1.0 Streetlytics Introduction

Streetlytics provides accurate traffic volume, speed, and home location data by roadway segment to Geopath and the Out of Home Industry (OOH) in order to meet the following two goals.

1. Increase accuracy and confidence in OOH marketing impression values.

2. Provide increased flexibility to the OOH industry for targeting specific demographic markets.

The primary objective of this document is to explain why the Streetlytics methodology is superior to traditional data collection methods and communicate the accuracy of the Streetlytics data product through quantitative validation graphics specific to each aspect of the deliverable.

2.0 Overview of Historical Traffic Data Collection Methods

Traditional methods for collecting segment level traffic data include: manual traffic counts, pneumatic road tubes, pavement detectors, video detection, and roadside driver interviews. The Citilabs team is experienced in these traditional methods and therefore has a detailed understanding of the strengths and weaknesses of each method. Here is a table summarizing the purpose, strengths, and weaknesses of each traditional data collection method.

Each traditional count method introduces varying types of uncertainty. In addition, these methods are all collected independently and do not consider data collected on adjacent segments. This often results in poor flow conservation within a corridor from one roadway segment to the next.

Open

Note, in the figure above, the count collected near the panel location (star) is 45,000 vehicles per day (vpd). However, when considering conservation of flow, this count is inconsistent with both upstream and downstream count locations. This type of inconsistency is common and due to day-to-day variation in traffic, variation in congestion, vehicle incidents, weather, event traffic, etc. Historical traffic data collection methods do not systematically address these inconsistencies.

Until recently, this count data has been the only data available, but it does not meet current OOH industry goals and is not nearly as accurate as one believes.

3.0 Streetlytics Process

Streetlytics provides an alternative to these historic traffic data collection methods. This section qualitatively addresses why the Streetlytics process produces a dataset that is superior, more efficient, and more robust than all historical data collection methods.

The Streetlytics process combines Referenced, Sampled, and Modeled Movement data to provide monthly Optimized Movement data for the full population within the Continental United States plus Hawaii and Puerto Rico. This process leverages a wide variety of complimentary source datasets. Additional information about each of these source datasets is provided in Exhibit A. The primary independent datasets used in the Streetlytics process are:

• Sampled Movement data derived by AirSage from observed device-level Geospatial Positioning System (GPS) sources

• Referenced Movement data including observed traffic counts from published government sources and travel speeds from HERE

• Data underlying Modeled Movement estimates including demographic, employment, and place of interest data from ESRI and a routable transportation network built on the HERE network dataset

Open

The Streetlytics process combines data from these and other sources, applies Citilabs’ technical expertise and mobility analytics software, and provides a comprehensive monthly travel activity dataset. More information about the Streetlytics process is provided in Exhibit B. The Streetlytics data product includes vehicle volumes, vehicle occupancy, vehicle speeds, pedestrian volumes, transportation system user home locations and traffic routing data by time of day, day of week, and month of the year.

4.0 Streetlytics Advantage

The Streetlytics process, source data, and team experience provides a much more accurate mobility dataset than was possible with historical methods. The following provides a list of primary advantages of the Streetlytics dataset with corresponding examples.

Advantage 1: Conservation of Flow

Vehicle traffic flow conservation is maintained across all the Streetlytics data products. Traffic flow conservation means that all vehicle traffic from one or more upstream roadway segments must be conserved on all downstream roadway segments unless a trip entrance (source) or exit (sink) is provided from the transportation system such as a parking lot, site entrance/exit, or driveway. Here is an example depicting typical location specific historical traffic counts which do not maintain flow conservation, followed by a description of why traffic flow conservation is important.

Example 1 Historical:

This example depicts historical source traffic counts collected at five unique locations on closed freeway system with one on-ramp and one off-ramp. The source count collected near the panel location (star) is 45,000 vpd and is inconsistent with both upstream and downstream counts. This type of inconsistency is common and results from a variety of issues related to the count collection day, time of year, count method used, and adjustment applied. These differences are due to day-to-day variation in traffic, variation in congestion, vehicle incidents, weather, event traffic, etc. Historical traffic data collection methods do not systematically address these inconsistencies.

