Daniel 🕹️

@strengejacke

He/she/it - 's' muss mit. We're lower than the world! R easystats project:

111 Following624 Followers503 Posts

YouTube Bluesky

Is this your profile?Claim it and strengejacke.dirty.social is yours.Claim profile

Daniel 🕹️@strengejacke· 14d

Still, "evidence" does not indicate "clinical importance". You should take these two into account (not a new insight, I know...).

Daniel 🕹️@strengejacke· 14d

Just for the record, this is a good example for the `post_process` argument in {modelbased}...

library(modelbased)
data(mtcars)
m <- glm(am ~ vs, family = binomial, data = mtcars)

estimate_means(
m,
"vs",
transform = effectsize::probs_to_odds,
predict = "response",
post_process = difference ~ pairwise
)
#> Post-processing difference ~ pairwise...
#> Estimated Marginal Means
#>
#> Parameter | Probability | SE | 95% CI | z
#> ------------------------------------------------------
#> (b2) - (b1) | 0.50 | 0.17 | [0.16, 0.84] | 2.88
#>
#> Variable predicted: am
#> Predictors modulated: vs
#> Predictions are on the response-scale.

Daniel 🕹️@strengejacke· Jun 27

Only payable with this plectrum

Daniel 🕹️@strengejacke· Jun 27

Daniel 🕹️@strengejacke· Jun 24

A new paper by @dominiquemakowski.bsky.social, @mattansb.msbstats.info, me and colleagues just out! We show how to choose informative priors in Bayesian regression models using a systematic simulation study and a practical step-by-step tutorial in #Rstats and #Stan! doi.org/10.3389/fpsy... >

Daniel 🕹️@strengejacke· Jun 15

Not a book, but a monthly magazine. Listings were available in Basic and ASM language.

Daniel 🕹️@strengejacke· May 20

Still not get it working 🙈 Can you rewrite this code with a different hypothesis argument and make it return the same results? (code in alt text)

set.seed(123)
n <- 200
d <- data.frame(
outcome = rnorm(n),
grp = as.factor(sample(c("treatment", "control"), n, TRUE)),
episode = as.factor(sample(1:3, n, TRUE)),
sex = as.factor(sample(c("female", "male"), n, TRUE, prob = c(0.4, 0.6)))
)
model2 <- lm(outcome ~ grp * episode, data = d)

avg_predictions(
model2,
by = c("episode", "grp"),
hypothesis = "(b1 - b3) = (b2 - b4)"
)

Daniel 🕹️@strengejacke· May 20

Code in Alt-text... You always have to look at the table of predictions, to find out the b-terms/rows. Since the order of coefficients is different between modelbased and marginaleffects, you see different values for the hypothesis-argument here.

library(modelbased)
library(marginaleffects)

set.seed(123)
n <- 200
d <- data.frame(
outcome = rnorm(n),
grp = as.factor(sample(c("treatment", "control"), n, TRUE)),
episode = as.factor(sample(1:3, n, TRUE)),
sex = as.factor(sample(c("female", "male"), n, TRUE, prob = c(0.4, 0.6)))
)
model2 <- lm(outcome ~ grp * episode, data = d)

estimate_contrasts(model2, c("episode", "grp"), comparison = "(b1 - b2) = (b4 - b5)", estimate = "average")

avg_predictions(
model2,
by = c("episode", "grp"),
hypothesis = "(b1 - b3) = (b2 - b4)"
)

Daniel 🕹️@strengejacke· May 17

Daniel 🕹️@strengejacke· May 5

And note the changes in syntax highlighting: (package names have a different color now... Sometimes it’s the little things...)

Daniel 🕹️@strengejacke· Apr 22

Here's a short vignette on within-between effects: easystats.github.io/parameters/a... From there, you can calculate the context effect by contrasting the (average) slopes of those two effects, see attached screenshots (code in ALT text).

library(parameters)
library(modelbased)
library(marginaleffects)
library(lme4)

data("qol_cancer")
qol_cancer <- datawizard::demean(qol_cancer, select = c("phq4", "QoL"), by = "ID")
mixed_1 <- lmer(
QoL ~ time + phq4_within + phq4_between + (1 | ID),
data = qol_cancer
)
# within- and between effects
model_parameters(mixed_1)

# context effect - difference between within- and between-effect
estimate_contrasts(mixed_1, c("phq4_within", "phq4_between"), comparison = "slope")

# context effect - using marginal effects
avg_comparisons(
mixed_1,
variables = c("phq4_within", "phq4_between"),
hypothesis = ~pairwise
)

Daniel 🕹️@strengejacke· Mar 3

I just realized the small post-it in the left corner... 🤣 maybe better to see in this hires-image

Daniel 🕹️@strengejacke· Mar 3

I'm pasting one of these memes into every manuscript where I'm asked to write the methods section after I did the analyses.

