I shared this elsewhere and received the following feedback:
"They found this decades ago, the "republican" states actually report when you expose children to die instead of calling them still births."
I have no idea on the truth of this or not but it seems like adding still births to infant mortality might be a valid thing to do. In particular it would be interesting to see if the R states report lower still births or not
As a follow up I did a search on the CDC site having got a hit there for still birth statistics and I'm completely failing to find state by state stats, just national ones. I may be looking in the wrong places though. On the whole I'd think the CDC would be a better source of data that wikipedia for all of your data assuming the CDC has relevant state level data
I would be curious to see if this persists when controlling for GDP per capita or median income. That could in itself be used as an argument for Democrats (our states are richer!) or Republicans (our states are a refuge for poor people who can’t afford Democratic states!). I’m not sure what policies specific to Democratic states would account for lower infant mortality given that even most GOP states ended up expanding Medicaid, but that would be an interesting follow up.
The bill in Ohio (passed in the 2024 session) provided more government services for pregnant women. I believe this includes check-ins and reminders to take prenatal vitamins, among other things.
There are mistakes here that are typical of academia. Don't go down that road - they mass produce stuff like this and you can do better.
> to my surprise, the coefficient on Republican was large and close to significance as well [...] The effect has a p-value of 12% [...] Republican policy probably does a have a real, causal effect which results in worse infant mortality
That's not how it works.
1. "close to significance" means "not significant". The P=0.05 threshold is arbitrary, but if you're going to adopt that framework then you don't get to claim a value above it is "close to significant", let alone P=0.12 which is well off. It means you have a greater than 10% chance of your hypothesis being an FP
2. You're clearly P-hacking. You have to pre-register your hypothesis before you start regressing hand-picked datasets against each other, otherwise you're just exploring the garden of forking paths and can be easily fooled by noise again.
3. Even if we generously assume these correlations are real and not noise, nothing you've done establishes causality.
Academics like to play these sorts of games, but that's why so many of us don't trust them anymore. Your pieces on resentment and space colonization were much more insightful.
1. Your ellipses are leaving out a lot. I concluded that being Republican was probably causal after I got a p value of 2%, not 12%.
2. I did state my hypothesis before I started the regressions! My initial hypothesis was that being Republican would not have an effect after controlling for race and obesity. I found the opposite.
3. A regression by itself never says anything about causality, so you have to use your judgment. I'm curious what you think the explanation is. What do you think is the missing variable?
So given that regressions don't show causality, why are you doing them and then claiming causality? You go straight from "here's some correlations" to "and thus Republican policy causes infant mortality" but that's not a valid inference from the way you're using statistics.
Speculating, it seems likely that the missing variable is just poverty. The Republicans have aligned themselves with the working classes for many years, and pick up a lot of support amongst them for that reason, Trump's famous "I love the poorly educated" being an example. But Republicans sticking up for the poor after the Democrats decided they were deplorable doesn't mean Republican policy creates poverty, let alone infant mortality. You haven't really even laid the groundwork for that, just asserted that your regression results imply there must be some policy, somewhere, that causes babies to die.
By P-hacking I mean you started with the hypothesis "infant mortality in Ohio is driven by the black population", and then after ruling that out the hypothesis evolved to be about obesity, and then after that Republicanism.
This sort of analysis is found everywhere in academia - regress ad-hoc things against each other, claim causality once P falls below the magic number. It's one of the reasons so few claims they make replicate. I don't think it's capable of proving much that's interesting (in any direction, I'd be arguing the same if the result had been that Republicanism lowered infant mortality).
What do you think about weighing in costs? Are you a hardcore libertarian, or are you up for infant mortality decreasing policies if they can be efficiently implemented with big results from little money? I think the government may have an easier time than private charities for things like this.
Post has been edited to correct a mistake in interpreting the second regression.
There's that map again
Except in this case the infant mortality map is NOT the same as the diversity map.
