Russell: "It is strange that the last men of intellectual eminence before the dark ages were concerned, not with saving civilization... or reforming the abuses of the administration, but with preaching the merit of virginity and the damnation of unbaptized infants."
@Ayjchan
@nytimes
Congrats, very nice article! I think my JRSSA paper should be linked along with the others demonstrating the weakness of the Worobey paper.
1/n My paper on the garbage dump that passes for causal analysis in Physics Ed. Res. (and I think in other education) has just been posted. Kudos to the editor for publishing this critique of papers from his journal!
4/n Big policy decisions, e.g. dropping standardized tests, are being made based on “research” with comically bad errors leading to conclusions opposite of what the data imply. (See also , )
1/few I've updated my Bayes estimate on Covid origins to use new simulations from Nod Nizzaneela. No change in the overall odds but really the end of any pretense that Pekar et al. was a legitimate paper. Despite giving them every possible break, fixing math errors makes the odds
@gadboit
@past_is_future
Perhaps there's a training camp where several young zoocrew learned to never state a clear hypothesis with quantitative evidence but always use a strange passive-aggressive pseudo-Socratic mocking style.
After a slight delay my short note on proximity case ascertainment bias in Worobey has now appeared on the arXiv. . I think it pairs well with Stoyan & Chiu's just published
My short paper on severe ascertainment bias in Worobey is now out. The short talk linked at the end will play in .ppt if downloaded, but not in google docs.
@gunsnrosesgirl3
One key is that the entropy change in ordering a few nails is utterly negligible compared to the entropy change in just slightly warming even one nail. They warm from the inelastic collisions. Plus the entropy change in your muscles burning fuel. No paradox at all,
Some of these papers have slipped into dishonesty. How can we claim to be scientists and accept this crap? How much is this just the tip of the iceberg? How much is driven by ideology and how much by bureaucratic rewards? I dunno but would love to hear anecdotes.
For those just tuning in, these are the probabilities of a certain pattern found in SARS CoV2 arising by natural accident starting from various natural sequences. The pattern's probability for a lab creation is ~1/2. "RpYN06" was added by request from zoonosis types. Whoops.
2/few of 2 spill vs. 1 drop from their 60/1 to 0.18/1. Huge errors revers the conclusion. Here's the breaks that they're given to keep the odds from dropping more.
1) No use of the known time-dependence of the ascertainment rate.
2) No use of the recently confirmed intermediates
I just realized that a simple near-rigorous math argument shows the Worobey proximity evidence for a Covid market spillover had to have serious ascertainment bias, based purely on their own data, no anecdotes needed.
My only original contribution.
3/few
3) Fine tuning 2 pre-spill parameters to maximize N=2 odds.
4) Allowing the host pool to be tiny. (Like a lab.)
It's not normal for an important paper to be wrong in so many major ways.
@MattBruenig
Yes, and it was very far-sighted of him to push $150 M to Moderna via DARPA and BARDA to develop the mRNA technique, back in 2013 and 2016. Otherwise this wouldn't have been possible. Oh, wait...
I've posted my first response () to DeBarre and Worobey's Reply to my JRSSA,
. A colleague who agrees with me thinks that responding shows I'm "Ein naiver dummkopf." Probably right.
@AliceFromQueens
It's unreliable people like you who doubted that comrade Kirov had been killed by Zinovievites paid by the Japanese and British empires. Political and epistemological chaos!
Re biased data in the key Worobey covid market origins paper, some journals now encourage having a supplementary talk to accompany a paper, so I've prepared (11.5 min) to go with
finding the best student selection method is trivial in one limiting case. If we do not try to maintain minimal standards of competence and transparency or even basic logic in our treatment of data, then the optimum group of students whom we should be educating is the empty set.
@mtmalinen
Got it. Instead you'll risk your health to a new disease whose long-term effects are unknown. But which is known to very often have serious medium-term effects even in younger people. While the vaccines are known to have none. Smart move.
