Lightnews — Scholar-powered news

Charly Marie, PhD @charlymarie.bsky.social · 2d

Bref: l’article 8 est respecté. L’usage, lui, l’est beaucoup moins (pour peu qu’il existe un usage, car la situation ne s’est jamais présentée de cette façon…).

Que vos attentes soient contrariées, que votre lecture ne coïncide pas avec celle du PR, voire que vous vous sentiez trahi, je l’entends.

1

Charly Marie, PhD @charlymarie.bsky.social · 2d

En l’occurrence, la règle est respectée, et aucune contrainte mécanique (l’assemblée) ne s’impose à Macron dans la nomination du PM.

Cela n’enlève rien au fait que la gauche a remporté une majorité relative, et aurait pu être mise à Matignon. Mais la situation n’est pas comparable à 97.

1

Charly Marie, PhD @charlymarie.bsky.social · 2d

Ouhla. Ne montez pas si vite en tension :).

Jospin avait la majorité absolue au parlement, via la gauche plurielle. Nommer un autre PM aurait mécaniquement amené une censure, Chirac n’avait aucune autre option.

Ici, on a vu qu’une censure n’est pas mécanique/immédiate.

1

Charly Marie, PhD @charlymarie.bsky.social · 2d

Et où est-il écrit que le premier groupe parlementaire (qui n’en est pas un, mais une alliance électorale) remporte les élections législatives?

En l’occurrence, le PR nomme le PM selon l’article 8 et rien de plus.

1

Charly Marie, PhD @charlymarie.bsky.social · 2d

(Also, note that X3 is not affected by collinearity, which highlights the point made by @dingdingpeng.the100.ci in her blog)

1

Charly Marie, PhD @charlymarie.bsky.social · 2d

It's funny (not) how some highly important assumptions, such as independence of the errors, are given the same level of importance as collinearity in many classes students take...

1

Charly Marie, PhD @charlymarie.bsky.social · 2d

What's the quality of your p-value? Asking for a friend.

1 2

Charly Marie, PhD @charlymarie.bsky.social · 2d

Regarding screenshots

If your covariates are perfectly correlated (r = 1), you can just drop one of them. You're (literally) measuring the same thing. Duh!

The same applies if r = 0.99: there's no point in including both covariates, since they are (almost literally) measuring the same thing.

1 1

Charly Marie, PhD @charlymarie.bsky.social · 2d

Great post, as always!

I'd just add that multicollinearity is *almost* never an issue... unless your predictors are so heavily correlated that it becomes statistically spaghetti (i.e., that it becomes algebraically impossible to disentangle their unique effects, leading to unstable estimates).

1 1

Reposted by Charly Marie, PhD

cosmoflash @cosmoflash.bsky.social · 5d

Lecornu II : Matignon Drift
Lecornu 3
Lecornu 4Ever
Lecornu V Infinite
Lecornu Origins
The Lecornu Tales
Legend of Lecornu

63 320 1.4K

Charly Marie, PhD @charlymarie.bsky.social · 5d

Glad I could help!

2

Charly Marie, PhD @charlymarie.bsky.social · 7d

You’re right, sorry. I meant to say that it changes them! One shouldn't type when half awake

1 1

Charly Marie, PhD @charlymarie.bsky.social · 7d

How can your confidence intervals be exactly equivalent? It is known that multiple imputation inflates them.

1

Charly Marie, PhD @charlymarie.bsky.social · 12d

That's great! Will you be sharing your workshop online like your courses?

1 1

Charly Marie, PhD @charlymarie.bsky.social · 20d

C’est une question de censure, il y a une grosse littérature sur ce sujet, cf. hbiostat.org/rmsc/surv

(Je ne sais pas répondre à la question, juste pointer vers une direction pour y répondre 🙃)

17 Introduction to Survival Analysis – Regression Modeling Strategies

hbiostat.org

1 2

Charly Marie, PhD @charlymarie.bsky.social · 20d

Why do you say that rm(list = ls()) is not a safe trick? To me, it does what it's supposed to do: delete all objects. Remaining package conflicts are not a problem as this is not what rm(list=ls()) is intended to do, right?

1 1

Charly Marie, PhD @charlymarie.bsky.social · 23d

Merci pour la veille.

Vraie question, y a-t-il un fondement juridique au double standard entre les drapeaux déjà affichées (ukrainien, États-Unis post 11 septembre, etc…) et le drapeau palestinien ?

1 1

Charly Marie, PhD @charlymarie.bsky.social · 23d

Poisson goes brrr.

But good old OLS is still my favourite.

Magnus Johansson @pgmj.bsky.social · 23d

For instance, yesterday I read a paper with a table describing participants' sickness absence days with a mean of 71 and SD = 88. Generating a random (gaussian) sample using these values produces ~20% participants with less than zero sick days.

A histogram with 10000 values using the mean+sd in the text, showing zero with a red line.

Charly Marie, PhD @charlymarie.bsky.social · 28d

Interesting — thanks! Could we create 100 silicon samples to mimic the sampling procedure and the associated noise? Would you then see a link with Rubin’s rules for multiple imputation?

1 2

Reposted by Charly Marie, PhD

Jamie Cummins @jamiecummins.bsky.social · 28d

This is when I realized: we’ve been here before! The garden of forking paths in well-trodden by us psychologists. And we already know that analytic flexibility can allow us to present basically any finding we want as significant.

So I wondered: how much flexibility can we get with silicon samples?

1 2 34

Charly Marie, PhD @charlymarie.bsky.social · 28d

LLMs are incredibly powerful tools.

The only limitation is that, in order to unlock their full potential, it is necessary to verify the answer.

In other words, you have to be able to do what the LLM is doing.

Kelly Bodwin @kellybodwin.com · 28d

The thing that struck me most is that even if you take away the code-writing necessity, knowing what to ask the bot for requires so much expertise.

Steps like "These points are overplotted, can we fix that" or "Let's look at only the top 50 restaurants" are incredible non-obvious to beginners...

Ted Laderas @tladeras.bsky.social · 28d

Hadley is demonstrating ggbot2 - an llm to help to transform a plot. As an analyst who has done endless iterations on a plot (can you make this gteen?), this is like a dream. #PositCONF2025

3

Reposted by Charly Marie, PhD

Julien Gossa @juliengossa.cpesr.fr · Sep 13

This would be the right time to push for the creation of an ‘old-school code school,’ proudly advertising that it doesn’t use AI — betting on the idea that programmers who actually know how to program will soon be rare and worth their weight in gold.

2 2 18

Reposted by Charly Marie, PhD

Vincent Arel-Bundock @vincentab.bsky.social · Sep 13

The new {marginaleffects} release for #RStats (0.30.0) comes with two new vignettes:

1. Speed up computation with automatic differentiation (often 10x gains) marginaleffects.com/bonus/perfor...

2. Power analyses with {marginaleffects} and {DeclareDesign}. marginaleffects.com/bonus/power....

37 Performance – Model to Meaning

marginaleffects.com

3 34 150

Charly Marie, PhD @charlymarie.bsky.social · Sep 12

Great plot !

1

Reposted by Charly Marie, PhD

Thomas Lin Pedersen @thomasp85.com · Sep 11

I am beyond excited to announce that ggplot2 4.0.0 has just landed on CRAN.

It's not every day we have a new major #ggplot2 release but it is a fitting 18 year birthday present for the package.

Get an overview of the release in this blog post and be on the lookout for more in-depth posts #rstats

ggplot2 4.0.0

A new major version of ggplot2 has been released on CRAN. Find out what is new here.

www.tidyverse.org

9 280 850