Week or two back the world was shocked to learn that white doctors were killing BLACK! babies. Not the ones that haven’t escaped their mothers’ wombs: that’s old, and even welcome, news. No. The white doctors were slaying BLACK! babies that made it past the goalie.
How these maniacal white doctors were doing the killing we never learned. But it was happening. Wee Ps said so.
Black newborn babies in the United States are more likely to survive childbirth if they are cared for by Black doctors, but three times more likely than White Babies to die when looked after by White doctors, a study has found.
The mortality rate of Black newborns in hospital shrunk by between 39% and 58% when Black physicians took charge of the birth, according to the research, which laid bare how shocking racial disparities in human health can affect even the first hours of a person’s life.
By contrast, the mortality rate for White babies was largely unaffected by the doctor’s race.
One of the authors of the work said “Black babies have been dying at disproportionate rates since as long as we’ve collected data. The time is now to change this and to ensure that Black infants are afforded the opportunity to thrive.”
Damning accusations! Let’s see the evidence.
The peer-reviewed paper is “Physician—patient racial concordance and disparities in
birthing mortality for newborns” by Greenwood and others in the very Proceedings of the National Academy of Science. You can’t get a better imprimatur than this. This paper must be real Science. (Since the paper and CNN capitalized both races, in this post we can dispense with our new racism-free orthography; i.e. BLACKS! vs. whites.)
Greenwood (standing in for all authors) got birth data, including the race of babies, from Florida. Then this happened:
We also receive access to information about the attending physician in charge of the patient’s care, e.g., name, specialty certifications, and date of licensure. Physician race is not coded by the data and is captured from publicly searchable pictures of the physician…all physicians not coded as White or Black, are dropped from the sample, isolating our examination to strictly White and Black patients and physicians.
Don’t laugh. This is Science.
The raw data says 289 dead per 100,000 live births for whites, and 784 for blacks, i.e. 2.7 times more, which is not unexpected.
Now comes the magic, which I present as a picture taken from the paper.
Before talking about that, what’s the deal with the “100”? Why not 0-1 and use a logistic model as is usual? “The estimator is an ordinary least squares (OLS) to avoid interpretation issues associated with nonlinear estimators like logit regression”.
Ah. They couldn’t figure out how to interpret linear regression. “Math is hard,” said Barbie.
What of estimates of y less than 0 or greater than 100? Dude. Never mind. And what would, say, y = 27.2 mean? A little more than a quarter towards dead? Science!
I wanted to leave the equation as a homework problem, but I know the internet is lazy. So I’ll break it down for us. We have:
- When x_i = 1 & x_j = 1, y = b_1 + b_2 + b_3,
- When x_i = 1 & x_j = 0, y = b_1,
- When x_i = 0 & x_j = 1, y = b_2,
- When x_i = 0 & x_j = 0, y = 0.
When i = j = 0, it’s white baby and white doc, a combination defined as having all babies survive. Now I ask you: if you run this model, even ignoring the possibility b_1 or b_2 can be less than 0 or greater than 100 (and similarly with constraints for b_3), which group is likely to have the most babies survive? The answer is: HA HA HA HA HA HA HA HA!
They didn’t laugh. They instead added the observed white rate as a “constant” term. They put in 0.290, but didn’t show it in the equation. But only for this model—others in which “adjust” for a host things, it goes missing. Science!
Now the effect for black babies and white doctors, the category that generated the headlines, is when x_i = 1 & x_j = 0, or when y = b_1. Thus the estimate for b_1, ignoring the other flaws, should tell us how evil white doctors are. This estimate should feature prominently in their results.
No, sir, it does not. It too has gone missing in their tables. The other betas are in their tables, and all have wee p-values, the sample size being large.
Anyway, are we ready for the R^2 of this model (0 means no predictive ability, 1 means perfect)? This is not a wonderful measure, but it’s not worthless either. Ready? Druuuuuuuuuuuuuuuuum Roooooooooooooooooooooool. It’s 0.001.
Without snark or exaggeration, this model is just like that of a student in Statistics 102 who did not pay attention to the lecture on regression.
This paper is terrible, and worse than terrible. It is nonsensical. It is embarrassing.
As I say in our forthcoming book, you can easily lose your job for being right in the wrong direction. ANd you’ll likely keep it for being wrong in the right direction. This model is wrong, but it says what people want to hear. It’s wrong in our “leading” journal, but right in its politics. Science is now just as political as anything else.
To support this site and its wholly independent host using credit card or PayPal (in any amount) click here