How accurate is a median price per square metre? Read the interval

In SW11, Battersea, the median price per square metre is £9,260/m², from 732 matched sales in the 12 months to August 2026. Beside it on the district page sits a second figure that most readers skip: the interval, £9,030–£9,490. That second figure is the honest one. It says how far the median could move if the same year had produced a different set of sales, and it is the number to read before you lean on the first.

No official house-price producer publishes one. The UK House Price Index prints a point, the ONS small-area statistics print a point, the lender indices print a point, and none of them tells you how many sales the point rests on or how far it might be wrong by. A postcode district is a small sample by construction, so on this site the interval is not an appendix. It is part of the figure.

What does the interval actually say?

Take every matched sale in the district over the window, divide each price by its floor area, and sort the results from cheapest to dearest. The median is the middle value. The interval is the pair of positions in that sorted list, one below the middle and one above, between which the true middle of the market would fall 95 times in a hundred if you could rerun the year. Which positions those are is a question the binomial distribution answers exactly, for any sample size, without assuming the prices are shaped like a bell curve.

Statisticians call this an order-statistic interval. Two things about it are worth knowing. It is exact rather than approximate, so a district with thirty sales gets an interval that means what it says, not one borrowed from a formula that only works at three hundred. And it need not be symmetric. Prices per square metre bunch up at the cheap end and stretch out at the dear end, so the arm above the median is often a little longer than the arm below it, and the interval is allowed to show that.

What it does not cover is as important as what it does. The interval describes sampling uncertainty only: the noise from which homes happened to sell. It says nothing about whether an EPC floor area was measured well, nothing about the fifth of sales that could not be matched at all, and nothing about the sales still waiting to be registered. Those are separate questions, and this site answers them separately.

How much does the interval narrow as sales pile up?

A lot, and on a predictable curve. Across the 2,235 districts that publish an interval, this is how its half-width, the distance from the median to either end as a percentage of the median, falls with the number of matched sales behind it:

Read the first and last rows together. A district with 30–49 matched sales typically carries an interval of about ±9.4% around its median; one with 500 or more carries about ±2.3%. The shape is the square-root law that governs every sample statistic: to halve the width you need roughly four times the sales, which is why the step from a thin district to a busy one is so much larger than the step from a busy one to a very busy one.

Two districts make the point better than the table does. CR0, New Addington in Croydon, one of the busiest districts in the country, publishes £4,860/m² from 956 matched sales, with an interval of £4,790–£4,940, a half-width of 1.3%. B10, Small Heath in Birmingham publishes £2,300/m² from 38 matched sales, with an interval of £2,000–£2,600 and a half-width of 11.0%. Same method, same confidence level, a different amount of evidence. The medians are printed to the same three significant figures, and only the interval tells you that one of them is far better known than the other.

Why does a wide interval not mean the figure is wrong?

Because width is about precision, not bias. A wide interval says the data can support only a loose statement; it does not say the statement is false. The widest intervals on the site belong to districts where few homes sell, or where the ones that did sell were unusually varied. BD1, central Bradford publishes a median of £2,270/m² inside an interval of £1,080–£4,230, and its page rates confidence Low for exactly that reason. The median is still the best single estimate the record can give. The interval is the record admitting how much it does not know.

The opposite failure is the one to watch for. A narrow interval around a biased figure looks like precision and is not. If most of a district’s detached houses never found a floor area, the median is a precise description of the homes that did, which is a different market from the one the reader lives in. That is why the district page prints the match rate beside the interval, and why the two are scored separately in the confidence rating below.

How does the interval become a confidence rating?

Every district page carries a one-word rating, High, Medium or Low, and the interval is the largest part of it. The score weighs three things: the precision of the median, read off the interval, at 0.60; the share of sales that carried a floor area, at 0.25; and how recent the sales behind the figure are, at 0.15. Those weights are a choice, and they are published on the methodology page so that they can be argued with. Across the current dataset, 1,392 districts rate High, 706 Medium and 137 Low.

SW11 is rated High: 89% of its sales matched a floor area, the sales are recent, and the interval is 2.5% either side of the median. The rating on a page is a shorthand for those three facts and never a substitute for reading them. A Medium with a narrow interval and a poor match rate is a different Medium from one with a wide interval and a full match, and the page shows you which you are looking at.

Is the interval itself checked?

Yes, and by an independent route. The site also computes a relative standard error for every median from the spread of the sales around it, and derives a second interval from that, the way a textbook would. Where the two agree, the order-statistic interval is doing its job. Across the districts, the ratio between the two half-widths sits at 1.06 at the median district, and 78.2% of districts have the two within a quarter of each other. Two methods with different assumptions landing on the same width is what a reader should want to see before trusting either, and both numbers are published rather than asserted.

There is one thing this cross-check cannot do, and the repeat-sales article makes the same admission about its own intervals: agreement between two measures of sampling noise says nothing about a bias they share. If the floor areas are systematically off, both intervals are precisely wrong together.

How should you use it when you are buying or selling?

1. Read the ends, not the middle. If you are pricing a home against its district, price it against the whole interval. A flat at the top of the range is not “above the median” in any sense the data can defend; it is inside the band the data cannot rule out.
2. Do not rank two districts whose intervals overlap. SW18, Earlsfield publishes £8,390–£8,730 and E17, Walthamstow publishes £7,580–£7,860. Put any two districts side by side like that and look at whether the bands touch before you decide one is dearer than the other. The comparison guide makes the same point about windows and tenure.
3. Treat a price outside the interval as a claim. An asking price that works out above the top of the district’s band is asking you to believe the home is better than the market’s middle, and the seller should be able to say why. One inside the band is a price the data is silent about, which is not the same as a fair one.

What will the interval still not tell you?

Three things, and they are the reason it is a starting point rather than a verdict. It cannot tell you what the sample is made of: a district median is a median of whatever sold, and if this year’s sales skewed towards flats the band has moved for a reason that has nothing to do with value, which is what the median-and-mean article is about. It cannot tell you about the most recent weeks, because registrations reach the record two to eight weeks after completion and the newest quarter is always partly missing. And it cannot tell you about your home, only about the middle of its district; the interval on a single sale is the whole distribution. And it cannot tell you where the market is going: the 2026 trends article reads the ten-year series for that, with an interval on every point of it.

Every figure on this page, the medians, the intervals, the counts and the shares, is read from the current dataset as the page is built, currently August 2026, so an example here and the district page it links to cannot disagree. Everything is rebuilt monthly from HM Land Registry sold prices matched to EPC floor areas, and the glossary defines every term used here. Every figure is per square metre; if you think in square feet, the unit article converts both the median and its interval.