I'm on an REB and have to review ethics applications pretty regularly. If you're doing a survey, don't list a few "sample questions" in your method, and then assume you get blanket approval to ask whatever later. You gotta like, know what the methods are before ethics approval.

Shitpost options for this poll: Code as 8, but add a bunch more 8s so it's not an outlier. Dox the participant, go to their house, and make them do it right this time. *squint* it kinda looks like a 3 so let's go with that NA, then ask chatgpt the to impute the missing value

Drunk statistics needs to be a new meme format. Explain some sort of statistical thing with no preparation, but only if you're very drunk🤣

I sometimes get confused on Twitter when people seem to think that inferential statistical methods give us claims to truth. Stats are useful ways to make guesses about uncertain/ unobservable phenomena via inductive reasoning. But you have to live in perpetual uncertainty.

Students’ least favorite academic subjects ranked: 10) you 9) can’t 8) rank 7) them 6) because 5) individuals 4) like 3) different 2) things 1) Statistics

Our early conversations for this study focused on how age confounds the stimulant-substance use link. Figure 1 from our paper (my original image below) shows that kids tend to stop using stimulants at about the same time as they begin experimenting with alcohol and drugs: /4

"The Performance of RMSEA in Models With Small Degrees of Freedom" journals.sagepub.com/doi/epub/10.11…

What would you call this kind of coding scheme for a binary categorical variable: -0.5 = Control 0.5 = Treatment There seems to be massive inconsistency across published sources. I'm trying to find a consistent / most common term for it to use in my teaching.

Regarding my recent poll, what about this for a naming scheme? Effect coding is an umbrella term for "coding so the intercept is the grand mean". Sum (-1, 1) and deviation (0.5, -0.5) are examples of effect coding. Pics below to disambiguate sum, deviation, and dummy codes.

I am losing my mind at this post on Reddit. Apparently he was mishearing the phrase "nice to meet you".

So...hi! I don't really post here as often these days, but I find myself stuck in an airport with a delayed flight only to find out that an "in-press" publication with my awesome coauthors Ed Kroc & Bruno D. Zumbo has come out. And this was a fun one! link.springer.com/article/10.375… 1/13

BLOCKBUSTER SUMMER MOVIE TITLE! For our next Quantitude summer fun contest, what would be a great quant/stat/academia-related name for a summer blockbuster movie? (Groan-worthy puns welcome.) (Please post by Wednesday.)

I hate that odds ratios are asymmetrical. It's so brutal for interpretation. Like: OR = 2.5 OR = 0.4 Same effect size, different direction. Attached picture shows a plot of how asymmetrical. I can deal, but why did math have to do me dirty like this? I gotta teach this stuff.

Keeping on the logistic regression theme, anybody out there use dominance analysis and pseudo-R2 as a logistic regression? I used it once before productively. The appeal of brute forcing an effect size solution in this area is palpable psycnet.apa.org/record/2010-07…

The unofficial official Barbie theme, theme_barbie(), is now available on my GitHub: github.com/MatthewBJane/t… Also comes with the scale_fill_barbie() for the gradient fill you see on the right. And hex color values like “text_color_barbie” for the text color you see below.

I've been feeling increasing cognitive dissonance being on Twitter as the platform gets worse in so many ways. Feels like this "X" rebrand is a sort of formal end to it all. I'm going to try being on Mastodon instead for a while since I have an account. Maybe Bluesky someday.