The Streetlytics process maintains conservation of flow at all times and therefore the Streetlytics data is more consistent and accurate from one location to the next improving both quality and reliability. The following example shows the Streetlytics volume at the panel location.

Example 1 with Streetlytics:

Note, this count is consistent with both upstream and downstream roadway segments. The resulting panel circulation of 60,000 vpd is more accurate than the previous 45,000 vpd.

Advantage 2: Comprehensive Coverage of Data

All Streetlytics datasets are comprehensive and include all transportation segments in the U.S. Specifically, vehicle traffic volumes are provided for each roadway segment that permits vehicle. Here is an example depicting typical traffic count coverage when traditional methods are used followed by a description of why traffic flow coverage is important.

Example 2 Historical:

This example depicts historical source traffic counts collected at six unique locations on a roadway system including both arterial and collector roadways. A source count was not available from traditional sources at this panel location, but counts were available nearby. This is typical since it is cost prohibitive to use historic data collection methods to collect counts on every roadway segment. Based on the counts available, the historical approach would have used the adjacent counts of either 35,000 vpd or 20,000 vpd to represent the traffic count at the panel location.

Since the Streetlytics volume dataset is comprehensive, it is not necessary to guess which count is similar. Therefore, the Streetlytics data provides a more reliable and consistent count at each panel location. The following example shows the Streetlytics volume at the panel location.

Example 2 with Streetlytics:

Note, the Streetlytics volume coverage is comprehensive and is consistent with both upstream and downstream roadway segments. The resulting panel circulation of 27,000 vpd is more accurate than either 20,000 vpd or 35,000 vpd.

Advantage 3: Multiple Independent Data Sources

The Streetlytics data product is based on a variety of complementary data sources. Historical traffic count methods rely on only one source of traffic count data from a single location. Here is a graphic depicting the source datasets used by Streetlytics.

Here is an example depicting why the multiple datasets leveraged by Streetlytics provides a more accurate traffic count than collected by traditional count sources.

Example 3 Historical:

This example depicts historical source traffic counts collected at four locations on a roadway system adjacent to a residential development with only one access point. The published source count at the panel location is 6,000 vpd. By referencing multiple data sources such as demographic data, government sources and nearby traffic counts, the Streetlytics vehicle volume is much more accurate than the historical count. The following example shows the Streetlytics volume at the panel location.

Example 3 with Streetlytics:

In this example, Streetlytics uses demographic data such as the number of households by income, size, auto ownership, and age of resident categories in combination with established trip rates determined from government travel surveys to estimate an approximate of the magnitude of travel to and from each neighborhood. Additionally, GPS location data is used to provide the typical sampled movement of trips and adjacent counts provide an estimate of the trip magnitudes at each location. Based on the household demographics, GPS data movement and adjacent counts, the source count of 6,000 vpd is unreasonable, while the 3,600 vpd volume provided by Streetlytics is consistent with multiple independent data sources.

Advantage 4: Additional Robust Data

Traffic counts collected with historical methods provide an estimate of the circulation observed passing the panel, but do not provide any additional information about where the travelers are going, why they are traveling, what route did they take, and who the travelers are. To collect this information using historical methods, a roadside interview would be required; however, roadside interviews are illegal in many states, expensive, and often provide a very small sample size since it is difficult to stop everyone traveling on a segment of roadway.

With a comprehensive set of trip paths covering every roadway in the U.S., Streetlytics provides a comprehensive understanding of each of the who, when, where, and why people travel. The Streetlytics data product connects person movement on each roadway segment with the traveler home locations. Understanding the home locations of travelers near panel locations, empowers Geopath to provide robust demographic characteristics of each market.

5.0 Streetlytics Product Validation

The Streetlytics product has been validated against a variety of third-party sources including over 1 million published annual average daily traffic counts (AADTs), approximately 5,000 automatic traffic recorders (ATR), the 2017 National Household Travel Survey (NHTS), and ESRI Demographic/Business Data. Additionally, more than 150 pedestrian source datasets totaling approximately 10,800 pedestrian counts were referenced including data from the NYC Times Square Alliance. The following sections quantitatively describes the accuracy of the Streetlytics year 2016 seasonal dataset with validation graphics and statistics.

