BLUE
Login
DE
David Epstein
@davidnotdave.bsky.social
Literature major, then neuroscience Ph.D., then addiction researcher. Enough-knowledge-to-endanger-myself in social sciences and statistics. He/him. Views my own.
110 followers118 following20 posts
OverviewPostsReplies
DEdavidnotdave.bsky.social

"Orange you glad we didn't maintain a data column labeled 'methamphetamine'?"

Screenshot from a published protocol called "Personalized Deep Learning for Substance Use in Hawaii: Protocol for a Passive Sensing and EMA Study." Excerpt: "As a precaution, the interface on the app will not be labeled as 'substance consumption' and 'substance use' but rather as 'banana consumption' and 'banana use.' Furthermore, our data will be stored and labeled as 'fruit' rather than 'substance use.'"
0
DEdavidnotdave.bsky.social

I'd been wondering whether the nearness of .48 to .50 in my example had meaning. Thank you!

0
DEdavidnotdave.bsky.social

And furthermore, I might be willing to die on the same hill as these fellas...

John Hoenig and Dennis Heisey: "We feel that researchers often do not need a rigorous understanding of confidence intervals to use them to good advantage. Although we cannot demonstrate it formally, we suspect that imperfectly understood confidence intervals are more useful and less dangerous than imperfectly understood p values and hypothesis tests. For example, it is surely prevalent that researchers interpret confidence intervals as if they were Bayesian credibility regions; to what extent does this lead to serious practical problems? .... If informally motivated confidence intervals lead to better science than rigorously motivated hypothesis testing, then perhaps the rigor normally presented to students destined to be applied researchers can be sacrificed."
0
DEdavidnotdave.bsky.social

Ah, but that brings us to whether a CI is a cat's-eye or a cliff. I'm a cat's-eye guy myself.

Two screenshots from competing papers. Geoff Cumming, 2013: "A value close to the center of the CI is about 7 times as likely to be u as is a value near a limit of the 95% CI. Thus, the black area is the likelihood profile, or beautiful 'shape' of the 95% CI. Our CI defines an interval of plausible values for u, but plausibility varies smoothly across and beyond the interval, as the cat's-eye picture indicates." John Kruschke and Torrin Liddell, 2018: "Notice that a confidence interval has no distributional information. There is no direct sense by which parameter values in the middle of the confidence interval are more probable than values at the ends of the confidence interval. This absence of distributional information is why the confidence interval in Fig. 3 is drawn as a flat line."
1
DEdavidnotdave.bsky.social

I kinda startled myself with this little experiment in power calculations. You can have zero ability to find the effect you're looking for, but more than zero "power," all of which reflects your likelihood of overestimating the effect.

If your sample size is 50, then the smallest correlation that will be statistically significant (at .05, two-tailed) is .28. An observed r of .27 in that sample cannot be statistically significant (at .05, two-tailed). And yet… A power analysis would tell you that you have 48% power to detect that correlation with n = 50. The power analysis actually means you have a 48% probability of getting a p value <.05 with n = 50. When that happens, it can only mean that the correlation in your sample is spuriously large: bigger than .27. It's impossible, with n=50, to detect the correlation at its true size (r=.27) and have it be statistically significant. Presumably, this problem is less extreme at conventional power levels (80% or more), but the underlying fact seems to remain.
2
DEdavidnotdave.bsky.social

I really liked the discussions of equity, privacy, and consent in this big-data study group at #ACNP2023.

The Science and Ethics of Measuring and Modeling Individual and Group Behavior. Cheryl Corcoran, Chair. Holly Moore, Co-Chair. Elizabeth Stafford, Moderator. Laura Cabrera, Justin Baker, Tingting Liu, Satrajit Ghosh, and Sara Berger, Participants.
0
DEdavidnotdave.bsky.social

I mostly love Mike Flanagan's movies and series, but his 𝘏𝘰𝘶𝘴𝘦 𝘰𝘧 𝘜𝘴𝘩𝘦𝘳 adaptation asserts that opioid analgesia is an essentially illegitimate aim ("there's no such thing as a painkiller"). Ugh.

0
DEdavidnotdave.bsky.social

Exception: Willie & Frankie sketches.

1
DEdavidnotdave.bsky.social

This is why I'm glad there's a section for confidential comments to the editor.

I don’t usually say “reject.” I usually say, “Just present the findings more tentatively.” And I’m sure the authors will want to do so. They’ll say something like, “Let’s publish this as a preliminary finding. It will inspire future research to resolve the ambiguities.” But it’s not a case of “no harm done.” People love genetics and they love to overread it. There will be a press release and a tweet and a TikTok dance, all trumpeting how genes determine...
0
DEdavidnotdave.bsky.social

My first-ever Bluesky post is an ask: Do any of you have procedures/advice for giving/coding a TLFB (timeline follow-back) for drugs that aren't alcohol? Especially opioids in the age of fentanyl, with which it's probably harder than ever for people to know the identity or amount of what they got?

0
DE
David Epstein
@davidnotdave.bsky.social
Literature major, then neuroscience Ph.D., then addiction researcher. Enough-knowledge-to-endanger-myself in social sciences and statistics. He/him. Views my own.
110 followers118 following20 posts
HomePopularExplore