How to validate a population estimate against utility connection records

Utility connection records are one of the few administrative data sources that update faster than a household survey and cover more ground than a sample frame. If you're building an inter-censal estimate for a district where the last enumeration is five, eight, ten years stale, the electricity utility's connection ledger is usually the first thing a reviewer asks you to reconcile against. Here's a working method, and where it breaks.

Start with the connection count, not the population figure

A utility doesn't hand you a population number. It hands you a count of active accounts, billed or metered, by feeder or substation zone. Your job is to convert that into an implied population, then compare it to what your structure-based estimate says.

The conversion is simple in form and messy in practice:

implied population = active residential connections × average household size × occupancy rate

Average household size should come from your most recent census or a recent household survey for that district. A national average smooths over exactly the household-size variation you're trying to measure, which defeats the point of the cross-check. Occupancy rate matters more than people expect in peri-urban growth corridors, where a meaningful share of new connections sit on plots with a completed structure but no one living there yet, or on structures under construction that got connected early because the developer wanted power on site before move-in.

Run that arithmetic for the smallest zone the utility will give you. District-level connection totals are close to useless for validation; they average away exactly the spatial unevenness you're trying to check. Ask for feeder-level or, better, enumeration-area-level connection counts if the utility's billing system can cut data that way. Many can, because they already bill by zone for tariff and loss-tracking purposes.

Where the cross-check agrees, and where it won't

When your satellite-derived structure count and the utility's connection count move together across zones, that's a reasonably strong signal your built-up area and structure detection are tracking occupied housing stock at the right scale. Discrepancies are where the actual work is, and they fall into a few recurring patterns:

• Shared and multi-tenant connections. One meter serving a compound with three or four households will undercount population if you assume one household per connection. This is common in informal and transitional settlements, less common in planned subdivisions with individual plot connections.
• Unmetered or informal connections. Fast-growing districts, especially on the urban fringe, often have housing that's connected informally before the utility's records catch up, or never gets billed at all. Your structure count from imagery will show these buildings. The utility's ledger won't. Expect your imagery-based estimate to run ahead of utility-implied population in exactly these zones, and treat that gap as information about informal growth rather than an error to split the difference on.
• Commercial and mixed-use accounts tagged residential. Billing categories drift. A shopfront with an apartment above it sometimes gets filed under whichever tariff class was easiest at connection time, not the one that matches actual occupancy.
• Lag between structure completion and connection. A building can show up in high-resolution imagery as complete and roofed well before the household moves in and the utility finishes the connection paperwork. In a fast-growing district this lag alone can produce a visible gap between the two counts that closes within a year or two without any error on either side.

None of this means the utility data is unreliable or that the imagery-based count is wrong. It means a single cross-check is a diagnostic, not a verdict. The useful output isn't "the two numbers matched" or "they didn't." It's a map of where they diverge and a short, defensible explanation for each cluster of divergence, which is exactly the kind of annex a national statistics office needs when it publishes an inter-censal figure and has to answer for it later.

Water utility connections, where available, are worth the same treatment and often track differently from electricity in informal areas, since informal power connections are more common than informal piped water. Running both where you can gives you two independent checks instead of one.

Population Estimation builds the other half of this comparison: a structure count and built-up area layer derived directly from satellite imagery over your area of interest, cut to the same zones your utility data comes in, so the two counts line up from the start instead of needing to be reconciled after the fact. If you're assembling the evidence for a district-level figure between censuses, that's the layer worth starting with.