Reverse Bayes:

Imagine you run a randomized controlled trial;

The results come with a p-value of 0.03 -- significant at the customary level of 0.05.

But are they credible?

1/

#CausalSky #EpiSky #StatSky #EconSky #ABTesting

Is Half of All Advertising Spend Wasted?

In this week's issue of Causal Python Weekly:

- David Rohde on why even granular digital data often can't tell you which half of your ad budget is working

1/

#CausalSky #EconSky #EpiSky #StatSky #MLSky

CUPED: Old Wine in a New Bottle? CUPED has become a popular feature in online experimentation platforms. Regression works fine too.

Is CUPED just regression adjustment in disguise? You be the judge: nothing-so-practical.com/post/cuped-v...

#ABtest #experiment #CausalSky #Causalinference

"We intervene and the intervention changes the market dynamics"

Two-sided marketplaces can be a challenging setting for impact measurement.

In the new episode of the Causal Bandits Podcast, we meet Lawrence De Geest (Zoox, ex-Lyft, ex-NBA), a former...

1/

#CausalSky #StatSky #EpiSky #EconSky

Causal Inference Is Hard How I learned to stop worrying and love assumptions

What makes #causalinference hard? It's not the math or the code: nothing-so-practical.com/post/causal-...

#CausalSky #ABtest

We Often Think About Randomized Trials as "the Gold Standard" for Causal Inference.

And yet we almost as often forget to ask what information they provide and under what circumstances.

1/

#CausalSky #EconSky #EpiSky #StatSky #MLSky #ABTesting

What If We Reversed Bayes to Test How Robust Our Findings Are to Skepticism?

In this week's issue of Causal Python Weekly:

- Lu Qian's new blog post on Reverse Bayes and evidence

1/

#CausalSky #EconSky #EpiSky #MLSky #StatSky

Is the effect heterogeneity in your data real?

Testing effect homogeneity and confounding with mixed data

In their new working paper, Ana Paula Armendariz and Martin Huber propose a framework for testing the...

1/

#CausalSky #EconSky #EpiSky #StatSky #MLSky

You Need an Analysis. Which Statistical Test to Use?

Hey there, why not ask Claude?

With LLMs, the bottleneck is no longer the speed of access to information or the speed of content generation, but the effort needed to verify the outputs.

1/

#StatSky #EpiSky #BioStats #CausalSky

Dropbox Reallocates $25M After a Causal Study

In a recent paper, Varun Chivukula et al describe a series of large-scale blackout experiments across marketing channels that inspired the company to reallocate $25M of their marketing budget to more effective channels.

1/

#CausalSky #EconSky #StatSky

Theory for Identification and Inference with Synthetic Controls: A Proximal Causal Inference Framework Synthetic control (SC) methods are commonly used to estimate the treatment effect on a single treated unit in panel data settings. An SC is a weighted average of control units built to match the tr...

www.tandfonline.com/doi/full/10....

#causalsky #causalinference #StatsSky

Individualized Dynamic Mediation Analysis Using Latent Factor Models Mediation analysis plays a crucial role in causal inference as it can investigate the pathways through which treatment influences outcome. Most existing mediation analysis assumes that mediation ef...

www.tandfonline.com/doi/full/10....

#CausalSky #causalinference #statssky

Is there a reason that risk difference isn't used more for time to event outcomes?

For example, estimate baseline hazard and hazard at 5 years, then take the difference (i.e., using a Royston-Parmar model)

#CausalSky #StatsSky

Dropbox Reallocated $25M of Their Marketing Budget After a Causal Study.

In this week's issue of Causal Python Weekly:

- Alberto D. Horner reviews a new End-to-End Estimation for Counterfactual Fairness by Yuchen Ma et al.

1/

#CausalSky #EconSky #EpiSky #StatSky #MLSky

I'm trying to find examples of replicable mistakes by Chat GPT. I seem to recall there was a thread on here about common causal inference mistakes made by AI. Does anyone have a tip? #causalsky

5 Causal Inference Ideas for the Age of Vibes

I'm working on a talk for an event this spring, and I want to ask you for your thoughts.

