TIL

What? And miss the amazing repartee?

Love the look of these! #croissants au levain for the office #sourdough #bakesky #breadsky

We are following two in the playoffs Nic Dowd with the LVGK and John Carlson with AD. Dowd got round 1 win. Carlson is playing the Oilers tonight. #ALLCAPS

Excited to add my first goal puck to the mantle. Nice to finally get it!

Sounds delish

My son's match starts at 10am.

Like sugar and salt presumably. I add “enhancer” to some bakes, I am guessing those bakes are UPFs.

Big Money Is Betting on Bagels

Some banger lines - who knew there was a chronic shortage of traditional bagels in the southwest? www.nytimes.com/2026/04/14/d...

How do they define ultra-processed?

Aussie national champs breaking some records.

How cool is this 🤩

Every player on the /NYRangers is wearing a Jonathan Quick No. 32 jersey ahead of his final NHL game!

Idk. With more curative treatments available it is becoming harder to rely on quickly accumulating events.

I would love that.

Currently teaching a 2nd year structural metrics class. The IO journals seem to do a particularly poor job.

First pass at a website that keeps track of the data availability status of various published papers:

paulgp.com/replication-...

Very much in Alpha mode, so data will be updated soon.

I teach one or two course each year and the project is always a replication project. But actually finding data is super hard work. This is a great service.

Wow!

Definitely don't want that many people living! It really messes with the stats.

I said surviving but you could do it for any event.

Sorry do you mean that the % who survive X time. Sure, I believe that, but I'm saying why not just report that then.

Remember we are talking about counterfactual that is unobserved.

What I like about these is that it is average treatment effect, and the criteria about when to report is determined only be recruitment and not events.

Anyway how 'bout them Mets.

Also, in order to report differences in median survival you need to half the events to occur in both arms, which is a very odd criteria for ending a trial.

We should present differences in the probability of surviving X years. At least that is an average treatment effect and it doesn't require strong parametric assumptions.

It is actually possible for the difference in median survival to be positive and for no actual person in the trial be better off. It is not likely, but is mathematically possible. Results such as log ratio tests are better but they rely on very strong assumptions.

The difference in averages is equal to the average of the differences. That is the average treatment effect is the treatment effect of the average person in the trial. Medians are used as alternative because the average is infinity. But they don't have the same interpretation.

God I hate differences in median survival. Who ever? What the actual...?

Mt Panorama? Isn't it in the name?