Charles Mercieca

[Charles Mercieca @Charles_Merc]

Just the live page without any commentary/fluff: https://t.co/3cH1NwFXSj

[Charles Mercieca @Charles_Merc]

🇲🇹 General Election forecast is back live!

Making any stats product understandable to the random person on the street is always hard, but the Ghibli GPT-4o craze did give me an idea.

👉 blog post on the highlights + some other musings: https://t.co/JCxZbk56Q4

[Charles Mercieca @Charles_Merc]

You can get from raw audio file to an image like this in circa 50 lines of python code. I play around a bit with NOAA APT here: https://t.co/e3Ly8kCpOW

[Charles Mercieca @Charles_Merc]

@OwenWntr In any case, wouldn't it need to be plotted acros multiple elections to be valuable? Maybe this narrowing has been happening every cycle.

[Charles Mercieca @Charles_Merc]

Bayesian search theory in action. This example is straight out of the literature from its use in the incredible story of the hunt for the wreck of the USS Central America. If finding old shipwrecks using statistics is your thing: https://t.co/a5gxHjG63F

[Charles Mercieca @Charles_Merc]

@AJax1729 @QPollsandnews Quite a bit. Nate Cohn had given the same raw data to different pollsters in 2016 and results were from Clinton+4 to Trump+1: https://t.co/V2I0nKAztS

[Charles Mercieca @Charles_Merc]

A few more of these in my latest post: https://t.co/xOytXrLvs4

[Charles Mercieca @Charles_Merc]

We don't usually get granular breakdown out of General Elections, but we do with Local Council ones, which means you can plug in all sorts of demographics and tease out effects like this: PN share of vote goes up in localities where households said English was their 1st language

[François Valentin @Valen10Francois]

French pollsters are so bored they literally tested Napoleon (yes that one).

He'd beat Macron 62-38.

[Charles Mercieca @Charles_Merc]

Putting that polling 'miss' in context, and a brief aside about methodology: https://t.co/uKPrMurkSh

[Charles Mercieca @Charles_Merc]

@Zemploid If donkey vote is so prevalent, why did 3 A's, 2 B's and 2 C's, all preceeding the A's, B's and C's that did get elected in rank on the ballot, not get elected?

[Charles Mercieca @Charles_Merc]

@MartinBonnici3 I do agree that some interventions will change voting, but realistically, even if third parties x2 or x3 current support that's still 2-3 seats out of 65.

On hunters, I'd say considering how some districts are already very rural, there's no reason they can't manage it as is :)

[Charles Mercieca @Charles_Merc]

https://t.co/yHswXw6HV9

[Jaehwan Cho 조재환 @hohocho]

Never expected this flying suit theme exit poll graphic for parliamentary election tonight here in S.Korea.

[Charles Mercieca @Charles_Merc]

Credits and thanks to @EuropeElects whose polls I used & @JorgeAlbuja whose electoral R package did a ton of heavy lifting.

[Charles Mercieca @Charles_Merc]

More of a self educating excersise on the European Parliament than anything, but I wanted to see how hard it is to have a country poll to EP seat forecast. If you have any ideas on a more elegant method, I'm all ears!

https://t.co/6DKrX64eBv

[Charles Mercieca @Charles_Merc]

But in summary, this approach is to:

1) smooth national polls
2) add historically observed error
3) get plausible election day vote share
4) cast this into seats
5) group local seats to EP group.

Moneyshot:

(Anything other than EPP or S&D is a bit overstated because 'Others' should really be resolved into one of those existing groups).

[Charles Mercieca @Charles_Merc]

@Izengabe_ @SeanTrende As long as you vote for your number 1, can't you leave the others empty? How would that result in winning due to random choice?

[Charles Mercieca @Charles_Merc]

@OSPpatrick I swapped geom_text for geom_raster. Not sure if it made it any prettier.

Really great intro into a usually very math dense subject!