Thanks to DAV for reminding me of a clarification. See yesterday’s comments. This post is mostly bookkeeping. Tomorrow we do more examples.
Sometimes, in contracts as in predictions, certain “unexpected turns of events” are agreed to, both before and after the fact, as negators of the prediction or contract. We might agree that rain-outs negate the baseball prediction and make it void. We need this “out” because the prediction had an explicit date and time which, if we did not create this “escape clause”, would have made the prediction a bust (the Tigers did not in fact score more runs on the stated date and time).
An unexpected turn of events isn’t considered a prediction failure when the parties (the statement) agree beforehand that “given there are no unexpected turns of events” as part of evidence. This is, of course, ambiguous, and we often find parties in contracts disputing (after the fact) whether the event that caused the prediction to fail was really “unexpected” or not.
We are finally at the meat: A prediction is a statement, conditional on clear provisos, about an event that can be empirically verified. Or “in principle” empirically verified. Not all predictions are verified: criminal trials come to mind. The jury, taking into account many provisos, rules guilty or not, which is a prediction of an event that might never be verified. Predictions are thus not just of events that happen in the future; but of events that are not yet known to the provisos; that is, evidence which forms the conditions of the prediction must not include the evidence “the event obtained” or “the event did not obtain”; for then the prediction is a circularity. For criminal trials this implies the jury does not start out assuming, and then continue to hold the assumption, “the guy did it” or “the guy is innocent.” Guilt or innocence is what they are trying to predict (prove).
Now if after the fact people disagree that the prediction failed or succeeded it is usually the “clear provisos” that are argued over and disputed, no matter how sharply the provisos seem to have been defined. Well, human nature and all that. It usually one side who says, “Conditions A are what the ‘clear provisos’,” meant while the other side says, “Obviously, conditions B are what the ‘clear provisos’ meant.”
Given conditions (or interpretation) A and the clear provisos, the prediction is either verified or not; similarly given conditions B. Which conditions are the forever, all time, perfectly correct ones? Neither. Or, that is to say, probably neither. Unless those conditions, A or B, can be deduced (in the rare, technical sense) then it is a subjective choice which of A or B (or C, etc.) to accept as part of the prediction statement.
If I predict that “It’s likely that Mitt Romeny wins the Republication nomination in 2012” then this isn’t a (hard) prediction because the proviso “likely” is ambiguous, and it turns the prediction into a probability. (The event is contingent, meaning it is not necessary true.) We haven’t covered these “soft” predictions yet. For the moment, accept that we have to make the prediction non-probabilistic. Change the prediction to “Mitt Romeny wins the Republication nomination in 2012.” Now, if a week from today Romney finally reveals to the world that, one lonely dark night a decade ago, he was abducted by a UFO and probed assiduously—that is, if Romney has a Ross Perot moment—then, except among a sliver of the electorate, support for the man will plummet and he will lose.
I may then protest that I had not considered otherworldly influence and that it was not reasonable that I should have done so: the prediction is voided. I will convince some of you that this was so, that it was unreasonable to imagine every possible contingency and that given my excuse the prediction is indeed voided. But I won’t convince everybody, and among those recalcitrants—who accept the original prediction as read—my forecast was a bust. Both sides are right, depending on which additional evidence is assumed as belonging to the original conditions.
The conditional nature of predictions is no different than that for logical or probabilistic statements. We know that since “All Martians wear hats & George is a Martian” that “George wears a hat” is true, its probability is 1, the prediction is deduced to be so, even though we know, via other evidence that “No Martians exist.” That is, if we accept the original conditions, George certainly wears a hat, but if we insist that “No Martians exist” then George certainly does not wear a hat. Different sets of conditions give different predictions (or probabilities or truth or falsities or even absurdities).
Next time: the IPCC example.