Wednesday, November 16, 2016

An Analysis of Why the Polls were Wrong on Trump

Philip Pilkington, “Why the Pollsters totally failed to call a Trump Victory, Why I (sort of) succeeded – and Why you should listen to neither of us,” Fixing the Economists, 14 November, 2016.
A good discussion.

Probabilities can be categorised into the following types:
(1) a priori probabilities (which are analytic a priori and necessarily true), and

(2) a posteriori probabilities (which are contingent and empirical), further divided into:
(a) relative frequency probabilities;

(b) epistemic probabilities.
To cut a long story short, even if the media have well sampled polls (not biased or oversampled), the numerical probability estimates they create from these polls for the probability of an outcome on election day are not objective probability scores in the way that a priori probabilities are objective. The numerical probability estimates that the media and pollsters give us are subjective, even if when they are based on good poll data.

And they certainly aren’t proper relative frequency probabilities of Type 1.a either.

Our estimate of how probable an outcome on election day is likely to be is essentially an epistemic probability.

A more technical discussion of probability here.

Realist Left
Realist Left on Facebook
Realist Left on Twitter @realistleft
Realist Left on Reddit
Realist Left Blog
Realist Left on YouTube
Lord Keynes on Facebook
Social Democracy for the 21st Century: A Realist Alternative to the Modern Left

Alt Left on the Internet:
Alternative Left on Facebook
Alt-Left on Google+
Samizdat Broadcasts YouTube Channel
Samizdat: For the Freedom Loving Leftist

I’m on Twitter:
Lord Keynes @Lord_Keynes2


  1. No discussion is complete without considering how people now view pollsters. This skews who responds, and when they respond, how truthfully they do so. I disapprove of pollsters, and our poll driven politics. I think most polls are intended to distort not inform. Thus when I am polled I usually decline to answer and when I do answer I ALWAYS lie.


    The USC/LA Times poll was spot-on through the election.

    1. You can only be spot on or off at the end. And they weren't, they had Trump up, I think by 5 at the end. The exact count is still undetermined, but Trump was not way ahead in the popular vote.

    2. You shouldn't forget the 3% Margin of Error factor. Investor' Business Daily called it pretty close, at least in a 2-way race:

      Clinton after all was ahead in the popular vote. However, polls don't seem to be very well able to predict how the EC will go.