5.1 Validation: Streetlytics Vehicle Volume

Streetlytics Daily Vehicle Volume vs. Source Vehicle Traffic Counts

The Streetlytics average annual daily vehicle volumes are validated against more than 1 million AADT counts. The AADTs are collected and adjusted by state, regional and local governments using a variety of historical methods. They provide a comprehensive nationwide dataset, which provide an approximate roadway segment count. Due to the data collection method, error and inconsistencies are common with this dataset. The following figure depicts the validation of the Streetlytics daily vehicle volumes more than 1 million AADTs.

As demonstrated in the figure above, the Streetlytics daily vehicle volumes validate well against more than 1 million AADT source counts. The corresponding R2 and % RMSE values for the Streetlytics daily vehicle volume validation are 0.96 and 38.3 respectively.[MM1]  The relatively small set of outliers in the scatter plot are expected and primarily due to error in the source count data collection method.

Alternatively, the Streetlytics daily vehicle volumes have been validated against a smaller set of count locations, which are collected using pavement detector (ATR) since these stations, when in operation, typically provide more accurate source counts.

As demonstrated in the figure above, the Streetlytics daily vehicle volumes validates much better against the ATR count locations compared to the larger set of source counts. The corresponding R2 and % RMSE values for the Streetlytics daily vehicle volume validation are 0.98 and 21.4 respectively.

Streetlytics Vehicle Volume by Season vs. Automatic Traffic Recorder Counts

The ATR source counts provide the best available source for measuring seasonal traffic variation by roadway segment. The following figure depicts the validation of the Streetlytics vehicle traffic seasonal variation vs. the ATR source counts.

As demonstrated in the figure above, the Streetlytics vehicle volume seasonal variation validates well against the ATR source counts. The corresponding R2 and % RMSE values for each seasonal dataset are provided to the right of the figure. The validation statistics remain consistent across all four seasons.

Streetlytics Vehicle Volume by Day Type vs. Automatic Traffic Recorder Counts

The ATR source counts provide the best available source for measuring day type traffic variation by roadway segment. The following figure depicts the validation of the Streetlytics vehicle traffic seasonal variation vs. the ATR source counts.

As demonstrated in the figure above, the Streetlytics vehicle volume day type variation validates well against the ATR source counts. The corresponding R2 and % RMSE values for each day type dataset are provided to the right of the figure. The validation statistics remain consistent across all four day types.

5.2 Validation: Streetlytics Person Trip Motivation

Streetlytics Person Trips by Motivation vs. 2017 National Household Travel Survey

The most comprehensive publicly available source describing travel behavior in the U.S. is the 2017 National Household Travel Survey (2017 NHTS). This survey is used to produce a dataset which describes travel by categories such as trip motivation, trip mode, trip occupancy, etc. The following figure depicts the validation of the Streetlytics nationwide total person trips by motivation vs. the 2017 NHTS.

As demonstrated in the chart above, the Streetlytics trip activity by trip motivation validates well against the 2017 NHTS.

5.3 Validation: Streetlytics Vehicle Occupancy

Streetlytics Occupancy vs. 2017 National Household Travel Survey

The most reliable and comprehensive vehicle occupancy source in the United States is the 2017 NHTS, and therefore, the 2017 NHTS occupancy was input into the Streetlytics analysis and data product. The following figure depicts the validation of the Streetlytics occupancy values by day type and motivation for region 1 vs. the 2017 NHTS occupancies in region 1.

As demonstrated in the figure above, the Streetlytics vehicle occupancy values by day type, motivation, and geographic region are identical to the 2017 NHTS source data. For region 1, the occupancy values vary between 1.06 and 1.99 for each day type and trip motivation category.

5.4 Validation: Streetlytics Vehicle Speed

The roadway segment vehicle speeds included in the Streetlytics product, are based on the Here navigation dataset. Here segment vehicle speeds are calculated directly from observed vehicle probe data. This probe data set includes a combination of the following data sources: connected vehicle data, fleet telematics data, in-vehicle navigation system data, and mobile app data. HERE uses a variety of techniques to validate and normalize this probe data. These checks include:

1. Quarterly evaluations of probe data providers are conducted to confirm the data meet HERE quality standards. Any source data provider deviating from these standards is removed from the sample.

