Food Deserts in Vancouver
Testing whether grocery access in Vancouver tracks census demographics, one neighborhood at a time.
Food access is not evenly spread across a city. Some neighbourhoods have a grocery store on every second corner. Others make you ride a bus to buy fresh vegetables. The interesting question is not where the gaps are. It is whether the gaps line up with who lives there.
That was the question behind this study: which Vancouver neighbourhoods are genuinely underserved for grocery access, and does that underservice track demographic patterns in the census data? We wanted an answer grounded in numbers, not assumptions about which parts of town "seem" underserved.
How we built it
We built the study in ArcGIS Online. First, a spatial analysis of food access measured against neighbourhood boundaries. Then a logistic regression to test how food access correlated with demographic variables from the Canadian census. The output was a choropleth map plus summary visuals, and we built them so the findings would hold up for a non-technical review, not just a GIS audience. The full project files are on Drive.
Two things made this harder than a straight mapping exercise. Grocery access is a continuous, messy thing, and you have to commit to a definition before you can measure it against neighbourhood boundaries. And a regression result is not a map. Turning coefficients into a choropleth that a non-specialist can read at a glance takes as much design work as the analysis itself.
The pattern held. Food deserts clustered in Vancouver's lower-density, lower-income neighbourhoods, the ones with higher visible-minority populations, concentrated across South and East Vancouver, while the dense core stayed well served. And access was improving: the share of residents within a 15-minute walk of a grocery store rose from about 85% in 2012 to about 88% in 2024. The gaps that remained were the ones that tracked who lived there.
This was my first real use of logistic regression on spatial data. What stuck with me was not the statistics. It was the translation step: a statistical result only matters if someone who has never heard of a p-value can look at the map and understand what it says about their city. That principle, analysis in service of a readable map, has shaped every project I have done since.