Charles Driver
charlesdriver.bsky.social
Charles Driver
@charlesdriver.bsky.social
Asst Prof UZH Psychology - quant methods, dynamic systems, human development, psychology. @CharlesDriverAU
#4 Relations between depression and self esteem across 3 different time scales / datasets, with @peterhaehner.bsky.social et al. Besides the multiple time scales, extending VARish models to cross-effects that influence over different time scale is unusual I think.

psycnet.apa.org/record/2025-...
September 25, 2025 at 9:48 AM
#3 Examining links between and within social and cognitive development, and how these differ with behaviour problems. Some big moderated growth + fluctuations models. With @mkdeserno.bsky.social et al.

osf.io/preprints/ps...
September 25, 2025 at 9:36 AM
#2 "Personality functioning as generalized correlated changes in personality traits" -- basically a large factor model over time with some common factor nuance. With @chopwood.bsky.social et al.

osf.io/preprints/ps...
September 25, 2025 at 9:36 AM
Prepping for @dgps.bsky.social Methods, few recent papers pushing ctsem and dynamic models in interesting directions

#1 Developmental changes in twin cognitive correlation (ACE) across age, test instrument, zygosity, with @evangiangrande.bsky.social @ent3c.bsky.social et al

osf.io/preprints/ps...
September 25, 2025 at 9:36 AM
Ok. I can go home this evening.
August 21, 2025 at 9:07 AM
dear @quarto.org using friends, lost too much time on this already and I'm very stubborn please save me:

---
title: "wrap text"
format: pdf
---

```{r}
print("I just want this text to wrap to a new line I just want this text to wrap to a new line I just want this text to wrap to a new line ")
```
August 21, 2025 at 7:23 AM
June 26, 2025 at 11:12 AM
ctsem is updated on CRAN to 3.10.3 (in time for SMiP Summer School next week). Useful updates include a model fit check of moderated lagged correlations, and better importance sampling for the optimize then sample Bayes variant.
June 26, 2025 at 8:47 AM
Postdoc position open in Zurich -- Prof. Martin Tomasik and I have a joint SNF project on interpretable neural network approaches for large scale, complex item / temporal structure, online learning / cognitive development data.

Please retweet.

tinyurl.com/PostdocGNNSNF
May 28, 2025 at 11:16 AM
I think you can, what do yo think I miss with the following? top plots show age/period/outcome relationships when outcome is linear + interaction (shading is given by 3rd, non axis, variable), bottom is a more complex function. Looks distinguishable to me.
May 28, 2025 at 8:37 AM
Visualizing uncertainty in uncertainty (distributions of distributions?) gets weird sometimes.
December 1, 2024 at 9:43 PM
I first thought you could just free covariances between indicators and predicted non-indicator to maintain measurement. But then it's apparently unidentified. Removing the variance of the predicted variable 'works', re measurement, but I assume std errors of prediction are broken. Interesting :)
September 25, 2024 at 11:04 AM
Teaching rbind... #rstats
August 29, 2024 at 2:12 PM
Curious, does *anyone* understand this wonderful double barreled question from Microsoft? How do I a) stay signed in and b) keep my computer to myself? Do I read the questions top to bottom, with answers following questions, or is there some kind of spatial hierarchy?
June 13, 2024 at 6:53 AM
Happily, I just accepted a TT asst prof position at UZH Psychology. Excited to have more scope to support (longitudinal) modelling, also to stay in Zurich :) Some positions available soon, curious to hear from anyone passionate about modelling -- fit, ML, AI, theory, applications, software, etc!
June 5, 2024 at 9:22 AM
But some approaches are thankfully more tractable :)
May 8, 2024 at 1:02 PM
Likelihood surfaces of approaches to building positive definite covariance matrices are ugly monsters.
May 8, 2024 at 10:54 AM
Trying to sample n probabilities that sum to 1 using n-1 parameters, such that distributions of the n probabilities are equivalent. Any easy known solutions?
April 18, 2024 at 9:10 AM
I generate data for y (2 categories) & x (n categories) where the true model is no relationship, chi^2 test for independence, and repeat lots, p-values are uniform. Same if I subsample the data without replacement. Problematic if I bootstrap (with replacement). Struggling to get the intuition why!?
February 1, 2024 at 3:56 PM
(Late reply!) Haha yes you did ;) In the paper a case where one variable has twice as much error as genuine variance, and the other has half as much, is shown. Looks like this (ie very similar) when both have half as much error as true variance. will update paper with details ;)
February 1, 2024 at 9:40 AM
more frequent observations lead to 'seemingly' faster processes when measurement error isn't accounted for in such models.
osf.io/preprints/os...
December 16, 2023 at 11:08 AM
Discussing emotion co-regulation (dyad coupling) in class tomorrow. Little shiny app in case anyone wants to visualise coupled damped linear oscillators with a disturbance event: charlesdriver.shinyapps.io/shiny/
December 7, 2023 at 3:46 PM
December 7, 2023 at 3:08 PM
Thanks both, yes I overlooked that the measurement residual e was part of the new latent, so it's more flexible than the basic logistic regression approach, just computationally uglier :) given a certain level of measurement error the expectations / error variances line up for my arbitrary test.
November 17, 2023 at 9:34 PM
Ok, so you think this comparison is wrong? expectation of logistic regression style ordinal (red) versus what I understand of SEM style ordinal (black), conditional on latent state and some arbitrary set of thresholds.
November 17, 2023 at 4:33 PM