I recently gave a 15′ Talk at CuttingEEGX, and a 50′ talk at the Gießen SFB Perception Colloquium. The topics were “The modern regression toolkit”, with the later talk having a focus on eye-movement EEG combined data. Link to pptx for the 15′ slides (25mb) Link to pptx for the 50′ slides (49mb)
I often had the problem in the past, how to make complex plots with typical latex-elements like Matrices, Formulas etc. They frequently break when in Illustrator (plugins exist, but still… buggy for me), PowerPoint is not really re-usable etc. For a talk at https://cuttingeegx.org/ I tried to push what Julia / https://makie.org/ can do for me on this issue. I wanted to have a quite complex plot, that I’d typically do in Illustrator, completely in Julia – and succeeded 🎉 What follows is a short tutorial on how I plotted the matrix. Note that in the following I am using…
I got awarded an Emmy Noether research group on “EEG in motion” by the DFG 🎉! I’m therefore hiring 2 x 100% TVL13 Positions for PhD or PostDoc In this Emmy Noether funded project, we will investigate conceptual, methodological, and physiological foundations of EEG combined with eye-, self- and object-motion. One position will focus on the methodological and physiological problems when combining smooth pursuit eye-movements and EEG. The second position will focus on methodological and computational problems when combining object motion (e.g. video watching) with EEG. Both projects are closely related to the core of the lab. Further details can…
I recently read this paper, on the measurement of onsets of clusters in EEG data from Guillaume Rousselet and later got asked to be a reviewer of the paper. The main point it raises is a different one to what I adress in the paper: Most people do not explicitly test for an onset, they fall victim to the interaction-fallacy. In principle, you need an explicit test, testing e.g. timepoint 100 vs. 150 to check whether the activity changed significantly. But because I recently implemented the ClusterDepth algorithm in Julia, I thought it would be nice to add this to…
I recently gave a workshop @ cuttingGarden / cuttingEEG Frankfurt on Linear Mixed Models and EEG. You can find all workshop materials here. They are pluto-notebooks for slides and exercise, and a prerendered HTML. A recording is in the works. Some highlights: MixedModels.jl is 100x faster than lme4 Intuition behind MixedModels / Item Effects Mass-univariate random effects over time Likelihoodratio-test against time
After an innocent question of a student on how to measure occular dominance, I was let down a rabbit hole. In a more recent paper I found it referenced as the “Dolman Method” Li et al 2010. A good starting point! Indeed, it pointed to Durand and Gould (1910) which developed an aparatus to measure occular dominance. But this is clearly not a Hole-in-card test, also neither of those are called “Dolman”. Let’s dig deeper! Google didn’t really help, gpt4 offered me fake citations and blamed me that I can’t find them — but google-scholar offered me Miles 1929. It…
Like Interactive Explorations?Try out the interactive Type-1 errors LMMs Demo here In repeated measures designs, we commonly repeat trials within a subject. This leaves us with a problem, though: trials from within one subject are typically more similar compared to trials across subjects. This requires us to use repeated-measures ANOVAs, Hierarchical, Multi-Level, or as in the case of this blog: Linear Mixed Models. I commonly see analyses for within-subject designs with LMMs, that use formulas like: y~1+condition+(1|subject) or y~1+condition+(1|subject) + (1|item) Type-1 Error of omitting condition random slopes As can be seen in this graph, the type-1 error of omitting…
Dear EEG/MEG practitioners, beginners or experts We invite you to participate in our ~15-30 min ERP visualization tool survey. Our results will be freely available, thereby, we hope to improve the M/EEG visualization ecosystem. http://eeg-survey.s-ccs.de/ As a thank you, you can win one of three Muse EEG headsets! Please share this survey in your labs: we are looking for diverse input from novices and experts from all domains using EEG. This is collaborative work under the lead of Vladimir Mikheev, together with René Skukies and myself.
I am currently setting up a lecture on multiple comparison correction (for related posts see here or here). In a nutshell: If you apply a statistical test, that allows for 5% of false-positives (a ‘wrong’ significant finding), many many times, you are more or less guaranteed to find a significant effect (because p(at_least_one_positive) = $1 – (1-0.05)^{100} = 0.994 = 99.4$%) False-Discovery-Rate (FDR) is one way, to try to adapt to this. In this post I will give you a visual intuition behind it, not walk you through the math. Note that FDR has a different goal to e.g. Bonferroni…
Note: This blog is just explaining the basis sets, not how to actually fit models / get parameters etc. Recently, we used the unfoldtoolbox (Matlab or Julia; access it from python!) to analyze some fixation-related ERPs. The approach we used (multiple regression with deconvolution) allowed us to include this circular-predictor: absolute saccade-angle. Why can’t we model saccade-angle using a linear predictor? The issue is straightforward: Look at this plot. Ok, why can’t we use a standard non-linear spline regression? Wait – what even is a standard non-linear spline regression? Great that you asked. Instead of fitting a straight line with…