Morning all -- and hello new followers!
There are now twelve (and counting?) different models making seat-by-seat predictions for next Thursday's election, and while they all indicate a Labour landslide, there's a lot of disagreement in the details.
Hello everyone, less than 48 hours until the polls open...
This morning I've added a "coefficient of correlation" to the aggregator, to indicate how much the models agree or disagree in each seat:
Fellow nerds: I built a thing to compare general election predictions from different constituency models (Britain Predicts, Electoral Calculus, Economist, FT)
So, while there are only 11 constituencies where all models agree the Conservatives will win, there are 57 where at least nine models predict they will. (cc
@zoenora6
)
To my bafflement, and my family's amusement, I've been invited to talk about election predictions (and to share my own?!) with
@rosiewright99
on Times Radio at 5.45 tomorrow morning...
Lunchtime update: a couple of new little features to play with this evening...
Firstly, a new page that breaks seats down by predicted margin -- look how many Conservative seats are predicted to be held by less than 5%
Just a quick update this morning: to reflect the thin margins of many of the predictions, I've made a tweak to the colours on the site.
For instance, here are all the seats where there's a prediction of the Greens winning tomorrow:
@colinrtalbot
Right! I've got plans to evaluate the predictions, based not just on who got the most seats right, but on who was best able to model trends across the country. Watch this space!
Now on the site:
More in Common have always been the MRP that's given the Tories the biggest vote share, and even they are now only predicting 126 seats
Hello everyone, less than three days until the polls open...
This morning I've added a new feature to my site that aggregates seat-by-seat predictions: you can now rank constituencies by the vote share or majority for each party
🦕 Here is our
@TheEconomist
/
@wethinkpolling
MRP!
And
@JamesFransham
has updated our Britain prediction model methodology.
This is how we're thinking about the election (with maps + charts!) 🧵
Was in a pub last election night with lots of Labour activists who'd spent the day only knocking on doors of probable voters, and there was genuine shock when the exit poll was announced at 10pm
Also: I'm expecting several more final predictions to be released today, and will do my best to get them onto the site as quickly as I can, around family and work commitments.
I don't really know what's going on in Exmouth and Exeter East, but it's the only seat where four different parties are predicted to win -- anyone got any local knowledge to share?
As a anecdote have driven around Somerset and rural Glos the last day or so, passing through 5 marginals and.... I've not seen a Tory sign. Loads of Green/Lab and, in particular, Lib ones. But not one Tory.
Consituencies: Forest of Dean, North Somerset, Wells, Hanham, Weston
Actually as much as this is a pain in the arse after coming in from the pub: hats off to Survation for acknowledging the limits of their model like this
I've just pulled in the latest data from several sources, including tonight's Survation MRP and the latest Electoral Calculus data, which seems to have calmed down a bit -- they're no longer claiming North Cotswolds for Reform...
NEW MRP: Labour 99% Certain To Win More Seats Than in 1997
Labour on Course to Win 484 seats.
The Conservatives and Liberal Democrats are in a close race to form the official opposition.
Probabilistic seat count:
LAB 484
CON 64
LD 61
SNP 10
RFM 7
PC 3
GRN 3
34,558 interviews
Excited to be running a coding competition at
#PyConUK
this September. It should be accessible and fun for new programmers, and a bit of a challenge for those with more experience.
The Datalab is hiring engineers:
Our work is important and interesting, and has a real impact. We get stuff done.
Please spread the word, and let me know if you have any questions.
WE ARE HIRING
We need a DJANGO developer for
@EBMDataLab
We are fun, our outputs are awesome and change the world. We make and many more similar in pipeline.
TELL YOUR FRIENDS.
E: hello
@ebmdatalab
.net
In normal circumstances, obtaining permission to look at sensitive health data would take months or years. A new team has gone from idea to publication in just 42 days
There's still one seat outstanding, and it's a fitting one -- I wrote much of the code that supports interrogating the data in a tent on Skye only three weeks ago
My first company made firewalls for telecom systems. I was a load tester. I kicked off a long-running load test and went off to get married. Returned from honeymoon to discover that I'd routed all traffic through company's own corporate firewall. Oops.
This is your unscheduled reminder that telling early-in-career engineers stories of times you messed something up real bad is a good way to help them combat their own impostor syndrome.
