Mark Donoghoe

Statistics ∪ Football

[Mark Donoghoe @MarkD_4]

@official_lesdog Almost spot on

https://t.co/bBTfJCWm7t

[Mark Donoghoe @MarkD_4]

Found John McEnroe's contribution to the #Rstats source code

[Terrible Maps @TerribleMaps]

Using the Mercator Projection on a sphere is bold

[Mark Donoghoe @MarkD_4]

@sgilbertdavies @clairemcnear I need to know whether this would've earned a "good for you" from Trebek 🥲

https://t.co/dfHb0fKZ6g

[Billy Ford @BillyFjord]

my favorite part of Jeopardy is when contestants tell an awful story and Alex Trebek is like “um ok good for you”

[depths of wikipedia! @depthsofwiki]

The "Ship of Theseus" article has been edited 1792 times since it was created in July of 2003. At present, 0% of the phrases in the original article (seen below) remain.

[ABC SPORT @abcsport]

Three to win from the last ball, and what follows is peak European Cricket League once again.

Bravo, fellas. 😂

📹 - European Cricket Network

[Mark Donoghoe @MarkD_4]

@jershbot @NextGenStats @awscloud No argument here, just find it a little silly to quote the output of the model down to the nearest 0.1%, while in the same sentence saying that there is not enough data to assess the accuracy of the model towards that tail

[Mark Donoghoe @MarkD_4]

@NextGenStats @awscloud If there had been no previous catches with probability under 4%, how can you assign Sutton's 3.2% probability? The model can't possibly be calibrated at that level...

[Mark Donoghoe @MarkD_4]

@mostlymanthings @danielricciardo @F1Struggle "of course I know what a honey badger looks like, why do you ask?"

[Mark Donoghoe @MarkD_4]

@camjpatrick Can confirm - great roles in a great team

[Gordana Popovic @GordyPopovic]

We @StatsCentralAus are recruiting 3 level A/B statistical consultants. Contribute to a huge verity of impactful research at UNSW. Awesome collaborative team. Flexible hybrid work. Full-time or Part-time.
https://t.co/EWMJfV3CMV

[Mark Donoghoe @MarkD_4]

@stephenjwild Pretty sure the fitted coefficients should just be rescaled (by -log2(e)) versions of what you'd get from a usual log link, i.e. relative risk regression.

I'm not sure if people would find "bits of information" to be a more intuitive effect measure than a relative risk.

[Mark Donoghoe @MarkD_4]

Remembering when I was at the crease and our injured no 11 bravely hobbled out to the middle, a la Lyon.

But instead of swatting 130kph bouncers for six, I got clean bowled next ball, and he had to turn straight around and head back off 🫠

[Adam Peacock @adampeacock3]

15 runs added because of what Lyon did. Selfless.
Think it was more about that than getting a concussion sub….. ?

[Mark Donoghoe @MarkD_4]

@BruceExclusive I focused on the expectation thing (I'm a statistician, it's a reflex 🙃), but agree with what I now realise was the main point (I hope): that the chance of winning the SB --- even as the favourite --- is pretty low.

[Mark Donoghoe @MarkD_4]

@BruceExclusive I dunno about that last one (seems like a low threshold), but since sending my previous tweet, I think I was wrong about saying "only" a 33% chance.

If we split the first 56 Super Bowls into 8 groups, it means there should be 2 or 3 seven-year stretches where no favourite won.

[Mark Donoghoe @MarkD_4]

@mueblesfeos @colin69161558 You could well be right that the authors don't understand the distinction between data & sampling distributions, but being charitable, the statement itself is strictly true and kind of makes sense in the context of the paper (i.e., "this is why people do normality tests").

[Mark Donoghoe @MarkD_4]

@mueblesfeos @colin69161558 The t distribution requires that the sample variance is chi-squared; see the last sentence of your wiki screenshot.

I see what you're saying, but "the test makes this assumption" is true and is different from "the test strictly requires this assumption for it to work at all".

[Mark Donoghoe @MarkD_4]

@mueblesfeos @colin69161558 A more thorough description

https://t.co/U6fqMk3rOX

[Mark Donoghoe @MarkD_4]

@mueblesfeos @colin69161558 If we're being pedantic, the sample mean is only *approximately* normal for finite n, by the CLT (not to mention the assumed distn of the sample var). It's true this means the test is robust to deviations from the assumptions, but doesn't mean the highlighted statement is *wrong*