Shock Horror - Katie Hopkins States Something Everyone Can Agree With ...

[andy-hughes]

“Pollsters have demonstrated - once again- why they are an utterly defunct source of information.”

You can't argue with that!

[andy-hughes]

I managed to read it effortlessly, and I wear seeing and reading glasses.

Specsavers perhaps?

[jim360]

Just in case it wasn't clear, I am talking above about random polling error. There is also the possibility of systematic errors due to the way polls are conducted, or how samples are chosen, or even risks of pollsters who massage the raw data to generate results they were wanting to get. All of these could also lead to errors.

The fact remains, though, that the 2016 miss is not as huge from a statistical point of view as it's been made to look. Clearly the difference in practice is a lot more significant. Polls predicted that a narrow win for Clinton was more likely; we got a narrow win for Trump instead.

[jim360]

haha, AOG :P

It's a bad habit I have to go on maybe a little too more than I need to. Generally though I think it's better to say too much, and cover everything, than too little, and miss things that may be important. Hence the long posts I make.

[naomi24]

Jim, //Clearly the difference in practice is a lot more significant.//

As a practical guide the difference is all that is significant.

[Khandro]

jim; //Polls predicted that a narrow win for Clinton was more likely; we got a narrow win for Trump instead.//
So what use are polls? - I told you Trump would win a few months ago.

[ludwig]

// Polls predicted that a narrow win for Clinton was more likely; we got a narrow win for Trump instead. //

I suppose what I'm struggling with Jim is what purpose any of the polls actually served other than to fill newspaper space with inaccurate graphs and pie charts.

[ludwig]

// Polls predicted that a narrow win for Clinton was more likely; we got a narrow win for Trump instead. //

I suppose what I'm struggling with Jim is what purpose any of the polls actually served other than to fill newspaper space with inaccurate graphs and pie charts.

[pixie374]

Similar to weather reports probably. Use them as a vague guide, but don't bank on them.

[jim360]

In this case, for everything other than discussing the accuracy of polls, I'd agree with you for sure. But what bothers me is that, by implication, a poll -- any poll -- that predicted a Trump win was "more accurate", no matter the scale. There were some highly unscientific, poorly-conducted, opt-in polls available around the internet that would have indicated a thumping Trump win by a landslide -- 538-0, even. But such polls are clearly bogus. Are we seriously suggesting, though, that because they predicted a Trump win, and we got a Trump win (at a vastly different scale), they are somehow *more* accurate or trustworthy?!

We have to pay far more attention to the uncertainty attached to predictions, and far less to the central value, than previously. In answer to ludwig -- yes, newspapers are guilty of that. As I suggested in my long post, polls have perhaps become too great a part of the narrative.

Incidentally, Khandro, I also predicted a Trump win (in conversation with several friends on or about June 24th). Doesn't make me superior to pollsters.

[Khandro]

jim; // I also predicted a Trump win (in conversation with several friends on or about June 24th). Doesn't make me superior to pollsters.//

Err.... you and I, (you privately and me publicly) made an accurate prediction, how can that not be 'superior' to the wrong prediction of the pollsters?
Is your scientific acumen finally failing?

[jim360]

Because, as I've explained above, there was a lot more to the prediction. And anyway, there's such a thing as being right for the wrong reasons -- I can't comment on your own prediction, but for myself it's obvious that there was more than a little paranoia, or anger and sadness over the then-fresh referendum result, that was colouring my thinking. That I turned out to be right doesn't detract from this.

Actually, same with the Nate Silver thing. I was tracking that throughout the build-up to the election, and it was always troubling me that Trump's probability of winning was somewhere in the 15-35% range, and I thought it needed to be much smaller than that for Clinton to be secure. But this never happened, Trump's chances ended up being around 30% based on the modelling, and lo and behold, he won. Brilliant political analysis on my part? Absolutely not. Just my pessimistic nature speaking. That doesn't change just because it happened to be correct.

We shouldn't allow dodgy, specious, or just fundamentally broken reasoning to go ignored just because it happens to be right (and even then, only right in a very superficial sense). And we shouldn't toss aside the methods that lead us to reasoned, impartial, and well-researched arguments if they sometimes make mistakes.