Artwork by DonkeyHotey via Flickr (CC BY-SA).
A. Why am I writing this?
I know, I know. Just my second blog post in, without a secure reader base, and already I’m talking politics. I hope you’ll stick with me here though, because I don’t mean for this to be a divisive or inflammatory post.
For those of you who haven’t kept up with where I’m at in my life, two years ago I started a PhD program in chemistry at UC Berkeley. And while it has been a great experience so far and I have learned a lot, as a chemist working only with other chemists it can be easy to get trapped in an intellectual bubble. We go into work everyday and do chemistry for 12 hours, talk chemistry at lunch and coffee breaks, and go home and think about chemistry, because that’s what we’re expected to do as graduate students. But what this means is that non-chemistry topics like politics don’t often enter into our discussions. I realized as I was following this primary season that my views on the best candidate for the Democratic nomination1 were being formed almost completely free from debate or discussion with other people. I was inhabiting my own intellectual echo chamber with the entrance blocked off, and with nobody around for miles.
So here, I attempt to lay out my rationale for why I plan to be voting for Hillary Clinton in the California Democratic primaries on June 7. The nature of this post is not persuasive; I didn’t approach this post with the intent of convincing you, the reader, to vote for Hillary. In fact, just the opposite; I encourage all of you to give me the best reasons why I’m wrong. I’m inviting you into my echo chamber, not so you can hear my views, but so I can hear yours. Look for logical flaws. Question my reasoning. Tell me what factors I’ve not considered. I’m putting my reasoning under the microscope of peer review, and I’m open to making revisions.
Let’s get started.
B. Rise of the algorithms
Now, on matters of issues, I tend to agree with Hillary more than I agree with Bernie, but I won’t be talking about issues here. I’m guessing that, if you’ve followed the Democratic primaries, you’re already familiar with both candidates’ positions, with plenty of other media outlets debating those positions’ relative merits. If you want to talk issues, I’d be happy to indulge in the comments section. But here I want to present reasoning that is perhaps a bit unorthodox, and hopefully informative.
First, we have to take a detour and talk about Nobel Prize-winning psychologist Daniel Kahneman, well-known for his work on heuristics and cognitive biases and for jump-starting the field of behavioral economics.2 In 1955, Kahneman was a young lieutenant in the Israeli Defense Forces. He had a tall task in front of him: He was to come up with an interview system for new recruits for the Israeli army, which was to be used to determine the likelihood of their future success. The system that was in place before Kahneman’s arrival was a standard fifteen- to twenty-minute interview, at the end of which an interviewer would have formed a general impression of the interviewee. Unfortunately, it was found that this general impression had almost zero predictive value for the recruits’ future success. And so Kahneman, who at the time only had a bachelor’s degree in psychology, was brought in to change things up.
This is Daniel Kahneman. He is smiling because he has a Nobel Prize.
It turns out that Kahneman wanted to change things up so much that his suggestions were met at first with open rebellion. Specifically, he proposed scrapping the standard interview system, and instead, having the interviewers ask factual questions in order to objectively score the interviewee on six traits that he regarded as relevant to performance in the army, such as “responsibility” and “sociability.” He would then take these scores and plug them into a super complicated statistical formula, and the magic number that popped out was to be used as a predictor of future success.
Interviewers didn’t like having their role be reduced to robotically asking factual questions, but they eventually went along with it. After a few hundred interviews, Kahneman’s magic number for each recruit was correlated to evaluations of that recruit’s performance by commanding officers. The result? Kahneman’s method was far superior to the old interview system. It wasn’t perfect; in fact, Kahneman himself only described it as being “moderately useful.” That may be an undersell, however, given that the Israeli army still uses mostly the same interview system to this day.
The moral of this story is that statistical algorithms tend to outperform expert intuition in domains where the predictability of the outcome is low. If you think I’m cherry-picking one anecdotal example, I’m not; this turns out to be one of the more robust findings to come out of the social sciences. Since Kahneman’s application of this principle to the interview system of the Israeli army, statistical algorithms have been applied to predictions of outcomes ranging from life expectancy of cancer patients to winners of football games; from success of new businesses to future prices of Bordeaux wines; from recidivism of juvenile offenders to evaluations of scientific presentations. In all of these cases, algorithms either matched or exceeded predictions by experts in the relevant fields. Overall, across about 200 studies on this phenomenon, around 60% have shown that predictions from statistical algorithms significantly outperform predictions of trained professionals; the remaining 40% show a statistical tie. No convincing study has shown experts outperforming an algorithm.
