Cook 9

1

2 3 FIGURE 3. Population Density 4 5 6

7 8 FIGURE 4. Rural Population Ratio 9 10 4. Create Peer Groups 11 The methodology could conclude with four separate grouping schemes based on individual 12 factors. However, combining the four schemes to create one overall grouping method is more 13 valuable for researchers and practitioners. It was determined that there should be between 5 and 14 25 transportation systems in each peer group to best classify similar transportation systems while 15 having a manageable total number of groups. Four peer groups were attempted, but that method 16 did not provide a fine enough distinction between peer groups and one group’s size did not meet 17 the goal of 5-25 systems in each group. Therefore, five peer groups are used to categorize the 80 18 transportation systems in the study area. 19

Consideration was given to assigning more importance to certain factors. Equal weights 20 are used because it is not possible to scientifically determine whether a factor should be weighted 21 higher or lower than other factors. 22

TRB 2011 Annual Meeting Original paper submittal - not revised by author.

Monast, Zorio, and Cook 10

To establish the peer groups, a transportation system’s assigned score for each factor is 1 summed to determine the overall geographic and demographic profile. The Jenks Natural Breaks 2 method is repeated to classify the systems into five peer groups based on their profiles. The 3 resulting peer groups consist of transportation systems that share similar combined geographic 4 and demographic profiles. Members of each peer group are considered to experience a similar 5 degree of difficulty in providing transportation service in their service area. Figure 5 displays the 6 proposed peer groups. 7

8 9 FIGURE 5. Proposed Peer Groups 10 11 Selected descriptive statistics for the proposed peer groups are displayed in Table 3. The 12 descriptive statistics have large variations within each peer group, which would be expected 13 because the peer groups are determined based on uncontrollable geographic and demographic 14 factors instead of factors controllable by the transportation system. 15 16 TABLE 3. Peer Group Descriptive Statistics for Selected Characteristics 17

Peer Group 1 2 3 4 5

Group Size (n=80*) 6* 14 21 25 14 Daily Service Miles Average 8,357 2,370 2,591 1,709 1,113

Standard Deviation 6,084 1,165 1,673 1,054 475 Minimum 1,900 381 704 358 505 Maximum 15,737 5,634 6,574 5,004 2,258

Fleet Size Average 32 22 24 17 11 Standard Deviation 10.8 5.9 15.2 9.3 2.9 Minimum 21 13 7 2 8 Maximum 42 34 69 37 17

Passengers per Service Mile

Average 0.13 0.16 0.12 0.19 0.17 Standard Deviation 0.08 0.08 0.04 0.35 0.10 Minimum 0.04 0.08 0.07 0.06 0.05 Maximum 0.26 0.36 0.23 1.84 0.39

Passengers per Service Hour

Average 2.14 2.94 2.31 3.58 3.18 Standard Deviation 1.08 2.22 0.65 4.53 1.40 Minimum 1.18 1.46 1.20 1.17 1.38 Maximum 3.96 10.38 3.58 24.76 6.75

* One transportation system in Group 1 does not report performance data18

TRB 2011 Annual Meeting Original paper submittal - not revised by author.

Monast, Zorio, and Cook 11

ADAPTATING THE METHODOLOGY 1 The methodology may be adapted to determine peer groups for other study areas. It is believed 2 that the four factors are appropriate for establishing peer groups in other areas, but it may be 3 necessary to adjust the number of groups within each factor. Increase the number of groups used 4 in each factor if there is a large variation among the transportation systems’ characteristics. 5 Decrease the number of groups if there is little variation. 6

It may be necessary to adjust the weights or eliminate one of the factors. Researchers are 7 encouraged to use professional judgment when determining factor weights. 8

For study areas where some transportation systems have long distances to major medical, 9 educational, and employment centers, it may be necessary to include a factor that will account 10 for these distances. One potential factor is miles to such facilities. Those data may need to be 11 calculated manually. 12

Finally, care should be taken when selecting multiple factors, to avoid those that have a 13 close correlation. 14 15 CONCLUSIONS 16 This research proposes categorizing rural demand responsive transportation systems based on 17 uncontrollable geographic and demographic profiles. This process can provide operators and 18 researchers with peer groups that experience similar levels of geographic and demographic 19 challenges within their service areas. 20

This scheme groups transportation systems so each system in a peer group has a similar 21 opportunity to perform as well as the highest performing member of its group. By accounting for 22 uncontrollable factors, the differences between the transportation systems’ performance will be 23 primarily due to controllable factors. The transportation system director, staff, and/or governing 24 board can work together to adjust the factors under their control to improve performance. 25

In conclusion, this peer group methodology for classifying rural demand response 26 transportation systems based on uncontrollable factors can be applied to other study areas 27 because of the general availability of the data and simplicity of the process. 28 29 FUTURE RESEARCH 30 Once a peer grouping methodology has been established, the next step is to identify the 31 appropriate measures used to compare transportation systems within a peer group. Passengers 32 per service hour is an industry-wide factor of performance that could be used. Other measures 33 may be based on financial performance, such as Federal and State subsidy per passenger trip. 34

Using the performance of the highest-achieving system in a peer group as a target, the 35 lower performing peers could identify and address controllable factors that may be hampering 36 their performance. If improvements in performance were realized by these efforts it would 37 strengthen the legitimacy of these results. 38

Finally, this methodology should be applied to other study areas to test its validity and 39 adaptability. 40 41 REFERENCES 42

1. Ripplinger, D. Classifying Rural and Small Urban Transit Agencies. Presented at 89th Annual 43 Meeting of the Transportation Research Board, Washington, D.C., 2010. 44

2. Vaziri, M. and Deacon, J. A. Peer Comparisons in Transit Performance Evaluation. In 45 Transportation Research Record: Journal of the Transportation Research Board, No. 961, 46 Transportation Research Board of the National Academies, Washington, D.C., 1984, pp. 13–21. 47

TRB 2011 Annual Meeting Original paper submittal - not revised by author.

Monast, Zorio, and Cook 12

3. Ellis, E. and McCollom, B. Guidebook for Rural Demand Response Transportation: Measuring, 1 Assessing and Improving Performance. TCRP 136, Transportation Research Board, (2009). 2

4. ArcGIS Resource Center. What is the Jenks optimization method? June 17, 2010. 3 http://resources.arcgis.com/content/kbase?fa=articleShow&d=26442. Accessed July 28th, 2010. 4

5. Wikipedia. Jenks Natural Breaks Optimization. April 29, 2010. 5 http://en.wikipedia.org/wiki/Jenks_Natural_Breaks_Optimization. Accessed July 29th, 2010. 6

7 ACKNOWLEDGEMENTS 8 Special thanks to Debra Collins, Judson Lawrie, Jeremy Scott, and Bastian Schroeder at the 9 Institute for Transportation Research and Education for contributing knowledge and expertise to 10 this paper. 11

TRB 2011 Annual Meeting Original paper submittal - not revised by author.