Disparities in pedestrian crossing and driver yielding behaviors: evidence from a large-scale observational study at urban intersections.

Male pedestrians are significantly more likely to cross non-compliantly and to encounter lower driver-yielding rates.

• Finding derived from binary logit models estimated on over 20,995 pedestrian crossings and 3,124 pedestrian-vehicle interactions.
• Data collected from over 1,000 hours of video footage at two intersections in Austin, Texas.
• Two separate models were estimated: one predicting non-compliant pedestrian crossings (NCPC) and one predicting driver unyielding (DUY).
• Male gender was a significant predictor in both the NCPC and DUY models.

Black pedestrians are significantly more likely to cross non-compliantly and to encounter lower driver-yielding rates.

• Black pedestrian identity was a significant predictor in both the NCPC model and the DUY model.
• DUY behavior is also more likely when the pedestrian is Black or Brown.
• Data were collected via manual annotation of video footage from two urban intersections in Austin, Texas.
• The study included over 20,995 pedestrian crossings and 3,124 pedestrian-vehicle interactions.

Pedestrians displaying visible signs of housing insecurity (VHI) are significantly more likely to cross non-compliantly and to face driver failure to yield.

• VHI was a significant predictor in both the NCPC and DUY binary logit models.
• DUY behavior is more likely 'when the pedestrian in question is older, Black or Brown, and exhibits VHI.'
• VHI was identified through manual annotation of video footage.
• This finding highlights the intersection of marginalization and pedestrian safety risk.

Pedestrians crossing alone (solo) are significantly more likely to cross non-compliantly and encounter lower driver-yielding rates compared to those crossing in groups.

• Solo crossing was identified as a significant predictor in both the NCPC and DUY models.
• Social context (crossing alone vs. in a group) was captured through manual video annotation.
• The dataset included over 20,995 pedestrian crossings at two Austin, Texas intersections.

Runners exhibit higher non-compliant pedestrian crossing (NCPC) rates than walkers, with peak non-compliance occurring during late night and dawn periods.

• Running behavior was a significant predictor of NCPC in the binary logit model.
• Time-of-day effects were significant, with late night and dawn periods associated with peak non-compliance.
• Pedestrian mode (running vs. walking) was captured through manual annotation of over 1,000 hours of video footage.

Non-compliant pedestrian crossing (NCPC) behavior is the strongest predictor of driver failure to yield (DUY).

• 'Pedestrian NCPC behavior is the strongest predictor of failure to yield' according to the DUY binary logit model.
• The DUY model was estimated on 3,124 pedestrian-vehicle interactions.
• This suggests that pedestrian behavior substantially mediates driver responses at crossings.

Driver failure to yield is more likely during morning periods and among drivers of personal (non-commercial) vehicles.

• Time-of-day (morning periods) was a significant predictor in the DUY binary logit model.
• Vehicle type (personal vs. commercial) was also a significant predictor, with personal vehicle drivers more likely to fail to yield.
• These findings are based on 3,124 pedestrian-vehicle interactions observed at two Austin intersections.

Older pedestrians are more likely to experience driver failure to yield.

• Pedestrian age (older) was identified as a significant predictor of DUY behavior in the binary logit model.
• DUY is more likely 'when the pedestrian in question is older, Black or Brown, and exhibits VHI.'
• Age was coded through manual annotation of video footage covering over 1,000 hours of intersection activity.

The study analyzed over 1,000 hours of video footage documenting more than 20,995 pedestrian crossings and 3,124 pedestrian-vehicle interactions at two intersections in Austin, Texas.

• Data were collected at two urban intersections in Austin, Texas.
• Manual annotation of video footage was used to capture pedestrian and driver behaviors and sociodemographic attributes.
• Two binary logit models were estimated: one for NCPC and one for DUY.
• Individual attributes, social context, and time-of-day/weather conditions were included as predictors.