Correlates with Covid-19 death rates

We correlate county-level COVID-19 death rates with key variables using both linear regression and negative binomial mixed models, although we focus on linear regression models. We include four sets of variables: socio-economic variables, county-level health variables, modes of commuting, and climate and pollution patterns. Our analysis studies daily death rates from April 4, 2020 to May 27, 2020. We estimate correlation patterns both across states, as well as within states. For both models, we find higher shares of African American residents in the county are correlated with higher death rates. However, when we restrict ourselves to correlation patterns within a given state, the statistical significance of the correlation of death rates with the share of African Americans, while remaining positive, wanes. We find similar results for the share of elderly in the county. We find that higher amounts of commuting via public transportation, relative to telecommuting, is correlated with higher death rates. The correlation between driving into work, relative to telecommuting, and death rates is also positive across both models, but statistically significant only when we look across states and counties. We also find that a higher share of people not working, and thus not commuting either because they are elderly, children or unemployed, is correlated with higher death rates. Counties with higher home values, higher summer temperatures, and lower winter temperatures have higher death rates. Contrary to past work, we do not find a correlation between pollution and death rates. Also importantly, we do not find that death rates are correlated with obesity rates, ICU beds per capita, or poverty rates. Finally, our model that looks within states yields estimates of how a given state’s death rate compares to other states after controlling for the variables included in our model; this may be interpreted as a measure of how states are doing relative to others. We find that death rates in the Northeast are substantially higher compared to other states, even when we control for the four sets of variables above. Death rates are also statistically significantly higher in Michigan, Louisiana, Iowa, Indiana, and Colorado. California’s death rate is the lowest across all states.

That is from a new NBER working paper by Christopher R. Knittel and Bora Ozaltun.

Comments

So as did McLaren, they found that more public transit and more Blacks are associated with higher death rates. The public transit association is not surprising; the association with Blacks, hmm.

We might need data on individuals rather than on counties, so we can have a more precise look at some of the other factors that people have mentioned (diabetes, obesity, etc.)

Does everything always have to be about BLM?

who said anything about BLM? Sounds like someone is a little defensive, or did that trigger you... snowflake.

Calling someone snowflake is racist. Why does it always have to be about BLM, Nathan?

Respond

Add Comment

Respond

Add Comment

Respond

Add Comment

ACE-2 has been known to have multiple polymorphisms, some of which are clustered based on ancestral geography. This likely has real implications for physiology (e.g. ace inhibitors have historically been less effective at controlling hypertension in populations of African descent). It is quite possible that differences in binding affinity or expression level changes the dynamics of Covid infections and certain ancestral populations have the short end of the genetic stick.

I do not know all the molecular pathway fine detail, but when I first heard that this uses ACE-2 for endocytosis I just assumed that African descent populations would either be much more at risk xor much less. It could certainly all be some cultural or social factor, but my first hunch would be genetics.

Respond

Add Comment

Respond

Add Comment

We also find that a higher share of people not working, and thus not commuting either because they are elderly, children or unemployed, is correlated with higher death rates.

on the other hand sociologists found
"higher amounts of commuting via public transportation, relative to telecommuting, is correlated with higher death rates."

-both commuting and not commuting correlate with higher death rates

Respond

Add Comment

Respond

Add Comment

This is precisely the sort of "study" that gives Andrew Gelman nightmares. Are there authors aware that if you run one thousand regressions and report the p<0.05 ones, you'll get 50 "significant" ones?

No, this study is perfectly fine, they are just reporting all the correlations they found. Correlations are a thing, and might be a step towards understanding causation. What should give Andrew Gelman nightmares, whoever he is, is if they picked one "significant" result and wrote the paper about that.

Respond

Add Comment

21 variables, several with p<0.01. Many more with p<0.05.

Whether these correlations provide useful information I’m not sure, but they’re not (mostly) due to random chance.

They are not random because everyone knew going into this exercise that dense urban areas had higher death rates than sparsely populated rural areas. So it would have been surprising if a few variables associated with places like the New York metro area did not come up as statistically significant.

The problem is in the interpretation of the p-values. These are not meaningful when you have independent variables that correlate with one another and/or if your standard errors are not normally distributed.

Respond

Add Comment

Respond

Add Comment

+1

This is garbage. Variables that are significant at the county level, but otherwise insignificant, what could that possibly mean in a medical context? Correlations that are positive but not statistically significant. Commuting to work leads to significantly higher death rates, but so does unemployment. Was this "study" conducted by bots?

Respond

Add Comment

Absolutely. They're fishing for correlations and seeking media (and online) attention. Also curious that they left out housing density, occupation, and demographics. Here in MA these have been identified as driving factors of transmission. One would think that the authors would be well aware, given that they are at MIT. IMO this is another agenda-driven preprint.

Respond

Add Comment

Respond

Add Comment

"However, when we restrict ourselves to correlation patterns within a given state, the statistical significance of the correlation of death rates with the share of African Americans, while remaining positive, wanes. We find similar results for the share of elderly in the county."

