Jedit posted:

And Corbyn should be fighting on that basis: that May has no mandate to broker a deal which takes us out of the single market. This is something the Tories couldn't legitimately protest and would shore up the assertion that Labour want what is best for Britain as well as respecting the will of the people. But he isn't. He's letting her do as she pleases because it's easier.

TinTower posted:

TinTower posted:

I get 18%. Labour are on 30% with Remain voters.

I'd be interested in seeing new VI figures after the bill. Might have turned off a lot of Remain/Labour voters.


The most important bill possibly since the Human Rights Act.

There's also the Snoopers Charter, where the former head of Liberty failed to vote against…

Saith posted:

You're a Liberal. If we cut through the bullshit, what's left?

Fangz posted:

Well, like, I said, the fiscal rule stuff. Labour's adopted the position, contrary to all economic sense, that governments need to commit to deficit balancing.

http://www.taxresearch.org.uk/Blog/2016/08/01/a-note-to-john-mcdonnells-economic-adviser-on-fiscal-rules-and-their-desirability/


So once the tax take reduces due to Brexit (and note that tax take reduction here is *not* cyclical), Labour's own fiscal rule means spending must be constrained. Labour will doubtlessly try to push for a balanced program of tax increases and spending cuts (like Osborne's punishment budget) but this will be rejected, and then we'll be in a pretty fair approximation of the brexit situation where the only way to stick to Labour's own rule would be to vote in favour of the Tory austerity plan.

Fangz posted:

Stop pretending you understand what those terms you are using mean.

TinTower posted:

The polls in the American election were actually relatively accurate in predicting Clinton's popular vote share.

Spangly A posted:

if corbyn's support removed immediately from the a50 bill, would it make any difference?

would it make any difference to the outcome if he whipped against?

If so and only if so can you reasonably say he "propped them up". A rock holding a door open is propping it. If you take the rock away and the door doesnt budge, it wasn't propping it.

Lord of the Llamas posted:

Go on then, "educate" me this'll be a laugh.

Lord of the Llamas posted:

Nationally yes but on on a state level no. The distribution of DKs and their demographics goes a long way to predicting the actual result more accurately.

Private Speech posted:

I look forward to Labour voting with the government on every bill they can push through, then.

This is a pretty poo poo argument whichever way you feel about Labour and Brexit.

Fangz posted:

For a start, what do you mean by this.

Fangz posted:

For a start, what do you mean by this.

Or actually, explain what 'margin of error' means to you.

Spangly A posted:

no, we have a pretty poo poo government


undecided voters still vote and massively outnumber the educated. Their likelihood to vote and the constituency they are in, if calculated, combine to form a better reading of the climate than raw decideds. Anyone who has "decided" to be a tory is clearly a loving idiot. Most people just tag along because the tories they know are abrasive loudmouths

quote:

are you defending statistical methods that arbitrarily exclude the larger parts of their sample?

Fangz posted:

For a start, what do you mean by this.

Fangz posted:

Or actually, explain what 'margin of error' means to you.

Fangz posted:

Congratulations, this is why you look at the headline figures that make use of voter likelihood models.

Private Speech posted:

I look forward to Labour voting with the government on every bill they can push through, then.

This is a pretty poo poo argument whichever way you feel about Labour and Brexit.

Lord of the Llamas posted:

Essentially that because a large number of the undecideds (of which there was a lot more than your average US election) in the states that Trump picked up the most last minute support not accounted for in the final headline results were soft R leaners that the potential variation of Trump's performance was higher and skewed towards outperforming the poll number.

quote:

It's how close we expect the actual result to be given a certain likelihood.

Fangz posted:

No, *you are*. If you are trumpeting raw values for a party as a percentage of votes with the denominator including undecideds then you are making the very strong assumption that undecided voters don't vote.

Fangz posted:

Nope. It's an estimate of the inherent variability of the results on repeated sampling subject to assumptions. Voter likelihood methods and demographic reweighting are introduced to *reduce* the contribution of sampling variation to MoE.

Spangly A posted:

no I'm the one saying that raw values are dumb

Spangly A posted:

no I'm the one saying that raw values are dumb


are you trained in probability?

Fangz posted:

As a lazy hypothetical, is there actually a point at which Corbyn would be perceived to have crossed some kind of red line for people?

TinTower posted:

Most polls in the final week had Clinton winning the popular vote by 2-4%.