2. Here develops a real-time traffic product from the same set of probe data that is used to develop the Traffic Analytics Speed Data product. The quality of the real-time product is monitored closely and ranked high by both internal HERE studies and HERE customers.

5.5 Validation: Streetlytics Pedestrian Volume

Streetlytics Daily Pedestrian Volume vs. Source Pedestrian Counts

A figure depicting the validation of the Streetlytics annual average pedestrian volumes against the daily source counts is provided below. Additional pedestrian volume documentation is included in Exhibit C

As demonstrated in the figure above, the Streetlytics daily pedestrian volumes validate well against approximately the 10,800 pedestrian source counts obtained by Citilabs. The corresponding R2 and % RMSE values for the pedestrian volume validation are 0.99 and 35.6 respectively. All counts greater than 100,000 pedestrians per day were input into the product directly, and therefore were not included in this validation figure.

Streetlytics Pedestrian Volume by Season vs Times Square Alliance Pedestrian Counts

The Time Square Alliance pedestrian counts provide the best available source for season pedestrian volume variation in and on roadway segments near Times Square. Therefore, the Times Square Alliance counts were input directly into the Streetlytics product for these segments. The following figure depicts the validation of the Streetlytics pedestrian volume seasonal variation vs. the Times Square Alliance counts.

As demonstrated in the figure above, the Streetlytics pedestrian volume seasonal variation is consistent with the Times Square Alliance Counts for an average location. Each season has perfect R squared value of 1.0 and a perfect % RMSE of 0% when comparing each respective count data point.

Streetlytics Pedestrian Volume by Day Type vs 5^th Avenue, NYC Pedestrian Counts

The pedestrian counts along 5^th Avenue, NYC provide the best available source for season pedestrian volume variation on roadway segments along 5^th Avenue. Therefore, the pedestrian counts were input directly into the Streetlytics product for 5^th Avenue segments. The following figure depicts the validation of the Streetlytics pedestrian volume day type variation vs. the pedestrian counts along 5^th Avenue.

As demonstrated in the figure above, the Streetlytics pedestrian volume day type variation is consistent with the 5^th Ave Counts for an average location. Each day type has perfect R squared value of 1.0 and a perfect % RMSE of 0% when comparing each respective count data point.

Streetlytics Pedestrian Volume by Hour vs 5^th Avenue Pedestrian Counts

The following figure depicts the validation of the Streetlytics pedestrian volume hourly variation vs. the pedestrian counts along 5^th Avenue.

As demonstrated in the figure above, the Streetlytics pedestrian volume hourly variation is consistent with the 5^th Ave Counts for an average location. Each hour has perfect R squared value of 1.0 and a perfect % RMSE of 0% when comparing each respective count data point.

5.6 Validation: Streetlytics Home Location

Streetlytics Segment Traveler Home Location vs. ESRI Demographics

The Streetlytics home location data trip allocation percentage is used to compute the number of trips generated from each block group to compare the trips and population of block group.

From the above chart, we can see that majority of the segment’s home location trips does not exceed the number of people living in the area.

Streetlytics Times Square Traveler Home Location vs. Times Square Alliance

The 2017 Times Square Alliance Advertising Study represents a reliable source for validating the distribution of Home Location of people traveling in Times Square. The following figure depicts this validation. The following assumptions were made to support this validation.

1. NYC Locals were assumed to reside in one of the 5 Boroughs of New York City. (Bronx, Brooklyn, Manhattan, Queens, and Staten Island)

2. All survey participants were observed walking in Times Square and through vehicle trips were not specifically included in the survey.

3. The survey was conducted on Friday, Saturday or Sunday.

4. Since International home locations are not included in the Streetlytics product, a consistent international home location percentage was assumed.

As demonstrated in the figure above, the Streetlytics home location distribution between NYC locals and domestic visitors validates well compared to the 2017 Times Square Alliance Survey.

Streetlytics Segment Traveler Distance from Home (Reasonableness Check)