Tim Morris

@timpmorris.bsky.social

2443 Followers 465 Following

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

Spoof paper titled ‘Extending target trial approaches to case series,
case reports and opinions shared on facebook’

The ‘BATS AREN’T BUGS’ Calvin & Hobbes with class yelling ‘TARGET TRIAL EMULATION ISN’T A “STUDY DESIGN”!’

Screenshot of title page of linked paper. Title is “Why most psychological research findings are not even wrong”.

Email subject screenshot: “Reminder: Invitation to join the Metascience Novelty Indicators Challenge”

Suggestions for my possible affiliation. It’s selected ‘LETS Go UK, Portsmouth, United Kingdom’, but then offered some others:
‘We found potential matches. Please review the suggestions below:
- UK Athletics, BIRMINGHAM, GB
- UK Private Investigators, LONDON, GB
- United Kingdom Anti-Doping Ltd, LONDON, GB
- United Kingdom National External Quality Assessment Service, SHEFFIELD, GB
- The United Kingdom Environmental Law Association, BRISTOL, GB
- Search for a different affiliation’

Picture of gravestone that says:
RIP 6 7
died: 1 January 2026
born: September 205
a meme that was popular in 2025

Banner for blog post, showing the title and part of the image in the post (with its own alt text in the post).

Bernie meme: I am once again asking you to avoid “meta-models” when analysing your simulation study

Screenshot of my google search for "univers font"

Screenshot of google search for "univers"

Screenshot of a pdf with a random big title page before the paper begins

Two slices of gluten-free bread with large air pocket at the bottom making the shape of an intensely sad mouth

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

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”

Cover picture with blog title & subtitle, and results graph in the background

Goose chase meme. Goose asks "Unbiased for what estimand?" then yells "Unbiased for what!!"

Foreground text says “Marginal” and “conditional” are relative descriptions of an estimand
In the background is a smokey mollusc

A poster for The Hunger Games but with the word Hunger changed to Funder

Scene from Twisters. Sitting on the bonnet of a pickup truck, Daisy Edgar Jones asks “What if it’s the model that’s the problem?”

Doing the pose like The Son of Man painting

Screenshot of my coffee order with my alter-ego

Some people claim that you should never add very hot water to an espresso, and always brew the espresso on top of the hot water. I don't think it makes a great deal of difference, I just find the resulting coffee is cleaner and looks better.
The one disadvantage of diluting espresso is that you increase the perceived bitterness slightly. For that reason, as soon as you finish brewing the americano, I would recommend scooping the crema off the top of the drink and discarding it. Crema is delightful to look at, but there are many tiny pieces of ground coffee trapped in the foam so the crema contributes additional bitterness to the drink. Removing the crema before stirring and drinking definitely improves the flavour of an americano.
(I would also recommend tasting an espresso after you remove the crema - the difference in flavour is dramatic.
While I actually prefer the taste of an espresso without crema, I don't want the extra work and am happy enough to enjoy it as it is. With an americano, however, I really do think the extra work is worthwhile.)

Screenshot of abstract from “GRADE guidelines 6. Rating the quality of evidence—imprecision” https://www.jclinepi.com/article/S0895-4356(11)00206-X/fulltext
with highlighting on “An exception to this rule occurs when an effect is large, and consideration of CIs alone suggests a robust effect, but the total sample size is not large and the number of events is small. Under these circumstances, one should consider rating down for imprecision.”

Bar chart showing the number of studies targeting a relevant,
useful estimand using defensible methods of design & analysis:
zero.

Screenshot from a website’s twitter scroller: ‘Something went wrong with the twitter.’

Picture shows title of post ‘How to read a zip plot’ in front of a greyed-out image of the zip plot used in the post.

Photo of Adam Kucharski’s recent book ‘Proof The Art and Science of Uncertainty’

Screenshot says ‘To a systematic-reviewer, the results of groups A to D may feel like turning up with a dustpan and brush after an earthquake because:’

A wall with a shadow on is the background for a young man staring thoughtfully. The article’s title is ‘Young men and their worst fears’.

A Hackney carriage actual ambulance

Just a straight-up hottie going for a run wearing Stata sunnies and Stata quarter-zip top

Screenshot from Angrist et al. 2025 NEJM evidence. The abstract says ‘Many randomized clinical trials fail to play out as intended: some participants assigned to the treatment group remain untreated, while others assigned to the control group cross over and receive treatment. In such settings…’ and I have highlighted ‘intention-to-treat analyses that compare participants by treatment assignment are diluted by noncompliance,’

View of St Mary Le Strand, Waterloo Bridge, London Eye, the Thames, Big Ben…

The “16” comes from the following four statements:

1. When estimating a main effect and an interaction from balanced data using simple averages (which is equivalent to least squares regression), the estimate of the interaction has twice the standard error as the estimate of a main effect.

2. It’s reasonable to suppose that an interaction will have half the magnitude of a main effect.

3. From 1 and 2 above, we can suppose that the true effect size divided by the standard error is 4 times higher for the interaction than for the main effect.

4. To achieve any desired level of statistical power for the interaction, you will need 4^2 = 16 times the sample size that you would need to attain that level of power for the main effect.

Statements 3 and 4 are unobjectionable.

Text from ‘The Penguin Guide to Punctuation’ 1997:

that good-night is a single compound modifier. Without the hyphen, the reader might easily be misled:
* She gave him a good night kiss.
Here the reader might be momentarily flummoxed into thinking that she had given him some kind of "night kiss', whatever that means. Here are some further examples:
Her dress is light green.
She's wearing a light-green dress.
This book token is worth ten pounds.
This is a ten-pound book token.
She always turned up for the parties at the end of term.
She always turned up for the end-of-term parties.
This essay is well thought out.
This is a well-thought-out essay.
Her son is ten years old.
She has a ten-year-old son.
Use hyphens liberally in such compound modifiers; they are often vital to comprehension: a light-green dress is not necessarily a light green dress; our first-class discussion is quite different from our first class discussion; a rusty-nail cutter is hardly the same as a rusty nail-cutter; a woman-hating religion is utterly different from a woman hating religion; and a nude-review producer is most unlikely to be a nude review producer! You can mislead your reader disastrously by omitting these crucial hyphens: She

‘Revision based on user testing

User testing with junior systematic reviewers revealed a serious problem with the initial core item related to intention-to-treat analysis: when applying the instrument to different trials, four out of the first five reviewers made incorrect assessments for this item (see supplementary appendix 2 for details). This problem led the panel to drop the intention-to-treat item from the core items list and to modify an existing core item related to missing data to deal with the problem of participants whose outcome data were not included in the analysis for whatever reason

Photo of Nintendo World at Universal Studios in Osaka

April fool’s edition of Phoenix is named Total Brick and is just full of brick related stuff. Back cover is an ad for a brAIck, with some good deets.

A flux construct from TOTK

Screenshot of the top left of a grand-looking Wiley ‘Top Cited Article‘ certificate. A more accurate description might be a Wiley ‘Article’ certificate, since everyone seems to get them.

The front of The Ship Tavern in Holborn. A lamp obscures the top of the p in the name so it looks like it’s called The Shit Tavern.

$Screenshot of the following text (Latex code): $\widehat{\sigma}^2$ hey, we don't square an estimator $\widehat\sigma$! $\widehat{\sigma^2}$ accurate… but ugly$

'Missing baseline data when analysing change-from-baseline'
tpmorris.substack.com