Guidebook on Pedestrian and Bicycle Volume Data Collection (2014)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

21 C H A P T E R 3 This chapter describes how to plan and implement a pedestrian and bicycle counting program and provides guidance on the various steps of the counting process, including technical and logistical considerations. This guidance is drawn from a comprehensive literature review, surveys of and interviews with transportation agencies that conduct pedestrian or bicycle counts, and field testing of counting devices. Case studies are referenced to illustrate key points. The full case studies are provided in Appendix A. 3.1 Chapter Organization This chapter is divided into two main sections that address (1) planning the count program and (2) implementing the program. The subsections listed below describe the specific activities involved in each phase of the program. In the PDF version of this guidebook, click on any of the section names to go directly to that section. Planning the Count Program Implementing the Count Program Specify the data collection purpose Obtain necessary permissions Identify data collection resources Procure counting devices∗ Select count locations and determine Inventory and prepare devices∗ the count timeframe Train staff Consider available counting methods Install and validate devices∗ Calibrate devices∗ Maintain devices∗ Manage count data Clean and correct count data Apply count data ∗Steps that only apply to counts using automated counting techniques. Data Collection Planning and Implementation Chapter 3 Topics • Planning a non-motorized volume counting program • Implementing the count program

22 Guidebook on Pedestrian and Bicycle Volume Data Collection 3.2 Planning the Count Program Planning is an important first step in developing an efficient and useful pedestrian and bicycle counting program. Although it is possible to relatively quickly collect manual counts or to pur- chase and install automated counting technologies, this course of action may not produce useful, long-term results. Planning a count program typically involves the following steps: • Specifying the general data collection purpose, • Identifying data collection resources, • Selecting count locations and determining the count timeframe, and • Considering available counting methods. The following sections present these steps in a particular order, but they are often con- sidered iteratively. For example, count managers may reconsider the resources needed for data collection after they realize that they would like to count additional locations. Similarly, managers may revisit the number of count locations after recognizing that they would like to gather continuous counts over a long time period (which may require purchasing additional counting devices for more locations, or rotating existing devices among locations). Organizations planning a pedestrian and bicycle count program for the first time should expect that their program will be modified in the future. Although most programs benefit from having some core data that have been collected consistently from start, many programs revisit their stated purposes, reassess resources, consider new or different count locations and time periods, and inte- grate new counting methods. This chapter provides general guidance for planning count programs, but the most effective programs are tailored to meet an organization’s unique needs. Successful count programs result from experimenting and refining the approach over time. Several other sources also provide useful guidelines for establishing pedestrian and bicycle count programs. These include Chapter 4 of the Traffic Monitoring Guide (TMG) (FHWA 2013) and the Guide to Bicycle & Pedestrian Count Programs (Portland State University 2014). 3.2.1 Specifying the General Data Collection Purpose Chapter 2 describes reasons why transportation agencies and other organizations collect pedestrian and bicycle counts. These purposes include • Measuring changes in pedestrian and bicycle activity relative to baseline levels; • Documenting changes in activity levels after projects are implemented; Important Definitions Counting product (equipment). The counting product is the complete set of equipment used to collect volume counts. Products incorporate • The counting technology that detects pedestrians and bicyclists; • Components to house, mount, and power the counting technology; • Devices to store the count data; and • Tools to download or export data. Counting technology (device, sensor, method). A counting technology is the specific device or sensor that detects pedestrians or bicyclists. The counting technology is one component of the counting product provided by a vendor. Manual data collection uses human sight as the counting technology.

Data Collection Planning and Implementation 23 Considering Multiple Data Users A transportation agency wants to document how pedestrian volumes change after sidewalks are added along 10 roadways. Initially, it plans to use staff to con- duct 4-hour weekday afternoon counts at each location, both 1 week before and 3 weeks after sidewalks are installed. This plan would achieve the agency’s imme- diate need: comparing volumes before and after the sidewalk installation. However, there may be reasons to continue to count into the future. In particular, installing an automated counter at one location (situated where it will not be affected by the sidewalk construction) would allow the agency to develop a con- tinuous set of count data that records variations in demand by hour, day of week, and season of year. These data could be used to • Estimate annual pedestrian volumes from future short-term counts in other locations; • Evaluate whether the sidewalks continue to influence pedestrian demand beyond 3 weeks, and • Control for exposure when evaluating pedestrian crash risk. • Informing the general public about pedestrian and bicycle activity and trends; • Monitoring variations in pedestrian and bicycle activity levels by time of day, day of week, or season of the year, and under different weather conditions; • Identifying variations in activity in different types of locations (e.g., considering land uses and/or facility types) and calculating context-specific expansion factors; • Assessing local and system wide activity to prioritize locations for new pedestrian and bicycle facilities; • Quantifying exposure, as part of an analysis of pedestrian or bicycle crash risk at specific loca- tions; and • Developing models to predict future pedestrian or bicycle volumes at different locations throughout a community. All of these purposes can be achieved—at least in part—by collecting continuous pedestrian or bicycle volume data over time. The ability to collect counts over an extended period of time is one of the most important benefits of automated pedestrian and bicycle counting technologies. In turn, the broad availability of non-motorized count data is an important part of ensuring a multimodal (or “complete streets”) approach to transportation issues within a community. As an example, Figure 3-1 shows how continuous count data can be used to identify sea- sonal fluctuations in bicycle activity, as well as to document general increases in activity over time. This figure shows that bicycle activity on this trail is highest in the spring and summer months, and that bicycle volumes in 2011 were generally higher than in comparable months in 2010. As another example, Figure 3-2 shows how continuous count data can reveal the effect of spe- cific factors—in this case, uncleared snow between February 6 and 19—on bicycle activity. Identifying the specific counting purpose (or purposes) is a critical first step in the data collection planning process. The count purpose drives decisions about where, when, and how to collect data. Practitioners should consider both current and potential future uses of data, as illustrated in the “Considering Multiple Data Users” sidebar.

24 Guidebook on Pedestrian and Bicycle Volume Data Collection Source: Arlington County, VA (2012). Figure 3-1. Continuous count data from Custis Trail, November 2009– January 2012, Arlington County, VA. Source: Arlington County, VA (2012). Figure 3-2. Illustrative impact of uncleared snow on bicycle volumes. Case Study Examples The case studies in Appendix A illustrate other ways that continuous count data can be used: • Identifying typical pedestrian or bicycle activity patterns in different parts of a community (Alameda County); • Informing multimodal planning efforts (Arlington County); • Developing planning-level demand models (San Francisco); and • Estimating crash risk (UC Berkeley).

Data Collection Planning and Implementation 25 3.2.2 Identifying Data Collection Resources Available resources determine the scale of an organization’s pedestrian and bicycle counting program. According to the practitioner survey conducted during the development of this guide- book, the most common barriers to collecting more pedestrian and bicycle data were • Lack of staff time and volunteer interest, and • Funding limitations or cutbacks. Funding for count programs can come from internal agency budgets, external grants, or facil- ity improvement projects (e.g., installing counting devices as a part of roadway or multi-use trail reconstruction, providing budget for before-and-after pedestrian and bicycle counts). However, even agencies with minimal staff time and funding can establish a count program. This outcome can be achieved, for example, by organizing volunteers or by creating partnerships with other agencies (including those, such as health agencies, not traditionally thought of as generators or users of count data). Keep in mind that although more data can be collected by using volunteers, staff time must be budgeted to provide high-quality training, coordinate volunteers, and enter data into the count database. Even if an organization has no dedicated budget for pedestrian and bicycle counts, it may be possible to take advantage of other existing data collection efforts. For example, many transpor- tation agencies and consultants routinely conduct or commission intersection traffic counts for traffic studies. Even though these efforts may be focused on automobile operations, pedestrian and bicycle count data are often collected, because non-motorized volumes influence traffic signal timing and automobile operations. In addition, agencies may be able to mine existing data to identify instances where bicycle and pedestrian volume data were collected but not stored, because of a lack of formal means for report- ing. Prior counts may not be consistent with current methods (due to lack of common standards or protocol), but they can give a community an initial sense of walking and bicycling activity and can be used to demonstrate the value of having good data. Existing counts also show where and when pedestrian and bicycle activity is missing, which helps build the case for filling those gaps. Agencies with limited resources may be able to take advantage of volunteers. Most communi- ties have citizens who are interested in pedestrian and bicycle issues, and some of these people are willing to be volunteers if they learn about an opportunity to assist the local pedestrian and bicycle program. Volunteers are frequently recruited for manual counts, where it is advantageous to get short-duration counts at multiple locations. These volunteer counts are often done on the same day or week of the year, as specified by the NBPD Project (www.bikepeddocumentation.org). When using volunteers, it is important to train them and monitor the quality of their counts during a training session before using them for official counts (for more information, see Section 3.3.4). Options for Agencies with Limited Resources Agencies with minimal or no dedicated budget for counting can still establish a count program through • Organizing volunteers (see the Washington State DOT case study); • Creating partnerships with other agencies (see the Columbus case study); and • Compiling motorized traffic counts for intersections conducted for other purposes—these counts often include pedestrian and bicycle volumes.

26 Guidebook on Pedestrian and Bicycle Volume Data Collection In some cases it may be appropriate to use volunteers to assist in installing, moving, and down- loading data from automated counting devices. However, given the complexities associated with some technologies, this approach requires identifying volunteers with some degree of technical capacity, and it may be best suited for students at a local college or university. Most automated devices call for professional expertise to install and use, and quality control will be optimized by hands-on staff training and involvement, regardless of the count approach or technology selected. By showing the utility of counts for producing policy-relevant results, a transportation agency can make the case that counting pedestrians and bicyclists is an important function that should be a routine part of their activities. Therefore, useful count results can lead to additional resources for counting. Even data collected infrequently and opportunistically can lead to a permanent count pro- gram. Nevertheless, as initial decisions about counting activities affect subsequent decisions about how, when, and where to collect data, it is important for agencies to think about how their programs may grow in the future. Both geographic (where) and temporal (when and how long to count) expansions can be planned systematically, so that the program ends up representing key geographic areas (or the whole community) and all important time periods. Any expansion should keep some original count sites and time periods so that past trends can be monitored into the future. 3.2.3 Selecting General Count Locations and Timeframe Pedestrian and bicycle data collection programs can benefit from combining two approaches: 1. Gathering short-duration counts (typically less than 1 day to several days, but potentially up to several months) at many locations; and 2. Gathering continuous counts over multiple years at a small sample of locations. The short-duration counts capture spatial variation in pedestrian and bicycle activity in different parts of the community. The continuous counts identify specific types of activity pat- terns and are used to adjust the short-duration counts. The TMG (FHWA 2013) recommends this approach (see TMG Chapter 2 for general guidance on selecting count locations and times, and Sections 4.4 and 4.5 of TMG Chapter 4 for more specific guidance on non-motorized counting). Approaches for Selecting Count Locations The following are possible options for selecting count locations: • Random locations. Sites are selected randomly. This approach may not capture strategic locations, nor select sites appropriate for automated counting. Selecting randomly from within categories of desired characteristics (stratified random sampling) is an alternative. • Representative locations. This approach balances available resources with spatial coverage. Identified sites, in aggregate, are representative of the community as a whole. • Targeted locations. Sites are selected on the basis of being associated with particular projects, facility types, or locations with particular characteristics (e.g., safety concerns). • Control locations. This approach compares sites affected by a project with unaltered sites (control locations) to determine how much of the observed change in demand can be attributed to the project.

