Timeline for Do epistemic oughts exist?
Current License: CC BY-SA 4.0
14 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jun 29 at 19:34 | comment | added | user6527 | this isn't a Bayesian-v-frequentist thing - both sets of tools are useful, just for different tasks. They do get misused, but misunderstanding something cannot be the basis for an "ought". | |
| Jun 29 at 16:30 | comment | added | Annika | @DikranMarsupial -- OK, but people do take it as potentially prescriptive "ought" anyway. I'm not debating different views of proper stats practice, but people use both systems to try to answer the same questions (whether you and I agree that is correct or not). All I'm trying to show is that even within a theory, the choice of inference methods will differ in what one ought to conclude. Bayesians things frequentists are making logical errors, but nonetheless both camps can come to different conclusions about whether the data support the hyopthesis. | |
| Jun 29 at 14:04 | comment | added | user6527 | no, it is more that frequentists and Bayesians define probability in different ways, which means there are questions that a frequentist cannot answer directly using a probability. "boring ol' NHST, oughts are subjectively set by p-value cutoffs." actutally, not really, it is a tradition, but there is no real connection between the p-value and the probabiity that the null hypothesis is false. | |
| Jun 28 at 20:21 | comment | added | Annika | @DikranMarsupial its just that Bayesian's and frequentist approaches use probability at different places (I've used both approaches in my own work), and so can give different views to "ought". Even within boring ol' NHST, oughts are subjectively set by p-value cutoffs. | |
| Jun 28 at 20:19 | comment | added | Annika | @DikranMarsupial I do agree with this statement "frequentists cannot assign a probability to the truth of a proposition, so with a frequentist null hypothesis test there is no mathematical/logical connection between the p-value and the probability that the null hypothesis is false." However, I don't think that frequentists would say this prevents them from reasoning about hypotheses - they can still say, epistemically, I outght not reject the null, or I ought to reject the null, based on what I've learned from the data. | |
| Jun 28 at 20:13 | comment | added | Annika | @DikranMarsupial The reason I mentioned statistics is because its a well defined area where you would think we could appeal to for epistemic oughts but I was simply explaining the even here we don't get to oughts. I think step back and say even if within a fixed theory we could get oughts from statistics, we would still be faced with the underdetermination issue. The "ought" here is related specifically to what ought I to infer given this data. The fact that diff schools answer diff things misses the point of bringing up statistics. | |
| Jun 28 at 5:30 | comment | added | user6527 | "bayesians use degrees of belief, and frequentists focus on severity and error control (per Mayo)" is totally missing my point. I don't think the frequentist-Bayesian thing is doing more than obscuring the valid part of the answer. | |
| Jun 27 at 22:59 | comment | added | Annika | @DikranMarsupial -- agree on epistemic probabiltiy wrt bayes vs frequentist philosophy. Don't want to get too far into phil of stats. We use statistics to make inferences - bayesians use degrees of belief, and frequentists focus on severity and error control (per Mayo). Regardless, we can frame all inference using decision theory and the act of believing/accepting/not accepting etc as implictly an ought (e.g., if p-val > 0.05 we ought not reject the null). I agree there are more nuances here but justpointing out there are many ways to use data to get inferences. | |
| Jun 27 at 21:36 | comment | added | user6527 | Just to clarify, frequentists cannot assign a probability to the truth of a proposition, so with a frequentist null hypothesis test there is no mathematical/logical connection between the p-value and the probability that the null hypothesis is false. Which is why we just "reject the null hypothesis" or "fail to reject the null hypothesis". There is no basis for an "ought" for what we infer about the truth of the hypothesis. There is for a Bayesian test as we have directly made a statement about the plausibility of the hypothesis under a set of assumptions and some evidence. | |
| Jun 27 at 21:31 | comment | added | user6527 | "empiricism and certainty don't go together unfortunately" I completely agree with that - it definitely agrees with my experience! ;o) | |
| Jun 27 at 21:31 | comment | added | user6527 | I wasn't referring to asymptotic guarantees. p-values and posterior probabilities are not answers to the same question, rather like credible intervals and confidence intervals do not describe quite the same things, even though they are often treated as if they were (causing no end of statistical mistakes). The point I was making was that Bayesians and frequentists disagreeing does not rule out epistemic oughts - I wasn't making a comment about Quine-Duhem part. | |
| Jun 27 at 20:02 | comment | added | Annika | @DikranMarsupial - not really, even the asymptotic guarantees of convergence doesn't get us to ought. Frequentists select p-values, bayesians posterior credibilites and priors. IF they happen to agree then I think we'd have a happy situation where a Frequentist and Bayesian would both say we ought to infer something. But as my second paragraph shows, even if statisticians all agreed on oughts for every hypothesis test, we'd still be stuck in theoretical underdetermination .. empiricism and certainty don't go together unfortunately.. | |
| Jun 27 at 6:29 | comment | added | user6527 | Bayesian and frequentist frameworks can also give the same answers to "what ought we to infer from the data", in which case that might justify an "epistemic ought"? Note when Bayesian and frequentist answers differ, it is often because they are answering subtly different questions (due to the frequentist definition of probability being limited to things that can have a long run frequency), although the difference is often ignored (often leading to misunderstanding). | |
| Jun 27 at 1:23 | history | answered | Annika | CC BY-SA 4.0 |