Björn Siepe

Congrats, Omid! 🎉

🚨 New preprint: We compared 13 methods for detecting momentary careless responding in the WARN-D data (206k+ obs.). Tutorials guide you through each method. The takeaway? Diverging results, inherent subjectivity (to varying degrees), and a clear need for further validation.
osf.io/preprints/ps...

Two new preprints on multilevel HMMs! Time series data is now pervasive in psychology and new methods are needed to model the dynamics in such data. Hidden Markov Models (HHMs) are powerful models for dynamics in which a system is switching between a number of discrete states.

Adopt Registered Reports at Psychological Methods Can you spare a minute to help this campaign?

Help us show there is support for offering registered reports at one of Psychology's leading method's journal! chng.it/TwwnVBScVb

Fair coins tend to land on the same side they started: evidence from 350,757 flips.

That's the title of our paper summarizing ~650 hours of coin-tossing experimentation just published in the Journal of the American Statistical Association.
doi.org/10.1080/0162...

Smartwatches offer little insight into stress levels, researchers find Academic study suggests devices cannot differentiate between someone being overworked and being excited

New paper by @bsiepe.bsky.social & WARN-D team out now in JoPACS, on the (lack of) relation between EMA selfreport and Garmin passive data.

Guardian @theguardian.com coverage:
www.theguardian.com/technology/2...

Paper:
psycnet.apa.org/doiLanding?d...

🧪 #PsychSciSky #StatsSky #EpiSky

How Accurate & Consistent are Self-Report responses via Visual Analogue Scales (VAS) in Ecological Momentary Assessment & Digital Studies?

Examined in a massive EMA study (N = 3,761), our new #OpenAccess piece on this is just out:

online.ucpress.edu/collabra/art...

#PsychSciSky #AcademicSky

🧵:

Sure! Thanks for that.

It was logistic regression in scikit learn (scikit-learn.org/stable/modul...). Apparently, you could not fully turn off regularisation there for some time.

I'm excited that our preregistration template for simulation studies is now available on OSF!
See below for a brief interview with the Center for Open Science about the template and why we created it..

Does your smartwatch say you’re stressed? It may often be wrong Consumer grade smartwatches may not be as accurate as promised when measuring tiredness or stress. That conclusion is drawn by researchers Björn Siepe and Eiko Fried based on a comparison between smar...

1/2
Press release @unileiden.bsky.social about our upcoming JoPACS paper led by @bsiepe.bsky.social, showing that the overlap of similar constructs assessed via wearable vs EMA data is surprisingly low (n~850, 3 months).

www.universiteitleiden.nl/en/news/2025...

Congratulations!

I'm very happy to see meta-science on methods research getting more attention in the form of a great new call for papers!

I recently gave some talks on the existing literature on the quality and transparency of sim studies & ideas for future improvement. Slides: bsiepe.github.io/talks/2025-0...

intervalpsych: Analyzing Interval Data in Psychometrics Implements the Interval Consensus Model (ICM) for analyzing continuous bounded interval-valued responses in psychometrics using 'Stan' for 'Bayesian' estimation. Provides functions for transforming in...

Today, my first R-package was released on CRAN. 🎉
You can use it to analyze interval responses collected via dual range slider scales.
Thank you to my colleague @bsiepe.bsky.social for being such an enthusiastic co-author.

cran.r-project.org/web/packages...

Self-control in daily life is more than willpower — but how well do commonly used trait scales capture this recent, broader view on self-control?

New preprint with @kaihorstmann.bsky.social & @mhennecke.bsky.social: osf.io/preprints/ps...
Short summary below.

emailed you :)

🚀Postdoc position @unimarburg.bsky.social in the project:

"Bridging the Gap Between Verbal Psychological Theories & Formal Statistical Modeling with Large Language Models"
(funded by @volkswagenstiftung.de)

📅Start: 01.10.2025 | ⏳4 years
🔗 Apply now: uni-marburg.de/jhbCen
🔄 Thanks for sharing!

