Cassie is a startup CEO and Google's former Chief Decision Scientist. This is her channel for cheekily simple explanations of fancy-sounding things. ❤️ Stats, AI, data, puns, art, sci-fi, theatre, travel, decision science.
Decision Intelligence = Data Science + Decision Science
You said at the beginning you would flip the coin until it turned up heads, and that sounded like it was going to be interesting. And then you dwelled on exactly that one coin flip. It would have been more interesting if you spoke more instead about how many more coin flips might be necessary from a data set before you get 50 heads. This would be evaluated from the already-gathered coin flip data. I feel like a single data point of a single coin flip isn't a useful frame when you're anywhere close to statistics.
the subtle humor in between makes the course so much more engaging.. One of the greatest instructors i've ever seen..i can't believe i completed a 1.5 hr video in one go.. kudos to Cassie, admirable job..
Jem Corcoran, A math professor at Colorado University says probability is about the future and statistics is about the past. I think, frequentists will say that one of the pasts either happened or didn't happen and there is no probability and they have nothing to say about the past. Am I correct? The random variables, is it a frequentist idea or a Bayesian idea or some-other-statistian's-name-ian idea?
Is it correct to say all frequentists will have the same answer given the same data and all Bayesians will have the same answer given the same data and the same initial beliefs?
Bayesians may have a notion of error. In fact they should at least based only on your video. If I ask a Bayesian what the error is, they should have an opinion about it. To say there is no error is frequentist because they feel the error is 0 or 1. Since they can't know for sure, there don't want to concern themselves about it. I am not sure I am right about the third last sentence of the paragraph. Maybe, if we ask about their opinion about the probability that error is greater than some percentage, they should have an answer, I believe. But, there are 2 types of Bayesians. One kind that uses Bayesian methods to calculate probabilities. Other kind that actually believes that the mean of a population or some other statistics is actually random.
Hi & thank you for that motivational tutorial. Question: how do I correctly specify identify the distribution of the data to simulate for univariate, bivariate and multivariate data situations?
Bayesian statistics presume the world is uncertain and there are things we don't know about. To make any sense of the world requires reasoning and ability to adapt one's views to experience. If the coin landed tails 50 times in a row, a Bayesian would check if the coin had tails on both sides. To be a good bayesian statistician requires good ability to reason and a lot of experience; there is no shortcuts. Bayesian statistics is logically very unpopular among academics. To them, statistics is a tool to avoid dealing with feelings of uncertainty. This is why they often ask wrong questions, exclude significant datapoints, miss confounding factors, their studies often fail to reproduce, and they come up with nonsensical theoretical models. Frequentist statisticians are only useful if being led by Bayesians and kept on short leash. Case in point: Warren Buffett vs academic proponents of efficient market hypothesis, who keep counting sigmas in poverty.
That's a tough one if for no other reason than because my understanding at this time in my life, is at least in the context of a decision justifiably claiming prospectively less value away from deeming the null hypothesis as necessarily true, does not also, in and of itself ipso facto, pertaining to the prospective manifestation of a statistically significant p-value upon and after calculations resulted thereby, also _necessarily_ justify contending an alternative hypothesis is tantamount to a premise worth fully accepting and agreeing with, at least insofar as potentially reliably make use of such an alternative hypothesis insofar as potentially soundly asserting the existence of ipso facto commensurate truth per se, insofar as what the alternative hypothesis reasonably connotes numerically insofar as how results of calculations thereabout of a samples(s() data in fact be a prospectively reliable reflection (if you will) of a population's emplacement of the purported relationship being investigated, - but also given that to get a better notion of such matters a context and demonstration is arguably a necessary prerequisite insofar as a deep dive from an analytical perspective and not the _executive level_ that was presented with that video. And I can't imagine it would be summed up (pun intended) if doing so in brief form or fashion, to put it mildly. Now, I shall consider embarking on a moment of silence and while thinking positive thoughts about the video's maestro. I got well more than I paid for on my student RU-vid Premium subscription therefrom its viewing and commentary herein insofar as thinking about it wherever it may have gone or went, at least insofar as how quick and dirty estimates in the mind shake out accordingly, based upon my experience and observation(s) so far, such as they may have been. Basset hounds may be the superior species but also, we've been known to chase our own tails. 🤠
Chicken or egg problem / Dealer's choice I reckon, oh yes I do, because, duh, duh, duh -- If you understand the purpose and use of null hypotheses before attempting to convey p-values vis-à-vis to them, _and_ have some basic understanding of probabilities then it seems to me you're well on your way to estimating something somehow.
After she’s flipped the coin but before she’s revealed it, she says there is now a true answer but we just don’t know it yet. But I would say, even before she’s tossed the coin, there is a true answer, assuming we don’t live in a multiverse. Before she’s tossed the coin, it will still come to pass that the coin will either land heads up or tails. It’s still a truth that exists but which we don’t have any way of knowing. As we saw, the coin ended up landing tails up. So before she tossed the coin, it was true that after the toss it would it would turn out to be tails, even though at that time there was no way of knowing it.
There's one thing I don't quite understand about the Frequentist view. If the answer is set once the coin has landed on your palm, then surely it was also set before you flipped the coin, and surely from a deterministic point of view, the answer could have been predicted at the Big Bang. From that perspective, the idea that "there is a set True answer to that question, therefore there is no point assigning probabilities to it" doesn't make sense. There is always a True answer to any question but it is never knowable due to our lack of information. That lack of information could either be your hand covering your palm or it could be not knowing the exact position of each atom in your body and in your brain before you flip the coin: It's still just a lack of information, and probabilities are just here to fill the gap between our limited understanding of the world and its "True" state. Did I miss something?
When you let AI think for you you allow the people that created it to control your mind. If the people controlling AI want to brainwash a nation that is what they would use. We would be like a communist country and only obey without question.
Why do I have to belong in one of two opposing camps? What is wrong with using a t-test when it is what is going to give me the answer i want, and do the same with calculating a posterior probability when this is what i want to know? Scientific thinking should be rid of ideologies.
What's the frequentist perspective on getting the model right? It always seemed to be like bayseans try to make that more explicit but I struggle to see a real distinction, but I'm kind of convinced (after adjusting my prior ;-) I just don't understand the semantic distinction that the frequentist perspective had with respect to (implicit) assumptions