@gadboit
@past_is_future
Even before you get to phylogeny, ~0.0001 of China's mammalian wildlife trade is in Wuhan, compared to most of the coronavirus FCS-adding research plans. Big likelihood ratio. Non-market zoonosis is less unlikely, but some crappy papers headed down the market path.
@pauldauenhauer
Einstein on thermodynamics: “It is the only physical theory of universal content, which I am convinced, that within the framework of applicability of its basic concepts will never be overthrown.”
@alchemytoday
@ban_epp_gofroc
"previous analysis" is Pekar? Nod didn't change their model or data, just fixed the fng errors in their code & used equal observations for the conditional probabilities for the 2 hypotheses, tweaking parameters to favor N=2. Pekar was still off by x330. A20 etc. are not involved.
@legend9616
@RuthHasOpinions
@bgluckman
Everything there is wrong. Viruses do zero "breeding" outside your cells. Even shitty masks catch many of the drops in which the viruses travel, including drops much smaller than the fiber spacing. All well documented in the literature for years.
1/2 Phys Rev Phys Ed Res has just again rejected my critique of 3 grossly erroneous articles they just published, on the grounds that calling things "incorrect" isn't "positive" enough, and that if I keep doing this I'll have to write articles "until the end of time".
3/n It looks like the field of physics education research (NSF funders, editors, reviewers, funded researchers, National Academy panels …) has no clue about causal inference and little clue about even more familiar stats methods.
Of course. There's just something satisfying about taking a famous paper and showing that almost everything in it is wrong, even granting its dubious premises.
@acritschristoph
@gadboit
@past_is_future
My take: pre-knowing DEFUSE, both LL & ZW were pretty diffuse, not very strong likelihood ratios. DEFUSE focused LL prior on narrow parameters that fit observation, increasing the likelihood ratios --> strong LL odds. Some found it hard to change views in response.
@deaneckles
A physics colloquium speaker here had a badly overcrowded 1st slide. He explained that he'd just given the talk in Tel Aviv, where they had a rule of no interrupting the 1st slide.
@emilyakopp
& even before getting to the analysis, Worobey has strong internal evidence that the unlinked cases have far too much proximity ascertainment bias to allow *any* conclusions based on their locations. The unlinked cases include all the ancestral ones.
@mbeisen
"scientists ahead of profit". Not science or truth. "scientists", like any other interest group.
Meanwhile, AAAS has dropped the "technical comment" category for correcting published errors. (I got one of the last ones in.) He needs to be replaced.
@CaitlinPacific
@nytimes
I always let my kids play with toy guns, just as I had. They would have found a way even if we hadn't allowed it. I've seen boys make toy guns out of chewing the right parts out of a P&J sandwich. And there's one thing I absolutely didn't want them to become: arms manufacturers.
@julianbarnes
@nytimes
There are good diplomatic reasons for governments not to officially say it was probably a lab leak. Future cooperation with China is needed, e.g. on climate. But it's too much to ask for the rest of us to pretend that DEFUSE, 2xCGG in FCS, location, etc. are not evidence.
@nizzaneela
@Biorealism
@EmaNymton90
@jbkinney
Worobey's recent talk moved the probability of 2 spillovers from 99.5% to 97% He can't even read his own paper right! The original claimed odds were 60/1, not 200/1. After your first correction, they were 10/1, not his 30/1. Now 2.2/1 & dropping. & still irrelevant & crap model.
@JamesFallows
@propublica
1/2 These fuzzy rumors are side issues compared to the core, reliable scientific evidence. Which also strongly points to lab leak. It's painful to see this question so consistently seen as a political, particularly since it gives the bad guys a chance to be right. See e.g.
@trvrb
Although I believe that lab leak is somewhat more likely (mainly due to location), I hope that no definitive answer is found. Both mechanisms have occurred and will occur. Uncertainty about this case helps keep up the pressure to defend against both.
