The SAT subtracts a quarter of a raw point for every wrong answer (every right answers gets you one whole raw point). This wrong-answer penalty is meant to counteract any lucky guessing you might do: in theory, if you guess randomly on a selection of questions with 5 possible answer choices ( (a), (b), (c), (d), and (e) ), mathematical probability says you should be right 1/5 of the time and wrong 4/5 of the time. If you gain one raw point for each correct answer and lose 1/4 point for each incorrect answer, then your overall guessing shouldn’t affect your score. For every right answer that gets you one point, you’ll have four wrong answers that lose you 1/4 point each.

Many popular SAT companies, like Kaplan and Princeton Review, teach you a “guessing strategy” that supposedly exploits this penalty. This guessing strategy says that if you can eliminate one or two answer choices from a question, you should go ahead and guess from the remaining answer choices. If you can do this consistently, then you should come out ahead, according to the theory, because you’ll be guessing from fewer than five answer choices–even though you’ll only lose 1/4 of a point for every wrong answer, as though you were guessing from five choices. In this way, you can up the odds. At least in theory :)

The problem with this approach is that it relies on probability and randomness in order to work. But nobody chooses an answer choice on the SAT at random–you choose an SAT answer based on your ideas about the question. That means the entire notion of relying on probability is flawed, because probability relies on random selection.

Because SAT questions are written to trick people who don’t know how to analyze them, it’s very hard to use the guessing strategy. This is why high-scoring SAT-takers never guess.

Is this something people want to hear more about? I’d be happy to write a longer article on the subject. Leave me a comment.

great! i’ll post a longer article on the subject later today. when you get a chance, could you post some of your calculations, either as a response to this thread or to that one? i’d be interested to see what you’ve come up with.

thanks!

mike

The expected value calculation works fine in this setup and concludes that guessing is advantageous provided that F is convex and disadvantageous if F is concave.

The trouble is that in reality we have a discrete function that is more complicated than either side of that dichotomy and the analysis starts to become fuzzy. Your objection about the guessing being nonrandom can be overcome in various ways, but even in the simpler case that one doesn’t account for it, the model does not lead to clear conclusions.

thanks,
mike