Skip to content

What A Prediction Is And What It Is Not: Part I

I say that the Detroit Tigers, the baseball team—baseball being the most sublime of all sports, and this team being the game’s most stalwart representative—will beat the Boston Red Sox when they meet on Opening Day, Thursday, 5 April 2012. This is a prediction.

I also claim that the global average temperature in 2013 will be less than it will be for 2012. This is not a prediction, but a scenario. Predictions

And then, once we think about it, the first statement is not quite a prediction either. So why did we initially think it was? Why would anybody think the second one is a prediction?

Here is what a hard prediction is: first, it is not a probability, but a bet (of a sort); second, it is a statement capable of unique empirical verification. (We will discuss “soft” predictions, which are probabilities, another time.) Both criteria must be met for a proposition to be a prediction. A prediction is synonymously a forecast, a guess, prognostication, a prophecy or divination. A prediction is most decidedly not a “scenario,” though clever people often use scenarios in the place of predictions to fool their audiences.

Saying (what is true) that baseball is “the most sublime of all sports” is not a prediction because it is not ultimately capable of unique empirical verification. And this is because the statement is dependent on the aesthetic or moral term “sublime”, a term which rests on something deeper than any empirical observation.

What’s wrong with the proposition that “The Detroit Tigers will beat the Boston Red Sox on 5 April 2012”? Not much, but something; it is nearly complete, yet it misses the information about how one baseball team can “beat” another. We know this information so well—one team scores more runs than the other—that we do not consider that we should supply it to the statement. It is tacitly there—but only for fans of baseball, for those who know the sport.

Those unfortunates who live in places where this great game does not exist will not know how to turn the statement into a genuine prediction. For these folks the statement is not a prediction. Of course, the statement is easily fixable by changing “beat” to “score more runs than.”

Then what happens if it rains or, God help us (this being Michigan in April), snows? The games will be postponed. The prediction then fails. Not in the sense that it is no longer a prediction, but in the sense that the event predicted did not obtain. The prediction is a bust, a miss. It was wrong.

You protest? “Doesn’t count! The game didn’t go off. The prediction is voided.” But words about a game “going off” were not in the prediction. The statement merely said the Tigers would score more runs than the Red Sox on a certain date. This did not happen, so the prediction is a miss. We can fix this by adding the proviso, ahead of time, “If the games goes off, the Tigers will score more runs than the Red Sox.” We might call this a conditional prediction, but all predictions are conditional; the conditionality is just more explicit here.

So, it’s the fifteenth inning, 1 to 1, and the skies open: a cataract. The radar gives no hint that the storm will stop. The umpire, his soul seized by underworldly forces, declares the game a tie and then heads for the nearest pub to warm himself. The prediction is a bust once more. Why? Well, the Tigers did not in fact score more runs than the Red Sox. A miss!

More outrage? “Doesn’t count! The game was a tie!” But there were no words about a tie, vile as these things are in baseball (or in any sport), in the statement. The statement was not empirically verified, and that is that. Of course, if ties were a concern, or a negator of the prediction, then this information can ahead of time be added to the statement in the obvious way.

Get it? Every possible contingency has to be imagined and must be part of the statement for the statement to be a prediction. No wiggle room can be allowed in its interpretation. No possibility of dispute, or disparate explanation; no fuzziness, no nuance. The thing must be clear and agreed to beforehand. Imagine writing a prediction like one writes a contract and you’ll have the idea. Both sides must be satisfied that everything that could happen has been agreed to in advance.

15 thoughts on “What A Prediction Is And What It Is Not: Part I Leave a comment

  1. Frankly, I think golf is more subliminal than baseball.

    Every possible contingency has to be imagined and must to be part of the statement for the statement to be a prediction.

    Is a tautology a prediction? Presumably it would cover all contingencies.

    An unexpected turn of events shouldn’t count as a prediction failure either. Even in contracts, Acts of God are considered mitigating circumstance. And there is such a thing as tacit understanding in contracts.

    Whatever happened to rain dates?

  2. DAV,

    Excellent questions, comments as always. See tomorrow’s post for an answer. Short answer: predictions, like probabilities, like all statements of knowledge, are all conditional on explicitly stated conditions/evidence.

  3. Briggs,

    Here’s the way you do it.

    “The Tigers will score more runs than the Red Sox, plus or minus ten runs.” Then there is a high liklihood that the result “will be consistent” with your prediction/scenario.

    If by some miracle, the difference is more than ten runs, then you just say, “based on new data, I overestimated the ‘run sensitivity’ of my model, but otherwise, I was correct.”

