How accurate are small-area population figures?

Small-area population figures are estimates, and their accuracy depends on scale. The figure for one tiny cell of a city can be noticeably off, because it is spread from larger official counts onto the ground. Add up a few hundred cells, say everything within 1 km of a site, and the errors tend to cancel out, so the total is far more dependable than any single cell in it. Use these figures to compare areas and to rank them, and be careful about using one cell, or one exact number, as if it were a head count.

This piece is for anyone about to base a decision on how many people live near a spot: an investor checking a pitch, a planner sizing a service, a retailer weighing a lease. It explains where the uncertainty comes from, how to tell when it matters, and simple checks that take minutes.

Where the figures come from

No one counts every household in every block every year. Official censuses count people in districts that are often much bigger than a street, and only every several years. A population grid takes those counts and spreads them over the land where homes are, giving each small cell its share. In xMap Flows that grid is made of hexagons of about 0.1 km², with residents, men and women and nine age groups for each.

The method is sound for the purpose, but it brings three kinds of error that are worth knowing by name.

• Where inside a district. A district's people have to be placed on the right blocks. If one block holds a tall apartment building and the next holds low houses, the split between them may not be exact.
• When. The grid reflects the counts and building patterns it was made from. A housing estate finished last year may be under-counted; a block cleared for redevelopment may still show residents.
• Who counts as living there. Students, seasonal workers and people in care homes or barracks are counted differently from place to place, so areas full of them are less certain.

Why bigger areas are safer

Think of each cell's error as a small push up or down. In one cell the push is the whole story. Over two hundred cells, some are pushed up and some down, and most of those pushes cancel. That is why a figure for everything within 1 km (a circle of about 3 km², some thirty cells) is more reliable than any one cell, and a figure for everything within 3 km is more reliable again.

A worked check in Bangkok

This example is made up to show the checks. A clinic group is considering a shophouse unit off Sukhumvit Road and is told "about 9,000 people live within 500 m". Before relying on it, the analyst runs three quick checks.

1. Change the distance. She measures 500 m, 750 m and 1 km around the same point and gets 9,100, 19,800 and 33,500. The circle grows by a factor of about 2.25 and then 1.8 in area; residents grow by roughly 2.2 and then 1.7. The figures rise in step with the area, so nothing looks odd.
2. Move the point. She shifts the centre 200 m along the street in each direction. The 500 m figure moves between 8,300 and 9,600. That spread, about 15 per cent, is a fair idea of how precise the 500 m number is at this spot.
3. Look at the cells. On the hexagon map, one cell inside the circle is far darker than its neighbours. A look at the street shows a large condominium tower; the dark cell is believable.

Her conclusion for the investment memo: "about 8,000 to 10,000 people within 500 m, and about 33,000 within 1 km". A range, honestly given, survives questions better than a single number to the person.

When the uncertainty matters, and when it does not

A guide to using small-area estimates; it is not a measured error rate.

Mistakes to avoid

• Quoting exact numbers. "9,137 residents" claims a precision the method does not have. Round to the nearest few hundred, or give a range.
• Reading age groups in tiny areas. Nine age groups in one cell can be a handful of people each. Group them (children, adults, seniors) or widen the area.
• Treating a blank as zero people everywhere. Blank can mean no homes, or a country the grid does not cover. Population figures exist in some countries only, among them France, Canada, India, Thailand, Egypt and Australia, and not the United Kingdom or the United States.
• Comparing with a different source. Another provider's figure for the same spot will differ, because the method differs. Compare areas within one source.

Doing the checks with an AI employee

The checks above take seconds when an AI employee runs them for every site on your list.

1. Put your candidate sites in a My data table.
2. Add three Demographics blocks one after the other, each with its own distance: 500 m, 1 km and 2 km. Rename each block's fields with a Calculated field after it, for example people_500m = population, so the next block does not overwrite them.
3. Add a Calculated field that rounds each figure, such as people_1km_rounded = round(people_1km / 500) * 500, and flags sites where the figures do not grow with the area, such as check = people_500m * 2 > people_1km.
4. End with Email or a Google Sheets app block.

What lands in the inbox is each site with its residents at three distances, rounded, and a yes or no flag for any site worth a second look on the map. Ten minutes on the flagged ones is usually enough.