gsimpson.bsky.social
Gavin Simpson
@gsimpson.bsky.social
2.6K followers 680 following 98 posts
(Palaeo)[ecologist | limnologist] & #fakeStatistican, #rstats user, wielder of #GAMs. He/him/his. Opinions mine…
Posts Media Videos Starter Packs
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · 2d
I have very fond memories of feeding the giraffe there; have fun enjoying Nairobi (taking breakfast at the national park at dawn, in one of the picnic / viewpoints on the high ground overlooking the rest of the park was a real treat, as was the elephant orphanage)
1
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · 9d
If you want to check out the examples from the Physalia course, the materials from the last running are available on GitHub

github.com/gavinsimpson...
GitHub - gavinsimpson/physalia-gam-course: Generalized Additive Models; a data-driven approach to estimating regression models
Generalized Additive Models; a data-driven approach to estimating regression models - gavinsimpson/physalia-gam-course
github.com
1 1
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · 9d
I’d be happy to add/use a psych example or two if you have suggestions for papers or analyses where the data are open?
1 2
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · 9d
As I’m a geographer by PhD working in ecology, environmental science, and now animal science, the examples tend towards the natural and life sciences, but they are quite varied so attendees from a wide array of backgrounds usually find them relatable.
1 3
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · 11d
I’ll be running a 3-day one at AU Viborg (about an hour from Aarhus, Denmark) June 9th through 11th that is in person. Dates are TBC but registration should be open in the next couple of weeks

I’m also running an online one with @physaliacourses.bsky.social in December this year.
1 1 3
Reposted by Gavin Simpson
vincentab.bsky.social
Vincent Arel-Bundock @vincentab.bsky.social · 16d
The Pink Book of #MarginalEffects (aka Model to Meaning) ships next week and I've got a backlog of Zoolander memes.

Hope you're hungry for some spam in your timeline.

#RStats #PyData
1 18 90
Reposted by Gavin Simpson
vincentab.bsky.social
Vincent Arel-Bundock @vincentab.bsky.social · 25d
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
Reposted by Gavin Simpson
teunbrand.bsky.social
Teun van den Brand @teunbrand.bsky.social · 27d
We're glad to finally bring you this update!
thomasp85.com
Thomas Lin Pedersen @thomasp85.com · 27d
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
1 3 28
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · 27d
Congratulations Thomas; I know this release hasn’t been an easy one
5
Reposted by Gavin Simpson
thomasp85.com
Thomas Lin Pedersen @thomasp85.com · 27d
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
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · 28d
😱
1
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · Sep 6
It is also usually an example inappropriate calculation-first-estimation-second when we should be doing things the other way round; estimate model then calculate differences/ratios from the model, not the data
1
Reposted by Gavin Simpson
physaliacourses.bsky.social
Physalia-courses@Online @physaliacourses.bsky.social · Sep 2
If you’d like to go beyond updates and truly master GAMs in R, join our online course with @gsimpson.bsky.social, 8–11 Dec.

📊 Learn how to fit, interpret, visualise, and communicate GAMs in practice, guided by one of the leading experts in the field.
www.physalia-courses.org/courses-work...
1 2
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · Aug 25
🐛 A couple of bugs squashed along the way.

📄 Full changelog: gavinsimpson.github.io/gratia/news/...
Changelog
gavinsimpson.github.io
2
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · Aug 25
🛠️ partial_derivatives() improvements:

• handles multiple smooths better
• ignores random effect & univariate smooths
• clearer error messages
1 2
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · Aug 25
📈 quantile_residuals() supports more mgcv 📦 families:

• scat()
• nb()
• betar()
• tw()
1 1
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · Aug 25
✨ Plots: residuals_hist_plot() (and appraise()) now align histogram bins more sensibly, centering at 0.
1 1
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · Aug 25
🚀 gratia 0.11.1 is out!
Compatibility with ggplot2 4.0.0 + usability improvements + bug fixes.

#Rstats #mgcv #gam #statistics
1 3 28
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · Aug 19
You could try the gamlss2 package next time; it’s a lot quicker than gamlss (it’s a redevelopment of the original by same authors plus others). Only on GitHub at the moment
1 1 1
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · Aug 19
X_s being the model matrix for the specific smooth (i.e., the basis functions evaluated at the covariate values) and b_s being the estimates of the basis coefficients for that smooth
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · Aug 19
No that I'm aware; adding families to mgcv is tricky if you want all the standard automatic smoothness selection methods as those require many fourth-order derivatives. I think you could get close by adding observation-level random effects with `s(id, bs="re")` with standard binomial()
1
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · Aug 19
Thanks VIncent
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · Aug 19
On the latter point, I was re-reading Hastie & Tibshirani's GAM book recently where they restrict plotting the surface to the convex hull of the data, which I could implement as an alternative.
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · Aug 19
I evaluate (X_s %*% t(b_s)) the smooth at a grid of points evenly spaced in each marginal covariate, then set to NA any coordinate that lies further than `dist` from a data point when everything is mapped to a unit square to avoid extrapolation too far beyond support of data, like mgcv::plot.gam().
2 1
gsimpson.bsky.social
Gavin Simpson @gsimpson.bsky.social · Aug 18
I also added code to generate probability integral transform (PIT) & quantile residuals for GAMs. Right now I only handle binomial(), poisson(), gaussian(), & Gamma(), but other families can be added simply by writing a function for the CDF (& possible translating from mgcv's parameterisation)
1 3