How scoring works
Every neighbourhood score on PickYourPlace is a percentile: a 0–100 rank against the other neighbourhoods in the same city, computed from public data through a four-layer pipeline. A 50 is the city median. An 80 ranks above four fifths of the city. This page explains the machinery those numbers come from — and exactly what a percentile cannot tell you.
Last updated
What we measure
Scoring runs at the dissemination area — Statistics Canada's smallest published geography, typically 400–700 residents — across roughly 2,700 dissemination areas in the four covered cities: Calgary, Vancouver, Edmonton and Toronto. Five lenses are scored: Safety, Property Value, Accessibility, Census and Climate.
Layer 1 computes raw metrics per dissemination area from the source data: crime rates and drive times for Safety, assessed values and permit counts for Value, amenity counts for Access, income and employment for Census, hazard exposure for Climate — 34 sub-metrics in all.
Layer 2 normalizes. Each sub-metric is ranked with PERCENT_RANK against every other dissemination area in the same city, producing a 0–100 percentile. Metrics where lower is better — crime rate, emergency drive time — are inverted so a higher number always reads as a better result. Two families skip re-ranking because they already arrive on a meaningful 0–100 scale: the urban-noise score (an absolute acoustic scale) and all seven climate hazard scores (national percentiles, deliberately comparable across cities).
Layer 3 combines. Fixed, published weights turn each lens's sub-metric percentiles into five lens scores per dissemination area, and dissemination areas roll up to neighbourhoods by population-weighted averaging — the score follows where people live, not land area. The overall score is the unweighted mean of the five lens scores: an editorial choice to privilege no lens, published so you can disagree with it.
Layer 4 builds a 34-dimension vector embedding per neighbourhood, used for similarity search ("neighbourhoods like Kensington, but in Edmonton"). Each lens's sub-scores are z-scored nationally and length-normalized before comparison, so similarity measures the shape of a neighbourhood's profile across lenses, not whether its scores are high or low.
Data sources and update cadence
- The five lens pipelines
Every input to scoring is documented on its lens page: Safety, Property Value, Accessibility, Census and Climate each name their sources, cadences and limits
CadenceLens data refreshes on each source's cadence; scores recompute when the pipeline runs, typically on a weeks cadence
- Statistics Canada 2021 dissemination-area boundaries and populations
The scoring geography and the population weights for every neighbourhood rollup
CadenceFixed for the 2021 census cycle
- Municipal neighbourhood boundaries (Calgary, Vancouver, Edmonton, Toronto open data)
The neighbourhood polygons that dissemination areas roll up into — the same boundaries each city publishes
CadenceRefreshed when a city changes its boundaries
How the score is computed
PERCENT_RANK is the core operation, so its meaning is worth stating precisely: a neighbourhood's percentile is the share of that city's other neighbourhoods that rank below it. 50 is the median, not a grade — a 50 does not mean "mediocre", it means "typical for this city". 0 and 100 are the city's extremes, whatever the underlying values happen to be.
The within-city partition is deliberate. Crime categories, assessment methodologies and amenity densities are not measured consistently enough across cities for cross-city percentiles to be honest, so the pipeline never computes them — with one stated exception: climate hazard scores are national percentiles, because their inputs are consistent national datasets and cross-city comparability is exactly what a hazard question needs.
Missing data is handled by neutrality, not invention. A dissemination area with no noise model, no climate row or a suppressed census figure carries a 50 for that input — "typical", the least-informative honest value — rather than a guessed number or a zero that would masquerade as a measurement.
Scores refresh when the pipeline recomputes, on a cadence of weeks, tracking the slowest-moving source rather than the fastest. Every neighbourhood page and report shows the last-refreshed date and the methodology version it was computed under.
The overall score weighs the five lenses equally. Sub-metric weights within each lens are published on that lens's page.
What this score does not capture
Stated plainly, because a score is only worth trusting if its limits are.
- A percentile is a rank, not a measurement. It cannot say a neighbourhood is safe or affordable in absolute terms — only where it stands among its city's neighbourhoods on the stated inputs.
- Scores do not transfer between cities (climate excepted). An 80 for safety in Calgary and an 80 in Toronto mean the same rank within different distributions, not the same crime rate.
- A high rank in a city with a widespread problem is still a rank within that problem. Percentiles compress the underlying scale, and two adjacent ranks can hide a large real gap or almost none.
- Equal lens weighting in the overall score is a judgment, not a fact. If safety matters to you three times as much as value, the overall score is the wrong number for you — the per-lens scores and the preference-weighted tools exist for exactly that reason.
- Neighbourhood averages hide within-neighbourhood variation. A dissemination area is small, but a neighbourhood score is still an average across several of them.
- Scores lag their sources. A refresh cadence of weeks on top of source cadences of weeks to years means a sudden change on the ground appears in the scores with a delay.
Definitions
- PERCENT_RANK
- The SQL window function behind every percentile: for each value it returns the fraction of other values in the same city partition ranking below it, scaled to 0–100.
- Dissemination area (DA)
- Statistics Canada's smallest standard geography, typically 400–700 residents. All scoring happens at DA level before rolling up.
- Population-weighted rollup
- A neighbourhood's score is the average of its dissemination areas weighted by their populations, so empty land does not dilute the number and dense pockets count proportionally.
- Inversion
- For metrics where lower is better (crime rate, drive times), the percentile is flipped — (1 − rank) × 100 — so that higher always means better across every published score.
- Neutral 50
- The value assigned when an input is missing for a dissemination area. 50 is the median — the least-informative honest assumption — so gaps neither punish nor reward a neighbourhood.
- Overall score
- The unweighted mean of the five lens scores. An editorial choice to weight no lens above another, stated so you can substitute your own weighting.
- Similarity embedding
- A 34-dimension vector per neighbourhood (8 access, 7 census, 7 climate, 7 safety, 5 value dimensions), normalized so similarity compares the shape of a profile across lenses rather than its level. Powers cross-city "neighbourhoods like this one" search.
- Methodology version
- The version stamp shown on neighbourhood pages, identifying which pipeline revision computed the scores on display.
Frequently asked questions
What does a score of 50 mean?
The city median: half of the neighbourhoods in the same city rank above it on that lens, half below. It is not a failing grade — it means typical for that city, on the stated inputs.
Why can't I compare scores between cities?
Because each score is a rank within one city's distribution, and the underlying data is not measured consistently across cities — police services count crime differently, assessment authorities value property differently. The one exception is climate: its hazard scores are national percentiles built from consistent national datasets, so they do compare across cities.
How often do scores change?
The scoring pipeline recomputes on a cadence of weeks, and each input refreshes on its own source cadence — weekly for permits, annual for assessment rolls, a census cycle for demographics. Every page shows its last-refreshed date and methodology version.
Who decides the weights?
PickYourPlace does, and publishes every one — the sub-metric weights on each lens page, the equal lens weighting in the overall score. They are editorial judgments about what typically matters, stated so you can disagree; the preference tools let you re-weight lenses to your own priorities.
How does neighbourhood similarity search work?
Each neighbourhood is embedded as a 34-dimension vector of its sub-scores, normalized per lens so that each of the five lenses contributes equally. Similarity is the cosine between vectors, which compares the shape of two neighbourhoods' profiles — strong on access, weak on value — rather than whether their scores are high. That is what lets the tool answer "like Kensington, but in Edmonton" across cities where raw percentiles do not transfer.