In 3 you keep the full model, so the hypotheses and error structure stay the same, you’re not changing what was tested.
In 2 you’re actually changing the model (and therefore the hypothesis), so you’d need a different experiment for those p-values to be valid.

I’d choose 1 or 3.
If there’s no strong reason to subset, show all the conditions you tested, inefficient treatments are still informative.
If there is a pre-defined reason to group some conditions, keep the stats from the full model and report those tested together.
I would say 2 is p-hacking.

Like two drops of water.

I think you can use Google Translator to translate “being French” into Japanese?

Sent you a DM!

Do you see changes in markers of Alzheimer’s? Increased Abeta or p-Tau? If so, I have a couple of suggestions.

Waiting for @dieworkwear.bsky.social ‘s opinion on this.

Els meus veïns ♥️

Vas entregar PEC?

Oh you didn’t receive the newsletter? finding fancy acronyms is part of our job now

I hope after the ladder they focus on the window.

Such a beautiful place. I remember running from La Madrague to Béouveyre and crossing to Marseilleveyre through le Col des Chèvres. Can’t count how many times I did it.
Absolutely stunning. Having that a bus ride away was a luxury. I miss it so much!

No one is stopping you 👀

Et faries creus de com han de posar en majúscules i negreta a les convocatòries de beques que es tingui en compte el factor sexe en els projectes.

Pere, sé que fa 15 anys m’hauries dit que t’hi pots trobar una cosa molt específica.

Per què aixeques pedres del camp, Pere?