The share of elderly finding doesn't make sense. If you look at how high the fatality rate skews with age, properly accounting for age must increase deaths per capita. In Italy, there have been 32k deaths, 22k are over 80 and 21 are under 30.

https://en.wikipedia.org/wiki/COVID-19_pandemic_in_Italy#Lethality_by_gender_and_age

The throwaway reference to children in the context of higher death rates does not inspire confidence.

Respond

Add Comment

Respond

Add Comment

Interesting throughout, even if correlation is not causation. I like this one: "Counties with higher home values, higher summer temperatures, and lower winter temperatures have higher death rates." As readers of comments know, I spent part of last week in the low country, where they have high home values, higher summer temperatures, and lower winter temperatures and a large population of anti-science, Trump-loving, morons who don't wear masks and don't practice social distancing. There, I added some causation.

wayward never wants to talk about the arson. just like crazy old leftist
wingnut uncle paulie

Respond

Add Comment

It's in the news literally *today* that protests in South Carolina (i.e., "the low country") had to cease because a bunch of the organizers were infected. And yet here's rayward, as always obsessed with falsely blaming Trump and his supporters.

My nephew, who lives in Charleston and has an office on King, shifted from a Trump supporter to a never Trumper when he observed first-hand the Boogaloo and other far right-wing extremists running up and down King destroying property while wearing all-black clothing and masks to give the impression of being black.

That's antifa, Uncle Paulie.

Yep, and what the hell does that have to do with Trump? If nothing else he should have switch from anti Trump to Pro.

Respond

Add Comment

Respond

Add Comment

My uncle steny hoyer confided to me that he’s going to vote for Trump.

Respond

Add Comment

Respond

Add Comment

Respond

Add Comment

Respond

Add Comment

Strange findings. Just today I read in our German paper, that our medics find HIGH correlation for obesity with higher rates of hospitalization, need of tubing, deaths. No, not all Germans are obese, nor a tiny group. And you won't need binominal regression to think about causation. - I am with James, one can overdo the numbers. Should be reviewed.
What is this "black" thing, btw.? Obama's mother is "white". There is a huge bandwith of "blackness". Probably differing from county to county. How Chris and Bora account for that?

Want to know who is black? Ask a cop: When you pull this person over for a traffic violation, would you:

a. warn this person in a friendly manner

or

b. drag them out of the car and choke them to death in the street?

If the cop answers "b" you know the person is black.

1. On a per arrest basis blacks are less likely to be killed than whites
2. In well over 80% of deaths by police the persob who died was armed.

American police are brutal but there is little evidence for your narrative

1)Black people are arrested more often for the same crimes.

2) That includes cases where the police did not see the weapon and it was not in a place where the person shot could reach it. (Wonder if it includes toy guns?)

Steve

Warning warning too much BBC gay porn. Critical failure imminent. Some of the actors in this porn look like children. Critical failure imminent. Please close tabs-330 tabs of gay porn is excessive. Hdhjuuhzdsxrrfdsllt.70p.exe.node

Respond

Add Comment

Respond

Add Comment

Respond

Add Comment

Respond

Add Comment

Respond

Add Comment

The paper really goes out of its way to exclude NYC from the analysis because apparently, it skews the results too much. I propose another factor that the author's did not:

The Governor Andrew Cuomo Factor. Having Andrew Cuomo as your Governor correlates with high COVID-19 death rates.

Respond

Add Comment

They should post the dataset on Kaggle and see what people can find in it.

Respond

Add Comment

"Counties with higher home values, higher summer temperatures, and lower winter temperatures have higher death rates."

So maybe NIMBY's are increasing the death rate.

Respond

Add Comment

Caveat: I have not read the paper, only the abstract as posted here.

It strikes me that the comment about the 'higher death rates in the Northeast' may only be a time artefact because it seemed to hit NYC et al before other areas of the country.

As time continues we may still see places like Arizona and Florida catch up - or not! Seems a little early to draw too much conclusion about that particular finding though (probably a lot of these correlations).

as with Daegu and Hubei, most likely never... New York's initial catastrophic policy failure (encouraging congregation despite the "rumors") will never be replicated, social distancing will remain until the crisis has passed

with some luck synthetic antibodies will effectively end the pandemic by fall

Respond

Add Comment

True. The column of what we do not know is way higher than what we do. It's way premature to make any hard comparisons.

Respond

Add Comment

Respond

Add Comment

still largely ignored:

Raharusuna's study reached a similar conclusion. The scientists estimate that the mortality odds in their sample increase by a factor of 13 for patients who are vitamin D insufficient and 19 for patients who are vitamin D deficient. Moreover, controlling for other effects known to influence COVID-19 mortality rates such as age, gender, and preexisting conditions, the researchers find that vitamin D-deficient patients are still 10 times more likely to die from the disease. By some estimates, over a billion people worldwide are deficient or insufficient in vitamin D.

Respond

Add Comment

"when we restrict ourselves to correlation patterns within a given state, the statistical significance of the correlation of death rates with the share of African Americans, while remaining positive, wanes."

FWIW, very strong correlations down to the neighborhood level in Denver, Colorado.

Respond

Add Comment

Respond

Add Comment