Picking a persons-per-structure multiplier for a population estimate
Every inter-censal estimate built on structure counts comes down to one number multiplied by another: structures times people per structure. The structure count is the easy part now that imagery can flag built-up footprints at scale. The multiplier is where estimates go wrong, because it's the one input that quietly ages out of date while the district keeps building.
Where the multiplier actually comes from
Most offices start with the last census's average household size for the district, or the nearest comparable enumeration area if the district itself changed boundaries. That number is a ratio: de facto population divided by occupied dwelling units, sometimes adjusted for vacancy. It's a reasonable starting point. It is also, by definition, several years stale by the time you're running a mid-decade estimate, and in a district adding housing stock every year, household size rarely holds still.
Three things move it in practice:
A shift in housing type changes occupancy on its own. A district that was mostly single-family plots five years ago and is now adding walk-up apartment blocks won't carry the same persons-per-structure ratio into the new stock, even if household size per se hasn't changed, because a four-story building holds several households where a plot held one.
In-migration tends to push household size down before it pushes it up. New arrivals often double up temporarily, which inflates occupancy in the short run; as housing catches up, that compresses back toward the household-formation rate for the area. Which direction you're in matters for which multiplier you pick.
Informal or incremental construction complicates the denominator. A structure footprint from imagery doesn't tell you whether it's one household or three sharing a compound, and administrative records for that housing type are often the thinnest in the city.
When one multiplier isn't enough
The honest answer for a fast-growing district is usually that a single district-wide multiplier is too blunt. Demographers who've been burned by this tend to stratify: one occupancy assumption for the established residential core where the last census count is still a decent anchor, a different one for new subdivisions built since, and a third for anything that reads as informal or mixed-use in the imagery. A building-height layer helps here, because it lets you separate single-story footprints from multi-story ones before you apply any ratio, rather than assuming every structure polygon is a single household.
Where you can, pair the multiplier with something independent to sanity-check it: utility connections, school enrollment ratios, voter registration density, or a recent household survey from an adjacent area with similar housing stock. None of these replace a multiplier built on structure counts and building heights, but they tell you whether the number you've picked is in the right range before it goes into a published figure.
Document the assumption the way you'd document any other survey parameter: what the baseline ratio was, what year it's from, what adjustment you applied and why, and what you'd need to see to revise it. An estimate built on a household size assumption inter-censal offices can trace back to its source holds up under the kind of scrutiny these numbers usually get, whether that's a parliamentary question about service planning or a humanitarian agency sizing a response.
What this looks like with current imagery
A built-up area layer, a structure count and a building-height layer over the district give you the denominator split the stratified approach needs: footprint area, structure counts, and story count, without a field count. That's the gap Population Estimation is built around: converting those three layers from recent satellite imagery into the inputs a persons-per-structure multiplier actually needs, for the years between censuses when the last headcount is too old to trust on its own.
If your district's last multiplier predates the subdivisions and apartment blocks going up at its edges now, it's worth checking it against a fresh structure count before the next published estimate goes out.