Skip to content

Delegate at Large

Asher Meir points to this delightful garden of forking paths, which begins:

• Politicians on the right look more beautiful in Europe, the U.S. and Australia.
• As beautiful people earn more, they are more likely to oppose redistribution.
• Voters use beauty as a cue for conservatism in low-information elections.
• Politicians on the right benefit more from beauty in low-information elections.

I wrote: On the plus side, it did not appear in a political science journal! Economists and psychologists can be such suckers for the “voters are idiots” models of politics.

Meir replied:

Perhaps since I am no longer an academic these things don’t even raise my hackles anymore. I just enjoy the entertainment value.

This stuff still raises my hackles, partly because I’m in the information biz so I don’t like to see people mangle information, and partly because I feel that these sorts of claims of the voters-are-shallow variety, do their small bit to degrade the prestige of democratic governance. I’m similarly bothered by the don’t-vote-it’s-a-waste-of-time thing, and the shark-attacks-and-subliminal-smiley-faces-thing and the fat-arms-and-redistribution thing etc etc etc.


  1. idiot says:

    Disregarding the fact that we only live in a “representative democracy”, where laymen are only limited to selecting their “dear leaders” who are then empowered to do as they please…and that there are other forms of democratic governance (“initiatives/referenda” and governance via lot) that could encourage potential voters to more “active” in politics…

    Voters *are* shallow. Or more accurately, voters have to be shallow, because the actual governance of a society is so complex, with numerous intricate policies, contradictory metrics, inter-locking effects, rent-seeking behaviors, etc., etc. to take into account, that any human attempting to fully understand a country will go insane before he can even walk into the voting booth.

    The exact degree and nature of our shallowness is obviously open to debate. But I think it isn’t controversial to suggest that voters cannot be trusted to make the right decisions…because *humans* cannot be trusted to make the right decision. Yes, this admission would reduce democratic prestige, but maybe it doesn’t deserve that prestige in the first place. All governments are flawed, due to their man-made nature (even in some sci-future where humans build Skynet and put It in charge of the US, Skynet would be ultimately fallible, due to the biases and faults of its human programmers).

  2. A.H. says:

    Dear Andrew,

    I am a regular reader of this blog and your other writings, including your academic research. I was quite puzzled about this post of yours.

    I do not quite get it: what is wrong with this piece of empirical work? Is its empirics somehow flawed and if so, why / in what way? Or is just the topic or the style of writing that you do not like? To which issues you referred, concretely, when you said that this is an example of garden of forking paths?

    • Andrew says:


      This question comes up from time to time. The short answer is that for something to be convincing science, it’s not enough that it avoid glaring flaws; it also needs to have clear paths connecting data, statistical model, and real-world claims.

      Over the years, it’s been necessary to discuss examples of horribly incompetent science such as beauty-and-sex ratio, ESP, ovulation-and-voting, power pose, air rage, etc. But remember that each of these cases looked reasonable to reviewers when it came out. We’ve learned a lot about science and statistics from studying these examples carefully, but all this focus on the horribles can lead to the incorrect impression that papers with unsupported conclusions are always obviously ridiculous. Sometimes something can seem like a reasonable piece of work, and it could even be sharing valuable data, but its conclusions are not supported by its evidence.

      In this particular case, my problem is not with the writing style, my problem is with the forking paths. There is a one-to-many mapping from the scientific theory to possible comparisons to study. See section 2.1 of this paper paper for a listing of some forking paths in a similar paper.

      Or, to return to the article discussed in the above post, there are many dimensions of forking paths, starting with the four bullet points shared above, which are actually the first four lines in the paper.

      To expand a bit:

      “Politicians on the right look more beautiful in Europe, the U.S. and Australia.” Lots of forking paths consistent with the theory: (a) same pattern in all 3 countries represents a common pattern; (b) opposite pattern is also consistent with theory, with the argument being that the political left is the default (from media, entertainment, etc), thus candidates from the right can be viewed as the “rebels” and the “conservative” choice is to stay with the left; (c) mixed pattern is also consistent with theory, e.g. if it appears in U.S. and Australia but not Europe, one could argue that the political left is the default and thus the “conservative” choice in largely socialistic Europe, while the political right is the default and thus the conservative choice in the U.S. and Australia; (d) or if the pattern appears in the U.S. but not Australia or Europe, one could argue that mainstream conservative parties in Australia and the Europe are traditionally conservative, while the main conservative party in the U.S. is radical and thus the Democrats represent conservatism; (e) or there could be differences within European countries, with a reasonable theory-based explanation for differences between northern and southern Europe, or eastern and western Europe, or rich and poor countries (the meaning of the political left and right depends a lot on what the government has to work with), or countries with a recent history of left or right governance; (f) and there’s also the option of comparing statistically significant to non-significant results.

      “As beautiful people earn more, they are more likely to oppose redistribution.” This is stated as a within-person comparison with a causal flavor (later in the abstract, “beautiful people earn more, which makes them less inclined to support redistribution,” but their data are entirely cross-sectional, so no information at all on causal effects of earnings for the individual. Also there are a zillion interactions that could be studied here, indeed have been studied in the literature, including interactions with sex, marital status, parental socioeconomic status, and characteristics of the election.

      “Voters use beauty as a cue for conservatism in low-information elections.” This is an interaction. Other interactions could fit the theory too, for example interaction with the closeness of the election. Also of course the main effect would do just fine: typically (but not always) we find interactions in this sort of paper only after the main effect fails to show up as statistically significant.

      “Politicians on the right benefit more from beauty in low-information elections.” Again, lots of forking paths, including the definition of what is a low-information election, the definition of what it means to be on the right, and possibilities of interactions with characteristics of the voters and characteristics of the elections.

      The trouble is that the evidence is all p-values, and these tell us very little in the presence of selection among endless forking paths.

      To say this is not to say that this paper is specially horrible or that the authors are bad people. It’s just standard operating procedure cargo cult science, and I see no reason to take its conclusions seriously.

      • ds says:

        Clearly you did not read past the abstract (probably thought it wasn’t worth your time), yet you feel comfortable claiming the paper’s “conclusions are not supported by its evidence”. Maybe so. Your forking paths concern makes sense in the context of this paper, but it seems to me that your attitude of automatically dismissing this type of study is inherently hostile towards work in the social sciences that tries to go beyond description and examine causal mechanisms in the data (ironic, given this blog’s title). As you point out, “there is a one-to-many mapping from the scientific theory to possible comparisons”, and given that this is observational (not experimental) data in the social (not natural) sciences, this will almost always be the case, because our theory will almost always be weak. I think the intended audience of this paper (economists) understands that papers like this are more exploratory than confirmatory, even if the inferential framework being used is designed for the latter. These people are trying to rationalize a pattern they found in the data, in a way that’s at least internally consistent. Perhaps they are chasing noise, but perhaps not, and in any case, this is part of how theories are built. You’re a political scientist and a statistician. Why not add something positive to the conversation instead of sniping at the abstract?

        • Andrew says:


          You write “Clearly you did not read past the abstract.” I don’t know what you mean by “clearly” here. I actually read the whole article.

          • ds says:

            I apologize. By “clearly” I meant, “it seems to me that you probably didn’t read the article, based on your response to AH, which was structured around the four bullet point summary at the top of the article, and didn’t appear to contain information that wasn’t available at the very beginning of the paper”. Clearly, that’s not what “clearly” means, so I was wrong to write that. Thanks for responding. Sorry for being a jerk.

Leave a Reply to Andrew