The one study you shouldn’t write

Data in society

I might have my own set of ideological prejudices,[1] while at the same time I am more sure than I am about any of these I am certain about this: show me proof that contradicts my most cherished beliefs, and I will read it, evaluate it critically and if correct, learn from it. This, incidentally, is how I ended up believing in God and casting away the atheism of my early teens, but that’s a lateral point.
As such, I’m in support of every kind of inquiry that does not, in its process, harm humans (I am, you may be shocked to learn, far more supportive of torturing raw data than people). There’s one exception. There is that one study for every sociologist, every data scientist, every statistician, every psychologist, everyone – that one study that you should never write: the study that proves how your ideological opponents are morons, psychotics and/or terminally flawed human beings.[2]
Virginia Commonwealth University scholar Brad Verhulst, Pete Hatemi (now at Penn State, my sources tell me) and poor old Lindon Eaves, who of all of the aforementioned should really know better than to darken his reputation with this sort of nonsense, have just learned this lesson at what I believe will be a minuscule cost to their careers compared to the consequence this error ought to cost any researcher in any field.
In 2012, the trio published an article in the American Journal of Political Science, titled Correlation not causation: the relationship between personality traits and political ideologies. Its conclusion was, erm, ground-breaking for anyone who knows conservatives from more than the caricatures they have been reduced to in the media:

First, in line with our expectations, higher P scores correlate with more conservative military attitudes and more socially conservative beliefs for both females and males. For males, the relationship between P and military attitudes (r = 0.388) is larger than the relationship between P and social attitudes (r = 0.292). Alternatively, for females, social attitudes correlate more highly with P (r = 0.383) than military attitudes (r = 0.302).
Further, we find a negative relationship between Neuroticism and economic conservatism ( = −0.242, = −0.239). People higher in Neuroticism tend to be more economically liberal.

(P, in the above, being the score in Eysenck’s psychoticism inventory.)
The most damning words in the above were among the very first. I am not sure what’s worst here: that actual educated people believe psychoticism correlates to military attitudes (because the military is known for courting psychotics, am I right? No? NO?!), or that they think it helps any case to disclose what is a blatant bias quite openly. In my lawyering years, if the prosecution expert had stated that the fingerprints on the murder weapon “matched those of that dirty crook over there, as I expected”, I’d have torn him to shreds, and so would any good lawyer. And that’s not because we’re born and raised bloodhounds but because we prefer people not to have biases in what they are supposed to opine on in a dispassionate, clear, clinical manner.
And this story confirms why that matters.
Four years after the paper came into print (why so late?), an erratum had to be  published (that, by the way, is still not replicated on a lot of sites that republished the piece). It so turns out that the gentlemen writing the study have ‘misread’ their numbers. Like, real bad.

The authors regret that there is an error in the published version of “Correlation not Causation: The Relationship between Personality Traits and Political Ideologies” American Journal of Political Science 56 (1), 34–51. The interpretation of the coding of the political attitude items in the descriptive and preliminary analyses portion of the manuscript was exactly reversed. Thus, where we indicated that higher scores in Table 1 (page 40) reflect a more conservative response, they actually reflect a more liberal response. Specifically, in the original manuscript, the descriptive analyses report that those higher in Eysenck’s psychoticism are more conservative, but they are actually more liberal; and where the original manuscript reports those higher in neuroticism and social desirability are more liberal, they are, in fact, more conservative. We highlight the specific errors and corrections by page number below:

It also magically turns out that the military is not full of psychotics.[3] Whodda thunk.

…Ρ is substantially correlated with liberal military and social attitudes, while Social Desirability is related to conservative social attitudes, and Neuroticism is related to conservative economic attitudes.

“No shit, Sherlock,” as they say.
The authors’ explanation is that the dog ate their homework. Ok, only a little bit better: the responses were “miscoded”, i.e. it’s all the poor grad student sods’ fault. Their academic highnesses remain faultless:

The potential for an error in our article initially was pointed out by Steven G. Ludeke and Stig H. R. Rasmussen in their manuscript, “(Mis)understanding the relationship between personality and sociopolitical attitudes.” We found the source of the error only after an investigation going back to the original copies of the data. The data for the current paper and an earlier paper (Verhulst, Hatemi and Martin (2010) “The nature of the relationship between personality traits and political attitudes.” Personality and Individual Differences 49:306–316) were collected through two independent studies by Lindon Eaves in the U.S. and Nichols Martin in Australia. Data collection began in the 1980’s and finished in the 1990’s. The questionnaires were designed in collaboration with one of the goals being to be compare and combine the data for specific analyses. The data were combined into a single data set in the 2000’s to achieve this goal. Data are extracted on a project-by-project basis, and we found that during the extraction for the personality and attitudes project, the specific codebook used for the project was developed in error.

As a working data scientist and statistician, I’m not buying this. This study has, for all its faults, intricate statistical methods. It’s well done from a technical standpoint. It uses Cholesky decomposition and displays a relatively sophisticated statistical approach, even if it’s at times bordering on the bizarre. The causal analysis is an absolute mess, and I have no idea where the authors have gotten the idea that a correlation over 0.2 is “large enough for further consideration”. That’s not a scientifically accepted idea. A correlation is significant or not significant. There is no weird middle way of “give us more money, let’s look into it more”. The point remains, however, that the authors, while practising a good deal of cargo cult science, have managed to oversee an epic blunder like this. How could that have happened?
Well, really, how could it have happened? I trust this should be explained by the words I’ve pointed out before. The authors had what is called “cognitive contamination” in the field of criminal forensic science. The authors had an idea about conservatives and liberals and what they are like. These ideas were caricaturesque to the extreme. They were blind as a bat, blinded by their own ideological biases.
And there goes my point. There are, sometimes, articles that you shouldn’t write.
Let me give you an analogy. My religion has some pretty clear rules about what married people are, and aren’t, allowed to do. Now, what my religion also happens to say is that it’s easier not to mess up these things if you do not engage in temptation. If you are a drug addict, you should not hang out with coke heads. If you are a recovering alcoholic, you would not exactly benefit from hanging out with your friends on a drunken revelry. If you’ve got political convictions, you are more prone to say stupid things when you find a result that confirms your ideas. The term for this is ‘confirmation bias’, the reality is that it’s the simple human proneness to see what we want to see.
Do you remember how as a child, you used to play the game of seeing shapes in clouds? Puppies, cows, elephants and horses? The human brain works on the basis of a Gestalt principle of reification, allowing us to reconstruct known things from its parts. It’s essential to the way our brain works. But it’s also making us see the things we want to see, not what we’re actually seeing.
And that’s why you should never write that one article. The one where you explain why the other side is dumb, evil or has psychotic and/or neurotic traits.


1 Largely, they presume outlandish stuff like ‘human life is exceptional and always worth defending’ or ‘death does not cure illnesses’, you get my drift.
2 For starters, I maintain we all are at the very least the latter, quite probably the middle one at least a portion of the time and, frankly, the first one more often than we would believe ourselves.
3 Yes, I know a high Eysenck P score does not mean a person is ‘psychotic’ and Eysenck’s test is a personality trait test, not a test to diagnose a psychotic disorder.
Chris von Csefalvay is a data scientist/computational epidemiologist specialising in the dynamics of infectious disease and AI-based, data-driven solutions to public health problems.

You may also like

Leave a ReplyCancel reply