iiinova — where your digital self keeps evolving

WeRateDAGs @weratedags.com · 20/12/2025

What's this? Another confounded tree? 10/10 and a very merry Multistage Simpson's Paradox Machine* from all of us here at WeRateDAGs! *Pearl, J. (2014) 'Comment: Understanding Simpson’s Paradox' The American Statistician, 68(1):8-13

Baubles, branches, some fancy decoration on Z4, and a Z1 that always knew it was a star.

063

WeRateDAGs @weratedags.com · 02/12/2025

A trampoline DAG, as seen by someone who hasn't quite got the flipping-over part down. 12/10 because this is one of those cases when a fully connected graph is good, actually.

Everything is connected to everything else. Twice, probably.

060

WeRateDAGs @weratedags.com · 01/12/2025

An amazing fifty-part thread of DAGs and albums: 80/50.

060

WeRateDAGs @weratedags.com · 12/11/2025

Today we have a mediated moderated suspension bridge, designed in the 90s and finally open. That mediator outcome arrow reinforcement probably increases future tensile strength but it's still a 2/10 because we can't encourage arrow-on-arrow crime.

Our other theory is that it's a bicycle frame.

1323

WeRateDAGs @weratedags.com · 08/11/2025

Continuing our drive to rate the classics, here are two maximally LA-coded suburban DAGs from Judea Pearl. There's thirsty lawns and sprinklers, burglars and earthquakes, and neighbors who call you if your alarm goes off. 9/10 because there's no mention of the traffic.

Is this grass wet because it rained or because the sprinkler was on? Trick question: this is Los Angeles. The sprinkler was on.

Pearl's alarm goes off. Is it a burglary or an earthquake? (Such a Los Angeles question). Who knows? John or Mary will have to tell him because it's bumper-to-bumper on I-405.

044

WeRateDAGs @weratedags.com · 31/10/2025

We love a good 'association as as fluid' metaphor, but this guy takes it to the next level. 12/10 for the reminder that conditioning on a collider doesn't just mess with your identification; it can also flood your basement.

0347

WeRateDAGs @weratedags.com · 24/10/2025

Om nom 9/10 for this Neapolitan-style DAG, packed with new arrows and new toppings. We couldn't manage more than one though.

Composite quanties can be tricky to represent in DAGs, being deterministic and all.

We don't rate cupcakes, but if we did we'd rate this -1/10 and immediately pass it off to the nearest five-year-old.

092

WeRateDAGs @weratedags.com · 24/10/2025

We don't mind a little unfaithfulness as long as the Causal Markov Condition holds. Conditionally independent children need some stability.

040

WeRateDAGs @weratedags.com · 21/10/2025

We don't know if these two DAGs are up to divine mediation problems like identifying the Father's Supernatural Direct Effect or Nicene Indirect Effect, but the Church is in the business of cross-world assumptions, so we're sure it will work out. 10/10 for whichever one is theologically correct 🤞

1173

WeRateDAGs @weratedags.com · 19/10/2025

Today, three adorable missing data DAGs. O* is missing when Rₒ masks it and O otherwise. Now, can we d-separate Rₒ and O? 12/10 Ten for the graphs, plus two for casually flipping "causal inference is a missing data problem" From Mohan & Pearl (2021) "Graphical Models for Processing Missing Data"

These DAGs are rescues and their former owner seems to have called them Mcar, Mar and Nmar. We think those names are a bit weird to be honest, but the DAGs seem happy enough.

1204

WeRateDAGs @weratedags.com · 18/10/2025

Conditioning on a match indicator S can generate collider bias exactly cancelling a confounding path and identifying the effect of Z on X. Here it is in a very good DAG. 9/10 because of the functional form specificity. From Steiner et al. (2017) "Graphical Models for Quasi-experimental Designs"

S is a match indicator used as a selection node to cancel the Z-X association. We think selection nodes in DAGs are underrated and we're going to fix it.

0224

WeRateDAGs @weratedags.com · 18/10/2025

A very aerodynamic DAG here from Jens, induced from polling data, linearity assumptions and weed. We're not sure that was a good idea, but there is a pub node, so we'll say 4/10

0121

WeRateDAGs @weratedags.com · 17/10/2025

We couldn't not rate this classic early DAG by our beloved collective granddagy, Sewall Wright. 14/10. Ten for the DAG, plus one for each cute guinea-pig node. From Wright (1920) "The Relative Importance of Heredity and Environment in Determining the Piebald Pattern of Guinea-Pigs"

Why are guinea-pigs the colours they are? Sewall's got a DAG for that.

110032

Reposted by WeRateDAGs

Matt Parkes @mattyjparkes.bsky.social · 17/10/2025

@weratedags.com Saw this one *years* ago, but it's lived in my mind rent-free ever since. Hi-res: assets.publishing.service.gov.uk/government/u...

5284

WeRateDAGs @weratedags.com · 17/10/2025

People sometimes ask why we want to rate DAGs. Couldn't we just rate statistical models instead? We like to explain it with capybaras and bicycles.

Causal inference with DAGs is just like capybaras on underwater bicycles. Any questions?

25612

WeRateDAGs @weratedags.com · 17/10/2025

More DAG ratings here to lighten your hearts and provide that warm feeling of d-connectedness.

0111

WeRateDAGs @weratedags.com · 17/10/2025

A palate cleanser: 'The causal structure of banana'. Robust, simply structured, and convenient. 8/10. The DAG is nice too. This is the first of three bananadags in Hitchcock (2016) "Conditioning, intervening, and decision".

"Fig 1. The causal structure of Banana"

(Apparently your favourite mid-morning snack is a mediation problem)

2297

WeRateDAGs @weratedags.com · 17/10/2025

Hello Bluesky! We rate DAGs. Some are great. Some are... not so great. But we rate them all. Let's start with a famous powerpoint hairball a.k.a. "the Afghanisdag", presented to Gen. Stanley A. McChrystal around 2010. His own rating? 1/10 "When we understand that slide, we'll have won the war"

"the Afghanisdag", a massive tangle of arrows and noun phrases, possibly describing a counterinsurgency problem: lethal within 20ft in briefing environments.

87525