@DrJesseMorse@SlinkECasey I read the studies you posted; they seem like interesting preclinical work, that warrant followup. To pretend they abrogate the need for chemotherapy seems misguided. Are you involved in clinical research in oncology? (2/2)
@DrJesseMorse@SlinkECasey Are you really claiming that the primary reason that oncologists use chemotherapy is to make money, and not the mountains of evidence that for the majority of cancers chemotherapeutic agents are extremely important as a first-line therapy…? (1/2)
@edwardhkennedy Charles stein actually proposed the crux of TMLE in “efficient non parametric testing and estimation” (1958) — last paragraph of section 2. Which is absolutely wild.
@Sarah_Explains @realmayo112@USA_Polling We also have a) primarily domestic debt which makes hyperinflation much less likely and b) the reserve currency, so it would be much easier for us to service foreign debts. (I assume you know this, but perhaps it can help relieve some ignorance of those you are responding to)
@karlrohe Exactly! It’s a really devious “puzzle question” to ask, since you can even give the answer, and just ask for the (impossible) justification !
@karlrohe It would be awesome if there was a simple/slick argument for it, but the route I have seen is actually a fair amount of work! https://t.co/Ud5f8wDw0J
@karlrohe I believe the expected number of points you need to draw on a sphere until they don’t all share a hemisphere is exactly 7; and I think the probably that 6 share a hemisphere is, coincidentally exactly 1/2.
@karlrohe My guess is HT strat is superior- it is the sum of negatively autocorrelated indicators; HH is positively autocorrelated and has at least a slightly longer right tail. (Both have the same mean) so i imagine that tail forces the bulk to shift left
@alexkyllo 1) this is sometimes referred to as the “mean/variance relationship” (you can use a score test to partially mitigate this — score test in this case is the chi-square test), 2) this piece can be calculated using the berry-Esseen CLT
@alexkyllo Two ways to think about this: 1) the variance of each sample proportion blows up in the tails, so the estimated variance is a poor proxy for true variance; 2) the error from CLT/normal approx has a term that looks like 1/(np(1-p)) which is large near 0/1
@RuiWang97@Jake_Elder52 Good question! Answer is a bit subtle --- I am pretty sure for any correctly specified parametric model (that is then marginalized to estimate a treatment effect) this will hold. However, linear regression is also robust to model misspecification.
@krishnanrohit Absolutely no offense to Mark (everyone has to start somewhere) but this is the beginner division. Winning the beginner division can (very reasonably) be a personal point of pride. But it’s odd to use as a talking point for how amazing someone is.
@alexkyllo I do remember those "revolutions" fondly! (and our roll!) I also quit in 2020 with covid, and then baby happened. And then I severely herniated a disk in my back powerlifting in my garage :(. We will see if I ever return! I hope your return is fun and painless (for you ;) )