Research Request - Pedestrian Crash Fatalities among People Experiencing Homelessness

[KatrinaMKaiser]

Research Question

To what extent are people experiencing homelessness (PEH) represented among KSI crashes on the SHN?

Caltrans PMP is interested in improving pedestrian safety on its system. People experiencing homelessness are a particularly vulnerable population. Would like to statistically test the correlation between KSI crashes on the SHN and the presence of unhoused people.

How will this research be used?

Will use to identify priority locations for pedestrian safety upgrades, homelessness outreach?

Stakeholders & End-Users

Working with Abby Jackson (PMP) and the Division of Safety

Metrics

EDA metrics:

• K/SI per overall pop (density) per tract touching SHN
• K/SI per unhoused pop (density) per tract
• correlation coefficients/significance between K/SI and unhoused pop density

Longer term metrics (need more data):

• K/SI per job density per tract (lower priority)
• Percent of deaths among PEH due to motor vehicle crashes, by tract or other aggregation geography - use to validate
• Percent of overall K/SI on SHN that are PEH
• Utilizing mode share and VMT to generate crash rates as opposed to counts

Data sources

• Cal-ITP data sources:

• External data sources:

  • SWITRS data
  • CT Encampments data

• Remaining data source questions:

  • We'd need coroner data focused on homelessness specifically?
  • Can we leverage count data aggregated up by census tract put out by counties/CoCs?

Deliverables:

Easy deliverable: ArcGIS webmap or 1-page notebook site Longer term deliverables: story map or parameterized notebook site with maps (a page for each county/Continuum of Care for which we have data)

Timeline of deliverables:

• [x] Interim check-in
• [x] Planning meeting
• [x] Presentation: April 19

[KatrinaMKaiser]

Did EDA in ArcGIS Pro today. Steps:

1. Loaded relevant data: Encampments, SWITRS (filtered to ped crashes), SHN, Bridges
2. Projected all layers
3. Buffered bridges 200m (short walking distance)
4. Summarized crashes and encampments within buffers
5. Used "Local Bivariate Relationship" tool to identify bridges (buffered) where there's a significant relationship between encampments and ped crashes (based on 30 nearest neighbors)

Lessons learned:

• At one freeway intersection there are multiple bridge points. Would like to dissolve some of these overlapping buffers, but in denser urban corridors this creates long caterpillar shapes. Maybe the move is a small buffer, dissolve, create a point at centroid, and then bigger buffer
• Doing a full join of summarized crashes to summarized encampments seems not possible? I would like a comprehensive dataset with bridge areas with all combinations of y/n crashes x encampments, with 0 counts where applicable. This might be easier in geopandas.
• In the encampments data, each row represents a cleanup work order. This might not represent dozens of separate encampments; it might represent the same people being swept repeatedly. Is it appropriate to summarize this to a dummy variable, e.g. "the presence of any encampment since July 2021"
  • The date ranges for SWITRS and for encampments don't overlap at all

[KatrinaMKaiser]

Additional work in ArcGIS Pro, Portal webmap link. Awaiting next steps re: circulating list of priority bridge areas.

[KatrinaMKaiser]

Initial pilot list sent to Div of Safety and Office of Housing/Homelessness on 6/2