Thread by @centrist_phone

These are the factors that will win or lose the election:

1) Turnout by age
2) Lab leavers
3) Tactical voting
4) Con remainers

And in terms of predicting the outcome we also need

5) The accuracy of the polls

Let's predict all five - and make a final call prediction.

1/19

Firstly though, tonight's polls.

Treat 'final polls' with the contempt they deserve.

The final polls are nearly always wrong. And they're always wrong in the same direction (too Lab or too Con, uniformly).

This shouldn't happen. But there's a reason it does: herding.

2/19

Throughout the campaign, we've seen huge variations in the polls.

I haven't seen all the final polls, but I'll make a prediction: you are unlikely to see that wide variation tonight.

They will all magically cluster, probably around the median of 9%.

Why? Fear.

3/19

Being the outlier showing a 3% Con lead, or a 16% Con lead, is scary for polling companies.

The polling company will be terrified of being ridiculed and losing business, because of damage to their reputation.

Much better to 'tweak' a few numbers.

Herding.

4/19

This is one of many reasons why the polling industry's reputation is trash.

But they brought it on themselves.

So - don't treat the final polls any more or less seriously than any other poll.

If anything, treat them with more scepticism because of herding.

5/19

Now - turnout by age.

In simple terms, the higher youth turnout and the lower 65+ turnout, the better Lab will do.

If we get GE 2017 turnout levels on current polling averages, we're into the edge of hung parliament territory.

So what might it be?

6/19

Signs are positive for GE17 levels of youth turnout, and lower than GE17 levels of 65+ turnout.

If you compare YouGov's crossbreaks from GE17 and GE19, you can clearly see there's been a voting enthusiasm drop in the 65+ demographic.

If this holds: ADVANTAGE LAB

7/19

Next - Labour Leavers

These voters will determine if Labour retains or loses up to 30 seats.

Lab started out with just 8% of Leave voters - that's now up to 16% (close to GE17 levels of 20%).

It needs to edge a little higher still to fully secure the Red Wall seats.

8/19

Signs are positive. Lab leavers ARE returning home.

In the last week, there's been a net +18 improvement in who Lab Leavers would rather have as PM, in favour of Lab.

But, there is still a bit more to do. So at the moment: ADVANTAGE CON

9/19

Onto tactical voting.

The YouGov MRP has picked up remainer tactical voting.

Labour's vote share is up considerably in more remainy Con/Lab marginals as LD's/Green lend Lab their vote.

Tactical voting is being encouraged by targeted social media ads as never before.

10/19

This makes the level of tactical voting hard to predict.

People in tight marginal seats are MUCH better informed about who is in second place anyway.

With that, plus constituency-specific tactical voting ads, we are in new territory.

11/19

Remainers have good reason to hold their nose and vote Lab or LD.

So voters have the information, and motivation, to tactically vote.

How much will there be? 43% of remainers have said they will vote tactically. Enough to potentially flip over 30 seats.

ADVANTAGE LAB

12/19

Onto Con remainers.

Con vote share in the south has plummeted.

It's opened the door for Labour to make up to 10 gains, and the LD's to make up to 8 gains.

18 less Con seats is a big deal.

So far, no indication of any remainer movement back to Con.

ADVANTAGE LAB

13/19

In terms of polling accuracy, I have already revealed that the YouGov MRP uses 2015 GE weighting for age-based turnout.

This is just one symptom of several potential polling errors.

- turnout
- too many leavers in the weighted data
- downweighting of new voters

14/19

I have trawled the polls and I can find zero potential errors that would favour Lab.

All potential errors that it is reasonable to identify favour Con.

So Labour only need one of these polling errors to be correct, and we're sliding toward hung parliament.

ADVANTAGE LAB

15/19

In terms of a prediction, we have to consider the likelihood of the following:

A) Will turnout equal GE17?
B) Will more Lab leavers return home?
C) Will there be high levels of tactical voting?
D) Will Con remainers remain 'disloyal'?

On the balance of probabilities...

15/19

On current polling, adjusting for errors, these are the predictions:

Scenario 1L Worst case for Lab
A) No
B) No
C) No
D) No
Result: Con majority (30ish)

Scenario 2: Bad for Lab but possible
A) Yes
B) No
C) Partially
D) Partially
Result: Paralysis (Con on ~325 seats)

16/19

Scenario 3: Evidence based balance of probabilities
A) Yes
B) Partially
C) Yes
D) Yes
Result: Lab minority government (Con 310 seats)

Scenario 4: Best case for Lab
A) Yes
B) Yes
D) Yes
D) Yes
Result: Lab largest party (Con 280 seats)

17/19

My probability estimate is

Scenario 1 (small Con majority): 10%
Scenario 2 (paralysis): 40%
Scenario 3 (Lab minority govt): 55%
Scenario 4 (Lab largest party): 5%

18/19

My 'best guess' specific prediction:

CON: ~310
LAB: ~255

Labour minority government, which lasts six months with SNP/LD support to deliver a second referendum.

There is cause for hope, and fear. I think I'll need a valium to sleep tonight!

What are your predictions?

END

CORRECTION (sorry, 110% isn't a thing)

My probability estimate is

Scenario 1 (small Con majority): 10%
Scenario 2 (paralysis): 40%
Scenario 3 (Lab minority govt): 45%
Scenario 4 (Lab largest party): 5%

REAL END