Cats are way ahead of us

I am so gonna copy this idea [conditional on future-me remembering to buy rubber ducks]

Dirichlet (distribution)

`httr2` (httr2.r-lib.org)

Sorry about that. I removed my reply. Thank you for letting me know. My typing fingers got ahead of my brain

📌

📌

You forgot to mention that all files should be in a git repo hosted on GitHub or even better - on your own personal server

Yep. You don't want to end up with scientific notation for your coefficient estimates because they are too small or too large. Not only because it's not easy to read for some people, but also because it can bring up numerical issues in estimation

I guess we end up with my most frequent response to any stats-related question - "it depends". In this case, it depends on the measurement scale of the DV, on the range of Age in the sample, and the specific goal of the analysis (e.g., what hypothesis needs to be tested)

Let's say the outcome is yearly income. This income can be measured in EUR, or it can be measured in another currency (i.e., HUF), with an exchange rate of 1 HUF = 0.0025 EUR, or 1 EUR = 400 HUF. When using the DV measured in HUF, rescaling age might not be needed.

Before rescaling or transforming X_Age, it is very important for authors to think carefully about the scale for `outcome` and how fine-grained it is, as this impacts the magnitude of the coefficients for Age, Age^2, Age^3.

I agree that the models would be equivalent, as in the corresponding p-values should be the same between models that use X or X/10 (or X/100). However, I did run into numerical issues, even with a 3rd degree polynomial. I vote in favor of rescaling X rather than mean-centering or standardizing X

sfdora.org/read/

Fraud all the way down