Data Collection Planning and Implementation 27 Select General Geographic Locations Resource limitations often prevent counting at every desired location, so particular locations must be chosen based on the primary purposes of the data collection program. Four approaches, described in more detail below, have been used for determining count locations: • Random locations, • Representative locations, • Targeted locations, and • Control locations. Random Locations Count locations can be selected randomly. For example, an agency can assign unique identifi- cation numbers to each of its intersections and use a random number generator to select which intersections to count. However, this simple random sampling approach may not capture strategic locations for counting. Additionally, random sampling may not identify locations suitable for auto- mated technologies, because numerous site-specific factors ultimately determine suitability for a count location (e.g., opportunities to install equipment and patterns of pedestrian and bicycle movements). Random sampling can also result in selecting locations with very low volumes, which tend to have higher levels of variation over time than higher volume locations. High variability pro- duces more error when estimating long-term (e.g., annual) volumes from short-duration counts. There are alternatives to simple random sampling. Potential count locations can be stratified into categories according to particular characteristics, such as commuting versus recreational route, land use type, income category, or proximity to attractors (e.g., schools, parks, and transit stops). Analysts consider each category separately and select locations within each category ran- domly. This process, called stratified random sampling, can be used to ensure that there are at least a few count locations with each key characteristic of interest. This strategy has been used to select count locations when developing predictive pedestrian and bicycle volume models and safety performance functions (Schneider, Arnold, and Ragland 2009a; Schneider et al. 2010; Griswold, Medury, and Schneider 2011; Strauss, Miranda-Moreno, and Morency 2014). See the San Diego County case study for an example of stratified sampling. Representative Locations Most communities would like to measure how pedestrian and bicycle activity changes over time in the community as a whole. This objective requires counting at representative sites throughout the community. Representative locations could be identified using a random sampling process. However, it is more common to select representative sites using a systematic approach guided by a count manager or advisory group. In order to be representative, count locations should be • Located in different geographic parts of the community; • Surrounded by different types of land uses; • Found on different types of facilities (e.g., multi-use trails, bicycle lanes, sidewalks); and • Reflective of the range of socioeconomic characteristics in the community as a whole. Limiting count sites to locations that are convenient, have the highest pedestrian or bicycle vol- umes, or are expected to have the greatest increases in walking and bicycling does not produce a representative sample. A set of representative sites can be used to compare changes in the number of reported pedes- trian and bicycle crashes with changes in overall pedestrian and bicycle activity levels throughout the community. This approach allows analysts to track the relative risk of pedestrian or bicycle

28 Guidebook on Pedestrian and Bicycle Volume Data Collection crashes (per pedestrian crossing, per trail user, per bicyclist, etc.). In other words, representative counts control for exposure across the community as a whole. Targeted Locations Specific locations can be targeted for counting, recognizing that the count locations, in aggre- gate, will not be representative of the community as a whole. These locations are often related to particular projects, particular facility types, or locations with particular characteristics. For example, some communities choose to count in specific locations with a high number of crashes (i.e., “hot spots”). If the community is interested in identifying the relative risk of one specific roadway segment versus another specific roadway segment, the agency may target counts at these two locations. After using the counts to control for exposure, the agency can determine which locations have the greatest crash risk and evaluate the roadway design and behavioral characteristics that might be making those sites dangerous. Communities also target counts at locations where specific projects have been or will be implemented, to document changes in walking and bicycling after project completion. For this purpose, it is important to count at locations at or near the project, and to select control locations for comparison, described next. Finally, “pinch points,” or locations where pedestrians and bicyclists must converge to cross a barrier (e.g., river crossings, freeway crossings, railroad crossings), are good locations to docu- ment large portions of a community’s pedestrians and bicyclists. One sampling strategy is to count at a series of pinch points (e.g., all bridges crossing a river that bisects a community or all pedestrian and bicycle crossings of a freeway loop around the CBD). Control Locations To get a true understanding of the effect of a specific project on pedestrian or bicycle activity or safety, it is also necessary to count at similar locations not directly affected by the project (e.g., at a location with the same number of roadway lanes and a similar surrounding neighborhood on the other side of town). These other locations are called control sites. Control sites account for broader influences on walking and bicycling (e.g., an increase in gas prices or a community- level pedestrian and bicycle promotion program), making it possible to quantify the change in walking and bicycling activity or safety actually due to the project of interest. Some of the users of a new or improved pedestrian or bicycle facilities may have shifted from nearby parallel routes. Counts can be taken on these streets and corridors to help distinguish between new (or more frequent) non-motorized travel generated by the project and existing non-motorized travelers who have diverted to the new or improved facility. Developing Factor Groups When a goal of the counting program is to use long-term volume patterns to extrapolate short-term counts to longer periods of time (e.g., a year), it is important to extrapolate based on long-term volume patterns at a site with similar patterns to the count location being extra- polated. This is referred to as using an appropriate factor group. The need to establish factor groups then becomes a consideration when selecting the continuous count sites that will be used to develop the long-term volume patterns. Research by the Colorado DOT (Nordback, Marshall, and Janson 2013) has used long-term automated counts from multiple locations to identify different factor groups (see the Colorado DOT case study for examples). Figure 3-3 shows how bicycle activity patterns from multiple automated monitoring stations can be used to identify a factor group on the basis of having similar daily volume patterns. For motor vehicle counts, the TMG (FHWA 2013) recommends having at least five to eight continuous count stations as the basis of each factor group.

Data Collection Planning and Implementation 29 Recent research has provided guidance on applying factor groups for pedestrian and bicycle counting programs. Several studies have described how to define specific factor groups. In gen- eral, this is accomplished by gathering data from multiple automated count locations to iden- tify which locations have “similar” volume patterns. A basic approach is to graph pedestrian or bicycle volume patterns by time of day, day of week, or month of year. Patterns that appear visu- ally similar (e.g., higher volumes on weekends than weekdays or higher volumes during morning and evening commute periods than mid-day periods) are grouped together. A statistically based approach to developing factor groups has been developed by Miranda- Moreno et al. (2013). This method uses two indices: • WWI: weekend traffic volume divided by weekday traffic volume, and • AMI: morning peak (7–9 a.m.) traffic divided by mid-day (11 a.m.–1 p.m. traffic). Source: Nordback, Marshall, and Janson (2013). Note: Inverse daily factor is the percentage of the average daily volume observed on each specific day. Each legend item represents a different bicycle monitoring location. The bold line is the average of all locations, which is used to represent the factor group. Figure 3-3. Colorado DOT commute trail factor group. Approaches for Developing Factor Groups Factor groups are count sites that experience similar daily, monthly, and annual pedestrian and bicycle traffic patterns. These patterns may be due to similarities in local commuting activity and surrounding land uses. The typical traffic pat- terns observed at these locations are used as the basis for expanding short counts at locations with similar characteristics. Approaches to developing factor groups include: • Visual comparison of volume patterns from continuous counts. • Statistical comparisons of volume ratios derived from continuous counts. • Applying criteria describing characteristics of interest (e.g., land uses) in select- ing count locations to include in a given factor group. A common element in all of these approaches is the need for continuous count data, which are only practically obtained from automated counting devices.

30 Guidebook on Pedestrian and Bicycle Volume Data Collection The relative values of these indices are used to classify count patterns into utilitarian (low WWI and high AMI), mixed-utilitarian (moderately-low WWI and moderately-high AMI), mixed-recreational (moderately-high WWI and moderately-low AMI), and recreational (high WWI and low AMI), two examples of which are given in Figure 3-4. A practical disadvantage of this approach is that it does not yet provide guidance on how to determine which factor group should be applied at a particular short-count location (i.e., what characteristics make a location demonstrate utilitarian or recreational patterns?). This guidance is needed to be able to extrapolate short-duration count data. It is likely that this connection will be made in future research. The “Classification of Bicycle Traffic Patterns in Five North American Cities” case study provides an example of this statistical approach. Another approach to developing factor groups was used by researchers who already knew the characteristics of specific sites where they wanted to estimate long-term volumes (e.g., annual) from short-duration counts (Schneider, Arnold, and Ragland 2009; Schneider et al. 2012). This approach grouped the short-duration count locations according to the characteristics likely to create the local activity pattern near the site (commercial land use vs. residential land use, employment density, etc.). Long-term automated counts were then collected at some sites in each category, and the average long-term count pattern for each category was used to extrapolate all short-term count locations in that category. This approach is limited by the restrictive criteria used to define a particular land use category. Determine the Count Timeframe Two aspects of timing are important for count programs: duration and frequency. The count duration will depend on the counting technologies and resources available to an agency. Most automated technologies can capture continuous counts for multiple months. These continuous count data can then be used to extrapolate short-term counts taken at similar locations (see the Alameda County case study for an example). Count Duration An important consideration when conducting counts is how long counting must occur to have a suitable amount of data for analysis. If data collection is to document an hourly volume pattern Source: Miranda-Moreno et al. (2013). Figure 3-4. Examples of utilitarian (left) and recreational (right) bicycle activity patterns.

