Matt Malkus Profile picture
Life actuary. Classical liberal. Currently examining COVID. Sometimes I do baseball things. Brooklyn expat - haven't picked up a southern accent yet, y'all.
17 Nov
THREAD: At the risk of beating a dead horse, I thought I'd compile all of the recent data and evidence associated with schools:
- Are they safe?
- How are other countries handling them?
- What are the costs of school closures?

OK, let's go!
1/ In an email to families, Nashville school director Dr. Adrienne Battle warned that schools may return to virtual learning for elementary students.
But she notes that "Nearly all the [MNPS] cases have been contracted outside of the classroom or school."…
2/ This echoes the findings of researchers & health authorities in the US & internationally.
Even Dr. Fauci has acknowledged that as of mid-Oct, despite 30 mil kids in US schools, "there has not been an indication" that this was driving community spread.
Read 20 tweets
15 Oct
1/ I earnestly looked into this, because I wanted to know how they arrived at these results.

"[B]ased on predictive models, COVID-19 impacts could be lessened by up to 47% if a majority of people wear masks."

Oh, boy, predictive models? I thought we'd learned our lesson here.
2/ "Predictive model" here is a fancy way of saying "our results are based on parameterized inputs, not empirically observed data."

Still, let's give them the benefit of the doubt, shall we? Here's the paper. Let's dive in!…
3/ As expected, the model is based on parameters derived from other literature.
On masks:
"We chose a conservative non-medical mask efficacy, eM = 20%, within the estimated range for reducing disease transmission during interactions between susceptible and infected individuals."
Read 17 tweets
14 Oct
Thanks, Meghan, for covering tonight's board meeting. I appreciated the opportunity to meet with concerned parents and teachers beforehand, and to lay out some of the facts surrounding the costs of school closures for the board.

References to statements I made follow (THREAD).
Count/list of Texas school districts who have announced an end to in-person learning, and selected statements from school districts which did so. Thanks @therealarod1984 for compiling and staying on top of this.
I made two references to independent studies of absenteeism in schools and the widening gap between students in affluent districts versus those in poor/impoverished districts. Here is the first, from the Dallas Morning News, covering 80 districts.…
Read 8 tweets
9 Sep
Hey, what ever happened to the "COVID kids develop Kawasaki disease" storyline that was used to terrify parents back in April or May?

Well, the CDC didn't forget - they published a report on August 4 on the... wait for it... *565* total cases in the US.…
Through September 3, there have been about 515K cases of COVID-19 in children.

There are roughly 55 million school-aged children in the US.…
So, to recap:

- Less than 1% of all school-aged children in the US have tested positive for COVID;

- Of those, about 0.1% developed Kawasaki-like disease (MIS-C);

- Of THOSE, only 1.8% (10 total persons) died.

So, I wonder why you never heard these numbers?
Read 5 tweets
22 Aug
Something that people have a hard time grasping:

Taking COVID-19 seriously and demanding transparent, accountable, science-based government are not mutually exclusive.

I'd demand answers to questions about our policy response even if the death toll were 100x what it is today.
Good public policy should:
- Properly assess the risk and the benefit of action
- Properly assess the cost of each action & weigh it against the benefit
- Be clearly communicated to the public, including specific goals
- Be non-arbitrary (i.e. grounded in logic, facts, relevant data)
- Account for ALL stakeholders, i.e. all members of society (rich, poor, young, old, etc)
- Be re-evaluated regularly as part of a discussion involving the public (not by decree)
Read 4 tweets
5 Aug
Meet Steven Manzo, a tragic casualty of lockdown. A recovering addict who overcame so much, and had so much life ahead of him.


Next time you advocate for the closure of a business, school, or other aspect of society, remember Manzo.

Then multiply his case by...
5,000: Roughly the increase in national overdose fatalities from March to May, based on these % increases above the CDC's 2018 baseline.…
10,000: The estimated increase in breast & colorectal cancer deaths over the next 10 years due to missed cancer screenings. "This analysis is conservative," says the author, director of the U.S. National Cancer Institute.…
Read 14 tweets
29 Jul
Let's take a trip down #SARSCoV2 memory lane, shall we?

Here's one of the big misses from early estimates which informed @IHME_UW modeling, and ultimately policymakers: "15-20% of those who contract [COVID] will need hospitalization," said Dr. Fauci.
As some started to realize that there would be no avoiding the virus here in the US, hospital capacity was the singular focus. It was never about suppression.

After all, we only have so many hospital beds, PPE, etc - we needed to "flatten the curve."
By March 6, we're approaching the critical days when the world began to change. We had compelling visuals on what it meant to "flatten the curve":
- Buy time
- Reduce peak health care utilization
- Prolong the epidemic at manageable levels
Read 17 tweets
28 Jul
COVID+ athletes have been big in the news recently. The Miami Marlins had an outbreak in the last 24 hours. It's the latest in a laundry list of news stories about athletes testing positive.