@drjonty
1. Handled prison for draft resistance better than I handled 2cd grade.
2. Started international boycott of program (SDI) heading toward a nuclear 1st strike & got nominated for Nobel Peace Prize.
Cheating here: # 3: actually got laid.
@senatorshoshana
@PrestonCooper93
@FREOPP
Apparently those degrees don't involve a first-semester stats course with some intro to causal inference. Cuz if they did, everybody would know this whole exercise is bullshit.
@CT_Bergstrom
Bruce Fireman's beautiful difference-in-difference study with N=4M showed that flu vax cut flu mortality ~50% in elderly. Saying N=1000 is underpowered is a real understatement.
@flodebarre
@ydeigin
@Ayjchan
I didn't say or think you had a direct personal interest. But you've gone way out on a limb (raccoon dog, Faul,...) to defend a claim that you must at least suspect is false. I'm guessing out of loyalty to The Science.
@kph3k
My new paper on the crazy stats methods used by GRExit advocates to get the result they want will come out Thurs. night in Econ Journal Watch. Bottom line: in physics GREs give ~x3 odds ratio increment beyond what GPA gives. E.g. 75% PhD prob vs. 50%.
@gdemaneuf
@nizzaneela
@niemasd
@JamieMetzl
@natashaloder
Nod's new correction concerns a much deeper logic error, not just coding errors or modeling choices. Pekar had compared likelihoods of *different outcomes* for N=1 and N=2, with N=2 given a bigger target, blatant slanted illogic. Now Nod uses the same constraint for N=1 or 2.
@MJnanostretch
She bragged about how routine it was to add FCS's, & pointed out correctly that there was no *guarantee* that adding an FCS would make a virus dangerous. Not sure she'll be happy reviewing the video.
@SolidEvidence
@Dissenting2020
@humourme101
The semantic distinction between "direct" and "circumstantial" evidence is unscientific, not to mention just plain silly. All of forensic science is "circumstantial", but can be pretty strong. That's how I helped catch 2 fake pollsters. See e.g.
A minor question to help as I struggle to prepare a big paper. Does anybody know why Pfizer and Moderna chose not to use CGGCGG in their spike codes at the one spot where nature did use that code?
@harishseshadri2
@nizzaneela
One needn't require similar-size clads, but they chose to do so for N=1 and not for N=2! An almost inconceivably bad fundamental error, not a judgment call.
1/few Here's an informal pre-registration of a methods tweak re what's left of Pekar 2022. Nod has done a great job fixing errors in their code, but one peculiarity remains to be fixed. I have no idea which way the fix will tilt the odds. Pekar used a smaller clad % in the 30-50%
@graceelavery
@lastpositivist
Russell: "It is strange that the last men of intellectual eminence before the dark ages were concerned, not with saving civilization... or reforming the abuses of the administration, but with preaching the merit of virginity and the damnation of unbaptized infants."
@mbalter
@Rebecca21951651
Once my neighbor's furnace blew up ~1 day after routine maintenance. Either
1) Maintenance fucked up and caused the fire.
or
2) Maintenance completely failed.
Most likely (1), but either way I wouldn't want to hire that company.
@cailinmeister
Some of this feeling may have fed into the persistent obsession with hand-washing, cleaning etc. All things good boys and girls do. HEPA filters? N95s vs. cloth? Just some technical stuff, no moral salience.
@acritschristoph
@gadboit
@past_is_future
1/ I don't know the people and might well like them. When scientists make big errors and don't try to fix them, that reflects on their follow ups. (The same as you're claiming re me.) E.g. recent high-Tc stories. It's no guarantee tho. I'd love to see your own Bayes analysis. BTW
On
#GRExit
& pseudoscience in physics education research, new trash paper has come out since my last tweet. New critique included here:. I'll submit to journal after hearing reactions.
Do we really have to just accept this crap? & act as if we believed it?
