Researchers have trained a machine to spot depression on Instagram

technologyreview.com

38 points·pmcpinto··6 comments

Previous Discussion

https://news.ycombinator.com/item?id=12278722

Good catch; thanks.

These are interesting results, and some factors like "number of likes" have pretty intuitive correlations with happiness.

The color thing makes me wonder about confounders, though - without filters a photographer in Norway will get more dark, monochrome photos than one in Jamaica. And sure enough, the Norwegian is more likely to be depressed. Is it possible that some of these variables are proxies for seasonal affective disorder and regions with high risk of depression?

Do you have some numbers for the Norwegians are more depressed than Jamaicans statement that control for the other differences between the populations?

Nope, I wasn't trying to assert that they are. My point was just that you can 'predict' depression to a limited extent by knowing where people live, and that the features of (outdoor, unfiltered) photos are a proxy for where people live. That's a pattern that applies at the country level, without controlling for anything.

S.A.D. is a thing, and there's an argument that depression risk is higher in cold, sunless places, but I'll leave more-knowledgeable people than me to have that discussion.

So out of 170 study participants found on the internet, of whom those with any disclosed history of depression were removed, 70 were found to be clinically depressed on the basis of a screening questionnaire (http://cesd-r.com/). That's 41% of the sample.

Reply on news.ycombinator.com