She’s angry because an algorithm beat her in predicting the number of unicorns that would be found on Mars.
Oh, so you know how I said Kahneman used a super complicated statistical formula to predict the future success of Israeli army recruits? I lied. It turns out all he had to do was take the scores assigned to the six traits and add them up. That’s right – a simple summation of the scores resulted in significant predictive validity. Again, this is a result that has since been confirmed by further research; even back-of-the-envelope calculations tend to have an advantage over experts. This means that any of us can apply this technique to make better decisions, whether you’re an employer looking to hire the best employee, a gambler attempting to predict the outcome of a horse race, or – well, you probably saw where I was going with this – a voter trying to decide who would be the best nominee for the Democratic Party.
C. Political figure-skating
OK, so I’m going to be using an algorithm here. But how do I come up with the traits or characteristics that I’ll be scoring? Well, ideally, I’d be looking for traits that are good predictors of being a good president. For a bank of traits I suppose it’d be reasonable to start off with a personality inventory that includes the Big Five personality traits (Openness to Experience, Conscientiousness, Extraversion, Agreeableness, and Neuroticism). And if I wanted an objective measure of the greatness of past presidents I suppose I could contact a bunch of presidential historians (maybe about 800 of them) and get their ratings of the presidents. I could then correlate the traits of the presidents (attributed to them by about 100 presidential biographers, who I could also contact) to their greatness ratings, and see which traits correlate the best.
If I did all that I would end up with this study by political psychologists Steven Rubenzer, Thomas Faschingbauer, and Deniz Ones. The authors broke down each of the Big Five personality traits into six facets, and correlated each facet with presidential greatness as determined by presidential historians. Based on that study, as well as this similar study by political psychologist Dean Simonton, I’ve identified five facets that are the most highly (positively or negatively) correlated with presidential greatness.3 They are listed below. As points of reference, Rubenzer et al. consider correlations of above 0.40 to be large correlations, and correlations between 0.25 and 0.40 to be medium correlations.
Intellectual brilliance 0.56
The scoring system I’ll use is similar to that used for figure skating. Both Clinton and Sanders will be scored on a scale of 1 to 5 in each of the five personality traits, with 1 meaning “This candidate does not exhibit this personality trait at all” and a 5 meaning “This candidate personifies this trait”. These five scores will be scaled by a factor equal to the corresponding correlations, and then these scaled scores across the five traits will be summed. Whoever has the highest total at the end wins.
OK candidates…ready, set, skate!
D. The showdown
While I see sections B and C as less controversial, this section where the scoring takes place is much more subjective. If you wanted to throw tomatoes at me, it would be over this section (but please throw them respectfully in the comments section). If you disagree with me on the scores, however, what’s nice is that you can still retain the same scoring system and plug in your own numbers. If you do so, resist the urge to play around with the numbers if you get the final score back and it goes against your favored candidate, as that would immediately negate the value of this exercise. Try and score as objectively as possible the first time around, and then accept the final weighted sums, whatever they turn out to be. But anyway, on to my own scores:
Intellectual brilliance: What is meant by intellectual brilliance is not made incredibly clear in Simonton’s paper, but it seems to be not just intelligence, but a combination of cognitive factors that also includes creativity, curiosity, openness to alternative ideas and values, wisdom, and more. While there are no (reliable) reports of the intelligence of Clinton and Sanders as determined by IQ tests, both have impressive educational backgrounds. Based on what I saw in the Democratic debates, I was impressed by Clinton’s wide range of knowledge on both domestic issues as well as on foreign policy. Her answers seemed nuanced and detailed, and I got the sense that she had given every issue a lot of thought. Even on Libya, widely viewed to be one of her greatest failures as secretary of state, you can’t fault her for not doing her homework. Meanwhile, I thought Sanders faltered in the debates when it came to providing details for his proposals, especially when it came to foreign policy. I’m also troubled by the lack of evidence to support his economic proposals. All in all, I don’t see the same receptiveness to evidence and intellectual rigor in Sanders as I do in Clinton, so I give Clinton the edge in this category.
Clinton 4, Sanders 3.
