Daniel 🕹️
LightNews LightNews
banner
strengejacke.de
Daniel 🕹️
@strengejacke.de
580 followers 110 following 410 posts
He/she/it - 's' muss mit.
We're lower than the world!

R easystats project:
https://easystats.github.io/easystats/
Posts Replies Media Videos
strengejacke.de
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
October 16, 2025 at 9:15 AM
strengejacke.de
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
September 8, 2025 at 5:30 AM
strengejacke.de
R is not Nintendo. #rstats
September 5, 2025 at 9:10 PM
strengejacke.de
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)
September 3, 2025 at 8:42 AM
strengejacke.de
from: Douglas M. Bateslme4: Mixed-effects modeling
with R
September 3, 2025 at 8:33 AM
strengejacke.de
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📦
September 1, 2025 at 6:21 AM
strengejacke.de
Something like:
August 23, 2025 at 11:30 AM
strengejacke.de
paramters::model_parameters() has the `ci_method` argument with similar options . however, broom.mixed returns NA for Wald-CIs?
August 6, 2025 at 9:17 PM
strengejacke.de
(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()
August 6, 2025 at 8:32 PM
strengejacke.de
- 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? 😎
August 2, 2025 at 7:06 AM
strengejacke.de
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.
July 17, 2025 at 1:15 PM
strengejacke.de
See the slides 21-23 from "Introduction into predictions and the {modelbased} package", where I tried to write up my understanding/definition: easystats.github.io/easystats/ar...
July 17, 2025 at 12:53 PM
strengejacke.de
That's ("answer is indeed ME") probably a too fast conclusion (depending on what you mean by "marginal effects" - if a single number, that's not always a good idea. In this example, we used cubic age, and the interpretation makes sense... 1/2
July 8, 2025 at 12:32 PM
strengejacke.de
Ok, works for simple models, must check for more complex like mixed effects / zero-inflated models.
m <- lm(mpg ~ wt + hp, data = mtcars) ci <- 0.95 dof <- insight::get_df(m) set.seed(123) crit <- mvtnorm::qmvt(ci, df = dof, tail = "both.tails", corr = cov2cor(vcov(m)))$quantile # 95% level% confint(m) #> 2.5 % 97.5 % #> (Intercept) 33.95738245 40.49715778 #> wt -5.17191604 -2.58374544 #> hp -0.05024078 -0.01330512 parameters::model_parameters(m) #> Parameter | Coefficient | SE | 95% CI | t(29) | p #> ---------------------------------------------------------------------- #> (Intercept) | 37.23 | 1.60 | [33.96, 40.50] | 23.28 | < .001 #> wt | -3.88 | 0.63 | [-5.17, -2.58] | -6.13 | < .001 #> hp | -0.03 | 9.03e-03 | [-0.05, -0.01] | -3.52 | 0.001 #> #> Uncertainty intervals (equal-tailed) and p-values (two-tailed) computed #> using a Wald t-distribution approximation. # sup-t adjustment confint(m, level = 1 - 2 * pt(-abs(crit), df = dof)) #> 1.04 % 98.96 % #> (Intercept) 33.31930882 41.135231417 #> wt -5.42443900 -2.331222484 #> hp -0.05384452 -0.009701374 set.seed(123) parameters::model_parameters(m, p_adjust = "sup-t") #> Parameter | Coefficient | SE | 95% CI | t(29) | p #> ---------------------------------------------------------------------- #> (Intercept) | 37.23 | 1.60 | [33.32, 41.14] | 23.28 | < .001 #> wt | -3.88 | 0.63 | [-5.42, -2.33] | -6.13 | < .001 #> hp | -0.03 | 9.03e-03 | [-0.05, -0.01] | -3.52 | 0.003 #> #> p-value adjustment method: Simultaneous confidence bands #> #> Uncertainty intervals (equal-tailed) and p-values (two-tailed) computed #> using a Wald t-distribution approximation.
July 6, 2025 at 9:09 PM
strengejacke.de
Would it be something like this?
m <- lm(mpg ~ wt + hp, data = mtcars) x <- 0.95 l <- mvtnorm::qmvnorm(x, tail = "both.tails", corr = cov2cor(vcov(m)))$quantile 1 - 2 * pnorm(-abs(l)) #> [1] 0.9797365 # 95% level% confint(m) #> 2.5 % 97.5 % #> (Intercept) 33.95738245 40.49715778 #> wt -5.17191604 -2.58374544 #> hp -0.05024078 -0.01330512 # new "95%" level confint(m, level = 1 - 2 * pnorm(-abs(l))) #> 1.01 % 98.99 % #> (Intercept) 33.30014759 41.154392645 #> wt -5.43202222 -2.323639269 #> hp -0.05395274 -0.009593154
July 5, 2025 at 10:56 AM
strengejacke.de
Time for a new wallpaper... #easystats #insight
July 1, 2025 at 1:09 PM
strengejacke.de
I mean, it even appears if you just select a single char.
July 1, 2025 at 7:44 AM
strengejacke.de
July 1, 2025 at 7:40 AM
strengejacke.de
Here's a way how I handled it in sjPlot, *if* you really have neutral categories. Such thing would be nice to have in a future update, because you don't have much tools for plotting "Likert" scales (especially not in Excel/Office diagrams, but who makes figures with office anyway?)
