The transit premium, measured: what 69 Toronto station areas actually did

Uncategorized

One · the measurement

Three corridors, one question each.

Line 5 was announced in March 2007 and construction began in July 2011, both of which we were able to verify against the public record. Line 6 shares the 2007 announcement; its construction milestone is a different matter, set out in the method note. The Ontario Line appears at announcement only, because this dataset’s milestones stop in 2019 and its construction started after that.

Drag to turn it. Switch the layer and watch the skyline change — because how busy a market was and whether it beat that market are not the same map.

Drag to rotate · ⌘/Ctrl + scroll to zoom · click a station

Loading measurements…

Ring800 m
100%

Two · the first objection

“You measured the wrong ring.”

This is the first thing anyone says, and it is a fair challenge. Transit effects are supposed to be a walking-distance phenomenon, concentrated tightly around the entrance and fading fast. Toronto’s own framework draws major transit station areas at roughly 500 to 800 m, so 800 m is a defensible outer edge rather than an absurd one — but if the effect really does concentrate at the doorstep, tightening the ring should sharpen it.

So tighten it. The slider moves the catchment from 1,000 m down to 100 m. Each dot is one station area; the bar is the middle half of them.

—Median differential (pp)
—Median sales, before
—Under 30 sales
Beat its municipalityTrailed itFewer than 30 sales to measure on

The central figure barely moves. What moves is the evidence under it. At 800 m the median station area has 245 sales in the before window; at 100 m it has 4, and 20 of 35 stations have fewer than ten. A median computed on four transactions is not a measurement of a market, it is an accident of which four homes happened to change hands — so the scatter at the tight end is not a signal emerging, it is a measurement coming apart.

We tested it rather than eyeballing it. Pairing each station’s inner ring against the 800 m catchment that contains it, across 14 comparisons, 13 are indistinguishable from a coin flip. The one that clears the 5% line lands at p = 0.033 — out of 14 tests, roughly what chance delivers, and nine times weaker than the threshold that would survive correcting for them.

Spread, meanwhile, moves with the ring: Spearman r = -0.84 across the ten radii. We would not call that a discovery. The ten catchments are nested and their sample sizes rise together, and the scatter of a median mechanically widens as the sample behind it shrinks — so the coefficient largely restates the arithmetic rather than revealing something about transit. It is a description of how fragile the measure becomes, not independent confirmation of anything.

That test is weaker than we would like, and in a direction worth naming. These catchments are cumulative, so the inner ring is inside the benchmark it is being compared against — the 400 m sample is a median of 26% of the 800 m sample. A real inner-ring effect would be partly diluted into the thing measuring it, which biases the comparison toward finding nothing. The stronger test is a disjoint band, 0–400 m against 400–800 m, and this dataset cannot produce one: counts subtract across rings but medians do not, and there is no transaction-level data here to rebuild them from. So read this as a distance effect we could not detect with a conservative test, not as a distance effect shown to be absent.

A quarter mile is roughly 400 m in a straight line — and a 400 m walk along real streets reaches a smaller area than a 400 m circle drawn on a map. At 400 m this data gives -3.0 pp after announcement and +2.2 pp after verified construction, and that ring is statistically indistinguishable from the wider catchment that contains it.

Three · the second objection

“You measured the wrong moment.”

The second challenge is about timing. Markets price in news, so the lift should land at announcement — years before service, when the map changes and the concrete has not been poured. By the time trains run it is old news.

It is a good theory. The measurement is more awkward than that, and the awkwardness is worth showing rather than averaging away.

After announcement59 station areas
-2.5 pp24 of 59 beat
After construction began26 areas · verified construction milestones only
+1.2 pp14 of 26 beat
Finch West, 2015 window17 areas · filed as construction by the portal; actually procurement
-5.5 pp5 of 17 beat
After opening12 areas · five unrelated openings, half on thin samples
-7.1 pp4 of 12 beat

Announcement comes in slightly negative and verified construction slightly positive, and both sit so close to zero that neither resembles a premium. The difference between them is not nothing — a slim majority of station areas beat their municipality after construction, a clear minority did after announcement — but a median of a point or two in either direction is not the effect anyone is claiming.

