What major topics do you want us to discuss on Friday?

[palday]

One topic per answer and upvote other answers that interest you!

[ansost]

Model Comparison

[yjunechoe]

What language to use when reporting (in publications) some of the strategies we discussed, especially:

• Profiling
• Bootstrapped confidence intervals

And how to cite MixedModels.jl :)

[palday]

There's a Zenodo link for citation on the MixedModels homepage.

[palday]

Confidence intervals were computed using the parametric bootstrap with n replicates.

where n is the number you used -- at the very least 1000, but preferably 5000 or even 10000.

You should also specify whether you used the equal-tail probability interval or the highest density (shortest) interval.

For profiling,

Confidence intervals were computed by profiling of the likelihood surface.

[pandongpeng]

Reparameterization in post-hoc analysis with examples.

[yjunechoe]

(If there's time) A worked out example (or at least a list of things to consider) for parametricbootstrap with a numerical covariate too! (Edit for clarity: parametricbootstrap w.r.t. power analysis)

[TeichPaula]

Principal component analysis

[ansost]

Including the difference between fit and predict:) Most explanations I saw are for train and test data but I am not sure whether I would have to use both fit and predict or just fit if I dont split my data.

[pianobacki]

non linear covariates:
more examples on how to deal with covariates (like time), if this covariate shows a nonliniear behavior.
e.g. ERP responses in 4 different tasks;
how to center (meaningful zero),
scaling?
how to transform? (inside the model or before in the dataset)
how to include in the model (only the transformed version, or also in original metric)
what can go wrong?

Examples were the dv has a non linear trend over time:

[cyhsieh-psy]

GLMM for different types of dependent variables (e.g. count data, proportion data, ordinal data)

[ebruevcen]

(If time allows) a worked out example of power analysis (maybe using one of the available datasets?), if such a thing is ever possible? If not, why? Also, how do I decide the values (e.g., β, σ) for a priori power simulation?