Slashdot Mirror
Voters Swayed By Candidates Who Share Their Looks
iandoh writes "Stanford researchers have found that voters are subconsciously swayed by candidates who share their facial features. In three experiments, researchers at the Virtual Human Interaction Lab worked with cheap, easy-to-use computer software to morph pictures of about 600 test subjects with photos of politicians. And they kept coming up with the same results: For the would-be voters who weren't very familiar with the candidates or in perfect lockstep with their positions or political parties, the facial similarity was enough to clinch their votes."
It drives me nuts that Slashdotters always bring up "correlation does not imply causation" any time any sort of experiment is mentioned even if no one is even trying to assert a causative relationship.
"Correlation does not imply causation" seems to be one of those ideas that a lot of people seem to somehow be proud of knowing and as such try to apply even when they aren't needed. Other examples of these sorts of ideas on Slashdot are the term "prior art" and car analogies.
That may be happening because the headline's misleading (as usual); it should be
/. wouldn't get as many "Um, wha??" clicks, and the more cynical of us would tag the story "noshitsherlock" ;-)
Undecided Voters Swayed by Candidates Who Share Their Looks
But then,
Hey -- There's your idea for a social experiment!
"correlation is not causation" is one of the most overused mantras of slashdot users who want to be more skeptical than thou. Yes it's true that correlation does not always equal causation, but causation does tend to result in correlation.
It's been shown that people are more attracted to people with similar facial features when choosing mates, it makes some sense that people would feel better about choosing a leader with similar facial features for the same biological reasons.
Now I'm not saying that this hypothesis is clearly true, just that we don't have to jump all over it.
It's sad when choosing an installation directory on your own qualifies you as an "advanced user."
I totally agree. Every single scientific article reporting "A linked with B" gets this ridiculous tag. Almost no scientist every says "A causes B" because they obviously already understand that correlation does not imply correlation. However, correlation also does not imply "not causation." Any reputable scientist and journal will report results of the form "Here is the data. A appears to be statistically linked with B. Here are several hypotheses as to why, however these are speculative and require further study."
Furthermore, causality is something that a lot of very smart statisticians do spend a lot of time studying. It's not inconceivable that in the future people will be able to make concrete statistical statements about causality with confidence intervals and the works. What will the mantra be then?
Anyway, correlation's not *that* good of a measure of (interesting and nonlinear) dependence between (non-Gaussian) variables anyway. Mutual information is the ticket.
Ok, done with my rant.
Nonperiodic Central Trajectory
Nah I meant generally ignorant of the world he lives in, pretty stupid, probably can't stay awake during a meeting on the economy or science, and is no where near qualified to run a country.
Slashdotters are right to point out the Correlation is not, and never will be causation. Never, never, never, never, never. If you want to show causation, then you must have a model and you must subject it to experiment. Experiment! Not statistical mumbo-jumbo.
I think you are wrong. Epidemiology and observational science have given up a lot without the need for experimentation (we know smoking causes lung cancer, though this has never been directly established through an experiment, since it would be massively unethical). Correlation does imply causation, as I've pointed out in an earlier comment, the hard part is working out what the causal relationships are (ie A->B, B->A or C->A and C->B, these are the ONLY explanations for statistically significant correlation).
The reverse possibility B->A here is nonsense, because voting patterns cannot affect your looks, and the way this study was conducted (you can read the details) pretty well rules out the confounding factor 'C', leaving us with A->B as the only plausible explanation.
I'd like to see you try to refute this (without resorting to insults or rhetoric), particularly if you can think of a way for variables to be correlated without some form of causal relationship as I've described.
What is that you were saying again?
Right, because WHITE (R) people don't like any BLACK people.
Could it be that more blacks aren't (R) because of the hatred spewed against those few black people who are (R)? The vile vitriol spewed against people like Clarance Thomas and Michael Steele is simply amazing.
Also, look at who the NAACP supports, and in cases where it is a white (D) vs a black (R), they go with the white (D) everytime. I guess it color only counts if you're a (D), huh?
Agent K: A *person* is smart. People are dumb, stupid, panicky animals, and you know it.
OK first thanks for taking me seriously. I am actually quite flattered that there is now a webpage dedicated to describing how much of an idiot I am.
I also hope you have the good grace to post my rebuttal to both your arguments, first that science cannot be conducted without experiments, and second that correlation does not equal causation. You have my permission to publish this so long as you do so in its entirity.
I'm going to give you a short CV, just so you know (not sure it's relevant but anyway). I have a degree in mathematics, a masters degree in mathematical statistics, a PhD in evolutionary biology (my thesis topic was along the lines of 'what can we infer from comparisons of gene orders of extant species') and I've worked for four years as an epidemiologist on a observational study of health and cognition. It's fair to say that over the past ten years I've thought about the ideas of correlation, causation, and inference for a living. If I had any doubt that what I was doing was fundementally flawed from a scientific point of view I wouldn't do it.
Regarding your Saturn example, well you've managed to find two things that increase with time, but are quite clearly unrelated in every other regard. You have calculated their correlation as 0.88, suggested that I would draw the conclusion that one causes the other, which is plainly absurd, therefore my argument that correlation implies causation is incorrect.
There are two ways I will respond.
The most obvious is that you did not read my argument. I claimed A->B, OR B->A, OR C->A and C->B. Clearly here we have a correlation, so one of these must be true. A->B and B ->A are both obviously silly, so we are left with C->A and C->B. Well what could 'C' be? Here it helps that you've not plotted A vs B as would be traditional to illustrate a correlation, but you've helpfully plotted A and B against a third factor, 'time'. In this case C=time, the passing of time has caused the stock market to increase and has caused Saturn to do whatever it did (I'm not an astronomer). If you do a regression of A vs B adjusting for time I'd be pretty sure you'll see the correlation would be gone.
Second, (and this is a more minor subtle complaint) there is the issue of statistical significance. I don't know but I'd bet the correlation you showed does not hold much outside of the small window you've showed it, and that you've selected this particular example to illustrate your point. If you give me any two time series I could probably find a small window in which they are both increasing, so that correlation is statistically meaningless because of multiple testing issues (note I qualified my initial claim with the words 'statistically significant')
Next, I absolutely agree with you that experimentation is the gold standard of scientific research. I cannot accept however that it is the only way to draw conclusions. Much of science cannot be tested experimentally because it would be impractical, unethical (as with most of the work I do) or just plain silly. My earlier example 'lung cancer is caused by smoking' is a good example of a purely observation finding that was totally unexpected at the time and was found simply on the basis of observing the smoking patterns of people in lung cancer wards compared with others. The big prospective studies came much later, and experiments will never be done, yet I'm sure you would accept this finding as true.
I also agree with you that most science posted on Slashdot is rubbish, for a variety of reasons, mostly because science progresses in very small increments, and so on its own no paper is ever really newsworthy, and has to have its significance bloated out of all proportion to get into the news (ie they fail the 'so what' test). However faulty causation is not often the culprit, because most scientists are very good at adjusting for potential confounders in their relationships (and journals are very go