Thank you :)

🚀 PhD Position in Psychological Methods at @unimarburg.bsky.social
📅 Start: 01.10.2025 | 💼 50% TV-H E13 | ⏳ 3 years

Focus on statistical modeling—Bayesian statistics, cognitive modeling, psychometrics.

🔗 Apply now: uni-marburg.de/r5dKTr

🔄 Thanks for sharing!

Here's a recent methods paper on casual effects if (micro-)randomized interventions in EMA: psycnet.apa.org/doi/10.1037/...

Thanks to the great Marburg-Groningen team: @matzekloft.bsky.social, @yongzhangzzz.bsky.social, @fridtjofptrsn.bsky.social, @bringmannlaura.bsky.social, @danielheck.bsky.social

💻The sim was preregistered, all code + Docker container to reproduce main results available on OSF

Main takeaways:
▶️Using features of dynamic networks may often come with considerable statistical hurdles
▶️Proper uncertainty quantification is important
▶️We hope that our new implementation of a Bayesian one-step mlVAR model helps

Limitations:
▶️Our simulation was computationally intensive, and we used one main data-generating process, limiting generalizability. However, we think that our data-generating process is realistic and provides somewhat ideal conditions
▶️Estimating BmlVAR can be time-consuming

A plot illustrating the posterior distribution for outstrength centrality of 6 items for two participants in our empirical example. The plot indicates that for one participant (65), "depressed" has the highest point estimate of outstrength and a very high posterior probability for being the most central node, whereas this is very uncertain for another participant (92) with the same most central node.

We highlight some advantages of a Bayesian multilevel approach:
▶️We can estimate the probability of a node being the most central node
▶️For some individuals, there may be a clearly "most central" node, whereas this is very unclear for others
▶️This information is crucial for practical applications

A plot indicating the proportion of correctly identified most central nodes (for outstrength centrality), split in four plots in a grid for the temporal and contemporaneous network with different sample sizes and number of time points for four different methods (GVAR, GIMME, mlVAR, BmlVAR).

Simulation study w/ different estimation methods shows:
▶️Treatment selection: Identifying the most central node is often difficult
▶️Outcome prediction: Using centrality for outcome prediction requires large n & t & large effects
▶️Using a one-step model such as BmlVAR can help under some conditions

The model equation for the time series part of BmlVAR (equation 1 in the paper).

▶️ We introduce a Bayesian approach (BmlVAR) that estimates a multilevel VAR model and between-person network feature regression in one step
▶️ This accounts for uncertainty that is usually ignored when first estimating individual networks and then using centrality point estimates in a second step

Previous research has suggested using:
▶️The most central node as an aim for treatment
▶️Network centrality/density as predictors of an outcome, such as psychopathology

However:
▶️We know little about the accuracy of centrality estimation
▶️Uncertainty of centrality estimates is commonly ignored

Image with three main boxes with recommendations for applied researchers on conceptualization, estimation, and interpretation when using network features. Conceptualization: - Discuss why a network feature should be relevant for a given treatment or outcome - Consider plausible effect size and sample size needed to detect an association with a distal outcome Estimation: - Carefully choose preprocessing and network estimation methods - Follow good practices for predictive models (e.g., cross-validation, out-of-sample validation) Interpretation: - Consider uncertainty in node selection and network feature regression - Compare network features with simpler time series features (e.g., person-specific mean or SD)

Can we use features of dynamic networks (e.g. centrality) to improve treatment selection and outcome prediction?

New preprint on the topic: We highlight the role of uncertainty & introduce a Bayesian multilevel approach for uncertainty quantification of network features 🧵
osf.io/preprints/ps...

Don’t Close the Door on Creatine Yet A recent meta-analysis suggested that creatine only has a “trivial-to-small” effect on muscle growth. However, that language might understate the actual impact of creatine.

I'm just an interested amateur on the subject, but this ~2 year old post on the relevance of small effects of creatine by one of my favourite science communicators seems relevant here: www.strongerbyscience.com/creatine-eff...

Can't wait for June 5th!

Spending an entire day talking about meta-science with fellow young researchers is one of my favorite things to do

Join us! The more the merrier 🚀