forgot to add: measurement invariance www.sciencedirect.com/science/arti...

Last, how will you follow up on this topic? Do you plan to focus on topics such as bidirectional constructive relationships (not uncommon in CB SEM and not possible in PLS SEM) or latent growth models? Not much more and you'll have translated everything from Depaoli's book into PyMC! Great job! :D 👏

Would it be necessary to examine if this could open confounding pathes in the model (backdoor criterion, etc.)? How would the DAG look like after introducing this new hierarchical structure? Like adding moderating Constructs?

However, I find it difficult to assess the extent to which causal mechanisms should (or must?) be applied more in the development and revision of SEM models. The causal DAG of your presented Job Satisfaction model changes once you introduce gender as a hierarchical construct.

Furthermode, I'm interested in your perspective on causal thinking in the field of SEM. There is this paper by Pearl on this topic you probably know (ftp.cs.ucla.edu/pub/stat_ser...
).

The scope for changing things in the model structure once the data is available always seemed very limited to me, which in a sense contradicts this iterative process.

Especially when writing papers, the process of model and hypothesis development always seems very linear and like a strict one way street to me: you have to have everything fixed before data collection and analysis. Pre-registration, for instance, would also consider this a given.

You described the iterative process of model creation and re-evaluation very nicely. I would be interested to hear your perspective on the usual scientific process, at least as I have come to know it.

You also responded very well to the question “can't you just summarize the 12 variables and calculate a normal regression?” I often feel that the concept of the inherently more complex measurement model of latent (as opposed to observable) variables/constructs throws people off.

Excellent presentation, thank you very much! I cheered in front of the screen when you introduced the hierarchical version with the example of two experimental treatment groups. That's exactly what I often need and what I've never seen in PyMC before, so thank you for that!

My talk focused on craft in statistical modelling with Bayesian workflows, using a case study on job satisfaction and what makes work feel compelling.

@pymc.io case study → www.pymc.io/projects/exa...

Slides → nathanielf.github.io/talks/pycon_...

awesome, will give it a read tomorrow, thank you :)

Man, klingt echt spannend, gibt aber vermutlich kein Recording irgendwann? 🥺