Monday, December 14, 2009

Political Science

Another Climategate post, indirectly. I keep wondering what it is about the current AGW narrative that bothers me so much. I bristle at bald statements that "the science is settled", but not so much at the arrogance of the statement as the fact that there's something qualitatively fishy about it. I don't have any problem with statements like, "the science surrounding the theory of evolution is settled." So what's the difference?

For some reason, this Ezra Klein post on excess deaths from lack of health insurance triggered a possible answer. In it, we are informed:
By now, you're probably used to hearing about the $900 billion health-care bill. But what about the 150,000-life health-care bill?

Oddly, that label hasn't made its way into the conversation. But it is, if anything, a conservative estimate. The Institute of Medicine developed a detailed methodology for projecting the lives lost due to lack of insurance. The original paper estimated that 18,000 lives were lost in 2000, and the Urban Institute updated that analysis with data for 2006, yielding an estimate of 22,000 lives. As for 150,000, well, that's almost certainly too low. That's just the 2006 number across 10 years, which is the time frame we generally use for health care, with a third of the lives saved lopped off, as we're not going to cover all of the uninsured. But since the population of the uninsured grows every year, and so does the death toll, it would surely be higher. So call it the 150,000-plus-life health-care plan.

Now, I'm sure that the folks at the Institute of Medicine think they are doing science, and I'm pretty sure that they're right. But they're not really doing clean, falsifiable, Popperesque science, are they? They're building a model. The model produced some conclusions, but the conclusions are ultimately not falsifiable. It is impossible to tell whether 18,000 lives were lost in 2000 from lack of insurance, because it's impossible to build a model where 18,000 lives weren't lost because everybody had insurance.

With climate research, we also have models, but the models are even more sketchy. Instead, we have lots and lots of measurements with very little predictive theory to accompany them. We have thermometric data from the last 150 years, and we have whole bunches of proxy data from the last couple of thousand years, which may or may not correlate with actual temperatures. Once again, this may be wonderful work, but it's not exactly falsifiable, is it? Nobody's going to construct a time machine any time soon so that we can go back and measure the temperature and turn the various proxy hypotheses into hard theories. This is all correlation, not theory.

Mind you, I'm not saying that we should stop trying to produce good translations between proxy data and inferred temperature. Rather, we should just be aware of the limitations of doing so.

I have the same problems with epidemiological studies. Designing studies, sampling them, and observing statistical correlations is very important and contributes enormously to clinical knowledge. But it doesn't really tell you why a statistical correlation exists. That can only be done with a set of falsifiable hypotheses. When you base policy on statistical correlations instead of biological theory, you're probably doing better than you would with superstition, but you're running a pretty big risk that the causation you infer from the correlation might be wrong.

So a lot of this class of science is running a bit open-loop. We should always be skeptical of observation without explanation.

Which brings us back to Climategate or, more accurately, to the set of statements that can be made about the world climate, based either on the data available, or on the current set of climate theories:
  1. Thermometric observation shows that the global climate has warmed over the past 150 years. (Very likely true.)
  2. Measured increases in global climate are strongly correlated with CO2 emissions. (True.)
  3. Human activity, which can be proven to cause the increased CO2 emissions, is causing the global climate to warm. (Based on a simple deduction from 1 and 2, very likely true.)
  4. Current warming is unprecedented. (False if the theories on the correlation of proxy data with actual global temperature are incorrect. Since those theories can't currently be falsified, this can't be presumed to be true. Indeed, the whole issue behind the "hide the decline" scandal is that the decline may in fact falsify some of the proxy theories.)
  5. Without aggressive reduction in CO2 emissions, global warming will accelerate to cause significant environmental and economic damage. (Predicted only by model. Can't yet be falsified.)
  6. Aggressive reduction in CO2 emissions will slow the rate of global warming. (Can't yet be falsified.)
So here's where things start to get nasty: In the absence of adequately tested theories, which, while not yet falsified, are yet to be subject to any tests capable of falsifying them, climate scientists and their fellow-traveler policymakers are merely pounding the table. The consensus is, indeed, that these models are correct, but the consensus is operating without the benefit of the full, closed-loop testing cycle that we usually feel is important for scientific theory to be robust. This takes things partially out of the realm of science and deeper into the realm of politics.

Let's make sure that we distinguish between the two.

No comments: