Vincent Arel-BundockNov 4, 2022
{marginaleffects} is an easy-to-usešŸ“¦to compute predictions, contrasts, slopes, and marginal means for 71 model classes in #RStats. v0.8.0 comes with powerful new features, bug fixes, and a website with 20k+ words of tutorials https://vincentarelbundock.github.io/marginaleffects/
Show threadJoe QuinnNov 17, 2022

@vincentab I canā€™t wait to use this!

Clarifying Q: the doc says ā€œmost [other R packages] do not actually compute marginal effects as defined above. Instead, they compute ā€˜adjusted predictionsā€™ for different regressor values.ā€ But when the doc defines marginal effects ā€œabove,ā€ it says ā€œin ā€˜marginal effects,ā€™ we refer to the effect of a tiny (marginal) change in the regressor on the outcome.ā€œ These two things sound the same - can you help a new user understand the difference?

Show threadJoe QuinnNov 17, 2022
@vincentab related - I would love to see a tutorial/blog post that walks through an empirical example step by step to demonstrate the distinction between ā€œadjusted predictionsā€ that other marginal effects packages deploy vs the derivative equivalent. So much so that I may even write one myself if I tinker enough šŸ¤£
Show threadVincent Arel-BundockNov 18, 2022
@jmquinn I meant that in the context of this package a "marginal effect" is defined as a "slope". Other packages use the same expression to refer to what I call "adjusted predictions" or change in adjusted predictions ("contrast"). So in effect, they do *not* compute slopes.
Show threadVincent Arel-BundockNov 18, 2022
@jmquinn There are 20,000k+ words of vignettes on the website, showing detailed walkthroughs with real data for each of the quantities of interest. Let me know if you read those and are still missing something. You may find the "Alternative Software" vignette particularly interesting, since it makes a bunch of explicit comparisons. Table of contents: https://vincentarelbundock.github.io/marginaleffects/index.html#table-of-contents
Marginal Effects, Marginal Means, Predictions, and Contrasts

Compute and plot adjusted predictions, contrasts, marginal effects, and marginal means for over 70 classes of statistical models in R. Conduct linear and non-linear hypothesis tests using the delta method.

Show threadJoe Quinn
@vincentab awesome - thanks for getting me past my first page confusion šŸ¤£
Nov 19, 2022 at 12:20amMastodon for iOS