    In golf, this would be called a “Mulligan”, but in the prediction/scenario industry, it’s called a “Hansen”.

  4. A scenario is conditional: If the Detroit Tigers play the Boston Red Sox on Thursday, April 5, 2012, they will win according to the formal rules of Major League Baseball.

    A prediction is not conditional: The Detroit Tigers will beat the Boston Red Sox on Thursday, April 5, 2012 according to the formal rules of Major League Baseball.

    Scenarios often have probabilities attached to them which are untestable in scenarios of a single event.

    A political commentator makes what is called a prediction but which is a scenario of the form: If A and if B and if C, then XYZ is likely to win the New Hampshire primary. In the one-off event of the New Hampshire primary, “likely” is not testable. In fact, if each A, B and C have a probability of 0.5 then the probability of A and B and C is 0.125 and therefore it is improbable the XYZ will win – the opposite of what most listeners (likely) took away from the interview.

    In its purest form, a bet is a prediction with probability one. The bettor implies a probability between zero and one by the size of his bet.

  5. Is it possible to predict a scenario and predict an outcome? Or is it only possible to describe a scenario but it si possible to predict an outcome?

  6. Speed, “according to the formal rules of Major League Baseball” is a condition, hence the prediction is conditional.

    George: yes, sort of. Stick around.

  7. What’s wrong with the proposition that “The Detroit Tigers will beat the Boston Red Sox on 5 April 2012″? … yet it misses the information about how one baseball team can “beat” another…

    After explaining how a baseball team can beat another, do we need to explain how to or what it means to score in a baseball game? How much detail do we need to go into?

    Every possible contingency has to be imagined and must be part of the statement for the statement to be a prediction…

    Darn, darn, darn, so, my prediction of Obama winning 2012 presidency in a previous post is not a prediction then! Something is amiss in my understanding of this post. Is this a claim or scenario intended to fool us?

  8. Prediction is to forecast or make inference about the unknown future. A scenario can be about the past. For example, if JH had decided to come to US to study computer science instead of statistics, she wouldn’t have had met Mr. JH. This is a scenario and can’t be empirically proven.

    Why is it important to differentiate between prediction and scenario?

  9. I predict one W Briggs will do a blog on how to teach a computer the difference between an outlier (valid data) and corrupt data in 2012.
    Based on history, what is the probability of my prediction coming true?

  10. JH,

    IPCC climate modellers have found “scenarios” very useful!

    you can be called out on a prediction’s accuracy.

    ‘projections’ are also dodgy. I recommend all statisticians hence forth avoid andy sums which deal with the future or the unknown. This will free up some of their time.

  11. Hmmm . . .

    I’m sure our illustrious host will enlighten us further (my prediction), but I’m not sure at this point about the distinction between a prediction and a scenario, as those two words are used in the English language.

    At least as used in the climate debate by climate scientists, the only difference between a prediction and a scenario seems to be that the former is thought to be that which will occur, and the latter is thought to be that which could occur. Perhaps this, at least in small part, relates to what our host is describing as the conditionality aspect.

    However, I would argue, tentatively, that there is no bright line between a scenario and a prediction. At some point I start to believe that my scenario (which up to that point had perhaps been a mere possibility) is likely (now a likelihood); at some further point on the spectrum I think my likely scenario has become very likely, or highly likely, or even certain, barring unforeseen circumstances. There seems to be a gradiant from sheer logical possibility, on the one end, to absolute certainty, on the other hand. Almost everything is logically possible (the sun could cease to shine at noon Pacific Time tomorrow), while very few things are certain (death and taxes). In between lies a vast gradiant of uncertainty, which we try to describe by attaching various labels. With the exception of specific fields (tax court) or specific mathematical statements (more likely than not), it seems most of these labels are for descriptive convenience and leave some room for interpretation.

  12. I’m looking forward to the next post, but this just seems to be getting lost in formalism for formalism’s sake. Or maybe it’s just a long joke, with the punchline yet to come. It smells a lot like the IPCC weasel words, “It’s not a prediction, it’s a forecast!” Either way, I suspect Briggs’ new reviewer gig is the inspiration for this post.

    I would say that it’s an error to try to compare the two. I’d define a scenario as a description, and a prediction as a statement of a scenario becoming true. To be more formal, a scenario with conditions and dependent outcomes inside of it is really multiple scenarios packed together to look like one.

Leave a Reply

Your email address will not be published. Required fields are marked *