Comments on: Voters use child-like judgments when judging political candidates

By: Warren D. Smith

Warren D. Smith — Tue, 10 Mar 2009 16:52:50 +0000

“72% prediction accuracy” among 57 pairs? That’s 41 right guesses and 16 wrong guesses. The chance a fair coin, tossed 57 times, comes up heads 16-or-fewer times, is 0.0632%.
So this result is significant at the 99.94% level.
This is horrible for humanity if your appearance is uncorrelated with your competence. But is it? Perhaps
humanity can survive because more-competent-looking people actually ARE more competent? I doubt it, but maybe.
So I suggest re-doing the experiment to find that out, using
fields in which competence can, in fact, be objectively judged. For example, do pairs of chessplayers with known
FIDE numerical chess ratings. How often can you tell from
their pictures, who has the higher rating?
I can tell you this: if I were asked to look at pictures of two people and asked to tell who’d finish first in a
10km footrace, or weightlifting competition, I bet I could achieve pretty damn good accuracy. So in some fields at least, predicting competence from appearance is quite possible.

By: Arno

Arno — Fri, 06 Mar 2009 10:11:23 +0000

I am talking about voters, that probably explains the diverging results. My naive individuals always pick the better looking person A, thus A is the predicted winner. The question I am asking is to what extend one can infer about the voters from the accuracy of the prediction.

By: Ned

Ned — Thu, 05 Mar 2009 11:22:26 +0000

Thanks Arno; however, you are making a big mistake here–we are not talking about voters but naive individuals whose data we use to predict voters’ choices. We are talking about statistical accuracy and not an absolute result.
Antonakis and Dalgas looked at predicting an election result from individual data. If you took a sample of naive raters, and using your logic, 510 would go for A and 490 for B. As I mentioned, the difference in terms of votes, using a chi-square test is not significant. Thus, using your logic, there is NO WAY a sample of individuals who are predicting a specific race would be able to do better than chance, which is what the chi-squared tests above shows. Thus, the Antonakis and Dalgas results are very impressive.
Sorry, but your logic is flawed.

By: Arno

Arno — Thu, 05 Mar 2009 00:29:48 +0000

With 20 voters always in favour of A, A needs atleast 481 votes from the remaining 980 voters to win. I plugged the numbers into a cumulated binomial distribution, which should yield a probability of about 70% for A to win.
You are right that the expected percentage of votes for A does not exceed that of B by great extent, but the actual numbers of votes are not relevant here, only which candidate will win.

By: Ned

Ned — Sun, 01 Mar 2009 15:49:02 +0000

As for the comment by Arno, 2% of 1000 is 20, right? If the rest vote randomly, we have 980 voters split equally between the two candidates, thus giving A 20 + 490 = 510. B gets 1000-510 = 490. Thus A and B are split 51-49%. With a chi-square value of .40, the differences is not significant between those proportions at conventional levels. How you get 70% for A. Could you explain your reasoning explicitly, showing your calculations?

By: Arno

Arno — Fri, 27 Feb 2009 22:30:10 +0000

I did a short calculation, assuming a total 1000 voters deciding between two candidates. A percentage of p always chooses candidate A (the “competetent looking” one), the rest of the voters decides randomly whom to vote for. If p is 2%, then candidate A will already win with a probability of 70%.
Of course, the model I used is incredibly crude, but it shows that the data given above is not sufficient to support any claims about the rationality of the individual voter, as a really small percentage of voters can level a great influence.

By: Ray Ingles

Ray Ingles — Fri, 27 Feb 2009 15:45:41 +0000

Reminds me of what Charles Stross wrote in Accelerando – “[Humans are] as stupid as it’s possible for an intelligent species to be – there being no evolutionary pressure to be any smarter…”

By: Arno

Arno — Fri, 27 Feb 2009 14:40:40 +0000

Given that 70% accuracy in a binary decision is not that much, and that the winner in an election is usually determined by a rather small percentage of the total voters; it would be interesting to see a calculation how large the effect of perceived competence on the individual voter’s decision probably is.

By: Ed Yong

Ed Yong — Fri, 27 Feb 2009 11:12:31 +0000

Firstly, the judgments were made on competence, not attractiveness. Secondly, you would indeed expect that if asked to pick between two faces, you’d pick the one you prefer. But obviously, voting is meant to be about more than facial preferences – you would expect voting decisions to change as you discover more relevant information about the people concerned. So the interesting bit is the fact that the snap-judgments based on the faces are a reasonable predictor of actual election outcomes.

By: elliot

elliot — Fri, 27 Feb 2009 10:56:01 +0000

Interesting – but not surprising. We choose the face we like; the more attractive one. The one we’d be more likely to choose as a mate or protector. is that unexpected?
Of course, many voters have no idea what the politicians on a ballot paper look like, and choose based on a nice name or bring from the correct party.