WP4·Qualitative and mixed methods real data

What residents report, and what the images show.

Before any walk-along interview, one open source of residents' own observations exists: the City's public-space reports.

Start ↓

01 / 05

223,592 reports.

June to October 2025, the panoramas' months. Here: obstacles on the sidewalk, per km of street.

02 / 05

What the model sees.

The share of sampled panoramas where the model reports an obstructed sidewalk.

03 / 05

Buurt by buurt.

Locations are coarse, so reports and images meet per buurt, never report by image.

04 / 05

Where they agree, where they cannot.

Obstacles agree, even with address density held fixed. A street-light report says nothing about whether a pole is visible.

05 / 05

Reports are behaviour too.

Litter reports follow income far more than what the images show. That is why the protocol pairs data with interviews.

What WP4 would add

Walk-alongs where the two sources disagree.

The protocol interviews about twenty participants on site and on walk-alongs, then triangulates the transcripts with the numbers. Buurten where reports and images disagree are the natural place to start: a resident can say what the camera missed, or why nobody reports what it saw.

Details

Data, matching and the full table

Source: "Meldingen over de Openbare Ruimte in Amsterdam" (SIA open data, api.data.amsterdam.nl/v1/meldingen), Creative Commons attribution. 99% of the reports in the window are closed. The City's buurt code matches CBS as "BU0363" + code. Spearman ρ across 360 buurten, 95% bootstrap interval; partial ρ holds address density fixed.