Why different companies give different percentages

Two sisters in Petaling Jaya each take a test from a different company. One report says 70 per cent one region. The other says 55. They have the same parents. "One of them must be wrong," says one sister. Not quite.

The story

A region estimate is not a measurement like height or weight. It is a comparison: your DNA against a set of reference groups. Change the comparison, and you change the answer. Three things differ from company to company.

1. The reference groups. Each company builds its own set of reference people. Mathieson and Scally (2020) explain that measures of similarity are very sensitive to the choice and the labels of reference populations. One company may have many people from southern China and few from Southeast Asia. Another may have it the other way round. Categories are also an arbitrary choice: one company may have a group called "East Asia", another "China" and "Southeast Asia" separately. The same piece of DNA can fall under different names.

2. The method. There is more than one way to turn DNA into a summary. Mathieson and Scally list three popular ones, called principal component analysis, ADMIXTURE and Chromopainter. They look at the same genes in different ways and give different pictures. None of them is "the family tree".

3. The coverage. The authors note that reference panels are small or biased for many peoples outside Europe, so a person can be placed with a better-represented neighbour. That is a bigger problem in Southeast Asia than in Europe, which is why local samples matter.

On top of all that, today's groups are an imperfect map of the past. Pickrell and Reich (2014) stress that migration, mixing and replacement have changed the genetic map, so no reference group is a perfect picture of the ancient people it is named after.

The sisters can both be right. Each report is a fair answer to a different question. They are two photographs of the same person, taken from different angles.

Where you can still see it today

In the names of groups, which differ between companies.
In the way broad regions tend to agree between companies, while fine detail differs. Mathieson and Scally say similarity is informative at broad scales and less so at fine ones.
In the "versions" of a report: when the references are updated, the numbers can move.

What DNA can tell, and what it cannot

A region estimate can give you a rough, consistent story at the level of large regions, such as East Asia, South Asia and Southeast Asia.

It cannot give an exact "true" percentage. When two companies differ, look for the parts they agree on, and treat the rest as detail that may change. A mother line or father line is a separate clue from the whole-DNA estimate.

DNA does not decide race, religion or bumiputera status. A percentage is a summary of similarity, and it is not who you are.