Lightnews — Scholar-powered news

Bert van der Veen @vdveenb.bsky.social · 10d

Plaatje van het kieskompas uit 2021. Valt dus best mee volgens mij.

Bert van der Veen @vdveenb.bsky.social · 20d

Cake?

1 1

Bert van der Veen @vdveenb.bsky.social · 20d

Yesterday was my last day at NTNU. It marks the end of a long and prlducutive period (2017-2025) with @bobohara.bsky.social.

I'm happy to be able to share that I will now be moving to @unibayreuth.bsky.social for a new position with Steve Higgins 😁.

3 13

Reposted by Bert van der Veen

Danielle Navarro @djnavarro.net · Sep 8

At the risk of stretching the imagination beyond the limits of human endurance, the reader is asked to pretend that the author is interested in splines #rstats

Splines, B-splines, P-splines, and a disapproving kitten – Notes from a data witch

No, I do not care about splines. But I am trying to learn about GAMLSS regression, and yes, it is to this dark place that this topic has taken me

blog.djnavarro.net

7 21 68

Reposted by Bert van der Veen

Jarno Vanhatalo @jarnovanhatalo.bsky.social · Sep 3

We are seeking a postdoctoral researcher in statistics to develop methods for analyzing large scale biodiversity and ecosystem function data at the University of Helsinki.

This is a great opportunity for someone who wants to do advanced Bayesian modeling with direct link to real world applications.

Research Centre For Ecological Change @rececochange.bsky.social · Sep 3

📢 Post doc! Want to join our team?

We are looking for 3 post docs in biodiversity science, genetic indicator assessment, and ecological statistics for a fixed term of 2 years!

Come unravel the drivers and consequences of global change on biodiversity with us!

DL 24.9.
#universityofhelsinki

3 Post doc positions in biodiversity science, genetic indicator assessment and ecological statistics

jobs.helsinki.fi

13 8

Bert van der Veen @vdveenb.bsky.social · Sep 2

#rstats For a paper I am looking for go-to examples or demonstrations of complementary log-log models in ecology, preferably with publicly available data. Any recommendations?

1 2 3

Reposted by Bert van der Veen

Prof. Vigdis Vandvik @vvandvik.bsky.social · Aug 28

Come work with us!

Fully funded #PhD position in global change ecology - working in a cross-disciplinary cohort of 8 PhD students studying mountains in transition. 🧪🌐🏔️ @btwnthefjords-uib.bsky.social @cesam-uib.bsky.social

www.jobbnorge.no/en/available...

PhD Research Fellow in global change ecology (285493) | University of Bergen

Job title: PhD Research Fellow in global change ecology (285493), Employer: University of Bergen, Deadline: Wednesday, October 1, 2025

www.jobbnorge.no

1 38 34

Reposted by Bert van der Veen

Dr Nadiah Kristensen @nadiah.bsky.social · Aug 25

Check out the graphical abstract for a paper recently published in Ecological Modelling.

I have papers published in Ecological Modelling. Never again, I guess.

A diagram titled "Ecological Network Optimisation" with attractive landscapes and networks and very clear AI-generation artifacts.

16 47 130

Bert van der Veen @vdveenb.bsky.social · Aug 26

This position sounds great! Except the teaching in Finnish part, that considerably narrows down the potential pool of applicants.

Anna-Liisa Laine @annaliisalaine.bsky.social · Aug 26

We are recruiting‼️A permanent university lecturer position in BIOSTATISTICS is available at @helsinki.fi Faculty of Biological and Environmental Sciences. Great colleagues, excellent infrastructures and startup support 🙌🏻
DL 14 September

jobs.helsinki.fi/job/Helsinki...

University Lecturer in Biostatistics

jobs.helsinki.fi

3

Bert van der Veen @vdveenb.bsky.social · Aug 24

Thanks, I'll keep that in mind! Learning the language shouldn't be a problem, given that I am Dutch.

1

Reposted by Bert van der Veen

Gavin Simpson @gsimpson.bsky.social · Aug 18

🚀 gratia 0.11.0 is out!

Now has a paper in JOSS — please cite 📄 doi.org/10.21105/jos...

Experimental parallel processing ⚡

New assemble() for building plots 🎨

Better support for complex families + new diagnostics 🧪

Lots of bug fixes + polish ✨

👉 gavinsimpson.github.io/gratia/

#Rstats

An R package for working with generalized additive models

Graceful 'ggplot'-based graphics and utility functions for working with generalized additive models (GAMs) fitted using the 'mgcv' package.

gavinsimpson.github.io

3 64 190

Bert van der Veen @vdveenb.bsky.social · Aug 17

1 7

Bert van der Veen @vdveenb.bsky.social · Aug 12

The struct() defines the RHS correlation matrix while the LHS formula defines (as usual) a covariance in terms of lme4-style for the REs.

