Geopath Movement Patterns

Movement Patterns summarize the device activity in the form of a Trip Matrix documenting individual trip Origin-Destination connections. The origins and destinations are aggregated to Census Block Groups, neighborhoods defined to contain approximately 400 homogeneous households. This resolution of neighborhood definition is the same as the resolution used when reporting home locations and calculating audience segments.

Current Methodology (2019, 2020)

Currently, the matrices being calculated are provided by Airsage and delivered directly to Bentley to use in the calculation of Geopath Uni-Modal Trip Paths (closed network). The matrices produced by Airsage include both:

• Observed device movement matrices, simply reporting the sampled device movements.

• Expanded device movement matrices that used the sample rate of devices vs. the 2010 population in the home blockgroup to scale the observed movement matrix.

While the directly observed matrix of trips provided to Bentley answers fundamental questions on travel movements and has allowed robust audience circulation, segmentation, and reach on all roadway segments nationwide, it has also had some limitations:

1. Geopath is not in possession of the input matrices to Bentley’s process, to use for data validation and post processed analyses.

2. The direct reporting of observed device movements, introduces a location-based sampling bias producing a sparse matrix by home location. When examining home locations for any particular roadway segment, the reach metrics are robust and intuitive. There is no issue with the existing measures Geopath produces. However, when adding any additional dimension (traffic by home location, origin, or destination) the sparseness of the matrix is apparent as only a fraction of likely destinations show observed movements. This limits the secondary analysis possible, like examining movement patterns for selected pieces of inventory. This is especially problematic with long distance trips that are sparse themselves.

3. The sparse matrix puts more pressure on the optimization the Bentley is doing in balancing observed traffic counts. This has caused some variation in the trip lengths and the density (data size) of the home location data that is being received as the Airsage sample sizes fluctuate.

Updated Methodology (2021)

Standard Data has been contracted to expand the methods that are being used for Geopath Activity Patterns to deliver circulation for place-based measurement and deliver a trip matrix to the same spec by the end of 2020, with the ability to produce seasonal matrices running forward in 2021. Specifically this work will provide the data to Geopath directly, rationalize the Geopath Mobile Device Panel to be driving roadside with the same data driving the place-based circulation and audiences, and enhance the expansion of the observed devices to the full population using a microsimulation platform to better replicate the movements of the population as a whole.

Data Sources

The data-driven simulation leverages commercially available third-party datasets along with open datasets from the U.S. Census and others.

• (seasonal data excluding holidays)

• (current year)

• 2015-2019 (pending release December 10, 2020)

• 2020 Q4

• (seasonal schedule excluding holidays)

• 2017 (or latest available)

• 2017 (or latest available)

Standard Data produces a dataset that captures where the population spends time hour by hour, year over year.  profiles the entire U.S. using a uniform grid of tenth-of-a-mile cells. This provides key information for residents of a particular neighborhood, revealing things like:

• What areas of the city do these households frequent and what areas do they avoid?

• How many people are commuting to work, in what areas, and at what time?

• How many people are going out in the evening or at other times of the day?

• How often do they travel and how far? 

In addition to profiling the activity generated by every neighborhood, the types of households as defined by the Claritas PRIZM Premier Segments are also profiled as described in . The behavioral PRIZM household segments help to explain and classify the activity observed from the Standard Data Anytime Population data.

This PRIZM drive household-level explanation is then supplemented by the . The ACS PUMS provides person-level survey responses with a complete household descriptions.  This allows a perfect understanding of how the detailed compositions of households in an area drive travel. For example, profiles of households with children and without, different levels of vehicle ownership, effects of multi-generational households, and employment all contribute to expected levels of travel on a daily basis.

Similar to how the detailed household segments drive travel, the LEHD LODES data characterize workforce dynamics and characteristics at a fine-grained Census Block geography nationwide.  This macroscopic information, when combined with the Standard Data Anytime Population “Work Activities”, documents commuting patterns of households as well as providing checks on employment levels at a detailed geography across the U.S.

Finally, the documents the personal and household travel behavior of the U.S. population.  This provides, for a very small percent of all people, an extremely detailed record of travel and time use. It inventories how activities are chained together at what times of the day in different area types (rural versus urban) along the dimension of household composition, worker status, and income most importantly.  

The activity patterns from NHTS as explained by the demographics and household characteristics can be systematically recorded and codified as defined “tours”.  A tour is a schedule of activities away from home connected by “trips”. The simplest tour is called a “simple work tour”, which consists of a person traveling from home to work, spending time at work, and then traveling from work back to home. The following table describes a mutually exclusive set of tour types that are commonly used.

Discrete Event, Agent-Based Simulation

Using the inputs above, the following questions outline the components that drive travel for the population by location, household type, and other demographics:

• What is the current year population of households, persons, businesses, and jobs?

• What types of activities do people do away from home and how often?

• When do they schedule activities in their day?

• Where do they go to do those activities?

• What are their options for how to travel?

• What are their preferences for how to travel?

All of this observed data is combined such that the final view of activities replicates all the dimensions of the observed data for the expanded to the full population.  Using a discrete event simulation, which replicates the interactions of complex systems as a series of interactions along a timeline and produces locations of activities for all people at specific latitude and longitude points, detailing the time spent at these locations with travel between them minute-by-minute throughout a week.

Simulation Validation

The resulting full population, their time at home and activities at other locations, and their trips between those locations are validated using comparisons of the person-level activity to the input data to make sure it replicates the observed data along the expected dimensions, including:

• Comparison of tour types by demographics to NHTS

• Comparison of time use (home, work, other, travel) per 15 minutes to NHTS

• Comparison of number of trips per person per day to NHTS

• Comparison of number of tours per person per day to NHTS

• Comparison of no-tour persons to NHTS

• Comparison of no-tour persons to Standard Data Anytime Population

• Comparison of count of arrivals, average occupancy, and average dwell time to Standard Data Anytime Population

Specifically, the expected national average trip length frequency distribution is closely monitored and compared to the Standard Data Anytime Population and the NHTS. Any deviation from expected results needs to be well explained in actual observations and the NHTS. Similar rigor is also applied to the distribution of trips within each day part bin listed below.

• 12:00 am to 6:00 am (early morning)

• 6:00 am to 10:00 am (AM peak)

• 10:00 am to 3:00 pm (midday)

• 3:00 pm to 7:00 pm (PM peak)

• 7:00 pm to 12:00 am (late evening)

These comparisons are especially valid for explaining the changes in travel that happened initially and continuing through COVID-19. The comparison and explanation using the demographics, work tours, employment, long distance travel, and daily observed travel is key to not only producing robust Movement Patterns but also for enhancing the real-time weekly impressions adjustments and future forecasts.

In a final phase of validation, the itineraries are compared against datasets and 3rd party travel insights not used as inputs to the simulation:

• Comparison of commuting patterns to LEHD LODES Employment Origin Destination patterns

• Comparison of commuting patterns and commuting travel times to Census Transportation Planning Package Journey to Work

• Comparison to other open datasets like the Humanitarian Data Exchange (HDX) Movement Range Maps – https://data.humdata.org/dataset/movement-range-maps

Trip Matrix Delivery

The detailed output from the microsimulation delivers the full daily itineraries. To produce a Trip Matrix, the detailed itineraries for all people nationwide are simply aggregated up into trip matrices along desired dimensions of home location, origin, destination, and purpose by hour of the day.