Dan Gardner is the New York Times best-selling author of Risk, Future Babble, Superforecasting (co-authored with Philip E. Tetlock), and How Big Things Get Done (co-authored with Bent Flyvbjerg). His books have been published in 26 countries and 20 languages. Prior to becoming an author, Gardner was an award-winning investigative journalist. More >

Why Politicians Should Experiment More, Not Less

Reading through the policy platform of the Ontario Progressive Conservatives, it's immediately apparent that PC Leader Tim Hudak loves families -mentioned 90 times in 44 pages -as much as he hates experiments. "Expensive experiments have sent hydro bills skyrocketing," it reads. Over and over the same phrase appears. Expensive experiments. Expensive experiments. And do you know what you'll get if the Liberals are re-elected? "More experiments with your money." The PC platform reeks of focus groups, which suggests the public is as offended by the idea of "expensive experiments" as Hudak. Which is fascinating, and depressing, because the critical problem with public policy is not that there are too many experiments. It's that there aren't nearly enough. To be clear, most of Premier Dalton McGuinty's big-ticket energy policies are not experiments. An "experiment" is something tentative. It's not known whether it will work. It is implemented on a small scale. It is carefully structured to deliver meaningful information. It is monitored. And the results are rigorously analyzed to determine whether it does what it is supposed to do. If it succeeds, an experiment can be turned into a permanent program, and possibly scaled up. If it fails, a lesson is learned and that's the end of it. McGuinty's energy policies may be new, but they share none of the other characteristics of an experiment. They're just ideas the premier thought would work. So he made them government policy. No tentativeness, no small-scale test, no monitoring and analysis. No experiment. (An important exception is the smart meter program, which began with small pilot projects that produced positive results and were subsequently scaled up.) This is typical of the policies proposed by politicians, including a long list of proposals in Hudak's platform. Hudak is sure, for example, that forcing provincial prisoners to do menial labour in the community will produce various positive results, including "protecting hardworking families." He says he will implement it, provincewide, at a cost of $20 million, if he is elected. No tentativeness, no small-scale test, no monitoring and analysis. No experiment. Is this approach to policy making sensible? Not in the least. In fact, it's remarkably stupid. Let's say the result we want is Y. How do we know that if we do X we will get Y? The fact that X is intended to produce Y doesn't help. We all know what the road to hell is paved with. Nor can we be sure that X produces Y because it seems reasonable that it would. Feelings are not a reliable guide to reality. It once seemed perfectly reasonable to let leeches suck the blood of sick people, after all. Nor can we be confident X produces Y because lots of smart people believe X produces Y and they have been doing X in order to produce Y for years and years. Everybody knew leaching worked. They did it for centuries. So how can we know if X produces Y? We must experiment. Until a surprisingly short time ago, a very large portion of the treatments used by physicians were similar to leaching in that they were well-intentioned, seemed reasonable, were widely believed to work, had been used for ages, and were scientifically unproven. Physicians gradually accepted that this wasn't good enough. Apply leeches to sick people and sometimes the patients get better, sometimes they don't. So does leaching cure people? Kill them? We couldn't be sure. What was needed was an experiment. Give randomly selected patients one treatment. Give others different treatments, including a placebo. Analyse the data. This is "evidence-based medicine." It would be nice to think governments practise "evidence-based policy." But all too often they don't. Policies are implemented because someone thinks they'll work. Afterward, the situation may improve or it may get worse. Did the policy cause the change? Politicians will say it did, or didn't, depending on the politics of the moment. But in reality, the truth often isn't known, and cannot be, because the policy wasn't set up as an experiment. "If we don't know whether we are doing any good, then we are not any better than the medieval doctors and their leeches," noted Esther Duflo, a superstar French economist. There's a wide variety of experimental designs and social scientists like Duflo are coming up with clever new ones all the time. As a result, the range of policies that can be subjected to experimentation is wide and growing. But politicians are seldom interested. They are sure X produces Y. They don't need experiments and carefully analyzed evidence. They just know. This is the "God complex," a term used by Archie Cochrane, a pioneering physician whose story is told by Tim Harford in his excellent new book Adapt. In Cochrane's day, physicians thought it was a waste of time and money to put treatments to the test. They knew they worked. But Cochrane pushed on -and sometimes discovered that treatments assumed to be effective were useless. Or worse. That's the thing about "expensive experiments." They can catch bad ideas before they are widely implemented, and they can identify good ideas which can be scaled up. Both outcomes can save money. Even lives. Or we can just assume we're right, make sweeping changes, spend huge amounts of money, and have no real clue if we are doing any good at all. It seems that like most politicians, Tim Hudak prefers the latter.