I think the issue is that we can’t know if it’ll give the same results, in our specific use case, until after doing all that work, which also sucks.

I remember the tabachnick and fidell book “experimental designs using ANOVA” that I used in my MA program being super helpful, but it’s been a while since I looked at it.

I learned in grad school to not rely on p-curve analyses from Will shadish and Jack Vevea. I was surprised to see them used so frequently after I graduated, but I could never find a paper backing up what I was taught until now!

Oh I see! Then I also agree lol

Which is why the causal inference we work with is probabilistic and not deterministic

I’m with @solomonkurz.bsky.social . Is this how the field is looking at causal inference? The causal inference class I took in grad school and the ones I teach focus on how we typically only have INUS conditions.

All this! I just did a causal inference workshop at Penn State and I started off with “stats can’t save you here. Causal inference is a qualitative inference given the data collected and the design of your study.” It was very Shadish, Cook, and Campbell heavy with a dash of DAGs

This book was so good! Have you read jonathan strange and mr norrell yet?

The tabachnick and Fidell book, analyzing experiments using ANOVA. has a nice section on a priori contrast for one way and factorial ANOVAs. They even include a table at the end for different contrast based on the number of groups you have. Linda collins MOST book is a good one too

If MCAR, I’ll just randomly impute 1s and 0s so I don’t lose power

Oh absolutely. Unfortunately it means you’ll have to double check these errors in lme4 before determining if they are erroneous though

Another thing to request, particularly when using lme4 or nlme in R, is the confidence intervals. If you just do a summary on the model object, you won’t actually know if your model has converged, particularly in the random effects. Asking for confint (in lme4) or intervals(in nlme) will check this

As a Sacramento kings fan, I am a fan of anyone that is a thorn in the Lakers side. Go Celtics!

Have you been telling participants to “git gud” in your studies?

Wow, this is an impressive three to have in one talk!

I once heard someone say that “they didn’t need to randomize because the data were all collected so close in time to each other”

Definitely this paper. It is written so clearly without a lot (or any equations). It helped me also see that the whole DAGs framework also fits with the threats to validity and the Rubin’s causal framework