@JRSS_A
@SNChiuHK
@TUBergakademie
@hkbaptistu
Stoyan and Chiu show that the Worobey analysis of the data is severely flawed but defer on the question of the data quality. Worobey also has strong internal evidence of proximity-based case ascertainment bias. My shortest paper:
@acritschristoph
@gadboit
@past_is_future
as I pointed out, the Pekar analyses are irrelevant to the LL vs ZW issue since either is compatible with shallow pre-spillover distance to MRCA. I discussed the paper primarily to show how shaky the work is. A Q 4 U: what % of CN wildlife mammal trade goes through Wuhan markets?
@harishseshadri2
@StevenSalzberg1
@kesvelt
@R_H_Ebright
This is good news. It shows they understand what happened last time and are taking at least some action to avoid a repeat. Asking them to also officially blame China would just be asking for pointless trouble. Next time it could start in Madison.
@acritschristoph
@gadboit
@past_is_future
"quite obviously" doesn't = "quite possibly". I made a point of not using FCS presence due to conditioning on pandemic. Its coding does still look somewhat weirder for natural than for DEFUSE. Mainly, together with DEFUSE it ties LL to the Wuhan location. As I specify.
@robinhanson
It's probably going to be terrible, but the idea of replacing parts of the old general curriculum (Euclid, trig, parts of algebra 2) with more integrated stuff about functions, probability, statistics, causal logic would be good if done right. Big if.
@gadboit
Haw. They count a "pass" as having distance to unlinked as less than distance to linked. But their own data said it was less than 0.7 of that. Just typing that 0.7 into the code gives this for their kappa=0.4. Not exactly the result they claim or want!
@DKlemitz
@MJnanostretch
@OdysseyBohemian
The problem: P(FCS) is small but N(people infected by bat coronaviruses) is big. Very hard to estimate the product except that in a few decades of chances before 2019 it didn't happen, & in a few months after DEFUSE it did. So timing likelihood factor favors L over Z.
@acritschristoph
@gadboit
@past_is_future
Wrapping thing up with the output of a crude model implemented with at least 3 major coding errors and one weird prior ratio may not give correct results.
@Wunsen08763219
@Rebecca21951651
@mattwridley
Oh shit- he's still pushing Pekar, as if he didn't know that it had *huge *coding errors and modeling problems. & even giving a bogus probability based on it. Looks dishonest.
@past_is_future
@gadboit
I understand that Kumar and Bloom looked closely enough to see that the vector space of very early sequences was not spanned by an {A,B} basis.
@MJnanostretch
@nizzaneela
Obvious fundamental error in unbalanced conditional probabilities. Peculiar priors. 3 admitted coding errors. Nod found another little one. Now repeated use of atypical results. Every error of the same sign.
@BlairWilliams26
4.5 days ago guy behind me on a plane was so sick & productively coughing so constantly that other people were muttering about it. I kept my N95 on & told him truthfully "I've got cancer. Can I offer you a mask?" He wore it, tho not well, the rest of the flight. So far so good!
1/? thread in which I stick my neck out re LK-99. Spoiler guess: the messy Korean samples have many little grains with a broad range of SC Tc's due to a range of local strains. Real but not ready for prime time applications. I think this is consistent with the diverse data types.
@flodebarre
@Ayjchan
Again, Deigin provided one reason for CGGCGG. It specifically allows easy tracking of whether the site has been lost in cell culture. Sounds like an easy convenient thing to do if you start with ENaC template. Not P=1 but P>>0.0007.
@acritschristoph
@gadboit
@past_is_future
True- the error-ridden analysis I described is not the same one used for the table. It just leaves little confidence in the "substitution process model in a Bayesian phylodynamic framework" in the follow-up from same group, since they never corrected the big initial errors.
@Ayjchan
Yes, one shouldn't pretend that DEFUSE has details it lacked. But it did have some specifics- 6 restriction segments with none too long- that can be compared with natural viruses and synthetic ones. That gives another factor in the odds.