The limits of our own data

Why publish this

We put numbers in public. Anyone doing that owes readers an account of where the numbers are soft, written before someone else finds it. A statistic with its limits attached is worth more than a rounder one without them.

This page is the standing caveat behind every data study on this blog.

1. It is a directory of public listings

We hold businesses that have appeared in a public source or been submitted to us. Nigeria has an enormous number of traders who have never appeared in either: the shop with no signboard, the mechanic working from a shared yard, the caterer operating entirely on WhatsApp.

So every count we publish is a floor. When we say a city has 4,400 businesses, the honest sentence is "we hold 4,400 businesses in that city". The undercount is not evenly spread, and it is worst exactly where informality is highest, which means our data systematically under-represents the smallest operators.

2. Closures lag

A business that shuts does not tell anyone. It stops answering the phone, and the listing stays up until someone reports it or a refresh finds it gone. In a market with high turnover among small businesses, a meaningful share of any directory is describing something that no longer exists.

If a listing matters to your plans, call before you travel. That advice applies to us and to every competitor.

3. We inherit review bias we did not create

Most ratings in our data originate with Google. That means we inherit every bias in voluntary reviewing: happy-or-furious selection, campaigns run by owners, and the enormous mass of businesses nobody has reviewed at all. We have written about how badly this distorts the star average, and the honest position is that we can measure the distortion but not remove it.

4. Classification is ours, and it is imperfect

We assign 1,249 categories. Category boundaries in Nigeria are blurry by nature: a pharmacy that sells provisions, a hotel with the busiest restaurant in town, a "hospital" that is one doctor and two rooms. When we publish category counts, some of the spread is real commerce and some is our own labelling.

We would rather say that than present a category table as though it were an official industrial classification.

5. Coverage is uneven, so comparisons need care

Where we have crawled and verified more thoroughly, counts are higher. That is a fact about our effort, not about the city. It is why our city comparisons use percentages within a city rather than raw totals across cities, and why we restrict city-level breakdowns to places where we hold enough listings for a percentage to mean something.

What we do not do

• We do not invent numbers to fill a gap. If a figure is not in the data, the sentence does not get written. A blog post is not worth a fabricated statistic.
• We do not sell ranking position. Featured and Sponsored placements exist, are labelled where they appear, and change nothing about the order.
• We do not treat a self-reported claim as verification. An owner telling us something makes it profile data. Evidence makes it verified. The two are shown differently on purpose.

If you find an error

Tell us and we will fix it, including in public where we published the wrong number. Corrections on a data page are cheap. A reputation for quietly editing them is not.