The third row is why this chart has four bars instead of three. The portal files Finch West’s July 2015 window under construction, and construction there did not begin until 2018. Pooling those 17 station areas with the 26 verified ones gave a combined median of −0.5 pp and buried the distinction. Separated, the verified milestones read +1.2 pp on 14 of 26 station areas, and the 2015 window reads −5.5 pp. We show it rather than dropping it, labelled for what it is.

The opening row cannot settle anything either: 12 stations across five unrelated openings between 2002 and 2017, half of them with under 30 sales, its figure carried by station areas with 1, 4, 7, 8 and 11 transactions. Require an adequate sample and it inverts. And the openings people actually care about are absent entirely — Line 6 opened 7 December 2025 and Line 5 on 8 February 2026, while these milestones stop in 2019. That is the next study, not this one.

Four · what is actually there

The spread is the finding. Nothing we tested explains it.

None of this says nothing happened. Something happened, and it was large. Eglinton spans 41.5 points, from Kennedy at -9.2 pp to Chaplin at +32.3 pp. Finch West runs wider still at 90.6 points, from Milvan Rumike at -33.6 pp to Sentinel at +57.0 pp, in a window the method note explains. Neighbours on one line, moving in opposite directions.

So we went looking for what sorts them, and found no relationship that holds. Not distance and not timing: neither section above turned up a consistent one under this design. Not how active the market was — on Eglinton the correlation between transaction volume and outperformance is -0.09 across 23 station areas (p = 0.67), and the busiest market on the line, Mount Pleasant with 883 sales, returned +5.3 pp while Chaplin managed +32.3 pp on 227.

Price level is the only candidate that even looks promising, and it does not hold up either. On Eglinton it runs +0.37 — but across 23 station areas that is p = 0.08, which is not distinguishable from zero at the usual threshold. On the Ontario Line it runs the other way at -0.76, which does clear it (p = 0.01), on 10 station areas. Two corridors, opposite signs, neither firm enough to carry a rule.

And those p-values are generous to us. At 800 m the catchments of neighbouring stations overlap — on Eglinton adjacent pairs share about 36% of their area, on Finch West about 59% — so the same transactions feed more than one station area, and neighbourhood price movements are correlated across space whatever the transit does. These are not 23 independent observations. How far short of 23 they fall is not something the geometry alone can tell you: establishing an effective sample size would need an explicit spatial-dependence model, which we have not built. So read every p-value above as the optimistic end of a range, and note that this weakens the one result that cleared significance more than the ones that did not.

The spread in observed differentials is real. Whether it reflects like-for-like appreciation, or what is being sold in each window, this measure cannot separate — and no explanation we tested accounts for it either way. That is a result about the explanations, and a reason to look at the station rather than reach for another average.

Five · what to do with it

Not a multiplier. A question about the station.

The evidence here does not establish a universal transit-related appreciation premium, and it does not rule out a standing proximity premium either — by construction, it would not see one. What it does show is substantial variation in observed station-area price performance, even within a single corridor. Across 114 windows the typical station area did not outgrow its own market, and the gap between two stops on one line routinely dwarfs anything you could call a corridor average.

So the useful move is narrower than a multiplier and more defensible than one. Benchmark the station area against its own municipality over the same window, so you are measuring the difference rather than the market. Check the sample before trusting the number — thirty transactions is a working floor, not a guarantee, and below it the figure is noise wearing a decimal point. Measure the station you are actually lending against, because its neighbour four stops away may have gone the other way.

That variation deserves property-specific investigation rather than an assumed corridor-wide multiplier. The next step is to combine station-level observations with property characteristics, market composition, local supply and accessibility measures, and test which relationships actually hold — which is a research programme, not a number. And treat the openings themselves as unmeasured: nobody yet knows what Line 5 and Line 6 did to the land around them, because the windows have not closed.