Should we be concerned?

You be the judge - here is the breakdown of symptom severity.
No doubt, in rare cases, COVID can cause health complications & lingering issues. This is the case of Red Sox pitcher Eduardo Rodriguez, who is experiencing lingering cardio issues. He's in the <2% of "Severe Illness" group (and <1% with lingering effects)
Far more common, though, are cases such as Russell Westbrook: Athletes who test positive are completely asymptomatic nearly 60% of the time - and that number is growing since May, as leagues return and regular testing is implemented.…
Read 9 tweets
17 Jul
START/ This thread contains all-cause mortality charts for children aged 0-14 for every country tracked by the Human Mortality Database (

Due to the large number of countries, this thread will be posted in two installments, one on 7/17, one on 7/18.
1/ Finland: We start with Finland because it's unusual. There was a tragic school shooting in 2007 which appears in this data.…

I post Finland first to provide a sense of scale: That very large spike is the result of 9 fatalities.

Very small numbers.
2/ Belgium: We continue with Belgium, which is among the highest per-capita mortality rates from COVID in the world.

Yet, there is no evidence of this when looking at the childhood mortality data.
Read 27 tweets
14 Jul
The Human Mortality Database, jointly run by UC-Berkeley and the Max Planck Institute for Demographic Research in Germany, is publishing weekly mortality rates by age group for the US and most European countries, dating back to 2000.

Here is Swedish data on school-aged children.
"Yeah, Matt, but what about the teachers?"

Well, if the teachers in Sweden were also age 15+ in 2015-16 (and they were), then they can take heart! They've lived through a higher-risk environment than the COVID-19 pandemic. This age group covers all working-age adults.
So, where is the excess mortality coming from?

Well, of course, from the age 85+ demographic, where 40% of Sweden's deaths reside. Note that even here, there is higher mortality as recently as January 2000 - but, this was before substantial medical advancements. Here's the risk.
Read 4 tweets
7 Jul
File under "Bad Model Predictions": This Vanderbilt School of Medicine model for Tennessee, dated April 10.…

"Lifting social distancing scenario: Under this scenario ... at the peak close to 50,000 Tennesseans are sick enough to require hospitalization."
50K hospitalizations at once is TERRIFYING! Look at the scale of this chart. This red line would skyrocket off the page fourfold until it reaches that level. Catastrophe!
You may notice in that chart that the lines start to diverge right around the end of April.

That's funny, because... @GovBillLee lifted Tennessee's stay-at-home order on April 30. A close match to when VUMC's model changed its assumption of behavior.…
Read 5 tweets
6 Jul
Most of my followers are new since this thread. One of the most egregious examples of doomsday epidemiology and public health came from the Univ. of Minnesota and the MN Department of Health. That model predicted peak infection on June 29. Let's revisit!
Model "Scenario 5" is defined as "Stay at Home order lifted 5/18."

That's exactly what MN did:…

What did the model predict through the present date?

12,000 cumulative deaths.

700 daily deaths at the peak of the epidemic, which was scheduled for... now.
So, what has actually happened in Minnesota?

700 deaths per day? No. A 7-day average of... 5.4 deaths per day.

12,000 cumulative deaths? No. A grand total of... 1,508 deaths.…
Read 9 tweets
3 Jul
So, something strange is happening in Maricopa County.

Back in early June, their hospitalization data was robust. They disclaimed a 6-day reporting delay from specimen collection until 6/11, at which time they changed to a 10-day reporting delay. No big deal here, so far...
Two weeks later, on 6/25, they changed their reports again. This time, they moved to a 17-day delay on hospitalization data. That's a long delay! Moreover, though, they changed the language on this data to say that the delay is "due to incomplete data."

Incomplete, why, exactly?
On Wednesday, Maricopa County moved to a PowerBI dashboard and stopped publishing the daily PDF reports. Makes it a bit harder to pull historical data, but whatever.

But look at what the new dashboard says about their hospitalization data: 21-day delay now!
Read 7 tweets
15 Apr
1/ So, the latest argument against reopening states which have not come anywhere close to health care capacity is that we need to ensure we're reducing R0 to <1. In fact, this is the official reason cited by Tennessee, who referred to a Vanderbilt model:…
2/ In Chart 1, Vanderbilt shows that mitigation & containment efforts have reduced R0 from an initial estimate of 5-6 (exact figure not given), all the way down to a current estimate of 1.4. That huge effect is attributable to social distancing, they say.
3/ I was curious, though - how are they arriving at these estimates of R0? Fortunately, the CDC has published some guidance in this area:…
Read 5 tweets