June 4, 2025 at 10:03 AM
strengejacke.de
I often have "neutral" categories, like "don't know" or similar, and these are neither positive nor negative, so it would be great to account for their proportion, but placing it in the middle would add half of their counts to both sides, which can be misleading when interpreting "totals".
library(tinyplot) tinytheme( "clean2", palette.qualitative = c("black", "sienna", "grey", "indianred", "goldenrod") ) hec <- as.data.frame(proportions(HairEyeColor, 2:3)) # Add a new factor level "Dont know" to the variable "Hair" hec$Hair <- factor(hec$Hair, levels = c(levels(hec$Hair)[1:2], "Dont know", levels(hec$Hair)[3:4])) hec$Hair[5] <- "Dont know" tinyplot( Freq ~ Eye | Hair, facet = ~Sex, data = hec, type = "barplot", center = TRUE, flip = TRUE, facet.args = list(ncol = 1), yaxl = "percent" ) library(tinyplot) tinytheme( "clean2", palette.qualitative = c("black", "sienna", "grey", "indianred", "goldenrod") ) hec <- as.data.frame(proportions(HairEyeColor, 2:3)) # Add a new factor level "Dont know" to the variable "Hair" hec$Hair <- factor(hec$Hair, levels = c(levels(hec$Hair)[1:2], "Dont know", levels(hec$Hair)[3:4])) hec$Hair[5] <- "Dont know" tinyplot( Freq ~ Eye | Hair, facet = ~Sex, data = hec, type = "barplot", center = TRUE, flip = TRUE, facet.args = list(ncol = 1), yaxl = "percent" )
June 4, 2025 at 10:03 AM
strengejacke.de
😎
June 4, 2025 at 6:38 AM
strengejacke.de
I often show students this figure and ask, how different is the green distribution (p < 0.05) from the blue distribution (p = 0.10)? Just to raise some awareness that the difference between "statistical significant" and "not significant" is not always that significant...
June 3, 2025 at 6:57 AM
strengejacke.de
You may also find `convert_to_na()` and `convert_na_to()` helpful (easystats.github.io/datawizard/r...), or maybe also `replace_nan_inf()` from the #rstats {datawizard} package.
May 21, 2025 at 1:28 PM
strengejacke.de
And although the scales are different for predictions and coefficients, the statistical test for the difference between levels of a factor (= p-values) are literally identical, meaning that if you work with predicted probabilities anyway, it almost doesn't matter which model you take.
data(heart, package = "glm2") start.p <- sum(heart$Deaths) / sum(heart$Patients) fit.glm <- glm( cbind(Deaths, Patients - Deaths) ~ factor(AgeGroup) + factor(Severity) + factor(Delay) + factor(Region), family = binomial(), data = heart ) fit.logbin <- logbin::logbin( formula(fit.glm), data = heart, start = c(log(start.p), rep(c(0.2, 0.4), 4)), trace = 1 ) modelbased::estimate_contrasts(fit.logbin, "Delay") #> Marginal Contrasts Analysis #> #> Level1 | Level2 | Difference | SE | 95% CI | t(65) | p #> ------------------------------------------------------------------- #> 2 | 1 | 0.01 | 0.01 | [-0.01, 0.04] | 0.86 | 0.394 #> 3 | 1 | 0.03 | 0.02 | [ 0.00, 0.07] | 2.13 | 0.037 #> 3 | 2 | 0.02 | 0.02 | [-0.01, 0.05] | 1.56 | 0.124 #> #> Contrasts are on the response-scale (in %-points). parameters::model_parameters(fit.logbin, exponentiate = TRUE, keep = "Delay") #> Parameter | Risk Ratio | SE | 95% CI | Statistic | df | p #> --------------------------------------------------------------------- #> Delay [2] | 1.06 | 0.07 | [0.93, 1.22] | 0.85 | 65 | 0.395 #> Delay [3] | 1.19 | 0.10 | [1.01, 1.39] | 2.13 | 65 | 0.034 #> #> Uncertainty intervals (profile-likelihood) and p-values (two-tailed) #> computed using a Wald distribution approximation.
May 20, 2025 at 12:17 PM
strengejacke.de
Nice video, as always! One conclusion would be to use marginal means / adjusted predictions, because this will give consistent results for both models.
data(heart, package = "glm2") start.p <- sum(heart$Deaths) / sum(heart$Patients) fit.glm <- glm( cbind(Deaths, Patients - Deaths) ~ factor(AgeGroup) + factor(Severity) + factor(Delay) + factor(Region), family = binomial(), data = heart ) fit.logbin <- logbin::logbin( formula(fit.glm), data = heart, start = c(log(start.p), rep(c(0.2, 0.4), 4)), trace = 1 ) #> logbin parameterisation 1/81 #> Deviance = 149.321 Iterations - 9052 modelbased::estimate_means(fit.glm, "AgeGroup") |> print(select = "minimal") modelbased::estimate_means(fit.logbin, "AgeGroup") |> print(select = "minimal")
May 20, 2025 at 12:17 PM