Thread by @yohaniddawela

Meta is using Facebook and mobile phone data to produce super-granular household wealth estimates.

Here’s what you need to know about the Relative Wealth Index:

The Relative Wealth Index (RWI) is a geospatial measure, which shows wealth disparities within countries.

Main features:

• It's open source
• It measures 'asset-based' wealth
• It covers 135 countries in the world
• Data is provided at 2.4km resolution

1. Methodolgy—Data Inputs:

It uses 'ground-truth' survey data from @DHSprogram (covering 1.4m households across 67,000 villages in 56 countries).

It then uses a range of alternative data:

• mobile phone data
• topographic maps
• aggregated Facebook connectivity data

@DHSprogram It also uses daytime satellite imagery, that's processed in a way I outline here:

https://twitter.com/yohaniddawela/status/1739983291032150192

@DHSprogram 2. Methodology—ML Models:

Machine learning (gradient boosting) is used to train a model on the relationship between the alternative data and survey data.

This is then used to predict wealth (out of sample) for 2.4km x 2.4km grids across 135 countries.

@DHSprogram Based on this, the RWI:

• is a relative index within each country at the time of the survey.
• has a mean value of zero and a standard deviation of one.

The scores cannot be compared across countries or over time.

@DHSprogram Model accuracy:

The ML model explains 72% of the variation in wealth, as measured with independent census data from 15 countries.

However, when compared to coordinate-level data collected by the Nigerian government, the model explained 50% of variation (at the grid-level).

@DHSprogram However there are a number of things to be aware of:

• larger errors in regions far from survey areas
• the model's accuracy is higher when data is aggregated to the local-government level

@DHSprogram Further questions on accuracy:

A study in 2023 looked at the RWI in Indonesia.

It found that using the RWI to pinpoint the poorest 14% of the population. showed mixed results.

The error rate was high—50.65%.

I.e. half of Indonesia's poorest regions were incorrectly identified

@DHSprogram And surprisingly, some areas RWI labeled as poorest were among the wealthiest.

@DHSprogram Takeaway:

Meta's RWI is a novel way of estimating asset-wealth at a household level.

But it still has limitations.

And this is the key message when seeking granular insights into wealth & GDP:

Nothing's perfect—it's just important to know what limitations each dataset has.

@DHSprogram If you're interested in other ways of measuring local GDP, check out this post:

And give us a follow @yohaniddawela for more breakdowns on geospatial and economics topics.

https://x.com/yohaniddawela/status/1782368514994581774

@DHSprogram Interested in going deeper?

I provide more in-depth tutorials and analyses in my newsletter.

You can subscribe here: yohan.so