Assertiveness: Assertiveness is defined by a candidate’s inclination to be a leader and take charge, so past leadership experience seems important here. Clinton’s resumé here is long: She transformed the role of First Lady by actively influencing policy, she served two terms as Senator of New York, and she served as Secretary of State in the Obama administration. Sanders also has plenty of leadership experience, having served for 25 years in the House and Senate. I give Sanders a lot of credit here for being an independent in a two-party-dominated system of government, for leading the way in showing that a serious campaign can be run without a Super PAC, and for locking horns at times with the DNC. Clinton, on the other hand, seems like less of a trailblazer when it comes to policy issues, and is much more likely to toe the party line. I give the edge to Sanders on this one.
Clinton 3, Sanders 4.
Achievement-striving/competence: This category involves both having high aspirations, as well as the ability to realize those aspirations. Sanders and Clinton both have one-half of the equation here. Sanders certainly has run a campaign with lofty goals, but I question his ability to implement those goals, especially given his poor legislative record in Congress. Clinton, on the other hand, has a penchant for compromise and pragmatism, but a platform built on compromise and pragmatism almost has to be non-ambitious. I’m calling this one a tie.
Clinton 3, Sanders 3.
Tender-mindedness: Tender-mindedness basically means having sympathy for others. This is a tough one to score, since I feel like this is a trait that would show much more in private than in public. Based on what I’ve seen in public, though, I don’t really have any reason to believe that one candidate is significantly more tender-minded than the other. I’ll call this one a tie, too.
Clinton 4, Sanders 4.
Vulnerability: Vulnerability is defined as having a general susceptibility to stressful situations. Note that this category has a negative correlation, so a lower score is better. I think Clinton has a pretty strong case for this category, having weathered conspiracy theories, multiple organizations dedicated to attacking her image, eight Benghazi investigations, and, at least according to one study, the most negative media coverage this election season (yes, even more than Trump). Even after all that, she continues to put herself in the center of public attention, and she doesn’t seem to have lost her appetite for public service. I really don’t have much evidence in favor of Sanders here; in fact, I thought he lost his composure at times during some of the (relatively tame) Democratic debates. I give Hillary a win here.
Clinton 1, Sanders 3.
Weighting all these scores and summing them together, we end up with final scores of Clinton 5.55, Sanders 4.88. So that settles it.
…OK, I’m being a bit facetious here. That doesn’t really settle it, because this is only a partial reasoning based solely on personality traits. As I said at the beginning of the post, though, I also have issues-based reasoning in favor of Clinton, and so those reasons in combination with this analysis gives me a solid foundation for why I’m planning on voting for Clinton. But as I also said at the beginning of this post, I’m broadcasting this line of reasoning to get feedback. Show me where you think my reasoning isn’t valid and I’ll update my beliefs accordingly (if you think the logic is sound, feel free to say that, too).
The California primaries are fast approaching, though. They’re on June 7. You have until then to convince me.
This is just an addendum section explaining why you might be wary of the reasoning I have laid out here:
- Coming into this analysis, I was already leaning towards Hillary, opening up the scoring to possible confirmation bias (although I tried to guard against that).
- I don’t have much of a background in statistics, and so it is possible that I am missing some of the nuances of the statistical analyses done in the political psychology papers by Rubenzer et al. and Simonton.
- This reasoning completely ignores electability in the general election, and only focuses on which candidate would make a better president.
1As for why I’m not discussing the Republican primaries here, I offer two reasons, a) I followed the Democratic primaries much more closely than the Republican primaries (it would’ve taken a ton of time to follow both), and b) at the time of the writing of this post, Donald Trump had already clinched the nomination.↵
2The remainder of this section is all drawn from Kahneman’s book Thinking, Fast and Slow, which mainly summarizes research over the past half century on cognitive biases and heuristics. I highly recommend it to anyone who is interested.↵
3Boring details on how I extracted what I did from the two papers: I picked the facets with the highest magnitude correlations, but only one from each of the Big Five personality traits, in order to keep the facets independent of each other as possible (having dependent traits would result in an overcounting of those traits in the algorithm). I’ve combined Achievement-Striving and Competence into one facet, since they had almost identical correlations and they are pretty related. Also, instead of taking a facet from Openness to Experience, I’ve substituted Intellectual Brilliance from Simonton’s paper, because it correlates better than any facets from the Rubenzer et al. paper and is supposed to be quite related to Openness to Experience. For correlations from the Rubenzer et al. paper, I averaged the correlations that came from the Ridings & McIver ratings and the Murray & Blessing ratings, leaving out the Ridings & McIver ranks, because that would be correlated with the Ridings & McIver ratings and it seemed to me that the ratings would just be more precise.↵