Crémieux Profile picture
Sep 16, 2024 14 tweets 4 min read Read on X
I have a pretty major update for one of my articles.

It has to do with Justice Jackson's comment that when Black newborns are delivered by Black doctors, they're much more likely to survive, justifying racially discriminatory admissions.

We now know she was wrong🧵 Image
So if you don't recall, here's how Justice Jackson described the original study's findings.

She was wrong to describe it this way, because she mixed up percentage points with percentages, and she's referring to the uncontrolled rather than the fully-controlled effect. Image
After I saw her mention this, I looked into the study and found that its results all seemed to have p-values between 0.10 and 0.01.

Or in other words, the study was p-hacked. Image
If you look across all of the paper's models, you see that all the results are borderline significant at best, and usually just-nonsignificant, which is a sign of methodological tomfoolery and results that are likely fragile.

With all that said, I recommended ignoring the paper. Image
Today, a reanalysis has come out, and it doesn't tell us why the coefficients are all at best marginally significant, but instead, why they're all in the same direction.

The reason has to do with baby birthweights.
So, first thing:

(A) At very low birthweights, babies have higher mortality rates, and they're similar across baby races;

(B) At very low birthweights, babies have higher mortality rates, and they're similar across physician races. Image
Second thing: Black infants tend to have lower birthweights.

MIxed infants tend to birthweights in-between Blacks and Whites, and there's a mother effect, such that Black mothers have smaller mixed babies than White mothers (selection is still possible)

Third thing:

(A) Black babies with high birthweights disproportionately go to Black doctors;

(B) The Black babies sent to White doctors disproportionately have very low birthweights. Image
If you control for birthweight when running the original authors' models, two things happen.

For one, they fit a lot better.

For two, the apparently beneficial effect of patient-doctor racial concordance for Black babies disappears:Image
At this point, we have to ask ourselves why the original study didn't control for birthweight. One sentence in the original paper suggests the authors knew it was a potential issue, but they still failed to control for it.Image
PNAS also played an important role in keeping the public misinformed because they didn't mandate that the paper include its specification, so no one could see if birthweight was controlled. If we had known the full model details, surely someone would have called this out earlier.
Ultimately, we have ourselves yet another case of PNAS publishing highly popular rubbish and it taking far too long to get it corrected.

Let me preregister something else:

The original paper will continue to be cited more than the correction with the birthweight control.
The public will continue to be misled by the original, bad result. PNAS should probably retract it for the good of the public, but if I had to bet, they won't.

So people like Justice Jackson will continue to cite it to support their case for racial discrimination.
They'll continue doing that even though they're wrong.

To learn more and to find the article linked, check out my post on this: cremieux.xyz/p/missing-fixe…

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Crémieux

Crémieux Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @cremieuxrecueil

Jun 28
You must pick one:

Double the productivity of the bottom 20% or double the productivity of the top 1%:
Double the productivity of the bottom 40% or double the productivity of the top 1%:
Double the productivity of the bottom 60% or double the productivity of the top 1%:
Read 7 tweets
Jun 27
Phenotyping is the vast, minimally-explored frontier in genome-wide association studies.

Important thread🧵

Briefly, phenotyping is how you measure people's traits. Measure poorly, get bad results; measure well, get good results.

Example? Janky knees. Image
The janky knee example refers to osteoarthritis, the most common form of arthritis, which occurs when the cartilage between bones is worn down, so bones start rubbing against each other.

This ends up being very painful. Image
Everyone with this condition isn't necessarily diagnosed with it.

This is especially true for men, who tend to just ignore this (and many other conditions) more often than women do.

This is, in a word, annoying, because it means that if you study it, sampling is likely biased. Image
Read 35 tweets
Jun 25
ADHD is a condition that's suffered from diagnostic drift: it's been defined more leniently over time, so more people are getting diagnosed.

One way to see this is to look at the benefits of taking ADHD medication. As prescription rates increased, the benefits have declined. Image
Another way to understand diagnostic drift is to look at the factors that promote it.

For example, school accountability laws lead to more diagnoses and, as a result, more psychoactive drug prescriptions.

Schools are pressured by law into making this happen. Image
An even more direct way to understand ADHD's diagnostic drift is to look at what types of diagnoses happen over time.

The increase has been more about non-severe ADHD than clinical ADHD. In other words, people with less and lesser symptoms are getting diagnosed. Image
Read 4 tweets
Jun 24
I have a story to break.

Columbia is still practicing racially discriminatory admissions in defiance of the Supreme Court's ruling in SFFA v. Harvard.

Newly-leaked data shows they still prefer less-qualified Blacks and Hispanics over more-qualified Asians🧵Image
Columbia has made a big show of "complying" with SFFA v. Harvard by noting that their 2024 batch of admits involved slightly less discrimination:

Fewer Black and Hispanic students, more Asian students.

That's what should happen, because Asian students tend to perform better.Image
But, with this leaked admissions data, we can see that race still predicts admissions.

With fair admissions, race should not have a significant effect, and it should not be directionally consistent.

And yet, in this data, it's clear Columbia still discriminates against Asians. Image
Read 14 tweets
Jun 21
Today's big biotech win is that we might be on the verge of a cure for type-1 diabetes🧵

Twelve diabetics were injected with stem cell-derived pancreatic islets.

They started producing insulin again.

One year in, 10/12 participants no longer needed to inject insulin. Image
In that chart, you can see the response to a meal.

At baseline, blood sugar levels go dangerously high (right) because participants don't produce insulin at all (proxied by C-peptide levels, left).

But notice the blood sugar and C-peptide levels after treatment: Image
With treatment, the patients kept getting better and better.

Their pancreatic function improved over time, and they became more and more able to handle food, and to do so without the need to inject insulin. Image
Read 10 tweets
Jun 20
About a year after this analysis came out, the Wall Street Journal published another one, with much clearer evidence🧵

It compares three adjacent counties located in three different states—Ohio, Pennsylvania, and New York. Image
These states are very differently partisan.

Ohio is Republican-controlled, New York is a Democratic bastion, and Pennsylvania? They split the difference. Image
These states vary as expected given their partisanship along many dimensions.

For example, Ohio has the lowest cigarette taxes in the group. Consistently, it also has the highest smoking-related death rate of the three. Image
Read 8 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us!

:(