Hacker News Reader

Comment by x86_64Ubuntu | Hacker News Reader

Even though there are only 3 comments in the thread, they are still rife with dismissive and bad-faith arguments. It's extremely telling how when black America mentions something about systemic racism in the context of the US, the response is "what's the p-value", "I don't like that sample size", "correlation != causation".

The article has studies with authors from the BIG 10 and with Ivy League heritage, but it's the mention of race that causes people to trot out canned and thinly supported reasons of doubt. Here is one of the studies if anyone is interested.

https://1410c6d1-d135-4b4a-a0cf-5e7e63a95a5c.filesusr.com/ug...

View on HN

I'm not sure. Psychology and sociology has such a horrible track record when it comes to reproducibility of studies that have some kind of political significance that we just have to be careful.

In case of these topics, there's a relatively strong publication pressure here.

Also: the raw data is unfortunately not enough without pre-registration. (For obvious reasons.)

"bad faith" is absolutely required in science. Especially in sociology and psychology, unfortunately.

I'm not sure there is a crisis of reproducibility in a study like this. We can tell with accuracy what the race of the physician is, the race of the baby and whether or not the baby dies or not.

We also can’t tell what is the death rate of all babies delivered by black doctors from this study

What does that have to do with anything?

Maybe they picked up outstanding doctors who deliver outstanding results regardless of patients race? I mean the dataset is so small, one off can derail any conclusion

> I mean the dataset is so small, one off can derail any conclusion

What do you mean? Per the Supplement, their data comes from a pool of ~4k white doctors and ~500 black doctors, with hundreds of thousands of observations/patients. This dataset doesn't seem "so small"?

[1] https://www.pnas.org/content/pnas/suppl/2020/08/12/191340511...

It really all depends on the methodology. I recommend some reading about 'p-hacking'. It is not that hard to find to find really shocking findings in large datasets -- the question is how hard you look. In this particular case, the question is how many hypotheses have been evaluated, i.e. has the author 'mined the data' for correlations. If yes, then the acceptable p value is much smaller than normal.

I think it's a bit of a reach to suggest that this is p-hacking. Some of the relevant authors have studied this hypothesis (a gap in pregnancy-related outcomes between black and white people) across multiple papers, and the effect found here -- ignoring any p calculation -- is pretty big. It's not like they're picking some extremely specific combination of factors.

And this is the narrative that I'm speaking of when I mention how these discussions go. It's a gish-gallop of saying that since we have imperfect knowledge, we can't draw a single conclusion about phenomena and systems that are well documented. First it's the lack of studies, then you get a study and it's a hidden bias in the populations used in the study, then it's the "low sample size" (that's a favorite), after that it's p-hacking or whatever else.

And it goes on and on, even though for hundreds of years, the country and society were tuned to ensure that the black community enjoyed degraded outcomes in all aspects of life.

Thanks for posting this, I must have misread the original article. The samples are x10 larger than I understood at first. However, please keep in mind that the subset of white ph/patients is almost 4 times larger than black ph/black patient. Data fitting seem questionable from there on. 4x is not that large of a scale but still analyses should have been more nuanced from here. For example, hospitals are not considered a variable at all. I would think that it is a major one

Are you sure you don't just have some unconscious bias that is keeping you from objectively seeing this study for what it is?

There are about 4x more white people in America than Black. So it makes perfect sense that you would have 4x as many white patients as Black in the study.

I am saying that in order to be accurate the report needs to be tied to specific doctors or at the very latest least specific hospitals.

I used to take these results at face value because I am against racism, but time and time again many of these analyses have been shown to be flawed so now I am more critical. Has nothing to do with not liking the conclusion.

Did you read the study? What are the technical reasons for the study being flawed.

My comment wasn't meant to be 'dismissive'. The idea that Black doctors provide a better standard of care for Black babies is a reasonable explanation for the data - perhaps even the most likely. I just thought other explanations should also be considered.

Reply on news.ycombinator.com