Data Collection Planning and Implementation 31 (how volumes change during the hours of a particular day), it may only be necessary to collect counts for a few weeks. Daily patterns (how volumes change by day of the week) may only need counts for a few months. Seasonal patterns (how volumes change by season of the year) are best identified with counts over multiple years. Benchmarking changes over time may require install- ing a counter permanently (or regularly rotating a counter to the site for a sufficient time period). Many count programs with limited resources use short-duration manual counts taken once a year, or less frequently, to document changes in non-motorized volumes over time. Often, these counts have been used to estimate volumes over multiple years. However, very short counts (e.g., 2 hours) at a particular location are subject to high levels of variation, so they may produce inaccurate estimates of annual volumes. Similarly, conclusions about increases or decreases in pedestrian or bicycle activity based on an annual 2-hour count may not be accurate. Although some variations can be accounted for using adjustment factors or models (e.g., to control for major weather conditions) (Nosal, Miranda-Moreno, and Krstulic 2014), others are very difficult to identify and correct. Increasing the count duration can improve the accuracy of extrapolation substantially (Milligan, Poabst, and Montufar 2013; Nordback et al. 2013; Hankey, Lindsey, and Marshall 2014; Nosal, Miranda-Moreno, and Krstulic 2014). Nordback et al. (2013) recommend a minimum count duration of 24 hours but suggest a longer period for more accu- rate annual estimates. In general, recent studies have suggested that counts should be taken for 4 days to approximately 1 week to reduce the error of the annual volume estimate to less than 20% (Nordback et al. 2013; Hankey, Lindsey, and Marshall 2014; Nosal, Miranda-Moreno, and Krstulic 2014). See Section 4.5.1, Temporal Adjustment Factors, for a more detailed discussion of these findings. Counting for 24 hours or longer necessitates using an automated counting device. These studies have also shown that extrapolation errors are generally lower when counts are taken at times with higher activity levels (e.g., summer months rather than winter months in most regions of the United States or sunny days rather than rainy days). In addition, it is more accurate to extrapolate short counts from a single day-of-year factor (i.e., daily volume as a How Long a Count Is Needed? The count duration depends on (1) the purpose of the data collection and (2) the available resources. Developing volume pattern data requires a few weeks (for hourly patterns) to a few months (for daily patterns) up to an entire year (for seasonal or monthly patterns). Extrapolating short-term counts to longer (e.g., annual) time periods can be done with durations as short as 2 hours, but with potentially highly inaccurate results. Recent research shows that 4–7 days of counts are needed to reduce the error in an annual volume estimate to less than 20%. However, these require automated counting devices. If it is not feasible to conduct counts longer than a few hours at a time, the fol- lowing helps minimize the error in the volume estimates: • Count at times with high activity levels (e.g., summer). • Count during good weather. • Conduct several short counts during different time periods. • Extrapolate using a single day-of-year factor, rather than using day-of-week and month-of-year factors.

32 Guidebook on Pedestrian and Bicycle Volume Data Collection percentage of the total annual volume) rather than using a two-step process to first estimate the weekly volume from a day-of-week factor and then estimate the annual volume from a month- of-year factor (Hankey, Lindsey, and Marshall 2014; Nosal, Miranda-Moreno, and Krstulic 2014). Collecting short (less than 24-hour) counts during several different time periods can also improve accuracy. A method for calculating the day-of-year factor is described in Appendix D. Despite their limitations for estimating annual pedestrian or bicycle volumes, collecting short counts at particular locations can be useful for other purposes. For example, counts collected during a single peak hour of activity have been shown to be highly correlated with 12-hour counts on the same day (Hankey et al. 2012). In addition, short counts can also be useful for comparisons over time if they are taken at similar times of day, on similar days of the week, during similar seasons of the year, and under similar weather conditions. Furthermore, taking short counts annually, using a consistent method, at many different locations in a community (e.g., 30 to 50 locations) can be used to track trends over time (see the Washington State DOT case study for an example). Although individual sites may have high variability, the average count across a large set of locations can provide accurate information about trends in walking or bicycling. In addition, counts at many locations can provide useful information about geographic differences in pedestrian or bicycle activity (see the Minneapolis case study). Figure 3-5 illustrates how manual counts at more than 300 locations have been used to create a citywide map of estimated average daily bicycle volumes. Figure 3-5. Minneapolis bicycle count locations and bicyclist estimated daily traffic. Source: Minneapolis Public Works Department (2013).

Data Collection Planning and Implementation 33 Count Frequency Count frequency is defined as how often counts are collected at a given site. For permanently installed automated counters, the count frequency is continuous. At other sites, whether they are counted manually or with temporarily installed automated counters, the frequency might be one to a few times per year, depending on how the data will be used. For motor vehicle counts, the TMG recommends collecting short counts at all locations throughout a roadway system at least once every 6 years. More important roadways in the system should be counted at least once every 3 years. Communities should choose a frequency for pedestrian and bicycle counts that allows these communities to achieve their counting purpose with the available resources. 3.2.4 Considering Available Counting Methods There are various methods available to count pedestrians and bicyclists. This section com- pares the most common counting technologies in U.S. practice at the time of writing, including manual counts, passive infrared, active infrared, pneumatic tubes, inductive loops, piezoelectric sensor, radio beam, and automated video. Specific details about each of these technologies, along with emerging technologies, are presented in Chapter 5. Pedestrian and bicycle data collection programs often use more than one method. For exam- ple, certain counting technologies only count bicycles, while others count people regardless of mode. Collecting separate counts of both pedestrians and bicyclists at a location (e.g., mixed traffic) often requires using combinations of technologies. The TMG (FHWA 2013, pp. 4-3 to 4-5) provides more information about this approach. Agencies also incorporate more than one type of technology into their counting programs because certain technologies can be relatively costly to install, preventing them from being installed at all locations where counts are desired. However, using several different technologies (and potentially working with several different product vendors) can complicate the process of managing and analyzing data. Counting at loca- tions with pedestrian, bicycle, and automobile traffic (e.g., intersections) is even more complex than counting just two modes, and additional research is needed to improve the technologies available for this purpose. Manual Counts vs. Automated Counts Manual counting, performed by human data collectors in the field, is a common data collec- tion method. This was evident from the results of the survey conducted for this project, which showed that most pedestrian and bicycle counts available in many communities were gathered manually. Although automated technologies have improved significantly in recent years, man- ual counts will continue to be used by organizations that lack the financial resources, technical capacity, or regulatory permissions for deploying automated detectors in the public right-of- way. There is a tradeoff between the cost of installing a long-term counting technology to pick up temporal variations in volumes at a particular site and the cost of using short-term counting methods to pick up spatial variation in volumes at sites across a community (see the Minneapolis case study). Manual counts will also be used where there is a compelling interest in documenting behav- iors and other attributes, such as age, gender, helmet use, and use of assistive devices, that are only possible with manual observation (either in the field or by reviewing video) (see the Washington State DOT case study). The accuracy of manual counts can vary greatly depending on many factors, such as the experience of data collectors, the quality and complexity of data collection training and instructions, the layout of data collection forms, and site conditions (Diogenes et al. 2007; Schneider, Arnold, and Ragland 2009b).

34 Guidebook on Pedestrian and Bicycle Volume Data Collection Identifying Sites Appropriate for the Counting Method Specific counting technologies require particular site characteristics. For example, passive infrared sensors should be directed across a sidewalk or multi-use trail facility into a wall or other object. They should not have other traffic in the background that is not using the side- walk or trail, given that other traffic might be detected by the sensor. Inductive loops need to be installed in pavement on a roadway or trail, so they cannot be used on soft-surface trails. Radio beam technology requires placing a transmitter and receiver on opposite sides of a facility, so there need to be walls or posts available on both sides (or the ability to install them) to mount both components. All counting technologies are limited by detection range and user volumes. One of the most important considerations for a site is the desired detection zone. This is the screenline or area where pedestrians or bicyclists will be counted. Each technology has its own type of detection zone, and it is essential to understand which pedestrians or bicyclists at a site will be counted and which will not be counted. For example, inductive loops only count bicyclists who pass over the width of the loops. If a multi-use trail is wider than the installed loop width, bicyclists who ride over a part of the trail not covered by the loops may not be counted. Given that the width covered by inductive loops and pneumatic tubes has some flexibility, the range of potential detection zone widths should be considered during site selection and installation. Many technologies have detection zones intended for screenline locations. That is, pedestri- ans or bicyclists are counted when they pass an imaginary line across a sidewalk, bicycle lane, or multi-use trail. However, organizations may wish to count the number of pedestrians who cross a street within a crosswalk, or the number of bicyclists who go left, straight, or right at an inter- section. These movements are complex. Therefore, with the potential exception of pedestrian detectors being installed where a crosswalk cuts through a wide median island, none of the auto- mated technologies available for counting at screenline locations can be used for counting people using roadway crossings or open spaces. Instead, counts are typically collected at inter sections by observing pedestrian or bicycle movements, or tracks, through an area-based detection zone. If a pedestrian or bicyclist track meets certain criteria (e.g., crosses the roadway centerline within the Questions to Consider when Selecting a Counting Method All types of counting methods can achieve the general purposes of a non- motorized count program listed in Section 3.2.1. However, some methods are more effective for specific purposes. Selecting a counting method involves asking questions, such as: • Who will be counted? (i.e., pedestrians, bicycles, both) • What types of sites will be counted? • What user characteristics (e.g., simple count, gender, behaviors) will be collected? • What is the desired count duration and frequency? • How well does a given counting method perform, given particular count site characteristics? • What resources (e.g., cost, training, procurement lead time, data reduction) are required to use the counting method? • How easy is it to work with the equipment? (e.g., durability, theft resistance, battery life) • Do other agencies have experience using a given counting method or vendor? Source: Tony Hull, Toole Design Group. Bicyclist riding outside the detection zone.

Data Collection Planning and Implementation 35 crosswalk), it is registered as a count. This is typically done intuitively by manual data collectors in the field, or through manual or automated video image processing. Other site characteristics that may affect particular counting technologies include • Peak-hour user volume; • Mix of users (e.g., pedestrians, bicyclists, automobiles); • Facility width (detection zone width); • Facility surface; • Vehicular traffic flows or presence of vehicular traffic; • Trees and other vegetation that could block a sensor or create background interference; • Locations of snow and debris storage that could block a sensor; • Windows that could deflect or distort a sensor beam; • Metal or synthetic surfaces that may experience variation in radiant temperature when exposed to sunlight; • Poles, walls, and other potential objects for mounting devices; • General security (potential for theft or vandalism of devices); • Built or social environment characteristics that create non-standard walking or bicycling movements (doorways, driveways, objects obstructing a travel path, bicycle racks, bus stops, food carts, “hangout” corners, etc.); and • Adjacent land use characteristics that may influence bicycling and walking travel paths (e.g., plazas or parking lots that may encourage short cuts that divert bicycles or pedestrians away from the detection zone). More details about the specific site requirements of particular counting technologies are pre- sented in Chapter 5. General Comparison of Counting Methods Ten common pedestrian and bicycle data collection methods are compared according to key criteria in Tables 3-1 through 3-3. These tables provide a general overview of each technology for the purpose of comparison. Detailed descriptions of each data collection method are given in Chapter 5. The technology labeled as “automated video” uses computer algorithms to identify and count pedestrians and bicyclists from video images and is different than counting pedestri- ans and bicyclists manually from a video monitor. However, existing automated video systems may not use a completely automated counting process. They may also involve manual counts or manual data checks of some automated video processing. The combination of passive infrared and inductive loops is the most common method used to count pedestrians and bicyclists sepa- rately. However, other combinations, such as passive infrared and piezoelectric strips, or passive infrared and pneumatic tubes, could be feasible. Table 3-1 compares user characteristics and site characteristics appropriate for each method. Methods may count pedestrians, bicyclists, or both types of users. Different counting methods are appropriate for particular types of facilities, such as multi-use trails, sidewalks, and bicycle lanes. Certain methods can also identify users’ direction of travel and differentiate between bicycles and motor vehicles. Table 3-2 compares volume, width, and duration capabilities. Maximum user volume ranges are classified into general categories depending on when the counting method tends to start providing less-reliable counts. This user count range also depends on specific site characteristics (e.g., average user group size, mix of pedestrians and bicyclists, and detection zone width). Data used to determine maximum user volume ranges are presented in Chapter 4. Detection zone width ranges are classified into general categories. Most methods have detection zones that cover the width of an entire facility (e.g., sidewalk, bicycle lane, or multi-use trail). However, because

