Fuck Nuance [pdf]

I would posit the social signalling theories of Robin Hanson as perhaps a theory of this magnitude.

The problem is, though most of Hanson's theories have plausible explanations and "feel" right, there is very little experimental evidence and very few studies are directed in that area. In fact, you might even say it is a symptom of that type of signalling itself: as homo-hypocritus, we don't want signalling theories to be proven true so we make up reasons for deriding signalling explanations as not rigorous enough to even merit being properly tested or studied. Of course, this too is no better, because then do we just actually have no evidence to believe in signalling explanations, or are we using signalling excuses to suppress the meaning of that evidence? It's somewhat self-defeating.

The next most important theory I would posit is Prospect Theory by Kahneman and Tversky. But there again, many aspects of it have not been successfully replicated, and too often limited experiments are generalized too much.

One set of social theories that I would love to see given more quantitative foundations is the research from the book Moral Mazes. Robert Jackall carried out some lengthy longitudinal studies in which he collected anecdotal narrative and interview data from employees at all seniority levels through a couple different firms. He was able to see people rise up the corporate ladder, people get fired, executives fired and replaced, large layoffs, etc., and conduct informal interviews with employees all up and down the ladder.

Based on his interviews, he put together a series of theories about how morality and ethical identity form within bureaucracy, and particularly how managers develop an understanding of ethical obligations to subordinates, and how certain factions of an organization (notably HR) come to function when there are competing concerns between ethical expectations and bureaucratic mandates.

In every job I've had, it's been a frustrating experience of basically saying, "Yep, I know this chapter from Moral Mazes ..." because companies operate almost as if they've read the book as a field guide.

Yet at the same time, it's still just a qualitative study that is scaffolded by academic sociology. If it were actually possible to collect that kind of data from corporations, I'd love to see some theories from the book explored more quantitatively.

At first "overfitting" was the concept that leaped to mind as I read it, but overfitting is sort of something different. You have a complicated data set, and when you overfit, you keep adding dimensions of freedom, and then setting those dimensions of freedom to produce the original values that were passed in, such that it fails to correctly predict any other instance drawn from some conceptual "same distribution".

In this case, the theories do the "adding many dimensions of freedom" part, but they don't really fix the values; they probably haven't got data to fix in the first place, and it's not mathematical enough to be that precise about the fitting.

I think it's more like these theories are often not so much hypotheses themselves, but the descriptions of hypotheses spaces, and the addition of "nuance" means that in a lot of cases, what sounds like a "theory" is really just the description of a space that happens to contain all or most of the possibility space in question. As much fun as it may be to do that, and as much as it may even be a necessary first step to a real theory, in practice that is not generally the difficult part of the process. (There are exceptions to that which may leap to mind but I submit they would be the exceptions, rather than the rule.) When one is working with fuzzy words (even if by necessity!) and not math, the fuzzy words make it very easy to make this mistake, I think. English does not have a strong distinction between a hypothesis space and a specific hypothesis in that space, as evidenced by the very fact I have to dip into jargon to even make that point!

From that perspective, the solution that leaps to my mind is implement the observation that in many ways, the important thing about a theory is not what it describes, but what it excludes. It is too easy and cognitively tempting to try to create a theory that, when you fiddle with the parameters, describes everything you see, but the real questions are, what is excluded by your theory? Can I fiddle with the knobs in your theory to describe something that you say ought to exist, but does not seem to? Is your theory even well-enough specified that these are meaningful questions to ask about it?

I don't know the answers to these questions for sociology. I have to admit I've tried to engage with it a couple of times, because what fan of Isaac Asimov doesn't want to learn about some real-life psychohistory, but I've always found it difficult to get traction with with my engineer mind. I can engage from a philosophical point of view, but that is not generally what I was looking for; I've got, well, philosophy itself for that. (Repeat previous sentence for history, psychology, and economics.)

To p4wnc6's question, I'd submit that as dismal as the Dismal Science may sometimes be, it does often meet this bar. I fear economists themselves may often be quite sluggish at recognizing when their own theories are contradicted by the evidence because the theories are saying that what is currently happening (in the general sense, not the literal March 27, 2016 sense) is excluded by the theory, but that is a human problem, not an economic theory problem. Economic theories do exclude things; if supply rises and demand holds steady, the vast majority of possibilities where price goes up are excluded, and if price were to go up, the theories would be useful in investigating how such a thing actually happened by examining what such theories would still permit (e.g., perhaps supply isn't really rising because the new supply is actually a higher quantity of lower quality, and too much of the demand is for a higher quality than the new supply is supplying which means "real" supply is dropping, or something like that). They certainly aren't perfect and a given theory may even be wrong, but, well, the very fact that a theory can be wrong is evidence that it is falsifiable, and thus not "nuanced" into infinite psudo-explanatory power.

This is possibly the most hacker-news-ish comment I've ever read.

I assume simple theories are very good for advancing the science itself. But I don't think they are as actionable as complex ones, because reality is complex, and you need to account for that when creating policy.

I thought of the map-territory jawn, as well.