Method and limits

What is measured, and in what units. The differential is the after-period station-area median minus the before-period median advanced by the municipality’s own growth ratio, divided by the before-period median. Algebraically that is one growth rate minus another, so it is reported in percentage points (pp), not percent: a station area growing 42.3% while its municipality grew 10% scores +32.3 pp. It compares growth against a benchmark. It is not a hedonic estimate and does not compare otherwise-similar homes at different distances from a station, so it cannot detect a price premium that was already in place and stayed in place. Read every figure here as relative appreciation, not as the value of proximity.

What it does not control for. This is a descriptive benchmark comparison, not a causal estimate: no matched controls, fixed effects, pre-trend tests or confidence intervals on the differentials. It also does not control for what is being sold. A station area that transacted more new condominiums in one window than the other can show a different median without comparable homes having moved at all, and some of the spread reported here is likely to be composition rather than appreciation.

Milestone dates, including one the source gets wrong. Eglinton’s milestones check out against the public record: announced March 2007, construction from July 2011. The portal dates Finch West construction to July 2015 — but Infrastructure Ontario records the Request for Qualifications in September 2015, early utility works in 2016, and construction beginning in fall 2018. We therefore read that window as an early-project period rather than a construction period, and do not treat it as equivalent to Eglinton’s 2011 milestone. Finch West’s 2007 announcement date is sound. The Ontario Line appears at announcement only: this dataset’s milestones stop in 2019, and its construction began after that — the City records works starting around the turn of 2021 into 2022, depending which milestone is counted.

Overlapping catchments. Station areas are not independent observations. At 800 m the catchments of neighbouring stations overlap — adjacent pairs share roughly 36% of their area on Eglinton and 59% on Finch West — so the same transactions contribute to more than one station area, and neighbourhood price movements are spatially correlated regardless of transit. Consequently the nominal p-values reported here may overstate significance. Establishing an effective sample size would require an explicit spatial-dependence model, which this study does not build; the overlap percentages describe the geometry, not the statistical dependence. Read every p-value as indicative, not as a definitive test of an independent station effect.

Rings. Catchments are cumulative, so rings cannot be subtracted, and a 400 m walk along streets reaches a smaller area than a 400 m circle drawn on a map. The portal assigns milestones per station, so an interchange can carry a neighbouring line’s dates; those stations are excluded from a corridor rather than averaged into it. Milestones in this dataset stop in 2019, so nothing here measures Line 5 opening on 8 February 2026 or Line 6 on 7 December 2025.

Which construction sample the radius figures use. Every construction figure in the distance section is the verified-milestone sample, the same 26 station areas as the construction bar. Finch West’s July 2015 window is carried as its own series and is not averaged in: pooling the two moves the 400 m figure from +2.2 pp to +4.3 pp, which is why the distinction is kept everywhere rather than only in the bar chart.

The four bars are not a paired comparison. Each milestone summary rests on a different subset of station areas and a different historical window — 59 at announcement, 26 at verified construction, 12 at opening. The differences between them are descriptive, and should not be read as a paired estimate of the effect of project timing. Nothing here follows one station area through all three.

Coverage, and why the counts differ. The dataset holds 69 station areas across the Toronto network; the three corridors drawn here account for 52 of them, the rest sitting on other lines and included in network-wide figures only. Within that, each figure is quoted on the subset that can actually support it, which is why the numbers move around: 59 station areas have a computable differential at announcement and 26 at construction, because not every station carries every milestone; 35 can be measured at the 100 m ring against 59 at 800 m, because the tightest catchments run out of transactions entirely. Any figure here is reported with the count it rests on.

Statistics. Correlations between station-level measures are Pearson, reported with their sample size and two-sided p-value; several are not distinguishable from zero and are described that way. The radius-versus-spread figure is a Spearman rank correlation across the ten rings. A test that fails to reach significance has not shown that an effect is absent, only that this sample cannot resolve one.

Sources. Market measurements — CUI Portal Transit Intelligence v6.8.0, price dataset. Alignments — City of Toronto open GIS, TTC Subway Route layer. A line still under construction has no published operational geometry, so its spine is drawn station to station and shown dashed: a schematic, not a surveyed alignment.

See your own corridor