I'm not sure if I am breaking/abusing the usual random-effect syntax, but it seemed sensible to me (does #glmmTMB do something similar @bbolker.bsky.social?).

Bert van der Veen @vdveenb.bsky.social · Aug 12

Still plugging away at this. I've now expanded this idea of a kronecker covariance over the whole random effect for all structures in the glmmVA interface, i.e., also for corExp, corMatern, corAR1 and corCS, so that you can do things like corCS(a+b|group).

I'm sure this is useful for some reason.

Bert van der Veen @vdveenb.bsky.social · Aug 9

Many bugfixes and updates later... This will probably happen a few times. I've also added a propto() structure, which can be combined with a correlation matrix for random effects, forming a kronecker covariance structure where the column matrix is fixed.

Bert van der Veen @vdveenb.bsky.social · Aug 7

New on the gllvm front: functionality for fitting univariate #GLMMs: gllvmVA(). You may wonder why this is useful, with great packages such as #lme4 and #glmmTMB at your disposal.

1

Bert van der Veen @vdveenb.bsky.social · Aug 9

There are papers that discuss the difference in the estinstors in more detail. Ormerod and Wand www.tandfonline.com/doi/full/10.... for example.

There are situations where the approximations are different, and one need not expect the same answer from a model.

Gaussian Variational Approximate Inference for Generalized Linear Mixed Models

Variational approximation methods have become a mainstay of contemporary machine learning methodology, but currently have little presence in statistics. We devise an effective variational approxim...

www.tandfonline.com

1

Bert van der Veen @vdveenb.bsky.social · Aug 9

That's an excellent question. I don't have anything prepared, but I've had a few use cases where laplace of zero-inflated models did not converge well.

A lot of our evidence is anecdotal, though we've done extensive simulation on LA and VA performance for LVMs.

2 2

Reposted by Bert van der Veen

Markus Wagner @wagnermarkus.bsky.social · Aug 9

An excellent course! Thanks so much for organising and running it, Bert & the whole team! Keeping my fingers crossed that there will be another one like it in the not too distant future, I can warmly recommend it!

Bert van der Veen @vdveenb.bsky.social · Aug 7

Two weeks ago I organised the first Summer school in Model-based Multivariate analysis for community ecologists.

We spent a week in Hjerkinn (NO), teaching 26 international participants about model-based multivariate analysis. Together with Jenni Niku, Pekka Korhonen, Francis Hui and Audun Rugstad.

1 2 4

Bert van der Veen @vdveenb.bsky.social · Aug 9

Thanks for the positive feedback Markus, and thanks for participating 😁

1

Bert van der Veen @vdveenb.bsky.social · Aug 9

Many bugfixes and updates later... This will probably happen a few times. I've also added a propto() structure, which can be combined with a correlation matrix for random effects, forming a kronecker covariance structure where the column matrix is fixed.

Bert van der Veen @vdveenb.bsky.social · Aug 7

New on the gllvm front: functionality for fitting univariate #GLMMs: gllvmVA(). You may wonder why this is useful, with great packages such as #lme4 and #glmmTMB at your disposal.

1 2

Bert van der Veen @vdveenb.bsky.social · Aug 7

Fingers crossed that I will be able to secure funding for another summer school next year(s).

Teaching material from the school is available at github.com/BertvanderVe....

5

Bert van der Veen @vdveenb.bsky.social · Aug 7

I tremendously enjoyed the event, especially interacting with participants and discussing analyses of their own data was amazing.

The event was funded by @nordicoikos.bsky.social, without which this would not have been possible.

1 2

Bert van der Veen @vdveenb.bsky.social · Aug 7

Two weeks ago I organised the first Summer school in Model-based Multivariate analysis for community ecologists.

We spent a week in Hjerkinn (NO), teaching 26 international participants about model-based multivariate analysis. Together with Jenni Niku, Pekka Korhonen, Francis Hui and Audun Rugstad.

1 1 16

Bert van der Veen @vdveenb.bsky.social · Aug 7

I keep writing "gllvmVA" in places, for obvious reasons, but it's aptly named "glmmVA" in gllvm. Darn abbreviations.

Bert van der Veen @vdveenb.bsky.social · Aug 7

In recent time we have greatly expanded the available distributions in #gllvm. Particularly with ZIB, ZNIB, ZINB, negative-binomial (1), and logit and cloglog link functions for the binomials.

We continue to expand the support for VA. But, if all else fails, LA is also available in #gllvm.

Bert van der Veen @vdveenb.bsky.social · Aug 7

VA tends to be faster and more accurate (in the large cluster sense) for highly discrete data. As a trade-off, it is more challenging to derive the approximation (needs to be done on a case-by-case basis, akin to Gibbs sampling approaches) and it is not possible for all distributions.

1 2