Characterisc Passive Infrared Acve Infrared Pneumac Tubes Inducve Loops Piezoelectric Sensor Passive IR + Inducve Loops Radio Beam (One Frequency) Radio Beam (High/Low Frequency) Automated Video1 Manual Counts2 Type of users counted All facility users Yes Yes Yes Yes Yes Yes Yes Pedestrians only Yes Yes Yes Yes Bicycles only Yes Yes Yes Yes Yes Yes Yes Pedestrians vs. bicycles Yes Yes Yes Yes Bicycles vs. automobiles Yes Yes Yes Yes Characteriscs collected Different user types Yes Yes Yes Yes Direcon of travel3 Yes Yes Yes Yes Yes Yes Yes Yes Yes User characteriscs4 Yes Yes Types of sites counted Mulple use trail segments Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Sidewalk segments Yes Yes Yes Yes Yes Yes Yes Bicycle lane segments Yes Yes Yes Yes Yes Cycle track segments Yes Yes Yes Yes Yes Yes Yes Shared roadway segments Yes Yes Yes Yes Roadway crossings (detect from median)5 Yes Yes Yes Yes Yes Yes Yes Yes Roadway crossings (detect from end of crosswalk) Yes Yes Intersecons (idenfy turning movements) Yes Notes: (1) Exisng “automated video” systems may not use a completely automated counng process; they may also incorporate manual data checks of automated video processing. (2) Includes manual counts from video images. (3) Technologies noted as “Yes” have at least one vendor that uses the technology to capture direconality. (4) User characteriscs include esmated age, gender, helmet use, use of wheelchair or other assisve device, pedestrian and bicyclist behaviors, and other characteriscs. (5) Roadway crossings at medians potenally have issues with overcounng due to people waing in the median. Median locaons were not tested during this project. Table 3-1. Comparison of common pedestrian and bicycle counting methods: user characteristics and site characteristics.

Table 3-2. Comparison of common pedestrian and bicycle counting methods: volume, width, and duration capabilities. Characterisc Passive Infrared Acve Infrared Pneumac Tubes Inducve Loops Piezoelectric Sensor Passive IR + Inducve Loops Radio Beam (One Frequency) Radio Beam (High/Low Frequency) Automated Video1 Manual Counts2 User volume3 ++ +++ ++ ++ + ++ ++ ++ +++ ++ Detecon zone width4 ++ +++ + ++ ++ ++ ++ + +++ +++ Count duraon5 +++ +++ ++ +++ +++ +++ +++ +++ + + Notes: (1) Exisng “automated video” systems may not use a completely automated counng process; they may also incorporate manual data checks of automated video processing. (2) Includes manual counts from video images. (3) +: provides consistent counts (although some accuracy adjustment may be necessary) up to approximately 200 users per hour, ++: up to 600 users per hour, +++: beyond 600 users per hour. These are approximate ranges under typical condions. The range also depends on specific site characteriscs (e.g., average user group size, mix of pedestrians and bicyclists, detecon zone width). The maximum user volume range for manual counts assumes a single data collector is counng one type of user and no addional characteriscs. Mulple manual data collectors can count more than 600 users per hour. Counts can be adjusted at user volumes above these levels. (4) +: typical detecon zone width narrower than 4 meters (13 feet), ++: narrower than 6 meters (20 feet), +++: 6 meters (20 feet) or wider. In the case of automated video and manual counts, the detecon width may be 25 meters (82 feet) or wider. (5) +: typically used for 48 hours or less, ++: typically used for non-permanent short- or longer-term counts, +++: oen used for permanent count sites. Most inducve loops are installed in the pavement, but there are also variees that can be installed on top of the pavement for up to 6 months.

Table 3-3. Comparison of common pedestrian and bicycle counting methods: resources. Characterisc Passive Infrared Acve Infrared Pneumac Tubes Inducve Loops Piezoelectric Sensor Passive IR + Inducve Loops Radio Beam (One Frequency) Radio Beam (High/Low Frequency) Automated Video1 Manual Counts2 Equipment cost3 $$ $$$ $$ $$ $$ $$$ $$$ $$$ $$ $ Preparaon cost4 $$ $$ $$ $$$ $$$ $$$ $$ $$ $$ $ Installaon me5 N/A Hourly cost6 $ $ $$ $ $ $ $ $ $$$ $$$$ Data collector training me7 Mobility8 +++ ++ +++ ++ ++ +++ +++ Pavement cuts No No No Yes Yes Yes No No No No Notes: N/A: not applicable This table presents generalized informaon specific to parcular counng technologies. Other aspects of counng products, such as baery life and communicaon interfaces, are also important to consider but are highly vendor-specific. See the text following this exhibit for more details. See Chapter 5 for specific details (e.g., typical costs) related to each technology. (1) Exisng “automated video” systems may not use a completely automated counng process; they may also incorporate manual data checks of automated video processing. (2) Includes manual counts from video images. (3) $: equipment (not including perming and installaon) typically cost less than $1,000 as of 2013, $$: typically costs between $1,000 and $3,000, $$$: typically costs more than $3,000. The cost of most counng technologies is subject to economies of scale, so the per site cost can be reduced by purchasing more counters. (4) Fewer dollar signs ($) indicate that it takes less me (and therefore fewer financial resources) to find an appropriate site and to obtain any required permits to install the counng product. Preparaon can range from less than one day for manual counts to several months for technologies with more restricve installaon requirements. (5) More clocks () are given to methods that require more installaon me (e.g., cut pavement, secure the data logger, test and adjust the equipment). Installaon can range from no me for manual counts and less than 30 minutes for passive infrared to more than half a day for inducve loops. (6) More dollar signs ($) indicate that the method is more costly for an average hour of counts, given the typical count duraon for a parcular method. These costs can range from a few cents per hour for automated technologies (the full equipment, preparaon, and installaon cost is spread across months of counts) to more than $50 per hour for manual counts (including training preparaon me, management, and on-site labor costs). (7) More clocks () indicate that more me is needed to prepare field data collectors to implement the counng method. A single data collector can be trained how to install or download data from a parcular automated technology in less than 30 minutes, but it oen takes more than one hour to thoroughly train data collectors to collect accurate manual counts. (8) More pluses (++) indicate that a counng technology is easier to move aer it has been installed. A minus sign (−) indicates that the technology is generally not intended to be used in more than one locaon based on the installaon being permanent.

Data Collection Planning and Implementation 39 inductive loops have specific width requirements, they may not cover the entire facility (in the case of a multi-use trail) or they may cover a greater width than the facility (in the case of a bicycle lane). Count duration indicates the typical length of time that the counting methods are employed to collect data. In this table, methods typically used for less than 48 hours are classi- fied as “short-term.” In particular, manual counts are commonly done for 2 hours and are rarely done for more than 8 to 12 hours. Methods often used to collect continuous counts for 1 year or more are classified as “long-term.” Table 3-3 compares start-up, data collection, and data analysis resources. Similar guidance is provided in the TMG (FHWA 2013, Table 4.1). Resource considerations include technology cost, preparation cost, installation cost, hourly cost, and data collector training cost. Mobility (portability among multiple sites) and pavement cuts also affect the resources needed for counts. Resource Considerations All counting methods require resources, such as money and time. This section provides addi- tional details about the resource categories listed in Table 3-3. Although the amount of resources required for most of these categories depends on the products and services offered by particular vendors, the resources required for each method generally have a similar magnitude, regardless of vendor. • Equipment cost. This cost represents the cost of equipment, not including preparation and installation. In general, for short-term counts, automated technologies are more expensive on a per site basis than manual counts because the equipment cost is higher than the cost of several hours of labor. The least expensive automated counting technologies are pneumatic tubes, inductive loops, and piezoelectric sensors. The cost of most counting technologies is subject to economies of scale, so the cost of equipment per site can be reduced by purchasing more counters. • Preparation. Preparation costs include the time required to identify an appropriate site, apply for an installation permit, and purchase a permit (if needed). Technologies that have more flexibility in their installation requirements tend to have lower preparation costs. In general, it is possible to count at nearly all potential sites using manual counts. In contrast, methods such as radio beam technology require installing a transmitter and receiver at an appropriate height in places that are free of obstructions between these components. • Installation. Installation cost includes time and budget (1) required to cut pavement; (2) secure equipment to install poles, posts, and boxes; and (3) test and adjust the sensor. Installation time considers the typical time required to install equipment at a site. In general, the longest installation times for counting products on the market are for inductive loops and piezoelectric sensors, because they require pavement cuts. Other technologies with more parts (e.g., transmitter and receiver), such as radio beam, also tend to have longer installation times. • Hourly cost. The cost per hour for manual data collection is higher than most other continu- ous counting methods. This is because once automated counting equipment is installed, it can count continuously for weeks, months, or longer periods. In contrast, the cost of labor for long-term, continuous manual counts is prohibitively expensive. The automated video count services available on the market at the time of writing are priced by the hour, so they also have a high hourly cost relative to other automated counting technologies. • Data collector training. Training costs include the amount of staff management time and labor time required to prepare field data collectors to implement a counting method. In general, automated counting technologies require relatively little data collector training; once a data collector knows how to install or download data from a particular product, he or she can implement it efficiently. Manual counts tend to require extensive data collector training (often involving many data collectors) to obtain accurate counts.

