Matthew Kahn, author of the cheeky book Climatopolis: How Our Cities will Thrive in the Hotter Future, likes to compliment our research (Schlenker and Roberts, 2009) on potential climate impacts to agriculture by saying it will cause valuable innovation that will prevent its dismal predictions from ever occurring.
Matt has a point, one that has been made many times in other contexts by economists with Chicago School roots. Although in Matt’s case (and most all of the others), it feels more like a third stage of denial than a serious academic argument.
It’s not just Matt. Today, the serious climate economist (or Serious?) is supposed to write about adaptation. It feels taboo to suggest that adaptation is difficult. Yet, the conventional wisdom here is almost surely wrong. Everyone seems to ignore or miscomprehend basic microeconomic theory: adaptation is a second or higher-order effect, probably as ignorable as it is unpredictable.
While the theory is clear, the evidence needs to be judged on a case-by-case basis. Although it seems to me that much of the research so far is either flawed or doesn’t measure adaptation at all. Instead it confounds adaptation—changes in farming and other activities due to changes in climate—with something else, like technological change that would have happened anyway, response to prices, population growth or other factors.
For example, some farmers may be planting earlier or later due to climate change. They may also be planting different crops in a few places. But farmers are also changing what, when and where they plant due to innovation of new varieties that would have come about even if Spring weren’t coming a little earlier. The effects of climate change on farm practices are actually mixed, and in the big picture, look very small to me, at least so far.
The other week the AAEA meetings in San Francisco, our recent guest blogger Jesse Tack was reminding me of Matt’s optimistic views, and in the course of our ensuing conversations about some of his current research, it occurred to me just why crop science surrounding climate-related factors is so difficult. The reason goes back to struggles of early modern crop science, and the birth of modern statistics and hypothesis testing, all of which probably ushered in the Green Revolution.
How’s all that? Well, modern statistical inference and experimental design have some earlier roots, but most of it can be traced to two books, The Statistical Manual for Research Workers, and The Arrangement of Field Experiments, both written by Ronald Fisher in the 1920s. Fisher developed his ideas while working at Rothamsted, one of the oldest crop experiment stations in the world. In 1919 he was hired to study the vast amount of data collected since the 1840s, and concluded that all the data was essentially useless because all manner of events affecting crop yields (mostly weather) had hopelessly confounded the many experiments, which we unrandomized and uncontrolled. It was impossible to separate signal from noise. To draw scientific inferences and quantify uncertainties, would require randomized controlled trials, and some new mathematics, which Fisher then developed. Fisher’s statistical techniques, combined with his novel experimental designs, literally invented modern science. It’s no surprise then that productivity growth in agriculture accelerated a decade or two later.
So what does this have to do with adaptation? Well, the crux of adaptation involves higher-order effects: the interaction of crop varieties, practices and weather. It’s not about whether strain X has a typically higher yield than strain Y. It’s about the relative performance of strain X and strain Y across a wide range weather conditions.
Much like the early days of modern science, this can be very hard to measure because there’s so much variability in the weather and other factors. Scientists cannot easily intervene to control the temperature and CO2 like they can varieties and crop practices. And when they do, other experimental conditions (like soil moisture) are usually carefully controlled such that no water or pest stresses occur. Since these other factors are also likely influenced by warming temperatures (like VPD-induced drought, also here), so it’s not really clear whether these experiments tell us what we need to know about the effects of climate change.
(An experiment with controlled temperatures and CO2 concentrations)
Then, of course, is the curse of dimensionality. To measure interactions of practices, temperature and CO2, requires experimentation on a truly grand scale. If we constrain ourselves to actual weather events, in most parts of the world we have only one crop per year, so the data accumulate slowly, will be noisy, and discerning cause and effect basically impossible. In the end, it’s not much different from Ronald Fisher trying to discern truth from his pre-1919 experiment station data that lacked randomly assigned treatments and controls.
I would venture to guess that these challenges in the agricultural realm likely apply to other areas as well.