Arman Oganisian

Statistician | Assistant professor at @BrownBiostats | Nonparametric Bayesian methods for causal inference.

[Arman Oganisian @StableMarkets]

I'm looking forward to teaching a 4-hour short course on Bayesian sensitivity analysis methods at the American Causal Inference Conference (ACIC) 2026!

It will have a distinct focus on practical implementation in @mcmc_stan w/ examples.

Register here: https://t.co/GWHmJobVYo

[Arman Oganisian @StableMarkets]

I’m worried that - especially with current students - over-reliance will cheat us out of the friction we need to grow even if can produce along axes we already know.

In other words, it presents a long-run exploitation-exploration tradeoff.

[Arman Oganisian @StableMarkets]

What makes for productive LLM use in research?

Imo: an ability to think in an organized & modular way about getting to a working solution.

One-shotting tasks usually gives me a complete mess. Modular usage keeps things tight and understandable + keeps my brain engaged.

[Arman Oganisian @StableMarkets]

I’m pretty good now at intuiting where I need to go and breaking it down into sub-tasks. So LLMs are great for me!

But I got there by doing things manually, messing up, thinking + reading, then trying again.

[Arman Oganisian @StableMarkets]

@RWJE_BA @5_utr @ildiazm @f2harrell You both made a claim, then refused to prove it on the grounds that this very claim is irrelevant.

Honestly based on response content and timing patterns I have to conclude @RWJE_BA and @5_utr are bot accounts. So best move for me is to leave this where it is.

[Arman Oganisian @StableMarkets]

@RWJE_BA @5_utr @ildiazm @f2harrell You first. You made the claim. Prove it.

[Arman Oganisian @StableMarkets]

@5_utr @RWJE_BA @ildiazm @f2harrell Conducting an RCT is not the same as proving they ��work.”

[Arman Oganisian @StableMarkets]

@RWJE_BA @5_utr @ildiazm @f2harrell ^that is not a proof.

An intro proof writing class: you start with one side P(outcome | treatment) - P(outcome | control)

And end with the causal parameter, QED.

Can you try again?

[Arman Oganisian @StableMarkets]

@RWJE_BA @5_utr @ildiazm @f2harrell Okay. Prove your claim.

[Arman Oganisian @StableMarkets]

@5_utr @RWJE_BA @ildiazm @f2harrell This is such a wild take.

Hey man, people stumbled upon probability facts using intuition and their gut about gambling. Kolmogorov was just being nerd by laying out axioms for no reason.

The Greeks had already vibed that something was off about √2. No need for real analysis!

[Arman Oganisian @StableMarkets]

@RWJE_BA @5_utr @ildiazm @f2harrell I never even mentioned observational data. I’m just saying POs allow us to prove that RCTs “work” - so if anything it’s a claim that supports RCTs.

[Arman Oganisian @StableMarkets]

@RWJE_BA @5_utr @ildiazm @f2harrell If that’s true then please prove that a contrast of average observed outcomes across arms in an RCT targets a causal effect. mathematically, starting w/ the contrast & ending in the last line with the causal effect, QED.

You’ll quickly see that it is impossible without syntax.

[Arman Oganisian @StableMarkets]

@5_utr @ildiazm @f2harrell What I said after “if so, ….” And the second post.

[Arman Oganisian @StableMarkets]

@5_utr @ildiazm @f2harrell There we go!

[Arman Oganisian @StableMarkets]

@5_utr @ildiazm @f2harrell And to define “causal effect” mathematically you need some syntax.

Some w/ intuition and experience, may not feel any need to formalize what can be explained with words, plots, diagrams etc. But imo intuition can fail and may not generalize which is why formalism is useful.

[Arman Oganisian @StableMarkets]

@5_utr @ildiazm @f2harrell It may be useful to define exactly what is meant by an RCT “working.”

Do we mean “comparing estimates of E[Y|A=1] vs E[Y|A=0] targets a causal effect.”?

If so, then to formally show it we must first define a “causal effect” then mathematically establish equality.

[Arman Oganisian @StableMarkets]

@DKThomp It depends on what you mean by "traditional" - this is an example of AI demolishing a take-home exam, which I would say is non-traditional method of evaluation that stuck after covid.

The traditional methods - in-class, closed book exams or oral exams are still just fine.

[Arman Oganisian @StableMarkets]

New paper in press at Biometrics by PhD Candidate Esteban Fernández-Morales

1) Develops Bayesian spike & slab and horseshoe models for causal inference under spatial spillover

2) Analyzes Philly's 2017 beverage tax accounting for cross-border shopping

https://t.co/vj5ggoQKi5

[Arman Oganisian @StableMarkets]

...you have a big monster hanging out in front of you that can make you cheat on everything ...contrast, aperture, lighting. It's your duty to decide how much you're going to let yourself fall down that well.

[Arman Oganisian @StableMarkets]

With rapid AI advancement, I've been thinking more about this conversation 13 yrs ago, btwn Jack White & Conan O'Brien, and how it's as relevant for statistics/data science education now as it is for music, comedy, film, etc.

https://t.co/ugRFvbrGLP

Excerpts below start @ 19:13

[Arman Oganisian @StableMarkets]

Jack: I'm a believer that new technology needs new responsibility. If you're a photographer you have photoshop now. And if you're going to call yourself a photographer and dedicate your life to it...