Nous nous sommes ainsi assuré de rendre le tout digeste par l'utilisation d'exemples, en limitant le jargon, et en répétant les concepts importants. 4/4

ainsi que les analyses de sensibilités en contexte descriptif. Le lecteur est ainsi amené à parfaire sa compréhension de la logique statistique derrière l'un des outils de recherche les plus courants en science politique qui est trop souvent omis de la formation méthodologique. ... 3/4

Nous abordons le concept d'ignorabilité des processus générateurs de données des sondages probabilistes et non probabilistes. Ceci nous permet d'illustrer les conditions permettant à la calibration de fonctionner; le lien entre la taille de l'échantillon, le PGD, et le biais de sélection; ... 2/4

There are times though where plots are superior to tables such as in multiple comparisons. Not to toot my own horn, but that is exactly what we tried to accomplish here:
www.cambridge.org/core/journal...

So, trying to add extra information contained in the table into the plot risk obscuring the message the plot was created for.

My reservation here is that to me, the goal of visualization exercises is to render immediately apparent to a lay person what is hidden away in regression tables. A visualization will never fully satisfy quantitatively inclined scholars the way a table would. It's not their point.

I'm with @dandekadt.bsky.social and @andrew.heiss.phd in that these looked like distributions of something at a first glance, but it's actually an ad hoc shading. Hence pairing them to posterior distributions would get the point across while being meaningful.

Forgot to mention, Prof. Armstrong also prepared open access seminars on the topic where he dives a bit more into the weeds of the issue and its implementation and usage in R and Stata. Links below!
instats.org/seminar/usin...
instats.org/seminar/usin...

This is all implemented our VizTest package. You can install the R version from here: github.com/davidaarmstr...

And the Stata version here: github.com/davidaarmstr...

Our procedure figures out what the visualized confidence level should be such that all pairwise tests are visually correct WHILE maintaining accurate comparison with 0 (the null hypothesis).

You might want to visually test whether the estimates for July 2020 and July 2022 are statistically different from each other by looking at the overlap of their CI. You might know that it's wrong, but it's so tempting!