40 Guidebook on Pedestrian and Bicycle Volume Data Collection • Mobility. Moving counting equipment to different sites makes it possible to capture a greater spatial variety of data with a single device. Using mobile counting methods enables agencies to reduce the number of individual equipment sets they need to purchase. Data collectors can move to any location easily, so manual counts have high mobility. Other methods have varying degrees of portability. Some devices can be easily unfastened or removed from poles; devices installed in pavement are very difficult to move or are not reusable; and other product components (e.g., the box with the data logger) may often be movable and reusable. • Pavement cuts. In general, counting methods that require pavement cuts tend to take more time to obtain permits and cost more to install. Costs can be reduced if agencies have the in-house expertise to make pavement cuts and install equipment. These considerations are included in the preparation and installation characteristics above, but it is useful for agencies to understand explicitly which technologies require pavement cuts. The need for pavement cuts also requires disrupting facility traffic for several hours or even a day while the installation (or replacement, when the loops eventually fail) occurs. There are many other important resource considerations when planning a count program. However, most of these resources depend more on the product and services offered by a particu- lar vendor than on the counting method itself. These resource considerations include • Testing and adjustment. Most automated counting technologies require testing and some adjustment in the field to optimize their accuracy and consistency. Vendors provide differ- ing levels of customer service to support testing and adjusting their product after it has been installed. For manual counts, testing and adjustment involves monitoring data collector accuracy. • Durability and theft-resistance. The containers used to house counting products, the materials used to seal counting technologies in the pavement, and the systems used to mount counting products onto structures affect overall durability and theft-resistance. Vendors use different materials and systems for each product. The product warranty and other agencies’ experiences with product reliability and lifespan are important to consider. • Expected product use life. Like all materials, counting equipment deteriorates over time. For example, although passive infrared sensors housed in boxes may last longer than 5 years, pneumatic tubes in heavy mixed traffic may need to be replaced in less than 1 month (e.g., one vendor recommends replacing tubes after 300,000 automobile hits). All automated counting devices should be checked regularly to ensure they are counting accurately. Equipment must be kept in a state of good repair, which depends on how devices are housed and installed, traf- fic conditions, weather and other site characteristics, and vendor maintenance services. • Battery life. Battery life affects the frequency of maintenance visits to a given count site and battery replacement costs. Vendors use different types of batteries to power their technolo- gies. These batteries can last from a few weeks to several years, and newer technologies tend to have improved power capability. For example, many use lithium battery packs that can have a lifespan of 1 to 2 years or more based on activity levels and device efficiency. Battery life also depends on battery manufacturers, which is especially important for devices that do not include a vendor-supplied power unit. In contrast, some products are hardwired, which require a site with electrical connections, but eliminate the need for batteries and battery- charge site visits. Some products also use solar power to charge the battery. • Data storage capacity. The data logger’s storage capacity determines how frequently data need to be downloaded from the count site. This timeframe also depends on the data logging technology, the volume of users at the site, and a product’s user interface. Storage capacities can range from less than 1 month to multiple years. For most devices, the interval between downloads can be lengthened by adjusting the data logger’s settings for how often events are recorded. For example, changing from 15-minute count intervals to hourly intervals can

Data Collection Planning and Implementation 41 reduce the amount of memory needed to store counts each day. (This is not possible when a product can only record counts in timestamp format.) • Downloading capability. Ease of downloading reflects how long it takes a person to gather data from each count site and depends on the product’s data logging technology and user interface. Some products require data to be downloaded on site using a portable storage device, such as a laptop or tablet; other products allow data to be downloaded remotely, usually requiring cellular transmission capacity that may entail monthly fees for service. Real-time data collec- tion systems are very rare. The products reviewed during this research provided data that were analyzed through downloading and post-processing. The combination of data storage capacity and downloading capability can dramatically affect costs: if frequent in-person downloads are required and particularly if counters are in many different locations, gathering data can take a substantial amount of time. Many agencies with large-scale automated counting programs interviewed for this research have, or are moving to, a remote data downloading capability. • Database creation. Creating a database from raw count numbers takes time. Counting prod- ucts can compile data automatically into a database that shows counts by time period (or time stamp). Manual counts done on paper must be compiled by entering hand-written informa- tion into a database; manual counts performed using a counting device or mobile application can be compiled automatically. • File format and count interval options. Converting data into a usable format also takes time. Vendors record data in various file formats. Some vendors offer data recording equipment that outputs time-stamped counts; others offer equipment that outputs the total count over a specific interval, such as 15 minutes or 1 hour. The specifics depend on the data logging technology and user interface developed by a particular vendor. Some products require the vendor’s own software to read and export the data to other formats. At the time of writing, few products exported data in the format specified by the FHWA Office of Highway Policy Information Travel Monitoring Analysis System (TMAS) 3.0. However, vendors may provide this option in the future so that it will be easy to compare data collected by different products in different communities. • Data cleaning. Cleaning data refers to removing incorrect counts or missing data (i.e., counts that do not represent what is supposed to be measured) from a count database. The time required to clean a database depends on how often anomalies are in the count pattern at a particular site. Anomalies may be related to the type of technology as well as specific site char- acteristics (e.g., large platoons of pedestrians or bicyclists, people tampering with the counter, objects temporarily blocking a sensor) (see Section 3.3.9). Time to train an analyst in data cleaning is also a cost to consider. When assessing locations where permanent counters are being considered, it can be helpful to initially collect volumes using short-term counting methods (e.g., manual counts or pneumatic tubes). Performing a test count can help confirm that volumes are roughly as expected and that a permanent counter would prove valuable. Short-term term counts can provide a better sense of travel patterns and help in stratification of sites. Count Accuracy and Consistency Count data accuracy and consistency are also important to consider when selecting a tech- nology. Data accuracy reflects the magnitude of the difference between the count produced by the technology and the actual (“ground truth”) count gathered manually from video data (or gathered using some other means of obtaining a precise estimate of the actual count). This dif- ference typically depends on user volumes, movement patterns, traffic mix, and environmental characteristics. The most common source of inaccuracy is occlusion, or undercounting, which occurs when multiple users pass a detector at the same time and the sensor fails to register more

42 Guidebook on Pedestrian and Bicycle Volume Data Collection than one count. Known sources of inaccuracy can be corrected, provided that the device errors are consistent. Data consistency reflects the remaining variability in the count data after being corrected for expected under- or overcounting, given specific conditions. This variability typi- cally depends on the counting technology itself, how a specific vendor uses the technology in a particular product, and the quality of the installation. Data accuracy and consistency for specific technologies are discussed in detail in Chapter 4. Emerging Technologies Counting technologies will continue to develop so that they are easier to use and perform better at a lower cost. Although the technologies presented in Tables 3-1 through 3-3 were com- monly available at the time of this report, practitioners should also consider emerging technolo- gies. This project’s research suggests that thermal, ultrasonic, and fiberoptic sensors may become more common counting technologies on the U.S. commercial market in the coming years. Other technologies that track user movements, such as GPS and WiFi/Bluetooth technologies, may be crowd-sourced to estimate counts at specific locations. 3.3 Implementing the Count Program After planning is completed, the count program can be implemented. Implementation typi- cally involves the following steps: • Obtaining permissions, • Procuring counting devices, • Making an inventory and preparing devices, • Training staff, • Installing and validating devices, • Calibrating devices, • Maintaining devices, • Managing count data, • Cleaning count data, and • Applying count data. The first eight bullets (through “managing count data”) are described in Sections 3.3.1 through 3.3.8, respectively. These sections cover the process of obtaining data from automated counters, from permission to installation to downloading raw data. The remaining subsections of Sec- tion 3.3 and subsequent material in Chapter 4 cover three important types of data adjustments used in traffic monitoring: (1) cleaning erroneous data, (2) correcting for systematic errors gen- erated by the counting method, and (3) expanding counts from shorter time periods to estimate pedestrian and bicycle activity over longer time periods. Cleaning erroneous data involves identifying when a count is likely to represent a time period when an automated counter was not observing the intended pedestrian or bicycle movements. Data from these time periods are discarded or replaced by better estimates. This concept is cov- ered in Section 3.3.9. Correcting for systematic errors involves applying a correction function that removes the expected amount of under- or overcounting for a particular counting technology (often due to occlusion). This concept is first discussed in Section 3.3.9, but it is the main topic in Chapter 4. Expanding short counts to longer time periods involves applying an expansion factor to a count collected for a short time period. The expansion factor is based on the type of activity pat- tern assumed to exist at the site, and it can account for the effects of weather on pedestrian and bicycle volumes. This concept is also covered in Chapter 4.

Data Collection Planning and Implementation 43 3.3.1 Obtaining Permissions Before installing counting devices at a site, check with local agencies, utilities, and other orga- nizations responsible for managing poles, signs, pavement, walls, or other features at the site. In many cases, permission is required. Permission may be given informally by email or letter, but often requires obtaining an official permit and/or posting a bond (e.g., roadway alteration per- mit, right-of-way use permit, and utility permit). Many different types of agencies grant permis- sion (e.g., public works departments, parks departments, and utilities), and it may be necessary Permitting Considerations The permitting process can be an unanticipated challenge to installing auto- mated counting equipment, especially for organizations or agencies that do not maintain the public right-of-way where devices will be installed. Therefore, it is important to anticipate permitting needs early in the process of planning a count program. The following actions may help make the permitting process easier: • Identify all organizations that own the right-of-way, poles, and buildings where equipment may be installed. Multiple organizations may need to provide permission. • Ask contacts at the organization(s) with jurisdiction over the site where equip- ment will be installed whether or not a permit is necessary. Some temporary devices may not require a permit; some installations can be approved informally by e-mail. • Understand exactly how the device will be installed, including specifications such as mounting devices and pavement cuts. This can help make communica- tion efficient and reduce approval time. Working with Other Agencies Sometimes an agency wants to perform counts on a pedestrian or bicycle facility owned by another agency. In these cases, inter-agency coordination becomes especially important. The Alameda County (California) Transportation Commission has worked with the East Bay Regional Park District to install counters when trails are constructed, as it is more cost-effective than installing them later. Arlington County, Virginia operates counters on multiple-use paths owned by the National Park Service, a regional park authority, and the District of Columbia DOT (the Key Bridge), and planned to soon have a counter on state DOT right-of-way. These counters typically involved filing some sort of permit, although the counter in the regional park was covered by an existing memorandum of understanding. Keep in mind that the partner agency’s permitting procedures may not have been developed with counters in mind. Developing relationships with partner agency staff and agreeing to share data can help smooth the process of getting the counters installed.