She won the popular vote by 2.1%.

Yorkshire Tea posted:

At this point he could probably murder babies in their sleep and people would claim that these are babies being brought up under the oppressive yoke of capitalism and that killing them is a mercy.

Nonsense posted:

Hillary lost horrifically and is probably one of the worst to ever attempt a run at the office. Corbyn finished.

Spangly A posted:

did she win the presidency

MikeCrotch posted:

Does anyone know how the Colombian government's decision to ignore the result of the FARC referendum and sign a peace treaty anyway went?

Fangz posted:

Incorrect. The undecideds were accounted for, but they were accounted for poorly (they used past elections as precedent, but turns out the past wasn't a great guide). The undecideds introduced additional unexpected variability into the system, but the whole point of variability - especially ones without a lot of precedent - is that you can't use the undecideds to generate better predictions.

Fangz posted:

Nope. It's an estimate of the inherent variability of the results on repeated sampling subject to assumptions. Voter likelihood methods and demographic reweighting are introduced to *reduce* the contribution of sampling variation to MoE.

TinTower posted:

Getting more votes for president than any white man in history = losing horrifically, gotcha.

kingturnip posted:

Last BBC article I read suggested that the UN were expecting the last of the FARC rebels to enter a designated zone by yesterday, I think.
In the meantime, the government have started negotiations with the second-largest group of rebels.

Because in some countries, the government realises that the populace are stupid, inbred ratarses and should be ignored for their own safety.

TinTower posted:

Getting more votes for president than any white man in history = losing horrifically, gotcha.

namesake posted:

This thread is really showing the truth of that Labour polling which said position on Brexit was more important than previous party loyalty.

Lord of the Llamas posted:

That's a very strong claim and strange to make. I would wager that past voting behaviour is indicative of which way an undecided might break. Also you could just ask them what way they lean. Any model that used the proportion undecideds to increase the variance of it's estimates even if it didn't move the means would be arguably more accurate than one that didn't. Otherwise you're basically claiming that a poll of 100 people with 0 undecideds has the same variance in its estimate as a poll of 1000 people with 900 undecideds.

quote:

How is what you're saying any different from saying it is how far our sampled estimate is expected to be from the population parameter a given proportion of the time? It basically sounds like you know some stats and have just memorised some frequentist terminology and don't actually get what I'm saying.

Fangz posted:

On the general point that polls suck especially in the UK system, I totally agree. But they are pretty much the only data we have at this point, and we need to be very careful about the temptation to throw out data because their conclusions are inconvenient. Also people who ridiculously mishandle the data suck.

The general picture of Con > Lab > LD = UKIP looks pretty good, so is the story that Lab is going down and LD is going up. Is the percentage Lab is going up and LD going down 'a lot', well that's a qualitative assessment that is pretty boring. On the question of how this will change over the next election, who the gently caress knows, I don't think things look good for labour.


This is just word garbage.

Yes, you can use assumptions on how undecideds vote to estimate the final result, also using prior voting data. Each company does this their own way. 'Increase the variance of its estimates... is more accurate' is just handwavy gibberish. Of course I don't claim that a "poll of 100 people with 0 undecideds has the same variance in its estimate as a poll of 1000 people with 900 undecideds".


It basically sounds like you are an idiot. This is the fundamental difference between between frequentist and bayesian stats, kid.

Also just between us, MoE estimates from polling companies are generally bullshit, but hey, whatever.

Fangz posted:

None of this says talking about raw values is better than talking about the headline values.

Fangz posted:

It basically sounds like you are an idiot. This is the fundamental difference between between frequentist and bayesian stats, kid.

Lord of the Llamas posted:

You missed my edit so here it is:

Maybe the easiest way to explain the general point is this: A binomial distribution does not have the same variance as a beta-binomial distribution. Headline polling figures are a result of a complex hierarchical model and therefore the nature of the variance of its estimates is also complex but not impossible to model.

Your position appears to be that these things cannot be understood and therefore we shouldn't bother and just ignore them.

And of course a model is more accurate if it has more variance when estimating an uncertain event than a more certain one. This mistake is why so many crappy polling aggregates gave Clinton a 99% chance of winning.

Fangz posted:

No, my position is that we should use the headline values which represent the polling company's attempt to correct for all sources of error using demographic data and voter intention data which they do not otherwise make generally accessible. Rather than... what, exactly?