Carl T. Bergstrom

Carl T. Bergstrom



1. I hate to invest precious time on taking apart the atrocious @aginnt article pictured below, but it is getting too much traction here and even in traditional media. This thread could be far longer than it is, but I'm doing my best to only discuss the most glaring flaws.

2. The introduction should be blaring red warning to any thinking person. The author begins by disputing that *context matters*. Without the background to put information in context, all the data in the world are not defense against misinterpretation.

3. You can give me all the stock market data in the world; I don't have the background to make the best use of it because I fundamentally do not understand how the market works or how to take advantage of that understanding. Infectious disease epidemiology is no different.

4. *Information gets lost in translation.* The author claims to be an expert in making products go viral. I suppose that field has borrowed some ideas from epidemiology. Now he's trying to back-infer how epi works from what he knows about that area. It doesn't work that way.

5. Imagine Shakespeare run through google translate into Japanese, then translated back to English by someone who'd never heard of Shakespeare. So much depth would be missing. Same here. We end up with loose neologisms like "virality" instead of a solid theoretical framework.

6. The author discusses the apparent decline in daily growth rate irrespective of control measures. He begins with some truism about small numbers being easy to move; this is irrelevant in the face of the exponential growth that he stresses in literally the previous sentence.

7. He fails to see that this drop in apparent growth rates is heavily driven by left censoring and shifts in testing strategy. Testing started at different times in different countries, was influenced by case density, and early-on, tests individuals in all stages of disease.

8. Next, inferences about "virality" and "viral capacity". I suppose he means "transmissibility". If so, we've spent 20 years developing sophisticated statistical methods to detect changes in transmissibility. With noisy, aggregate data this back-of-envelope stuff doesn't cut it.

9. Disaggregating data is essential to provide context, especially for transmission processes. That the virus can cross national boundaries does nothing to negate the importance of spatial structure and within-country analysis. Aggregating data obscures critical patterns.

10. I hate to ascribe to malice what can be adequately explained by incompetence, but using this lie to sweep away the disaggregated data is such utter nonsense that I wonder how a silicon valley guy could make this claim by mistake.

11. Then there's the bell curve business. If Hernstein and Murray gave the term a bad name, Ginn says "hold my beer". Most things in nature follow a bell curve, so viruses do too? Not science exactly. And do most things? What about log-normals? Exponentials? Etc etc etc.

12. But that's not the worst part. We have literally over a century's history of mathematical modeling epidemic progression. Some look somewhat bell-like. Others don't. It depends on the circumstances, details of the virus, behavior of the population, interventions, etc.

13. [pause to take beta-blockers]

14. This is unsubstantiated bullshit. IF the bell-curve were a "law of nature", it shouldn't necessarily apply to the vast range of human responses that people take to stop epidemics. Yet this assertion is supported with data from places where interventions slowed things down.

15. Wait, are already breaking the data down by country? We were cautioned against that as being misleading just a few paragraphs ago!

16. Ah, Farr's law. I don't know how the author could have more effectively discredited himself to the epidemiology community with any two other words. It's an old rule-of-thumb that suggests epidemics take a bell-curve shape. BUT....

17. When I teach ID epidemiology OR data science, I tend to have my students read this 1990 paper as a cautionary tail against non-mechanistic modeling. It uses Farr's law to predict the size of the HIV epidemic.

18. The authors conclude that the HIV epidemic will encompass roughly 200,000 cases before fading away in the mid 1990s. This graph is from the original paper. You can't make this shit up.

19. Next up a very, very basic fallacy about the effect of flattening the curve. Almost *any* reasonable epidemiological model you use, from SIR to all sorts of fancy spatial PDE or agent-based approaches, will show that decreasing transmission rate decreases total epidemic size.

20. This is common sense, as well as first-chapter-of-the-epidemiology-textbook stuff. It was also sadly predictable. See my note about severe #DKE19 strains, a day before @aginnt's medium post:

21. This claim needs citation. I am unaware of CDC plans that involve allowing the majority of the country to be infected. Because the author may be cherry-picking here, I won't call it an outright lie. But it's not the position of the organization that we allow this to happen.

22. Next the author claims that COVID19 will "burn off" in the summer, and quotes a paper from Beijing economists, posted to social science preprint server. Science is not a like a high school English essay, where you get to cherry-pick the quotes that support your point.

23. There's a big literature on the seasonality of respiratory disease, and consensus is that we have little grounds for optimism regarding #COVID19. @mlipsitch, an leading expert who advanced our understanding of flu seasonality, is admirably concise:

24. This is just misleading. Being tested is not the same as thinking you are positive. Did your doctor ever order a rapid flu test or strep culture or a chest x-ray for pneumonia? When you did, did you think YOU were positive? Same deal with COVID19, esp in places like S. Korea.

25. This single piece of bait-and-switch should be more than enough to discredit the entire article. @Aginnt claims that only 1% of cases are severe, and then shows a data graphic suggesting that 19% are severe or worse (critical). How on earth does he draw that conclusion?

26. He says "cases" in the headline, but "everyone who is tested" in the text. These are very different denominators, given the low positive rate he was just trumpeting about in the previous section. If a mistake rather than deliberate bullshit, it's amazingly sloppy.

27. Oh, and in the study that provided these numbers, "mild" cases included pneumonia short of hospitalization in a setting where hospitals are already overcrowded. Mild doesn't mean your ordinary cold.

28. Lastly on this point, I hate to go all MS-PAINT on you, but....

29. An aside: I realize that John Ioannides is trying to be clever and contrarian and has a hammer and damned if this pandemic doesn't look like a nail, but his latest STAT piece is just chumming the swimming beach for this particular kind of shark.

30. [Deep breath] It looks like this piece has now been taken down by Medium. I'm going to stop now, even though I'm only about half way thought. If it comes back to haunt us, I'll continue the thread. Thanks for reading, and stay safe out there.

31. So the piece has moved on to another right wing conspiracy-sphere website recently banned from Twitter for unfounded personal attacks on a scientist. Should I continue to go through the article, or get back to my positive efforts (modeling, sci comm) around the crisis?

32. Thank you so much to everyone who took the time to vote on the poll. I've been working on other disease modeling efforts this evening, and my twitter feed has been swamped by newly-formed troll accounts. I'll return to this tomorrow if necessary. Take care everyone!

Follow us on Twitter

to be informed of the latest developments and updates!

You can easily use to @tivitikothread bot for create more readable thread!
Donate 💲

You can keep this app free of charge by supporting 😊

for server charges...