🚨Final MRP estimates from
@focaldataHQ
is for a 238 seat Labour majority with Labour on 444 seats, Conservatives on 108 seats, Liberal Democrats on 57 seats, SNP on 15, Reform on 2 seats
I'm also going to be co-ordinating the John Pinner Award, where we'll recognise contributions to the UK Python Community in memory of John, who founded PyCon UK. More details on this very soon.
Poking around the data from
@inglesp
's excellent election prediction aggregation.
The seats where prediction models disagree most:
1. Ashfield
2. Islington North
3. Waveney Valley
4. North Shropshire
5. Cambridge
@JonnElledge
Down to 13 (JLP had NE Hampshire as a tie which I was showing incorrectly), and by one measure, Brentwood and Ongar is the most comfortable -- it's got the highest smallest predicted Tory majority of 3.7%, if that makes any sense
Was in a pub last election night with lots of Labour activists who'd spent the day only knocking on doors of probable voters, and there was genuine shock when the exit poll was announced at 10pm
If you watch back the 1997 election night broadcast you will see much discussion that looks exactly like this - “feeling on the doorstep much more positive than the polls.” Then the results come.
Using a combination of
@SlackHQ
, an RSS feed from
@PSNCNews
and our very own ebmbot (built by
@EBMDataLab
) we can get personalised price concession alerts to you within 7 minutes - if we ask ebmbot nicely. Sign up for yours at
North Shropshire had the massive swing to the LDs in the 2021 by-election; North Herefordshire and Waveney Valley are both Green target seats; Ashfield had a strong independent candidate in 2019 and has Lee Anderson's defection to Reform; and Islington North has Corbyn
OUR NEW PAPER.
Factors associated with COVID-19-related hospital death in the linked electronic health records of 17 MILLION adult NHS patients.
Largest study of its kind ever, anywhere.
The power of UK / NHS data, realised.
However, the increase in no. of MRPs makes me a bit uncomfortable, because they’re so open to misinterpretation (as is basically any form of modelling).
Differences in vote share by fractions of % point could make a big diff to seat outcomes, as many pollsters have already said.
@inglesp
@fascinatorfun
They cannot agree a model for the Claire Wright supporters from the East Devon part of the constituency. She polled very highly in 2019. Her supporters are almost certainly going to be LD/Labour leaning.
Evening all! I've been away from my desk for the day and so I've only just fixed a couple of bugs with how I was showing the results: two seats were incorrectly marked as being won by "oth" candidates, and three close races were marked as ties. Thanks to all who let me know!
Drumbeat of mortality: the first release of Python (20th February, 1991) was as close to the date that the first COBOL spec was approved (8th January 1960) as the present day...
So, while there are only 11 constituencies where all models agree the Conservatives will win, there are 57 where at least nine models predict they will. (cc
@zoenora6
)
I've not got around to presenting trends on my site but
@sib313
has been pulling in my data and has some some pretty charts where you can see changes in predictions over time
WE ARE HIRING AGAIN.
We're building openpathology and more features on openprescribing, so we need another RESEARCHER, and another SOFTWARE ENGINEER. We produce fabulous tools, services, papers, and policy shouts.
Fun team, epic mission.
JOIN US.
I'm looking forward to speaking at
@djugl
on 25/10. I'll be talking about mental models and leaky abstractions, and about how if we are to get the most out of the Djano ORM, we need to understand what's going on under the surface. Londoners: see you there?
***SPEAKER ANNOUNCEMENT***
I am delighted to announce that Peter Inglesby (aka
@inglesp
) will be one of four speakers at
#DJUGL10
with his talk on Leaky Abstractions and the Django ORM.
Tickets are still available here:
Hello world, I've been away from the internet for a week, climbing and dodging the rain on Skye. Many thanks to
@themadwort
for keeping the election prediction aggregator updated.
Also, you can now show the vote share and majority of the predicted winner (rather than the vote share and majority of a given party) -- click on the column heading to sort
Of the 123 seats (outside NI) left to declare, only 4 have any models predicting Reform winning (and I'm still baffled by North Cotswolds on this list)
I'll follow this up with a write-up evaluating how the models performed, and whether there's anything we can learn from the proliferation of predictions