Hacker News Reader

Comment by astazangasta | Hacker News Reader

This is a dataset designed to study genetics in a cohort of mostly white people from Framingham, MA. How can anyone claim to be able to draw conclusions about the genetics of friendship from this cohort? This is such a biased dataset I can't believe this got published.

For the record, I am Indian, my best friend growing up was Italian, my current best friends are a Russian Jew, a Scotch Irish mix from Woburn, and a Canadian of some uncertain European temperament. I have never had a best friend even close to being my fourth cousin.

View on HN

They explicitly call this out and apply population stratification later in the paper.

Yeah, which means the extensibility of these claims is exactly as good as those corrections, which can't possibly hope to model the missing information (the actual genetic distribution of friends).

Given that the US is a lot more racially heterogenous than the majority of the world (where the average person would have friends of a closer genetic makeup), I would bet that if anything these results are skewed the other way.

Exactly. These results would make sense in almost every part of the world, except the US.

    > Given that the US is a lot more racially heterogenous

The US? Sure, but this isn't about the US.

The population chosen (or available) for the study does not reflect that diversity.

https://www.framinghamheartstudy.org/participants/index.php

    > The Original Cohort of the Framingham Heart Study consisted of 5,209 respondents
    > of a random sample of 2/3 of the adult population of Framingham, Massachusetts

So I don't understand why you point to diversity in some wider but not included group to reason about results based on the much more limited one.

Stratification doesn't help much - it doesn't generate missing data, it still is based on that limited population.

The study isn't "wrong" of course, it clearly states what data it is based on and what methods they employed. If there's anything wrong it's the interpretation by others. This is merely a hint "maybe worth having a closer look at this", it isn't meant to be used to draw any final conclusions from. Papers like this are for "internal use", for other researchers in the field, it should never go to the wider media. It's like a programmer showing their very first lines of code for a new idea, and then trying to draw conclusions about the quality of the final software to be released five years later, written by different people in a different programming language. Bad example, because here at least you already know what it will be about while in the study this remains open, so I reframed it to be about "(software) quality".

I used to live in Malden, and IIRC Framingham and the other suburbs were pretty diverse. That's what I was basing my comment on.

Of course, my personal bubble could have skewed this -- looking it up it seems like Malden is more diverse than Framingham, but Framingham seems to still be close enough to the overall distribution in the US. Not the same of course.

shrug

> The study isn't "wrong" of course, it clearly states what data it is based on and what methods they employed. If there's anything wrong it's the interpretation by others. This is merely a hint "maybe worth having a closer look at this"

Right, that's all I'm really talking about -- the paper is very responsible about possible bias, and tries to correct it (as you said, that's not completely possible), very far from "I can't believe this got published" territory as the original commenter was mentioning.

Okay, here is what you wrote:

    > Given that the US is a lot more racially heterogenous than the majority of the world

"The US". You didn't say "Framingham". You know, it's really a PITA to have discussions with someone who can't just say "I made a mistake, I edited my comment" or something like that and instead try anything else.

Sure, I made a mistake. I was using prior knowledge of Framingham/the region that I didn't spell out in my first comment. I should have done that, but I was lazy and it didn't occur to me. Happy? I spelled it out in the next comment to explain the basis of the previous one, which I felt was sufficient. Evidently it wasn't; because you want explicit admission of fault. Here you go.

The problem is all they're measuring is the genetics of this locality, or any locality. You acquire your friends mostly through geographical proximity; in most places, the people nearest you are going to have a certain level of relatedness.

But this has less to do with friendship and more to do with geography; they might have achieved a similar result by comparing people to their neighbors.

The genes that you get from your ethnicity are a small fraction of your overall genome. There are tons of other genes that affect things like your height, your interest in music, etc, that could strongly affect your friend selection.

And I got those in some other way than my ancestry...?

I'm not sure you understand the whole genetics thing.

Reply on news.ycombinator.com