Por otro lado, la defensa de la tesis doctoral de Pablo Solana, titulada 🧠 "Grounding meaning in the sensorimotor system: behavioral, neurostimulation and meta-analytic studies", está programada para el próximo 30 de enero.

¡Felicidades a ambos investigadores por alcanzar este importante hito!

I did not intend to blame anyone for being ignorant. However, I do think that people should take the time to read the documentation and educational materials before using the package, especially if they are doing scientific research with such tools. Of course, the real problem is structural.

However, I think it is mostly the user's fault when they use marginaleffects incorrectly, given that its author has put in an insane amount of work to explain the "black box" behind the package.

If you can distinguish between by and avg_* and understand how newdata and variable work, then perhaps you should using marginaleffects at all.

Yes, I agree. In situations that matter, such as scientific publishing, I think such errors should be detected during peer review. Since that generally doesn't happen, there are fantastic initiatives, such as @error.reviews, that try to do that work ad hoc. However, it is clearly not enough.

Oh! I see. Fortunately, I've always been a "marginaleffects" user, and I think it works fine for repeated-measures designs. However, I'm really interested in situations in which things can go horribly wrong...

In which situation it can "trip you up"? Assuming that you are already incorporating the correct random effect structure in your model, but computating marginal effects whitout taking it into account. If you are interested in fixed effects (at least with the delta method), it should be fine.