One thing which may be newly possible in the last few years is getting satellite imagery of the country and using AI to count houses. With appropriate methodology, this is far more likely to be accurate than relying on bureaucratic reporting and/or projections, although there are the obvious pitfalls and probably some non-obvious ones too in backing this out to population. I believe MSF used to do something like this for their deployment areas but I haven’t heard of it attempted at a countrywide scale.
Small houses in a developing country may be occupied by between 0 and 20 people depending on abandonment, age profile and overcrowding (all of which tend to change more in the short term than construction), and that’s before considering apartments. You might be able to pick up on a few concealed trends like the emergence of new townships in a region which notionally has unchanged population but the margin for error is probably bigger than the apparently huge margin for error in Nigeria’s official population records.
To take the example from the OP: one of the major problems was a census permitting up to 9 people where some regions ended up with every household reporting nine people.Satellite data won’t tell you which houses have claimed more residents than actually exist (or indeed which houses underreport residents because the number of households containing more than nine people won’t be zero)
We use this for our healthcare mapping tool health AIM to accurately estimate the population in healthcare “black holes” where we launch health centers.
The problem is that house counts alone doesn’t get you to an accurate population estimate. You need to know the number of people living in each house, which varies wildly between direct staff. We used to use a world bank estimate of 1.8 per hut or something in our area but that’s far too loose to check population estimates.
An interesting method might be to check world pop’s estimates over specific small sample sizes and then physically visit those places and see whether the online counts were consistently higher, lower or similar to the real life counts. I would imagine with a few hundred samples of 15-30 households that might get close to answering the question (can’t be bothered doing the power calculation). Could probably do that for somewhere between 50k and 100k for what it’s worth.
One thing which may be newly possible in the last few years is getting satellite imagery of the country and using AI to count houses. With appropriate methodology, this is far more likely to be accurate than relying on bureaucratic reporting and/or projections, although there are the obvious pitfalls and probably some non-obvious ones too in backing this out to population. I believe MSF used to do something like this for their deployment areas but I haven’t heard of it attempted at a countrywide scale.
Small houses in a developing country may be occupied by between 0 and 20 people depending on abandonment, age profile and overcrowding (all of which tend to change more in the short term than construction), and that’s before considering apartments. You might be able to pick up on a few concealed trends like the emergence of new townships in a region which notionally has unchanged population but the margin for error is probably bigger than the apparently huge margin for error in Nigeria’s official population records.
To take the example from the OP: one of the major problems was a census permitting up to 9 people where some regions ended up with every household reporting nine people.Satellite data won’t tell you which houses have claimed more residents than actually exist (or indeed which houses underreport residents because the number of households containing more than nine people won’t be zero)
Worldpop has taken this a step further and combined census population counts with house counts to estimate population.
https://www.worldpop.org/
We use this for our healthcare mapping tool health AIM to accurately estimate the population in healthcare “black holes” where we launch health centers.
https://health-aim.onedayhealth.cloud/login
The problem is that house counts alone doesn’t get you to an accurate population estimate. You need to know the number of people living in each house, which varies wildly between direct staff. We used to use a world bank estimate of 1.8 per hut or something in our area but that’s far too loose to check population estimates.
An interesting method might be to check world pop’s estimates over specific small sample sizes and then physically visit those places and see whether the online counts were consistently higher, lower or similar to the real life counts. I would imagine with a few hundred samples of 15-30 households that might get close to answering the question (can’t be bothered doing the power calculation). Could probably do that for somewhere between 50k and 100k for what it’s worth.
Google has a great dataset that does exactly that: https://sites.research.google/gr/open-buildings/
It looks like they have even added building heights since I last checked!