@peterdrake so I can't say I prefer R, but I do like some aspects of it. So, a little bit of background: I did most of my undergraduate and masters research using R, particularly data mining biometric data for UX/UI experiments. I wasn't introduced to Python and Jupyter notebook until my last semester of my masters degree, and I've been using them ever since for my work and research.

R has what I call the anti-Pythonic philosophy: there are many packages/libraries that do the same thing. It was design for statistics first and foremost, hence the index starting at 1 and the multitude of statistical packages. This is particularly acute with statistical tests. One anecdote that I have was I found the same statistical test (believe it was Tukey's range test) in both Baylor's psychology statistical package and another stats package. Also, academia tends to make and publish their own statistical packages in R based on their research, though this has changed a lot, especially now that Python and Julia are more popular. However, despite R's anti-Pythonic philosophy, there is a more Pythonic push with Tidyverse ( https://en.m.wikipedia.org/wiki/Tidyverse).

One of the positives that I like about R is that it (probably) has every type of statistical model you could want. For my regression class, I created an ordinal logistic regression model for my final project, which was only possible to do in R at the time (I believe I used polr). Python now has ordinal regression in statsmodels, but this was very recent and may not be as expansive as polr. I will say though that rpy2 has made package deficiency in Python less of an issue, but it still takes downloading and understanding R packages.

tldr; R's one shining positive is that it (probably) has every statistical model and test you can name, no matter how obscure it is.