Lightnews — Scholar-powered news

LightNews

Tim Morris

@timpmorris.bsky.social

2.4K followers 460 following 1.4K posts

Biostatistician working on methodology at Novartis. Simulation studies, non-inferiority, missing data, estimands, covariate adjustment… He/him https://tpmorris.substack.com/

tpmorris.substack.com

Posts Media Videos Starter Packs

Tim Morris @timpmorris.bsky.social · 8h

Talk to us, what font is it?

Tim Morris @timpmorris.bsky.social · 10h

Hey, imagine how tough life must be striking the right balance between being top-2% and not-suspiciously-prolific, plus staying productive while people waste your time patiently explaining how you’re wrong!

Tim Morris @timpmorris.bsky.social · 14h

I’m thinking about the methods research that tends to get funded. Precious little late-phase methodology (often viewed as insufficiently novel etc.).

Tim Morris @timpmorris.bsky.social · 15h

I reckon not much – so many of the problems in funded applied research are so basic and obvious that any fundable methods research would be too sophisticated!

Tim Morris @timpmorris.bsky.social · 1d

New (short) post on hearing criticism of your work.

No, I’m not telling you who it was – they will recognise themself!

open.substack.com/pub/tpmorris...

Reposted by Tim Morris

Darren Dahly @statsepi.bsky.social · May 2

My favorite genre of fiction is Thesis Acknowledgments.

Tim Morris @timpmorris.bsky.social · 16d

Straight onto my reading list

arXiv stat.ME Methodology @statme-bot.bsky.social · Aug 22

Arman Oganisian: Untangling Sample and Population Level Estimands in Bayesian Causal Inference https://arxiv.org/abs/2508.15016 https://arxiv.org/pdf/2508.15016 https://arxiv.org/html/2508.15016

Tim Morris @timpmorris.bsky.social · 24d

The defenders of British Values, having completed their work assaulting police officers, have returned home, leaving London a Dystopian Hellscape once again

The morning sun peeping around some horse chestnuts (non-native) and hornbeams (non-native)

Tim Morris @timpmorris.bsky.social · 24d

Bonus marks for explaining why the quoted post appears to be wrong but the quoting post is in fact wrong.

Tim Morris @timpmorris.bsky.social · 28d

Why does everyone who picks up the mic at this conference feel they have to mention this talk, whether grudgingly or positively??

Tim Morris @timpmorris.bsky.social · 28d

Hahaha I enjoyed that! Yes, I agree on the sentence/s (these documents do not cite any references though, which always annoys people).

Tim Morris @timpmorris.bsky.social · 28d

Very much agree with that last point. I think what you’re politely saying translates to “come back when you’ve read my 🤬 paper!”

Tim Morris @timpmorris.bsky.social · 28d

The second sentence of your abstract!

Tim Morris @timpmorris.bsky.social · 28d

Isn’t this like complaining that the bus to the airport fell short when it didn’t fly you to your destination? 😜

It’s gives us a useful (important) start but is not a grand-unifiying-all-singing-all-dancing solution!

Tim Morris @timpmorris.bsky.social · 28d

😜

Tim Morris @timpmorris.bsky.social · 28d

I’m just a bit baffled about what you think this is “inferior to”. Writing an subsequent expression in terms of potential outcomes? If so, this should come from the natural-language version, not compete with it.
E.g. doi.org/10.1002/sim....

Tim Morris @timpmorris.bsky.social · 28d

Inferior to what? For the Addendum, the whole point is to give a structured description of the estimand in natural-language.

This should *not* set the estimand in stone – we cannot just skip past identification, estimation, implementation &c. – though practice isn’t as good as the intention.

Tim Morris @timpmorris.bsky.social · 28d

Yes, thanks. I disagree with your premise too as I disagree with Hernán’s grift: being able to say what trial you would like to do does not say what the estimand is, it’s something like “the estimand would be identified under this design”. But a causal estimand does not depend on study design!

Tim Morris @timpmorris.bsky.social · 29d

Definitely. I guess lots of useful approaches gain this problem once popular (estimands for one). A superficial label accompanied by no-thinking.

Tim Morris @timpmorris.bsky.social · 29d

Yes, agree!

Tim Morris @timpmorris.bsky.social · 29d

Thanks Cameron!
Dean, two key points:
1. It’s the structure in this pic that we all find so useful.
2. The “Intercurrent event strategies” mean interplay with the other four attributes.

Screenshot of a slide.
Title: “Addendum contributions (‘IMO’)”
Left hand column says
“Five ‘attributes’
1. Treatment condition of interest
2. Target population
3. Outcome variable
4. Intercurrent events
5. Population-level summary
measure”
Right hand column says
“Handling intercurrent events (LHS point 4)
1. ‘Treatment policy’
2. ‘Hypothetical’
3. ‘Composite variable’
4. ‘While on treatment’
5. Principal stratum”

Tim Morris @timpmorris.bsky.social · 29d

One more. In comparing “Trials” & “Target trials”, slide says something like “causal contrasts” vs. “observational analogs of contrasts”. Causal contrasts are of potential outcome distributions (see Hernán & Robins’ book), which do not depend on the study design – no “observational analog”!

Tim Morris @timpmorris.bsky.social · 29d

The target trials concept is great; I’m not slamming that. But even the poster-child examples leave a lot ambiguous (partly because they seem to rely on readers being able to reverse-engineer estimands).

Tim Morris @timpmorris.bsky.social · 29d

I’d say substance was largely absent. Most gripes were about minor semantic things. Seemed like sour grapes that the trials community didn’t think “real trials have target trials too” solved everything for them, and he wants them to turn back from this 🤷

Tim Morris @timpmorris.bsky.social · 29d

Having to infer the estimand from the design + analysis was one of the things that led to the E9(R1) addendum in the first place!
Also that the “ITT effect [sic] is never relevant” thing, avoiding engaging with how you would realise a different intercurrent event strategy in practice.