44 Guidebook on Pedestrian and Bicycle Volume Data Collection to obtain permission from more than one agency at a single location. In most cases, equipment is installed on public property, but permission from the property owner(s) will also be necessary if equipment will located on private property (e.g., a building wall). Ask for permission early in the process. Sometimes the permitting process can be so onerous that it is worthwhile to select a different count site. Use of video technology, either for counting or for validating other technologies, may be problematic in locations where privacy is a concern, and may not be permissible in some jurisdictions. 3.3.2 Procuring Counting Devices Choosing good equipment and a good vendor is important when using automated technology as part of a count program. The sensor technology itself is supported by other equipment, such as mounting devices and data loggers, that is also essential for the success of a counting product, and the vendor’s customer service record is important to consider. The Transportation Research Board does not endorse particular products or vendors. How- ever, based on the product testing conducted during the development of this guidebook, the following is a list of recommended questions that potential customers should ask vendors when selecting equipment. Some of these questions may help inform the selection decision; others will help in setting expectations and planning for the eventual installation. Agencies may wish to include some of the questions below in a request for proposals (RFP) when soliciting vendors to provide counting equipment. Providing specific questions and expectations in an RFP can help agencies obtain the best possible counting technology for their needs. Several of the issues behind these questions are discussed in more detail in subsequent sections. • Out-of-the-box readiness. In what form will the product be shipped? Will it require some assembly in an office or workshop before it is taken to the field? Will the device arrive mounted in an anti-theft box? After an order is placed, how long will it take to ship the equipment to the client? • Additional needs for the product to function. Is any extra equipment (e.g., wrenches, screws, fastening devices, batteries, or an electrical connection) required to install the product that is not included when it is shipped? What extra equipment is required to communicate with the device (e.g., SIM cards, smartphones, and tablets)? Are extra services required (e.g., a cellular data subscription or a service agreement with the vendor for downloading and processing data from the product)? • Warranty/expected use life. Does the equipment have a warranty? If so, what is its length? If there is no warranty, how long is the equipment expected to continue counting accurately? Most products tested for this study did not include a warranty, but vendors were willing to help with installation and troubleshooting. While there is little research on the accuracy of counting devices over long periods of time, the accuracy of pneumatic tubes and inductive loops may decline with age if these technologies are not maintained properly. For pneumatic tubes, proper maintenance includes replacing the rubber tube component of the equipment relatively frequently, especially when installed in mixed traffic. The frequency of tube replace- ment depends on vehicular traffic intensity; one vendor recommends replacing tubes after 300,000 automobile hits. For bicycle-specific inductive loops, proper maintenance includes replacing the inductive loop component occasionally (e.g., every few years). • Site specifications. What specific site or environmental conditions are required for installing the device so that it counts correctly? Some devices need to be installed at a particular height above the ground, pointed at a wall, or installed on a facility that has a specific maximum width. These requirements are typically set by the device manufacturer based on their testing and experience.

Data Collection Planning and Implementation 45 • Performance factors. What specific site or environmental conditions must be avoided so that the device counts correctly? For example, some devices are affected by moving branches, win- dows in the background, electrical lines, or other factors. • Installation complexity. How much time does it take to install the product and calibrate the device in the field? This includes getting the device to perform its basic function of registering counts. However, it also includes making sure the device is recording pedestrians and bicyclists as intended and making necessary adjustments (including modifying device settings for place- ment, sensor height, sensitivity, and other factors that may affect user detection) to ensure the data collection is reliable. • Contractor installations. Will it be necessary to hire a contractor to install the device? Some devices require more technical skills and electronics knowledge than others. Some devices require more disruption at the installation site than others. Devices that require pavement cuts or an installation at a height above the normal reach of a person are more likely to require a contractor. • Device security/durability. How secure will the device be after it is installed? How often does the vendor receive reports of stolen devices or replacement of vandalized units? Are any extra security options available for sites that may have a higher risk of theft or vandalism? • Purchase versus lease. Is there an option to rent or lease the equipment? If equipment can be rented, when does it need to be returned, and what is an acceptable condition of the equip- ment at the time it is returned? Is a “rent-to-own” or trial period purchasing option available? • Data downloads. What options are available for downloading the data? Some products require agency staff to go to the site and physically connect a laptop to the counter to download data. Other products have Bluetooth capabilities that allow agency staff to download data directly to a mobile device or computer. Some vendors can also capture data from their own product using cellular data technology and post it to a cloud storage space where it can be accessed by their clients. • Data formatting and compatibility. What data formats are used? Most vendors make it possible to download data to a spreadsheet (e.g.,.csv,.xls,.dbf). Some vendors also provide a propri- etary format that can be read by their own analysis software. The format of the count database itself can also vary from individual, time-stamped counts to the total number of users detected over a specific time period (e.g., 15 minutes or 1 hour). Can the data be exported in FHWA’s specified format (discussed in Section 3.3.9)? • Performance history. Has the equipment been used elsewhere, and has it gotten positive reviews? After going through the rest of the questions, it is helpful to contact other organiza- tions to ask for their personal experiences with the equipment and vendor. It is also recommended that potential customers ask the following general questions about the services that a vendor provides: • Installation support. Will a vendor representative be on site on the day the equipment is installed? If no representative can be present, will someone be available by phone to answer questions? Some vendors may expect a local agency to do the installation themselves. Other vendors may ask to install their own device so that it meets their standards. Having a vendor representative on site will likely affect the overall installation cost (either billed as a specific item or incorporated into the product cost). • Ongoing customer support. Is routine customer service included in the purchase price? Is routine customer service available, if it is not included with the purchase price? What is the general level of customer service provided (e.g., phone hotline, direct phone number and email of the vendor representative, site visits to fix problems with devices)? In some cases, purchasers may be expected to deal with problems and maintenance on their own after prod- ucts are installed.

46 Guidebook on Pedestrian and Bicycle Volume Data Collection • Calibration support. Will a vendor representative be on site to help calibrate the counting device when it is first installed? Will a vendor representative be able to assist with recalibrating the device in the future? How difficult is recalibrating a device, and what specific modifica- tions need to be made in the field in order to recalibrate? • Hardware and software upgrades. Does the vendor regularly upgrade the equipment or soft- ware associated with the product, and if so, are the current equipment purchases compatible and supported when upgrades occur? Are there costs or licensing fees associated with these upgrades? 3.3.3 Inventorying and Preparing Devices Equipment should be inventoried soon after it is received from a vendor. An inventory will document whether or not all of the expected equipment has been delivered. In addition, an inventory is useful for identifying any additional tools or supporting equipment that may be necessary to obtain in advance of field installation (e.g., wrenches, screws, fastening devices, and batteries). Preparing the counting equipment should be done in the office or shop, where the equipment can be laid out in an organized manner. Preparations include assembling specific components and testing to make sure that batteries, loggers, and other electronic components work. Advance preparations make the actual field installation effort much more efficient than opening the equipment boxes for the first time at the data collection site. The following is a checklist for preparing equipment prior to installation in the field: EQUIPMENT PREPARATION CHECKLIST Take pictures of the equipment immediately aer opening the boxes. Inventory the equipment received. Compare the equipment received to the product’s parts list. Make a list of the main pieces of hardware included in the shipment. Create a database that lists each counter. This informaon will help track the history of each counter, regardless of where it is installed or moved. This step can be especially helpful when compiling historical data from the device and when communicang with the equipment vendor. Equipment data: serial number, date of manufacture (if available), date of arrival. Installaon data: Date of installaon, dates of moves, locaon informaon (e.g., latude and longitude, site descripon). Review the full installa‚on instruc‚ons. Contact the vendor to clarify any installaon steps that are unclear. Idenfy any hardware and tools not provided with the product that will need to be obtained prior to installaon. Obtain any addional hardware or tools required for installaon. If necessary to communicate with the device, obtain a SIM card and set up a cellular data plan for the device. Label equipment with contact informa‚on. This provides informaon to cizens and police who may be concerned about an unknown device in a public space and will aid recovery in the event that the counter is removed or stolen. 3.3.4 Training Staff Staff training is important for both automated and manual counting, although the kind of training involved is very different for the two types of counting. Automated Counting Staff training for automated counting devices focuses on making sure the device is working properly and on downloading data from the device. Many organizations use more than one per- son to monitor their counters, so it is essential for all staff to be able to monitor and consistently Source: Tony Hull, Toole Design Group. Example of unpacking a counting product to inventory its contents.

Data Collection Planning and Implementation 47 adjust all of the equipment. The following is a checklist of things to consider when visiting auto- mated counters in the field (not all items are applicable to all counting products): EQUIPMENT MONITORING CHECKLIST Sensor height. Is the sensor sll mounted at the correct height? Sensor direcon. Is the sensor sll pointed in the correct direcon? Clock. Is the device’s clock set correctly? Cleaning. Remove dirt, mud, water, or other material that may affect the sensor or other vital components. Baery. Is the remaining ba
ery life sufficient to last unl the next scheduled visit? Obstrucons. For example, is vegetaon growing too close to the device? Unancipated site problems. For example, is the pole being used for bicycle parking, or are people congregang in the area (as opposed to walking past the counter)? Pedestrian or bicycle detecon. Are pedestrians or bicyclists passing through the counter’s detecon zone being counted? If not, can the counter’s sensivity be adjusted in the field, or does it need to be removed for repairs? Download data. Use the same export opon consistently to ease the data management burden back in the office. Securement. Are the installaon elements and locking devices sll secure and durable? Poorly secured or loose fi
ed devices are more vulnerable to the… and vandalism. Staff training increases efficiency, because a data collector who forgets one of these steps may need to return to a site more than once. In addition, well-trained staff members can detect exist- ing or imminent equipment problems while they are downloading data. Manual Counting To obtain the most accurate and consistent manual counts, data collector training is essen- tial. The benefits of manual count training apply both to people who are counting in the field (e.g., using clipboards, clickers, or mobile device counting applications) and to people who are tak- ing counts from video recordings. Training should help data collectors understand the following: • The overall purpose of the counting effort. Data collectors who understand how the data will ultimately be used are more likely to concentrate and take the job seriously. • Definitions of “pedestrian” and “bicyclist” (or “bicycle”). Make sure the data collectors under- stand the definitions of pedestrian and bicyclist used in the community. Tricky aspects of these definitions include skateboarders (typically counted as pedestrians), babies being carried or pushed in a stroller (typically counted as pedestrians), people walking their bicycles (typi- cally counted as pedestrians), dogs on leash (not typically counted, but their owners are), and bicyclists on a tandem bicycle (typically counted as two bicyclists but one bicycle). • Exactly when a person should be counted. For trails or roadway segments, pedestrians and bicy- clists are typically counted when they pass an imaginary line from either direction. At intersec- tions, some methods count pedestrians only when they cross the street (specifically, when they pass the centerline of the roadway being crossed). In this case, pedestrians who turn right or left at a corner but do not cross the street are not counted. Furthermore, should pedestrians be counted at an intersection if they cross outside the crosswalk lines? Some methods specify that all pedestrians crossing within 50 feet of the crosswalk lines should be counted. Finally, some methods count bicyclists when they arrive at an intersection, while others do not count a bicyclist until he or she goes left, straight, or right—an important consideration to correctly classify bicyclists who dismount and walk their bicycle after arriving at an intersection. • Priority of characteristics to count. While most automated technologies register a single count each time a pedestrian or bicyclist enters the counter’s detection zone, manual data collectors can observe pedestrian and bicyclist characteristics and behaviors. Manual observers may be asked to also document age, gender, helmet use, assistive-device use, turning movements, or behavior each time a pedestrian or bicyclist is counted. This information can provide a rich set of data for Source: Lindsay Arnold, UC Berkeley Safe Transportation Research & Education Center. Downloading data from an automated counter in the field. Source: Robert Schneider, UC Berkeley Safe Transportation Research & Education Center. Field data collector training.