I collected these 5 ideas in response to various queries I got from people across industries and...

1/

#CausalSky #StatSky #EpiSky #EconSky #MLSky

📊 Join us tomorrow for our #HealthData seminar, exploring causal inference challenges with examples from research in ACE inhibitors #CausalSky

📆 Tuesday 17 March 2026 (12:45-13:50)
🏛️ LSHTM | Online

More details 🔽
www.lshtm.ac.uk/newsevents/e...

The effect of SARS-CoV-2 testing on healthcare-seeking behaviour at primary care level AbstractBackground. Diagnostic self-testing for SARS-CoV-2 may lead to selection bias in test-negative case–control designs (TND) for COVID-19 vaccine effe

academic.oup.com/ije/article/...

#EpiSky #COVID19 #CausalSky #causalinference

Is the Effect Heterogeneity in Your Data Real?

In today's issue of Causal Python Weekly:

- A new paper by Ana Paula Armendariz and Martin Huber on testing for heterogeneity and confounding in high-dimensional mixed data (observational + experimental)

1/

#CausalSky #StatSky #EpiSky #EconSky

Very Handy!

The latest version of CausalPy introduces a new piecewise interrupted time series (ITS) class.

CausalPy is a popular Python library for modeling quasi-experimental data.

The new version comes with a brand new...

1/

#CausalSky #EconSky #EpiSky #MLSky #StatSky

What a Week!

We literally had too many topics to fit into one newsletter this week.

Here's what we picked:

- Alberto D. Horner reviews the brand new book by Quentin Gallea, PhD

- David Rohde on why policies are stochastic in reinforcement learning

1/

#CausalSky #StatSky #EconSky #EpiSky #MLSky

Royal Statistical Society Publications Summary. Conventional analytic results do not reflect any source of uncertainty other than random error, and as a result readers must rely on informal judgments regarding the effect of possible bias...

So does this one:

rss.onlinelibrary.wiley.com/doi/abs/10.1...

#causalsky #causalinference

The Optimal Split for an A/B Test Is 50:50

Unless...

I recently saw a post explaining why the optimal split between treatment and control groups should be 50:50.

The optimal split is indeed 50:50, but only under one assumption:

1/

#CausalSky #StatSky #ABTest #EpiSky #MLSky #EconSky

You Train Your Robot in August, It Trashes Your Garden in September.

In today's issue of causal Python Weekly:

- Causal POMDPs (Partially Observed Markov Decision Processes): Planning when the world changes - a review of...

1/

#CausalSky #MLSky #AISky #EconSky #StatSky

Should you use a t-test if n >= 30?

Perhaps. But most “which test should I use?” rules are wrong.

Most of these rules exist because we were trained to think in reverse: starting with procedures instead of questions.

The sad truth is that...

1/

#StatSky #CausalSky #EpiSky #EconSky

#causalsky 🌍

Science is a Crossword

A 2015 randomized controlled trial (RCT) demonstrated that chocolate speeds up weight loss.

P-value < 0.05

The results went viral and were discussed broadly in the press and social media.

Moreover, they likely...

1/

#CausalSky #EpiSky #EcoonSky #MLSky

Independent Causal Mechanisms Inside Your LLM?

In this week's issue of Causal Python Weekly:

- StochTree tutorial by Drew Herren, Richard Hahn, Jared Murray, and Carlos Carvalho (R and Python included)

1/

#CausalSky #MLSky #StatSky #EpiSky #EconSky #AISky

📊 Join us next week for a seminar with @georgiatomova.bsky.social on her new paper exploring how different modes of survey data collection can introduce bias #datascience #CausalSky

📆 Thursday 26 February (12:50-13:50)
🏛️ LSHTM | Online

Details 🔽
www.lshtm.ac.uk/newsevents/e...

It’s amusing that “Structural Equation Model” can refer to either:

1) a nonparametric model, y~f(x_1, x_2,… x_n), à la Pearlian DAGs, or

2) a massively parametric model, where every relationship is parameterized with a linear/factor model (the psychometrics approach).

#statssky #causalsky