Lightnews — Scholar-powered news

Sean Mackinnon @seanpmackinnon.bsky.social · 1d

Maybe the analogy is more like "Sure, climate collapse is happening, but hey, at least we'll get some good beach days"

I do think that generative AI is going to force most classes either to (a) abandon grading or (b) abandon online assessment / graded homework.

While I pine for a, probably b

1 2

Sean Mackinnon @seanpmackinnon.bsky.social · 1d

I wish lmao

1

Sean Mackinnon @seanpmackinnon.bsky.social · 1d

I think if Bluesky goes the way of X I might just quit public social media entirely, group chats only for life

Don't got it in me to pack it up again, and increadingly wonder if public social media was a mistake for society, tbh

1 3

Sean Mackinnon @seanpmackinnon.bsky.social · 6d

After a year they each produce reports. All 4 are spliced together into one document. No numbers ever double checked, hell no codebook produced. A 5th person wries up the intro as 1st author, the original analysts graduate. Someone asks for code so they say "No" because all the above is embarrassing

1 2

Sean Mackinnon @seanpmackinnon.bsky.social · 6d

Not principled, but workflow of some people is so disorganized, I think a lot of analysts find it hard to imagine.

Like, imagine you noodle around in "data (1) (1) (1).sav" for a while, write some numbers in a Word file, and send the dataset off by email to 3 different students who each work alone

1 1

Sean Mackinnon @seanpmackinnon.bsky.social · 6d

Me neither, which is kind of why I asked. If it's so prevalent, maybe it has some special benefit I'm missing even though it seems bad to me?

1

Sean Mackinnon @seanpmackinnon.bsky.social · 6d

Being the stats consultant on a lot of things, it kind of gets exhausting sometimes where everything has to come together on such a rapid timeframe, it feels like just holding together a project with duct tape.

I'd honestly love working with people taking it slow, and matching models to questions

1 1

Sean Mackinnon @seanpmackinnon.bsky.social · 6d

So "most scientists" are intellectually bankrupt then, you mean :p

(I suspect we agree on most points here, tbh. Everything in science just feels overly rushed, probably linked to general austerity and decreased funding in the sector as much as individual scientists' own knowledge limitations)

1 2

Sean Mackinnon @seanpmackinnon.bsky.social · 6d

Ah, methodological doomerism, the ultimate "all models are wrong" defense

(I get it though, lots of precise answers to the wrong question out there! I'm more personally comfortable making the best of bad situations, even if everything is strictly speaking, not 100% true)

1 1

Sean Mackinnon @seanpmackinnon.bsky.social · 6d

All good, I basically do something like that. The percent decrease effect size just seemed kind of useless to me, and just making sure I'm not missing something obvious.

1 1

Sean Mackinnon @seanpmackinnon.bsky.social · 6d

Real talk here: What analysis method that tests broadly similar research questions would you consider not intellectually bankrupt?

1

Sean Mackinnon @seanpmackinnon.bsky.social · 7d

We all have to lie to ourselves about that every year, only way to keep sane hahaha

1

Sean Mackinnon @seanpmackinnon.bsky.social · 7d

Walking the razors edge on deadlines just to feel something. Bodes well for the term lmao

1 1

Sean Mackinnon @seanpmackinnon.bsky.social · 7d

The problem is: I can't use % decrease as an effect size for power analysis, because there's no standard deviation.

Am I missing something which would allow me to use this in some capacity?

2

Sean Mackinnon @seanpmackinnon.bsky.social · 7d

So, if I'm doing sample size determination with a power analysis, I either need to know both means and the pooled standard deviation or the standardized mean difference:

(Pre - Post) / Pooled SD

When I ask MDs how big they think the effect is, they usually specify it as % decrease

2

Reposted by Sean Mackinnon

Joe Bak-Coleman @jbakcoleman.bsky.social · 7d

3 2 15

Sean Mackinnon @seanpmackinnon.bsky.social · 7d

When I'm consulting on medical #stats power analyses, MDs always seem to phrase effect sizes as a "percent decrease". For a t-test basically:

(PreTest - PostTest) / (PreTest)

I've never been able to make any use of this for power analyses. Does it have some sort of hidden use I'm not seeing?

3 2

Sean Mackinnon @seanpmackinnon.bsky.social · 9d

I was using notepad in Windows 11 and somehow clicked a button to format the text with bold, italics, etc.

It's notepad, I'm using it explicitly so it won't autoformat anything.

What kind of terrible world do we live in that I need to gesture to this @dasharez0ne.bsky.social T-shirt constantly.

Shirt with skeleton on computer, "I will never paste with formatting" Do not format my paste!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

I dunno why there is an american flag in the background, but I'm too lazy to find a better picture

5

Sean Mackinnon @seanpmackinnon.bsky.social · 10d

I dunno why, but sometimes music just makes it easier to process feelings for me too

1

Sean Mackinnon @seanpmackinnon.bsky.social · 11d

My hot take for online rating systems is use a 10pt rating scale (or 5 star with halves) but only report the trimmed mean.

All the zeroes and 10s are essentially contaminated data, imo

Reposted by Sean Mackinnon

tj ♏️ahr @tjmahr.com · 12d

Hank Hill

I do not want to save to onedrive.
I want to save to the documents folder... On my computer... That I own... In my house.

1 2 3

Sean Mackinnon @seanpmackinnon.bsky.social · 12d

I never want OneDrive as my default, burn all the servers

1

Sean Mackinnon @seanpmackinnon.bsky.social · 12d

Masterful post. Inspired, really.

Plebs over here arguing about p-values or mediation or some shit, and you just drop this into the discourse.

1

Sean Mackinnon @seanpmackinnon.bsky.social · 12d

Incredible, why does it have all positive reviews lmao

1 3

Sean Mackinnon @seanpmackinnon.bsky.social · 13d

This one sort of feels like a less pragmatic Monty Haul example. But, still kinda neat, ngl

If someone wanted to be REALLY annoying and pedantic, technically human birth rates of males vs females is not 50/50

2