But you said "little consensus and a lots of hard conceptual problems". I don't think that stats should pose a lot of hard conceptual problems for someone who's just trying to use it as a tool to evaluate the strength of evidence. This doesn't mean you can half-ass it though.
-
-
it would be nice if it didn't pose conceptual problems, but I don't know how one can expect to use inferential stats to support one's verbal claims without understanding where they come from and what they mean.
0 proslijeđenih tweetova 4 korisnika označavaju da im se sviđa -
Odgovor korisnicima @talyarkoni @bradpwyble i sljedećem broju korisnika:
take for example the distinction between fixed and random effects. failing to understand the distinction and specify one's model appropriately has momentous implications for "basic" tests, but it's hardly obvious or intuitive
1 reply 0 proslijeđenih tweetova 4 korisnika označavaju da im se sviđa -
Odgovor korisnicima @talyarkoni @bradpwyble i sljedećem broju korisnika:
ditto for the question of whether a null of zero is sensible, or if its rejection provides any meaningful corroboration of one's theory. that's absolutely foundational, but not easy to think about, and people have radically different opinions
1 reply 0 proslijeđenih tweetova 4 korisnika označavaju da im se sviđa -
If the rest of your generalizability criticisms are true, does it really matter if people interpret their statistical results with perfect acuity? i.e. Isn't a misuse of zero nulls the absolute least of our worries?
1 reply 0 proslijeđenih tweetova 2 korisnika označavaju da im se sviđa -
Odgovor korisnicima @bradpwyble @talyarkoni i sljedećem broju korisnika:
The reason I ask is because if concerns are that we have some incredibly hyperspecific and ungeneralizable theories/findings... isn't drilling down on these issues going to pull people even farther from asking questions that really matter?
1 reply 0 proslijeđenih tweetova 0 korisnika označava da im se sviđa -
not sure I follow. people are free to ask whatever questions they like. but I don't think it should surprise anyone if the difficulty of justifying verbal claims with statistical evidence tends to scale with the breadth of the question.
1 reply 0 proslijeđenih tweetova 0 korisnika označava da im se sviđa -
Sure people are free to choose but there are also only so many hours in the day. I just wonder if getting people to focus on deep conceptual problems in statistics will come at the expense of other problems (like generalizability)
0 proslijeđenih tweetova 0 korisnika označava da im se sviđa -
tbh I'm not sure what you're objecting to. all I'm saying is that learning from data via inferential stats is hard, and the fact that universalized push-button procedures exist does not mean that their mindless application can tell scientists what they want to know.
1 reply 0 proslijeđenih tweetova 2 korisnika označavaju da im se sviđa -
Odgovor korisnicima @talyarkoni @bradpwyble i sljedećem broju korisnika:
if your point is that time is finite and scientists can't all be statisticians, then sure, I agree. but reality is not under any obligation to care about the length of an Earth day or how much publication pressure a given scientist feels.
1 reply 1 proslijeđeni tweet 5 korisnika označava da im se sviđa
and it's not like this is specific to stats. e.g., most cognitive neuroscientists know almost nothing about system neuro, MR physics, etc. they operate on the assumption that those are just details they can gloss over without cost. sometimes they're right, and sometimes not.
-
-
Odgovor korisnicima @talyarkoni @bradpwyble i sljedećem broju korisnika:
all I'm saying is that the fact that someone doesn't like complexity (e.g., "I won't learn Bayesian stats because it seems complicated") doesn't make the complexity magically disappear. it's there whether you like it or not—even in the lowly p-value from a "simple" t-test.
1 reply 0 proslijeđenih tweetova 4 korisnika označavaju da im se sviđa -
Ok - trying to restate my original point: coming up with push-button consensus best practices for very simple cases of Bayesian statistics would increase adoption. The resulting inferences wouldn’t be perfect but pragmatically this step might have real benefits.
0 proslijeđenih tweetova 3 korisnika označavaju da im se sviđa - Još 14 drugih odgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.