48 Guidebook on Pedestrian and Bicycle Volume Data Collection analysis. However, collecting more characteristics and behavioral data increases the complexity of the data collection effort and can diminish counting accuracy, especially in locations with high pedestrian and bicycle volumes. Therefore, it is important for the data collector to know which characteristics are the most important to document. The highest priority should be to col- lect volume by mode (i.e., get the total count right). It may be necessary to use additional observ- ers to document pedestrian and bicyclist characteristics and to count complex sites effectively. If volunteers are used for manual counts in the field, training is essential. In addition to com- municating all of the key information, above, it is also important for data collection managers to recognize volunteers who should not participate in counting activities. During training, it is important for the count manager to evaluate the capacity of each volunteer for conducting the count. It may be necessary to exclude volunteers who may not be reliable for showing up at a particular count location and time or cannot demonstrate effective capability to follow the count protocol. The Washington State DOT case study provides an example of working with volunteers. Field data collectors should be trained to be aware of their surroundings and to be careful in and around traffic and should bring a letter or some other form of official documentation that describes the counting effort. (See Appendix B for an example of information provided at a manual data collection training session.) Having a letter on hand can help the data collector avoid being distracted from their task by anybody coming up and asking questions. Finally, when planning manual counts, it is important to identify when a site may need more than one data collector. In general, locations with higher user volumes or a greater mix of pedes- trians and bicyclists require more data collectors. 3.3.5 Installing and Validating Equipment Equipment installation can be one of the most challenging steps in the data collection process. Count managers should budget significant time for installation to ensure that it is done correctly. The installation process involves everything from obtaining necessary permits to planning and scheduling an equipment installation date to verifying that the equipment continues to work several weeks after initial installation. Unless an organization has significant previous experi- ence with the equipment, it is important to work closely with the equipment vendor during the installation process. The following is a checklist of the steps involved in preparation for an installation (not all steps apply to every counting technology or product): INSTALLATION CHECKLIST: ADVANCE PREPARATION Site visit to idenfy the specific installaon locaon. Specifically, note poles that will be used, where pavement will be cut, or where ulity boxes will be installed to house electronics. Verify that no potenal obstrucons (e.g., vegetaon) or sources of interference (e.g., doorway, bus stop, and bicycle rack) are present. Obtain and document necessary permissions. Permits or permissions may include right-of-way encroachment permits, pavement cung permits or bonds, landscaping permits, or interagency agreements. Obtaining these permissions may take up to several months, parcularly if other agencies are involved. Create a site plan. Develop a detailed diagram of the planned installaon on an aerial photo or ground-level image documenng the intended equipment installaon locaons and ancipated detecon zone (a€er installaon this will be useful for validang equipment either visually or with video monitoring). This diagram may be useful for obtaining installaon permissions and working with contractors. Figure 3-6 provides an example site plan. Arrange an on-site coordinaon meeng involving all necessary pares (e.g., staff represenng the organizaon installing the counter, perming staff, contractors). If possible, a vendor representave should be on hand or available by phone. It may take several weeks to find a suitable me when everyone is available. Check for potenal problems. Problems with the site may include interference from ulity wires, upcoming construcons projects, hills, sharp curves, nearby illicit acvity, and nearby insect and animal acvity. Some of these condions can be idenfied from imagery, but they should also be evaluated in the field. Hire a contractor if necessary (or schedule appropriate resources from within the organizaon). Source: Tony Hull, Toole Design Group. Technician initiating device for field deployment.

Data Collection Planning and Implementation 49 The following steps are recommended upon arriving at the site on the installation day: INSTALLATION CHECKLIST: ARRIVAL AT THE SITE ON THE INSTALLATION DAY Review the site with the vendor and other pares to verify there are no potenal problems with the site (e.g., interference from ulity wires, evidence of planned construcon, frequent obstrucons [e.g., delivery trucks] in the installaon area). Prepare the site. Perform any maintenance or preparaons for the installaon, such as clearing vegetaon or sweeping pavement surfaces where inducve loops or pneumac tubes will be installed. Record detailed notes on any aspects of the site menoned by the vendor as potenal issues that could affect accuracy. Take a picture of the site before the counter is installed. The following steps are recommended during the actual installation: INSTALLATION CHECKLIST: COUNTER INSTALLATION Maintain a safe work zone. If the installaon requires working within or disrupng the traveled way, be sure to establish a work zone, including required signs and detours if needed to avoid creang a safety hazard for the installaon team or passers-by. Install the counter according to vendor specifica­ons. Document any deviaon from the specificaons (e.g., difference in mounng height due to site constraints). Record detailed notes on any difficules with the installaon—this informaon may make future installaons go more smoothly. Take pictures during installa­on. Acon photos (e.g., cu­ng pavement, securing equipment to poles, installing ba€eries) are useful for documenng that the correct steps were followed. They are also useful for reports and presentaons. Sync the device’s clock with the actual ­me. The actual me can be obtained from many sources; for example, most smartphones regularly sync with the actual me. Verify that the device is working and recording data correctly. Ideally, this acvity will be done while the vendor is present. It may include watching counts register on the device, or taking manual counts for 15–60 minutes that can be compared with data downloaded from the device. If the test count is not sufficiently accurate, calibrate the device if possible by adjusng the sensor’s sensivity and repeating the previous step. Once the device has been installed and appears to be working as intended, the following steps are recommended prior to leaving the site on the installation day: INSTALLATION CHECKLIST: POST-INSTALLATION Take a close-up picture of the device. Consider collecng a GPS point to document the exact coordinates where the device is installed. Take pictures of the device vicinity. Take at least one picture each from the front in the direcon of travel, from the back in the direcon of travel, and perpendicular to the direcon of travel. Take a picture depicng the counter’s detecon zone. In the picture, have the vendor (or another expert) indicate exactly where the detecon zone should be, using chalk, paint, etc. This picture helps when comparing video or manual ground truth counts with the device’s counts, when assessing the device’s accuracy. Source: Tony Hull, Toole Design Group. Example picture documenting detection direction at a multi-use trail site. Example of a picture documenting the installation of inductive loops. Source: Ciara Schlichting, Toole Design Group.

50 Guidebook on Pedestrian and Bicycle Volume Data Collection Once the counter has been installed, it is important to follow up periodically to ensure that the counter is still working appropriately. This activity includes the following actions: INSTALLATION CHECKLIST: FOLLOW-UP ACTIVITIES Create a site descripon sheet or diagram containing the notes and photos from the installaon day. Revisit the site aer a couple of days to download data, to check that the recorded volume paerns seem reasonable. It is important to catch any systemac problems with the counter or site condions right away. Revisit the site at least every 3 months—sooner if required for baery replacement or data downloads—to make sure the device is sll working. This step is not necessary for temporary installaons. Monitor count data and pa…erns rounely to idenfy any significant anomalies or deviaons that could suggest an equipment malfuncon. It is advisable to conduct 1–2 hour manual validaon counts annually, or as needed based on data anomalies. This step is not necessary for temporary installaons. An important step in the equipment installation process is validation: determining whether or not a device is working properly. Validation involves testing the device both on the instal- lation day and several days after installation. Immediately after a device is installed, the instal- lation team should check that pedestrian or bicyclists are being detected and recorded. This check involves taking manual counts of all pedestrians or bicyclists who pass the detection zone during an initial test period (typically 15 minutes to 1 hour). These manual counts are compared with the total count shown on the device’s data logger or the total count downloaded from the device. Figure 3-6. Example equipment installation plan at a multi-use trail site. Source: Tony Hull, Toole Design Group. Validation of pneumatic tube (roadway) and passive infrared sensor (on pole) during installation. Source: Frank Proulx, UC Berkeley Safe Transportation Research & Education Center.

Data Collection Planning and Implementation 51 Another check should be conducted several days after the installation is complete. It should follow the same procedure as the initial check, comparing manual to automated counts. This second test can spot changes in the device that may have occurred over more than a day (due to sun and shade, heating and cooling, etc.). Importantly, the initial days of data should be down- loaded and reviewed to see if there are any strange patterns in the data (e.g., abnormally high hourly counts or unexpected zero counts during the middle of the day). These strange patterns may indicate a problem with the device, but they may also indicate abnormal behaviors at the site (e.g., a delivery truck parked over the sensor, a bicycle parked in front of the sensor, pedestrians or bicyclists walking back and forth in front of a sensor). Ideally, abnormal pedestrian or bicycle movements at a site will be identified before a site is chosen. However, if these site problems are revealed in the first few days of testing the device, it may be beneficial to move the device to a different location at the site or to a completely different site. While testing a device requires additional time, it will help ensure that high-quality data are being collected. It is much better to identify and fix problems early in the process than to collect many months’ worth of data that cannot be used. 3.3.6 Calibrating Devices The sensors used in some counting technologies, such as inductive loops and pneumatic tubes, can be adjusted to make the sensor more or less sensitive and thereby less prone to non- detections (undercounting) or false-positive detections (overcounting). The initial test period during installation can suggest whether or not a sensitivity adjustment is needed. Figure 3-7 compares the accuracy of a pneumatic tube counter before and after calibration—the count accuracy improved after the counter’s sensitivity was adjusted. Validation counts should also be taken in future months and years to monitor a device’s accu- racy. This step is particularly important for counters using inductive loops or pneumatic tubes, because it is possible that, if not maintained properly, these types of sensors may become less accurate over time (see Section 3.3.2 for more detail). If a counter appears to be too inaccurate, contact the vendor to see if the device can be recalibrated or if certain components of the device should be replaced. It is important to work closely with vendors when calibrating counters, because vendors are typically the most knowledgeable about how a device detects users and how Source: NCHRP Project 07-19 testing. (a) Before (b) After Figure 3-7. Illustrative comparison of pneumatic tube accuracy before and after calibration. Source: Tony Hull, Toole Design Group. Validation count at an inductive loop site.

