The Primary School Premium Is Negative Until You Hold the District Still

The Resale Project Scorecard believes buyers pay for schools. The Primary School Effect carries 20% of the published weight, and as last week's variance decomposition showed, it accounts for 41.5% of the spread between one condo and the next. It is the single biggest reason two projects sit a grade apart.

Tested against what people actually paid, that belief looks wrong. Across 884 scored projects with at least three resale caveats in the year to October 2026, the ones scoring 3 or below on the school factor have a median of 1,734 psf. The ones scoring 9 or 10 have 1,662 psf. The projects the scorecard rewards trade 4.2% cheaper.

That comparison is broken, and the way it is broken is the useful part.

The pooled number is a map of Singapore

Primary schools are not distributed like condos. The prime districts were built out before the catchments were drawn, and they are where the most expensive homes are. Of the 219 projects in the bottom band, 31 sit in D1, D2, D6, D9, D10 and D11, at a median of 2,555 psf. In the same 884-project sample, exactly three central-district projects score 9 or 10 on the factor.

So the pooled comparison asks two questions at once. Does a school catchment carry a premium, and is the buyer in Orchard or in Sengkang? The second question is worth about a thousand dollars a square foot, and it swamps the first. A number that mixes them is not a finding about schools.

Hold the district still

The honest test compares projects competing for the same buyer. Nine districts have at least four projects on each side of the split:

The median gap is 8.4% and it is positive in eight of the nine. The sign flipped because the comparison changed, not because the data did.

Note how uneven it is. In D16 and D18 the catchment is worth a quarter of the price. In D14, D23 and D15 it is worth nothing measurable. A premium that averages 8% and ranges from zero to 28% is not a rule of thumb. It is a thing to check locally.

What survives a tighter cut

Projects with good school scores are also newer and larger, and both lift psf on their own. The median high-band project reached completion in 2013 against 2009 for the low band, and holds 321 units against 242. Those are real objections, so here are the narrower cuts.

Restricted to 99-year leasehold, five districts qualify and the median gap is 16.7%, positive in all five. Restricted further to 99-year projects completed from 2000 onward, four districts qualify and the median gap is 10.9%, again positive in all four. Across the three specifications, 17 of 18 district comparisons are positive.

The point estimate moves between 8% and 17% depending on the cut, which is what happens when each filter costs you half the sample. What does not move is the direction.

What the factor actually counts

The Primary School Effect is close to binary in practice. Of 2,350 scored projects, 683 score 3 or below and 1,572 score 7 or above. Only 95 projects, 4.0% of the board, sit anywhere in between. The median project in the bottom band has no primary school within a kilometre. The median project in the top band has three.

So the factor is not a smooth ten-point scale of school quality. It is mostly a count of schools inside a radius, weighted by recent Primary 1 ballot demand, and most projects are clearly on one side of the line or the other. That is worth knowing before reading a 7 as meaningfully different from a 9.

One more thing the sub-score hides. 103 projects, 4.4% of the board, score 1 on this factor because their address could not be geocoded, not because the catchment was measured and found empty. On a factor carrying 20% of the weight, a floor score is expensive. Those 103 are excluded from every price figure above, and 316 projects in the comparison have a measured zero.

What this does not show

It does not show that a catchment causes a price. Families who want schools also want other things, and the same demand that fills a school catchment fills the shops and the bus routes around it. The high band also scores better on the Project's Historical Performance factor, 7 against 5, so some of this premium is demand the scorecard is already counting on another line.

It is medians, not a regression. Each project contributes one median psf and each district one median gap, with no adjustment for floor, unit size or bedroom mix. A district is also a coarse control: D15 runs from Tanjong Rhu to Siglap, and a kilometre radius does not respect a district boundary. And the central districts cannot be tested at all on this data, because almost nothing there scores well on the factor.

Nothing here changes the published weights.

How to read it

If two projects in the same district are a grade apart and the gap is in the school line, the market broadly agrees with the scorecard, by somewhere between nothing and a quarter of the price depending on where you are standing. Check your own district rather than taking the average.

And if you are comparing across districts, ignore the factor entirely. Between a Sengkang project with several schools inside the radius and a Novena project with none, the school line is telling you about Singapore's building order, not about either home.

Method

Scorecard data from rps-data-v2.json as served at tribesg.com/rps-data-v2.json, retrieved 21 October 2026: 2,350 scored resale condominium projects with seven integer sub-scores each. The published weights (Project's Historical Performance 25%, Primary School Effect 20%, Project Size 16%, MRT Proximity 13%, Rental Yield 10%, Tenure 10%, Future Transformation 6%) reproduce the published score to a mean difference of 0.013 and a maximum of 0.030 across all 2,350, which validates the factor mapping before anything is built on it. The 41.5% share of spread is the covariance decomposition cov(wᵢxᵢ, total) ÷ var(total) on recomputed totals.

Price data from the Resale Price Check dataset built from URA REALIS resale and sub-sale caveats, as committed to the repository on 13 October 2026 and covering caveats lodged through 4 October 2026. Projects are matched between the two datasets by normalised name, then a looser form, then the pre-"@" stem within the same district; 158 scorecard names did not match and are dropped. For each matched project, the median psf of its caveats in the trailing twelve months (November 2025 to October 2026) is taken, requiring at least three caveats; 884 projects qualify. The bottom band is a school sub-score of 3 or below with a measured catchment; the top band is 9 or 10. Median schools within a kilometre: none in the bottom band, three in the top. The 103 projects whose catchment could not be geocoded are excluded throughout. District gaps are the difference between the two bands' median project psf, reported only where each band holds at least four projects (three in the age-restricted cut). Only aggregates are used; no individual caveat is published. The scorecard does not score price or asking psf, and nothing here is a view on any individual project's value.