The perils of Phlogistonomics

In a recent post, Matt Yglesias complains about a kind of economic analysis that has always kind of bothered me. I've called it "label-the-residual economics"; on Twitter, Matt suggested "phlogistonomics" and Richard Thaler suggested "aether". I like "phlogistonomics" a bit better, despite its unwieldliness, for reasons I'll explain later. In the meantime, let me go into a bit more depth than Matt did about the dangers of the approach.

What is this type of analysis that I'm warning about? In economics, when explaining a phenomenon, you generally encounter a "residual". The residual is the part you can't explain in terms of things you can measure directly. For example, suppose you are trying to explain a country's GDP. You can (somewhat inaccurately) measure how much labor the country has. You can (somewhat more innacurately) measure the amount of capital it has. So you assume that GDP is a function of capital, labor, and something else:

GDP = TFP * K^a * L^1-a

Here, K is capital ("das kapital"), L is labor, and TFP is the "something else". TFP is the residual.

This residual represents how productive capital and labor are, which is why we call it "total factr productivity" (TFP). What determines TFP? It could be "human capital". It could be technology. It could be institutions like property rights, corporate governance, etc. It could be government inputs like roads, bridges, and schools. It could be taxes and regulations. It could be land and natural resources. It could be some complicated function of a country's position in global supply chains. It could be a country's terms of trade. It could be transport costs and urban agglomeration. It could be culture. It could be inborn racial superpowers. It could be God, Buddha, Cthulhu, or the Flying Spaghetti Monster. It could be an ironic joke by the vast artificial intelligences that govern the computer simulation that generates our "reality", putting their metaphorical thumb on the scales because they are bored underpaid research assistants with nothing better to do.

Many, many economists refer to TFP as "technology". They are usually careful to stress that what they mean is not "technology" as we usually think of it, but a "generalized technology" that represents all the ways that society has found to utilize capital and labor to pump out GDP. But despite this careful qualifier, using the label "technology" has implications for how people think about and evaluate models like this. "Technology" sounds like something exogenous, something that we can't predict and can't control. Hence, using the label "technology", which has outside meanings, instead of the more neutral term "TFP", puts us in danger of allowing semantic biases to cloud our judgment.

This is "labeling the residual". Another example is in development economics, where the things we don't understand are often labeled "culture". This can lead to semantic biases. For example, when explaining Japan's wealth, many people turn to old cultural stereotypes such as "Japan is conformist" or "Japanese people imitate foreign things but don't invent new things". These stereotypes are blunt and inaccurate. Many of them were manufactured by either Japan's fascist government in the 1930s (as a way to promote the idea of racial differences) or else by writers in the British Empire. Many may no longer be very accurate. Others may never have been very accurate. Others may be somewhat accurate but miss crucial details. Nearly all of them greatly annoy Noah Smith.

What is the danger of semantic bias? Semantic bias may discourage us from trying to delve deeper into the workings of the economy. For example, suppose we find that TFP looks kind of like an AR(1) random process:

TFP_t = p*TFP_t-1 + e_t

In other words, TFP looks like it has "random" shocks (e_t) that decay after a while. Most of the action is in the shock, e_t. If e_t is truly random - if it's like a quantum fluctuation or a sunspot - then we're done, we can't do any better. But if e_t depends on things that we can observe, then we can do better than this model. We can explain more than we have already explained. So the question of whether e_t is truly random is crucial to the question of how well we can explain the economy.

I argue that calling TFP "technology" biases us toward thinking that e_t is truly random. After all, things like the invention of the internet can't be predicted in advance with any kind of certainty. For all intents and purposes, they are truly random events, like sunspots or earthquakes. So even though TFP might include a lot of things other than what we normally think of as technology, using the label "technology" for TFP discourages us from trying to actually go and predict TFP. In reality, we might be able to predict other determinants of TFP, like terms of trade or government investment. I know some economists try to predict these things. Good!! But maybe if we didn't call TFP "technology", more people might try, and they might get more funding to develop new sources of data that would help explain TFP.

I also argue that calling TFP "technology" allows some models to get a free pass on the Lucas Critique. TFP (or the persistence parameter p) might be a function of all kinds of government policies, in which case the model presented above would not be policy-invariant. Remember that applying the Lucas Critique involves a judgment call - the only thing deciding whether a parameter like TFP is "structural" is the consensus judgment of macroeconomists. And I suspect that the label "technology" makes it easier for economists to just assume that TFP is structural...because hey, policy can't affect whether some genius invents something in his garage, right?

I think this semantic bias is even more evident when "culture" is the residual being labeled. Culture is assumed (wrongly, I believe) to be something ancient and immutable. Japanese "culture" (i.e. cultural stereotype) is usually explained in terms of ancient traditions and conditions - Japanese people are risk-averse because they live on an island with lots of earthquakes, Japanese people are imitators because of the dominance of China in East Asian culture hundreds of years ago, Japanese labor markets are a reflection of samurai-era feudalism, etc. This in turn implies that "culture" cannot easily change. (The assumption is clearly false. For example, Japan had very little lifetime employment or seniority-based pay before World War 2; so much for samurai feudalism!)

Labels like "technology" and "culture" may bias economists toward being lazy and sloppy. If we allow our beliefs to be guided by our labels, we risk reducing ourselves to speculating about the future trends of technology or invoking the same tired cultural stereotypes that fill internet forums, instead of searching out new sources of data to explain the heretofore inexplicable. I should caution the reader that I don't know how much of this sort of thing really happens; I am NOT alleging that the econ profession as a whole commits these mistakes. It just seems like a danger. If pressed for examples of when the "technology" label was over-influential, I'd point to the wide acceptance of the RBC paradigm in the 1980s. Regarding "culture"...well, let's just say the whole country of Japan is a serial offender on this count.

Another danger of residual-labeling is that it may bias us against parsimony. When we see a new phenomenon that is difficult to explain - for example, the slowness of Japanese cafes to offer free wi-fi - we may be tempted to postulate a new cultural trait ("Japanese businesses are suspicious about giving anything away for free") instead of looking for some incentive (Perhaps Japanese cell phone service is so good that few people own laptops?) to explain the phenomenon. We just point at stuff we can't explain and say, in the parlance of blogs, that it's "the culture that is Japan". But if you add a new parameter to explain a new data point, you haven't really explained anything.

So what should we call the practice of labeling residuals with semantically laden words? I like Matt Yglesias' proposed label of "phlogistonomics" a little better than Richard Thaler's suggestion of "aether". In the late 1800s and early 1900s, "aether" was something we could and did investigate; it was supposed to be the medium in which light waves moved. We proved it didn't exist. On the other hand, "phlogiston" was more like a residual - it was just "whatever makes stuff burn" (the error was in thinking it had mass). We know there is something that makes stuff burn. And we know there is something that makes some economies function differently from others. The danger in labeling this residual is that in doing so we may trick ourselves into thinking we understand the residual as well as we ever will. That would be a mistake. Eventually, we figured out what makes stuff burn. Someday we may figure out what makes economies different.