A regular reader, John Vokey, pointed out a very nice recent article in the British Medical Journal, by the arch skeptic Ben Goldacre and David Spiegelhalter, about how we know whether something that's obvious is actually true. Here is a link to that piece. He's a widely known writer and commentator, as well as a practicing physician in Britain, who has written a great deal about similar aspects of how we use data and how this affects medicine. He writes about 'Bad Pharma' to try to correct such things (see link below his picture for more).
 |
| Ben Goldacre |
While it is obvious that exercise is good for your health, and we have some good physiological and physical reasons and mechanisms to back up that statement, in our post we noted that the correlation between health and exercise may not be so simple. For example, you have to already be healthy to exercise so the correlation may be a result not a cause of better health.
Goldacre takes something bluntly obvious, that wearing a helmet when bicycling is good for your health (that is, in terms of injuries). He shows that even this is neither so obvious nor simple. Just to illustrate the point, if you ride more often or more often in traffic because you feel safer when you wear a helmet, even with the same per-mile (or, in Goldacre's UK, per kilometre) risk there will be more rather than fewer cycling-related injuries: the population at-risk has grown. Or drivers may cut closer to you seeing that you are helmeted. And so on. As John Vokey pointed out in his comment, that brief but to-the-point article is a fine lesson in statistical reasoning.
If something as apparently simple as the risk of cycling with vs without a helmet is not so simple, then how much more complex will other sorts of causation, epidemiological, genetic, and evolutionary are supposed cause-and-effect scenarios be? A due respect for this complexity should routinely temper conclusions from simple study designs (or, in the case of evolution, almost pure surmises about natural selection in the distant past).
Yet pressures, and perhaps natural tendencies in our boastful current culture, seem to be doing just the opposite: leading investigators to make ever-quicker and ever more grandiose claims about their findings. This is used for self-promotion in general, in seeking grant support, and in the rush to the media. And science journalists often show little, sometimes almost zero sense of skepticism or even circumspection, about such claims.
The issues we face in science are nowadays very complex and subtle, and we know from even simple examples, such as the one Goldacre used to illustrate the pitfalls of statistical reasoning, that our conclusions can be very wrong, even in very simple ways. We try to make conclusions in science, but we should do that by starting with respect for the complexity of the problem.
4 comments:
You're to be commended for this post. And there's no topic that fits this paradigm better than global warming. I've noted in a few recent tweets that for the last decade or so, the increase in global mean temp is completely uncoupled from the increase in atmospheric CO2, with the latter increasing at a nearly linear rate and the temp virtually constant. If one looks at a 100-year record instead, there are 3 periods of several decades where the temp is either flat or actually declining. Of course, as shrilled by politicians and the press, periods of warming connect these pauses to give an overall 150-year warming trend.
The important point here is that climate scientists have never made successful predictions of climate. None of their models predict this step-wise behavior, or explain its mechanism. The complexities of global heat transfer are only poorly understood, and there are enormous gaps in our simple models of the carbon cycle and the global heat budget. Simple extrapolation from prior decades would inform us that we can expect to have essentially the same or somewhat cooler temps as this year for 30 or 40 more years - and then warming (might) recommence.
One need not be grandiose, as the politically motivated climate watchers are, for they are more interested in the electoral thermostat than that of the globe. Any competent scientist who looks at the same data I do would immediately say - this is something we scarcely understand at all.
I can't tell from your message whether you're dismissing evidence for climate change (or human agency), but if so then the burden of proof will weigh heavily on you, given so much global and diverse evidence. As a former meteorologist myself, I realize the complexity of prediction, but also that even if the basic climate change theory is correct, the complexity of the atmosphere and earth and their physics does not lead to simple consequences. So whether the exceptions or uncertainties undermine the general conclusion is perhaps the question.
We did a post a few weeks ago about how climate-change advocates (including politicians) seize on any evidence,even what may be false evidence, to invoke on behalf of their cause. But one has to try as best one can to judge the totality of the evidence.
In a sense, it is the net result that is important. Even whether or not it's human generated, if CO2 is a danger and we're disgorging gobs of it, and we're in a general trend that can threaten many aspects of our lives, then it's worth taking seriously.
As I was named a few times in this piece (MT is way too generous!), let me respond to the provocateur here. What Drs. Goldacre and Spiegelhalter did in their editorial is to highlight brilliantly how difficult it is to isolate causality in seeming simple problems, such as the wearing of bicycle helmets. But each one of their example complications involved agents---goal-seaking agents. Now, it is true that these same agents are products of nature, but, and importantly, except for the human-sourced causes of climate change, the rest of the climate system is agentless, and that simple fact renders Maxey's arguments, at least as relevant in this context, specious. What makes psychology, sociology, anthropology, etc. and living-organisim-centerd biology (ethology) hard is precisely that: agents. Whatever merit Maxey's arguments may have regarding climate change (and I am skeptical about all of them), they are not supported in the least by the agent-based issues of causality raised here.
You raise a fundamental point that far too few wish to take into account. Indeed, one could go deeper with your point and note that many of the statistical models that are used in these areas of research are, in essence, about replicable agent-less phenomena.
Post a Comment