Median or average: which HDB price should you believe?

“The average HDB price” is two different numbers wearing one name. When a headline says average it usually means the mean: add every price, divide by the count. When this site says median it means the middle resale: half came in above, half below. As of August 2026 that middle resale went for S$630,000, across 24,590 resales in the national window. In a large calm market the mean and the median sit close. In a single district in a year when a handful of million-dollar flats transact, they do not.

Why does the mean drift upwards?

The mean uses every dollar of every resale, so one exceptional resale moves it. A district that trades a few hundred flats around S$550,000 plus a dozen resales above S$1.2 million gets a mean well above what a typical buyer paid. Nobody bought at the mean. The median barely notices: the middle resale is still the middle resale.

An invented street makes the mechanics plain. Say five flats change hands at S$500,000 each and a sixth, larger and higher, goes for S$1.3 million. The mean of those six prices is about S$633,000, a figure more than a quarter above what five of the six buyers actually paid; the median is S$500,000, the price the middle buyer paid. Now let the sixth flat sell for S$2 million instead: the mean jumps again and the median does not move by a dollar. One number describes the money that changed hands, the other describes the typical purchase, and only the second is what most people mean when they ask what flats cost.

Singapore has made this concrete over the past few years. Million-dollar HDB resales went from a curiosity to a regular occurrence, concentrated in a small number of districts and blocks. Every one of them pulls a mean upward and leaves a median almost untouched, which is exactly the behaviour you want from a headline figure.

What does the register itself publish?

Not a mean, anywhere on this site. What the dataset behind these pages publishes instead is the distribution itself, stated as percentiles, and the three national ones are worth reading as a row of positions rather than prices: the resale a fifth of the way up the market, the middle one, and the one four-fifths of the way up.

Look at the two gaps rather than the three numbers. The step down from the median to the 20th percentile is S$77 per square foot; the step up to the 80th is S$144. The top of the market sits further from the middle than the bottom does, which is the statistical shape called right skew, and it is exactly the shape that drags a mean above its median. The register shows you the skew directly; it never needs to print a mean at all.

Where is the median weak?

It is not a free lunch. A median throws away information about the tails, and the tails are where the interesting flats are. That is why the district pages here also publish the 20th and 80th percentiles, and a full distribution chart: the median is the headline, the spread is the context, and the individual resales are underneath both. Compare Bishan and Sengkang and the medians tell you one thing, the spreads tell you another.

How many resales does a median need?

More than most quoted figures get. A median over a dozen resales can jump by tens of thousands of dollars when one flat drops out of the window, which is a property of small samples rather than of the market. The pages here handle that with a ladder: each area is measured over the shortest of a 12, 18 or 24-month window that reaches 30 priced resales, and an area that cannot reach 30 even at 24 months publishes its count instead of a figure. Each area median also carries a confidence interval and a grade, so a tight-looking number over a thin sample announces itself. When you see a district median quoted without a count next to it, ask for the count: the same rule that protects a mean from one outlier protects a median from one quiet month.

Why can a median move when no price does?

Because a median has a second weakness the mean shares: composition. The register records 5 different products under one roof over the current window, and their medians sit hundreds of thousands of dollars apart.

Line the table up and the trap is visible. A month in which more Executive and 5-room flats happen to transact lifts any blended national or district figure, mean and median alike, with not a single flat becoming dearer. That is why the pages here cut every district by flat type, why the rankings use a mix-adjusted rate that reweights each district to the national mix before comparing, and why comparing two districts on their headline medians alone is the commonest way to read a mix difference as a price difference.

So which number should you use?

Judging one flat against its market: use the median for the right district, flat type and lease band, and read the 20th to 80th percentile spread around it as the range of normal. Reading a headline: check which average it is before you let it move you, and if it is a mean, ask what sold in the window before believing the change. Comparing two areas: median against median, same flat type, same window. Every figure on this page is read live from the published dataset and is current to August 2026.