Lightnews — Scholar-powered news

Aleksander Molak @alxndrmlk.bsky.social · 1d

"I like to think about causality as a spectrum"

Yesterday, Miguel Hernán shared with us experiences and lessons he learned while advising the Spanish government during the COVID-19 pandemic

I also like the perspective of "causality as a spectrum" and find it very useful in practice

1/

#CausalSky

1

Aleksander Molak @alxndrmlk.bsky.social · 3d

12 Challenges for the Next Decade

One of causal inference’s main strengths is also one of its biggest curses.

Causal inference is an interdisciplinary field and as such, it has greatly benefited from contributions from...

1/

#CausalSky #StatSky #MLSky #CausalInference

1 1

Aleksander Molak @alxndrmlk.bsky.social · 4d

It's also worthwhile remembering that the "hierarchy" discussions typically conveniently ignore the fact that there exist more than one (type of) causal questions we might be interested in answering.

2

Aleksander Molak @alxndrmlk.bsky.social · 4d

We'll start sending it today at 9am PT / 12pm ET / 6pm CET

To get it in your inbox (FREE): causalpython.io

3/3

Causal Python || Your go-to resource for learning about Causality in Python

A page where you can learn about causal inference in Python, causal discovery in Python and causal structure learning in Python. How to causal inference in Python?

causalpython.io

Aleksander Molak @alxndrmlk.bsky.social · 4d

1. This brand-new paper by Carlos Cinelli, guido imbens, Sara Magliacane, et al highlights a dozen of top challenges in causality
2. David Rohde on the Horvitz-Thompson estimator for causal inference
3. A new interview with Marcos Lopez de Prado on quantitative finance, AI and causality

2/

1 1

Aleksander Molak @alxndrmlk.bsky.social · 4d

What are the biggest challenges for causal inference?

In this week's issue of Causal Python Weekly ( causalpython.io ):

1/

#CausalSky #MLSky #StatSky

1 1 1

Aleksander Molak @alxndrmlk.bsky.social · 4d

Looks like this will be fun!

#CausalSky

Professor Melinda Mills @melindacmills.bsky.social · 7d

Join our first @oxforddemsci.bsky.social seminar of the term!
Who said causality couldn’t be fun?

Leverhulme Centre for Demographic Science, Oxford University @oxforddemsci.bsky.social · 7d

nonparametric. multiply robust. semiparametrically efficient.
💥the Avengers lineup of causal estimators

Felix Elwert, 8 Oct, 2pm
📍 Butler Room, @nuffieldcollege.bsky.social
Nonparametric Causal Decomposition of Group Differences
w/metrics-and-models.github.io
🔗 demography.ox.ac.uk/news/upcoming-…

1

Aleksander Molak @alxndrmlk.bsky.social · 5d

👏🏼👏🏼👏🏼

Aleksander Molak @alxndrmlk.bsky.social · 6d

Paper: https://arxiv.org/abs/2508.11727
Our weekly newsletter (FREE): causalpython.io

8/8

Aleksander Molak @alxndrmlk.bsky.social · 6d

The results come with causal identification guarantees, and experiments on both synthetic and real-world data show that the method effectively recovers causal structures despite the presence of unobserved subprocesses.

Thoughts?

7/

Aleksander Molak @alxndrmlk.bsky.social · 6d

The authors establish necessary and sufficient conditions for identifying latent subprocesses and propose a two-phase iterative algorithm that switches between inferring causal relationships between these processes and uncovering new ones.

6/

Aleksander Molak @alxndrmlk.bsky.social · 6d

This assumption is often violated in real-world scenarios.

In their new paper, Songyao Jin and Biwei Huang (UC San Diego) present a novel Hawkes process-based causal discovery method that works with unobserved subprocesses.

5/

Aleksander Molak @alxndrmlk.bsky.social · 6d

And indeed, Hawkes process can be used to describe and discover causal dependencies in multivariate time series, but...

Most existing methods operate under the assumption of causal sufficiency, meaning that all relevant variables (or subprocesses) are observed.

4/

Aleksander Molak @alxndrmlk.bsky.social · 6d

An important property of Hawkes process is that it's self-exciting: if an event occurs at any given moment, it makes it more likely that it will also occur in the future.

For many, the multivariate version of Hawkes process is a natural choice to describe causal structure of time series data.

3/

Aleksander Molak @alxndrmlk.bsky.social · 6d

A Hawkes process is a stochastic process (think a statistical model describing time progression of some phenomenon using random variables) often used in finance, epidemiology and seismology.

2/

Aleksander Molak @alxndrmlk.bsky.social · 6d

Granger causality is not causality, but...

Here's a new causal discovery algorithm for time series with latent confounders

1/

#CausalSky #EconSky #StatSky #MLSky

7 3

Aleksander Molak @alxndrmlk.bsky.social · 8d

Despite using reportedly "sophisticated systems", Amazon's counterfeit and fake review identification is far from perfect.

The same seems to be the case for other large-scale marketplace platforms and even social media content moderation.

3/3

Aleksander Molak @alxndrmlk.bsky.social · 8d

The raise of generative models trained on the entirety of the internet inspired a new wave of conversations about copyrights.

But there's also a challenge that we discuss less often.

2/

Aleksander Molak @alxndrmlk.bsky.social · 8d

Someone published my book under another name on Amazon.

I had three thoughts:

1. Thanks for the compliment, Ashaly
2. What a chutzpah, though
3. Where's the email to the copyright team?

1/

#CausalSky #Amazon #ripoff #book

2 2

Aleksander Molak @alxndrmlk.bsky.social · 9d

Take the survey: https://share.deftform.com/causal-python-community-2025-09-01-GrEg6x

Thank you for helping shape this community 🙏🏼

5/5

Aleksander Molak @alxndrmlk.bsky.social · 9d

- We'll analyze the results and see if we reached the the critical number of members needed to launch.

- If so, we will contact all the people who left their email in the survey or is subscribed to the community list with the next steps.

4/

Aleksander Molak @alxndrmlk.bsky.social · 9d

The survey is super simple and should take 2-3 minutes of your time. If you already filled it, you can ignore this email.

What happens next?

- The survey closes on Tue, Oct 30, 23:59:00 (anywhere on earth).

3/

Aleksander Molak @alxndrmlk.bsky.social · 9d

...we'd love your input on what matters most for you.

Your perspective will help us:

- Prioritize the right features
- Set a fair monthly membership for those who want to be paid members
- Pick the best features for those who are only interested in free membership

2/

Aleksander Molak @alxndrmlk.bsky.social · 9d

Can we do better? 20 hours left.

What if we could built a better, more direct and engaging way to communicate and learn about causality together?

We're in the final stage of collecting feedback for the upcoming Causal Experts Community, and...

1/

#causalSky #StatSky #MLSky

4 1 2

Aleksander Molak @alxndrmlk.bsky.social · 9d

❌ Tribal wars in statistics or pragmatism?

Mark van der Laan consistently chooses pragmatism and he freely combines inspiration form Potential Outcomes and Structural Causal Models frameworks.

Results?

See the full episode here: vist.ly/48pdx

1