52 Guidebook on Pedestrian and Bicycle Volume Data Collection the device may be adjusted. Vendors can and should provide feedback to customers on how to achieve accurate results. Counters that output data in a timestamp format (i.e., where each individual user detection is recorded), rather than as 15-minute or 1-hour totals, may allow for better troubleshooting of accuracy issues. A vendor or count manager can test how pedestrians or bicyclists passing the sensor at different speeds, in different locations, and in different group sizes affect the count accuracy. Some products allow the user to observe counts occurring in real time on site, even if they are only saved in a binned format. This is sometimes done with an LED indicator or by connecting to the device with a computer. When a device is recalibrated, it is important to differentiate the data collected before and after the recalibration. It may be possible to apply an adjustment factor to the data collected before the device was recalibrated. When there is insufficient data history, or where missed or inaccurate counts cover an extensive period of time, it may be necessary to exclude data from the time period where the count accuracy is suspect. 3.3.7 Maintaining Devices Counting equipment must be regularly maintained to ensure accurate consistent counts. In particular, staff should visit permanent count sites regularly (at least every 3 months) to check that devices are still present (not stolen or vandalized), pointed in the correct direction, and in working condition (see the follow-up activities checklist provided earlier). Staff should check for the accumulation of dirt, mud, water, or other materials that could affect the sensor or other equipment components. In addition, staff should download and review the count data to make sure that the equipment is working properly. Staff should also conduct accuracy tests to determine whether or not the counting technology is maintaining an acceptable level of accuracy. Accuracy checking may be done less frequently than routine site visits, but it should be done at least once per year. If the counting technology has become less accurate over time (this can occur after several months for pneumatic tubes or after several years for inductive loops), it may be necessary to recalibrate the device, as discussed in the previous section. Chapter 5 provides more detailed information about the maintenance needs associated with specific counting technologies. 3.3.8 Managing Count Data Various systems are available for managing data after they have been downloaded from the counter, including agency-developed spreadsheets, vendor-supplied software, software devel- oped in-house, and cloud-based repositories. Based on the practitioner survey conducted for this research, the most common count man- agement tool is using a spreadsheet to compile and analyze count data. Spreadsheets typically represent specific time periods as rows and individual counters or directions (e.g., “Counter A, Northbound” and “Counter B, Southbound”) as columns. Each cell contains the count from a specific counter for a particular time period. Populating the database with counts requires copying data from the counter’s output files into the spreadsheet, making sure that counts are pasted into the correct time periods. This approach requires staff time to download the data (often involving a field visit) and add the data to the spreadsheet. Because this effort can place a burden on staff resources, other options are also available to manage count data. Technician inspecting a counting device. Source: Tony Hull, Toole Design Group.

Data Collection Planning and Implementation 53 Some vendors provide custom software that imports the output files from their products into a spreadsheet. This approach can help avoid mistakes from manually copying counts between data files. In addition, some vendors’ software provides tools to create graphs and tables show- ing changes in counts over time. Practitioners should also be aware that some vendors’ products require sensor data to be analyzed on the customer’s computer using specially developed software. This requires an extra step for the analyst, but reduces computational power needed on site (which increases battery life and decreases equipment costs). Organizations with in-house programming expertise can develop their own software to com- pile data from the counting equipment. This approach may make it possible to automate certain calculations and graphics production, at the cost of the up-front investment to develop and test the software. This approach entails a risk that if the person who developed the software leaves the organization, it may be difficult to update the software in the future. Some vendors also offer a service (typically involving a monthly or annual subscription) where count data are automatically uploaded to a designated web page or cloud repository. This approach can save an organization from having to make field visits to download data directly, keeping in mind that some field visits will still be necessary to check and maintain the counters. Finally, approximately one-third of this project’s survey respondents stated that their pedes- trian or bicycle count data were included in, or could be easily linked to, an existing motorized count database. When possible, agencies should consider building on the expertise and data management systems they have already developed for their motorized count data. In addition to saving time and effort by using an existing framework, integrating pedestrian and bicycle counts into a motorized count database can help an agency create a fully multimodal traffic monitoring system. 3.3.9 Cleaning and Correcting Count Data Cleaning Count Data Once counts are in a database format, they should be reviewed for unusual data. Some unusual counts are incorrect measurements by the technology itself (i.e., missing counts or counts of movements not intended to be counted). Potential reasons for incorrect counts include • Blocked sensor (e.g., delivery truck in front of or on sensor, bicycle parked in front of sensor, person standing in front of sensor). • Multiple counts of the same person (e.g., where people often walk back and forth in front of the sensor, such as near a bus stop or near a corner of an intersection with high pedestrian activity). • Equipment malfunction (e.g., power is lost temporarily, component wears out, detection sen- sitivity changes, sensor is bumped or turned the wrong way). Understanding Unusual Counts On August 23, 2011, a magnitude 5.8 earthquake hit Washington, D.C., a region not typically known for its seismic activity. The Metrorail subway shut down for the day, resulting in extremely high volumes of bicyclists and pedestrians being counted crossing the Key Bridge into Arlington, Virginia. Without the contextual knowledge of the earthquake and transit closure, the counted volumes—roughly double the count on a normal day—could have been written off as a sensor error.

54 Guidebook on Pedestrian and Bicycle Volume Data Collection • Incorrect initial installation (e.g., sensor is pointed at a doorway or other background that creates false counts at certain times; sensor is at a location where people gather at particu- lar times; equipment is not sealed correctly, resulting in water damage, tampering, or rapid deterioration). Incorrect counts do not include regular undercounts due to occlusion. Occlusion is a limita- tion of the technology itself and is addressed by the correction factors described in Chapter 4. Incorrect counts also do not include temporary changes of activity patterns, such as re-routing of pedestrian or bicycle traffic due to construction or other rare events (e.g., festivals, field trip groups, and weather events). These changes in activity patterns may look like incorrect counts, but are actually measured correctly by the counter. Reviewing weather data, local event sched- ules, and other sources can help analysts understand why abnormal counts may have occurred. The level of effort used to clean data should be commensurate with the end use of the data. If, for example, the data are being used to develop day-of-week and month-of-year expansion fac- tors that will be used to convert short-term counts to annual volumes, then a careful check of the data would be warranted. If, on the other hand, the data will only be used to report monthly or annual volume totals for that location, then checking and correcting for short periods of missed data will likely not change the result in a meaningful way. Figure 3-8 shows a selection of raw count data from a passive infrared counter. In this case, the sensor was moved from pointing across the sidewalk to pointing across the roadway and was not discovered for a week. During the week when the sensor was pointing the wrong way, recorded counts were up to 10 times higher than normal levels. It is likely that the infrared sensor counted Figure 3-8. Example of incorrect pedestrian counts due to sensor pointed in wrong direction. Source: Robert Schneider, UC Berkeley Safe Transportation Research & Education Center. 0 500 1,000 1,500 2,000 2,500 Ho ur ly C ou nt fr om In fr ar ed S en so r Day of Week Count Data: University Avenue near Bonar Street (4/8/08 to 5/9/08) Week 1 Week 2 Week 3 Week 4 Tu W Th F Sa Su M Tu W Th F Sa Su M Tu W Th F Sa Su M Tu W Th F Sa Su M Tu W Th

Data Collection Planning and Implementation 55 a portion of motor vehicles that used the roadway, given that their engines and passengers are sources of heat. If the erroneous counts are not removed from the dataset, erroneous conclusions would result about the level of pedestrian activity at this location. The only certain way to identify the difference between an incorrect count and a temporary change in the normal activity pattern is to observe the site (either with a field data collector or reviewing a video) and then compare the automated count with the ground truth count. How- ever, analysts can review raw data for unusual spikes or dips (often counts of zero) and then decide whether or not to flag them as “probably incorrect” counts. One strategy to identify a “probably incorrect” count is to compare it with similar counts before and after it was collected. A count of zero pedestrians during 1 hour in the middle of a day averaging more than 200 pedes- trians per hour is “probably incorrect.” Although it is possible to review each individual count period for unusual counts, it is imprac- tical (and potentially error-prone) to attempt to manually identify “probably incorrect” counts from multiple months of raw data. Therefore, an organization may choose to establish standards for identifying counts that are “probably incorrect.” Different thresholds could be set for a single observation period and for multiple observation periods. For example • Single observation threshold. For the count in question, consider the counts taken at the same time of the week in the previous 4 weeks and in the following 4 weeks. The count is “probably incorrect” if it is more than two standard deviations above or below the average of the eight same-time-of-week counts. The same-time-of-week counts should exclude holidays. • Multiple observation thresholds (four consecutive count periods). For the four consecutive count periods in question, consider the counts taken at the same time of the week in the previous 4 weeks and in the following 4 weeks. For each of the four periods, calculate the average and standard deviation of the eight corresponding same-time-of-week counts. The four consecu- tive hours of counts are “probably incorrect” if each individual count in the series is more than one standard deviation above or below the average of its eight corresponding same-time-of- week counts. The same-time-of-week counts should exclude holidays. Another automated approach to identify incorrect counts was proposed by Turner and Lasley (2013). This method incorporates differences in directional counts (e.g., eastbound vs. westbound) and separates weekday from weekend data. Outliers are identified based on the interquartile range of all counts of a particular type. The authors recommended having an experienced professional perform a targeted manual review of portions of the data to identify possible anomalies not high- lighted by the automated process. After an incorrect count is identified, it can be omitted or cleaned. Omitting counts will leave gaps in the pedestrian or bicycle activity pattern when it is reported. Cleaning involves replacing the incorrect count with an estimate of the correct count (i.e., an imputed value). This replacement can be done using a process similar to the technique used to identify “probably incorrect” counts. For example, if the incorrect count was from a summer Tuesday between 10 a.m. and 11 a.m., an analyst can substitute the average value from the previous 4 weeks of counts on Tuesdays between 10 a.m. and 11 a.m. and the following 4 weeks of counts on Tuesdays between 10 a.m. and 11 a.m. Care must be taken not to use counts from time periods that had different characteristics (e.g., using data from sunny days to estimate an incorrect count from a rainy day; using data from a holiday to estimate an incorrect count from a regular workday). However, if the time period with incorrect data has different weather characteristics than the comparison time periods, it is possible to apply a regression model that includes weather and time period variables to estimate the count for that time period (Wang et al. 2014). Additionally, before determining that a count is erroneous, consider carefully whether or not the unusual count can be explained by a special event or other temporary change in activity pattern.

56 Guidebook on Pedestrian and Bicycle Volume Data Collection Clean pedestrian and bicycle counts can be integrated with automobile traffic count data- bases. Approximately 30% of the respondents to the practitioner survey conducted as part of the research behind this guidebook integrated non-motorized count data into motorized count databases (or developed a parallel and easily linked database). Most transportation agencies have georeferenced traffic count locations (e.g., mileposts or GPS coordinates), so pedestrian and bicycle counts can be added to the appropriate location in a count database. In some cases, additional data fields are needed to create a complete multimodal count database. These fields may specify the type or direction of the movement being counted, the pedestrian or bicycle facil- ity type, or the specific type of counting technology used. To make data easier to share among agencies and to report to a national repository, pedestrian and bicycle data fields should be consistent with the data format specified in the FHWA’s Office of Highway Policy Information Travel Monitoring Analysis System (TMAS) 3.0. Correcting Count Data After the raw counts have been cleaned, they are ready to be corrected (i.e., adjusted for sys- tematic under- or overcounting due to the specific counting technology used). The process of correcting data is the focus of Chapter 4. 3.3.10 Applying Count Data At this point, the count data have been collected, adjusted, cleaned, and stored and are ready to be used. Chapter 2 provides examples of how non-motorized count data can be put to use.