January 21, 2013 1 Comment
Any damn fool can predict the past.
Niven’s Laws, #8, Analog magazine Nov 2002
Interesting. Now, I have an open mind, but when I hear that a claim that a single variable completely controls a complex system, my skeptic hackles get raised. That picture sure looks like an extremely perfect fit – but in the times I’ve conducted experiments, I’ve never seen data which perfectly fit predictions (it should be within an error range, there should also be noise).
So retrospective analysis which precisely fits a specific variable makes me wonder about methodology more than it inspires confidence. What would impress me more is prospective predictions, made in a clear manner, saying that “we expect X increase in CO2 to correlate precisely with Y increase in temperature, unless there are Z volcanoes, and evaluate over W period”. That would be impressive. It’s also what hasn’t happened, and is why there’s so much sturm und drang about climate science – give me falsifiable predictions, à la Karl Popper (successor to the ideas of Francis Bacon et al.
Sadly, we’re not at the point yet where the medium term testable predictions have been formulated (which is extremely surprising to me), and the politics of the moment has injected itself into every part of the discussion, so any data point becomes the subject of partisan debate.
But the answer, as both Dawkins and Sacks would agree, to bad science, is good science, and this is what we need. But retrospective studies are great for validating whether a model has promise, but are not great for validating whether a model has predictive power, which is of course what matters.
Predicting the future is hard, but that doesn’t mean that it’s okay to predict the past and call it the same as predicting the future.
Note: I reserve the right to reevaluate any opinion based on any new evidence learned. I also reserve the right to be wrong, along with everyone else throughout human history.