I shared this elsewhere and received the following feedback:
"They found this decades ago, the "republican" states actually report when you expose children to die instead of calling them still births."
I have no idea on the truth of this or not but it seems like adding still births to infant mortality might be a valid thing to do. In particular it would be interesting to see if the R states report lower still births or not
As a follow up I did a search on the CDC site having got a hit there for still birth statistics and I'm completely failing to find state by state stats, just national ones. I may be looking in the wrong places though. On the whole I'd think the CDC would be a better source of data that wikipedia for all of your data assuming the CDC has relevant state level data
I would be curious to see if this persists when controlling for GDP per capita or median income. That could in itself be used as an argument for Democrats (our states are richer!) or Republicans (our states are a refuge for poor people who can’t afford Democratic states!). I’m not sure what policies specific to Democratic states would account for lower infant mortality given that even most GOP states ended up expanding Medicaid, but that would be an interesting follow up.
The bill in Ohio (passed in the 2024 session) provided more government services for pregnant women. I believe this includes check-ins and reminders to take prenatal vitamins, among other things.
There are mistakes here that are typical of academia. Don't go down that road - they mass produce stuff like this and you can do better.
> to my surprise, the coefficient on Republican was large and close to significance as well [...] The effect has a p-value of 12% [...] Republican policy probably does a have a real, causal effect which results in worse infant mortality
That's not how it works.
1. "close to significance" means "not significant". The P=0.05 threshold is arbitrary, but if you're going to adopt that framework then you don't get to claim a value above it is "close to significant", let alone P=0.12 which is well off. It means you have a greater than 10% chance of your hypothesis being an FP
2. You're clearly P-hacking. You have to pre-register your hypothesis before you start regressing hand-picked datasets against each other, otherwise you're just exploring the garden of forking paths and can be easily fooled by noise again.
3. Even if we generously assume these correlations are real and not noise, nothing you've done establishes causality.
Academics like to play these sorts of games, but that's why so many of us don't trust them anymore. Your pieces on resentment and space colonization were much more insightful.
1. Your ellipses are leaving out a lot. I concluded that being Republican was probably causal after I got a p value of 2%, not 12%.
2. I did state my hypothesis before I started the regressions! My initial hypothesis was that being Republican would not have an effect after controlling for race and obesity. I found the opposite.
3. A regression by itself never says anything about causality, so you have to use your judgment. I'm curious what you think the explanation is. What do you think is the missing variable?
So given that regressions don't show causality, why are you doing them and then claiming causality? You go straight from "here's some correlations" to "and thus Republican policy causes infant mortality" but that's not a valid inference from the way you're using statistics.
Speculating, it seems likely that the missing variable is just poverty. The Republicans have aligned themselves with the working classes for many years, and pick up a lot of support amongst them for that reason, Trump's famous "I love the poorly educated" being an example. But Republicans sticking up for the poor after the Democrats decided they were deplorable doesn't mean Republican policy creates poverty, let alone infant mortality. You haven't really even laid the groundwork for that, just asserted that your regression results imply there must be some policy, somewhere, that causes babies to die.
By P-hacking I mean you started with the hypothesis "infant mortality in Ohio is driven by the black population", and then after ruling that out the hypothesis evolved to be about obesity, and then after that Republicanism.
This sort of analysis is found everywhere in academia - regress ad-hoc things against each other, claim causality once P falls below the magic number. It's one of the reasons so few claims they make replicate. I don't think it's capable of proving much that's interesting (in any direction, I'd be arguing the same if the result had been that Republicanism lowered infant mortality).
Did you do income level?
No, because income level is plausibly affected by policy.
That doesn’t mean it doesn’t have independent explanatory power.
What do you think about weighing in costs? Are you a hardcore libertarian, or are you up for infant mortality decreasing policies if they can be efficiently implemented with big results from little money? I think the government may have an easier time than private charities for things like this.
I would prefer for it to be done by private charities but I don’t have a strong opinion about it.