Appendix A. Data Comparability

Comparison of Data Sources

The project team acquired output analysis for Seattle from two different sources: proprietary data were processed by Cruise Detector, but behind a data secure firewall. Output was provided in the form of particular metrics, including number of trips, number of cruising trips, date, and time of day for 15,000 units of analysis. The units were a mix of metered streets, streets within half a mile of a metered street, and for census block groups beyond that. In addition, the research team acquired and processed raw location data for several of the same date periods. The outputs of the two sets were compared to understand comparability and whether the sets might be used interchangeably.

The pre-processed output was aggregated into nine policy time periods, six of which overlap with the second data source (i.e., the raw location data). The first three time periods correspond to most of January and February 2020, divided into before and after the price change. Subsequent periods correspond to late March and early April 2020 (before and after the Seattle parking meters had been temporarily decommissioned). The two data sets varied widely in terms of the volume of trips identified. Trips reported by the third-party processor gradually increased in the first three periods, then dramatically dropped for the remaining time periods—consistent with local stay-at-home orders. The number of trips per day in the raw location sample increased as users (new data sources) were added to their data collection base.

Source: FHWA.

Figure 67. Chart. Volume comparison.

The frequency of cruising was also relatively consistent within each data source, but the raw location data showed consistently lower cruising than the processed data. There was also one notable drop in cruising in the third policy time period in the raw location data set.

Source: FHWA.

Figure 68. Chart. Cruising frequency by policy time period.

In terms of distribution throughout the day, both data sets show the bulk of trips between 8 a.m. and 8 p.m. but there are important differences between these extremes. The raw location data show trips climbing in the morning to an inflection point, after which they continue to climb but at a slower rate. There is a large spike in the afternoon. The processed data displays a diurnal pattern more similar to the expectation of a traditional morning and early evening peak.

Source: FHWA.

Figure 69. Graph. Time-of-day trip distribution Seattle data sources.

While the data sets differ temporally, spatially their distribution is much more similar. When the data were grouped by the neighborhood in which the trips end, the comparison is much more similar (Figure 70). This pattern suggests that both data sets are drawing from a similar cross section of Seattle neighborhoods, and the raw location set may be influenced more by when people use applications relative to the processed set, which may better describe when people make trips. The discrepancy is not important for some of the possible analysis, but it suggests the benefit of having independent trip counts in which to weight against. The need for such a step would depend on the kind of analysis required.

Source: FHWA.

Figure 70. Chart. Spatial distribution of trips Seattle data comparison.

These data sources offer policy analysts new opportunities to gain insights into trip making, cruising, and parking behavior. These analyses have demonstrated that although there is great potential, limitations in research design may arise from lack of data. Additionally, given the lack of transparency in how most data vendors obtain and process their data, researchers should use caution when drawing conclusions from any single data set. Nevertheless, Cruise Detector and location data can be used to identify potential issues that can be verified with additional data.