Daniel 🕹️@strengejacke· Feb 21

Since I had some experiences with tibbles (the colleague did not and thought function doesn't work), it meanwhile doesn't take that long to figure it out. Note the `as.data.frame()`.

Daniel 🕹️@strengejacke· Feb 21

One of the many, many issues I faced with tibbles in my 30 years of programming and 10 years of R experience (not joking, it's almost the same dates). What is wrong here?

Daniel 🕹️@strengejacke· Feb 20

Daniel 🕹️@strengejacke· Feb 19

@vincentab.bsky.social, the guy whose days have more than 24 hours. (ok, altdoc is mainly @etiennebacher.bsky.social, with support from Vincent)

Daniel 🕹️@strengejacke· Feb 3

Hm, doesn't work for me. I'm using the latest daily build: Positron Version: 2026.03.0 (system setup) build 14 Code - OSS Version: 1.106.0 Date: 2026-02-03T08:15:39.951Z Electron: 37.7.0 Chromium: 138.0.7204.251 Node.js: 22.20.0 V8: 13.8.258.32-electron.0 OS: Windows_NT x64 10.0.26200

Daniel 🕹️@strengejacke· Feb 2

You should switch to Positron! (and here's how to do it there)

Daniel 🕹️@strengejacke· Oct 16

These are my #Positron extensions I've currently installed. Are there any "must haves" missing in my list, what other extensions would you recommend etc.? (answers including "python" will be ignored... 😎) #rstats #rstudio #vscode

Daniel 🕹️@strengejacke· Sep 8

This weekend I dug out my very old #starwars #ccg cards (decipher) because my son really wanted to play the game. Now I'm thinking about checking out Star Wars Unlimited, because it's no longer easy to get cards for the CCG. Anyone experience with Star Wars Unlimited? Would you recommend it? #tcg

Daniel 🕹️@strengejacke· Sep 5

R is not Nintendo. #rstats

Daniel 🕹️@strengejacke· Sep 3

If this is not of particular interest, i.e. you're not investigating "treatment methods", I'd suggest adding "hospital ID" as random effect. As said before, in the "worst case", we end up with the same accuracy as for simpler models (that's again Gelman/Hill 2007)

Daniel 🕹️@strengejacke· Sep 3

from: Douglas M. Bateslme4: Mixed-effects modeling with R

Daniel 🕹️@strengejacke· Sep 1

This is how table printing in #easystats look like - nice tables out-of-the-box thanks to #rstats packages like {gt} or {tinytable}, which is now fully supported across easystats📦

Daniel 🕹️@strengejacke· Aug 23

Something like:

Daniel 🕹️@strengejacke· Aug 6

paramters::model_parameters() has the `ci_method` argument with similar options . however, broom.mixed returns NA for Wald-CIs?

Daniel 🕹️@strengejacke· Aug 6

(code in ALT text)

# variance components
parameters::model_parameters(m, effects = "random")

# BLUPs
parameters::model_parameters(m, effects = "grouplevel")

# fixef+ranef
parameters::model_parameters(m, effects = "random_total")

# same as
modelbased::estimate_grouplevel(m)

# dot plot of BLUPs
parameters::model_parameters(m, effects = "grouplevel") |> plot()

# dot plot of BLUPs
modelbased::estimate_grouplevel(m) |> plot()

Daniel 🕹️@strengejacke· Aug 2

- modern look'n'feel - fully customizable layout - absolutely easy to handle GitHub integration - code assist / LLM integration, if desired - rather simple and intuitive UI - hide/show relevant panes with a keystroke What's not to like about it? 😎

Daniel 🕹️@strengejacke· Jul 17

After several years, I noticed that the first author of a co-authored article had corrected what he believed to be a ‘spelling mistake’ in the name of an R package.