Geopath Mobile Device Panel

Owned by Kevin Narine
May 31, 2023
3 min read

Location data streaming from devices are observed and understood over the life of a device identifier.  However, many devices produce unreliable or infrequent location data such that the characteristics of travel can not be understood or used with confidence. Just like when doing survey sampling, the goal is to establish a panel of respondents that well represent the population and have enough information to control for bias in the sample. To this end, a Mobile Device Panel consisting of reliable, high quality devices are selected to create the representative sample on a weekly basis from which statistically significant insights on the population behavior can be gathered.

Panel Selection

It is often the case that a device is not actively observed enough to establish movement patterns with confidence. If a device location is only logged a few times per day, the ability to understand if it commutes to work or regularly goes out to restaurants for dinner cannot be established to the point of using the device as a sample of what the population as a whole is doing. Generally, ~80% of the GPS sightings belong to only 20% of the unique devices, leaving the bulk of actual devices observed during a day with too little activity to classify.

Standard Data culls down devices observed to an active set of panel devices for which the activity pattern and representativeness of the U.S. population can be documented. In short, for a device to be included as a panel device, it has to be:

  1. Generally Active 8 hours per day.

  2. Historically consistent home location, documented as a Census Block Group, to ensure privacy as one of approximately 400 households in a neighborhood.

The figure below summarizes the count of total and panel devices as well as their sightings and the total minutes of activity for the week of August 9, 2020 where the panel devices were 19.5% of the total processed, but were responsible for 78% of the locations processed.

Sampling

The objective of this step is to understand to what extent a device is a representative of a certain demographic group that lives in an area. Sampling rates of devices in the panel are documented as they represent the total current year population of their home neighborhood (Census block group) as provided by .

However, if this sampling rate was applied directly, the bias introduced by the sample would be enormous. Imagine the individuals visiting a mall today. We will observe a few mobile devices visiting from a specific block group, during a specific hour. If we only expanded to the full population using the sampling rate in the block group, we’d report an entire crowd of people visiting from that block group and possibly no one from a neighboring block group with very similar households.

To appropriately scale a device’s movement patterns to a population in addition to the sampling rates of their home neighborhood, we also document the sampling rates of the population of similar households to create the most reliable understanding of the population represented by each panel device.

The household types are defined by the .

Â