GMAT Strategy

AdamKnewton wrote:

sashah2006 wrote:Normally, how the GMAT CAT works is that if you answer a question right, the next question is more difficult, and if you answer a question wrong, the next question is easier. Also, answering easier questions wrong count more against you than answering difficult questions wrong.

Full disclosure: I work for Knewton, and am therefore biased. However, I, too, think the MGMAT CATs are quite excellent.

The only reason I'm posting here is that I have to chime in simply on the point above: I agree that MGMAT CATs work that way, however, the actual GMAT does not. The GMAT will not necessarily give you a harder question after every right answer. It has other factors to consider besides question-difficulty-level, including 1) the content it has to provide you (that is, it must give you a certain number of geometry questions, etc., and may not always have one available at the "correct" level; and 2) the appropriate differentiation it wants to make of your personal ability estimate at each point.

For example, the algorithm functions like this: After the first 10 questions, it thinks your ability level is a 0.8. As a result, it gives you a question that most students who are also 0.8 get wrong. You also get this question wrong; fine. It doesn't decrease your difficulty level now because it's only confirmed its theory and wants to try another, different question that can better distinguish between people who are estimated at 0.8

The difficulty level of questions is also wildly misconstrued. There is no such thing, on the actual GMAT, as a "650-level question," although MGMAT uses this notation for its CATs. The actual formula rates each question based at what ability-level-of-test-taker do 50% of them get it correct.

Basically, all of the GMAT algorithm calculations are about its estimate of your ability level, NOT a back-and-forth game played with difficulty levels. Do not confuse the idea that the test is Adaptive with the idea that you must get a certain number of easy/medium/hard questions right. And do not ask if a question is a "600-level question" -- there is no such thing!

Night reader wrote:Hi Adam, can we consider the GMAT Prep algorithm identical to that of a real GMAT exam? I know that the GMAT Prep doesn't have experimental questions; it counts all your correct and incorrect answers towards the scaled scores. Is it rational not to credit the answers which are experimental but answered correctly - a test taker invests one's time and efforts into the solution process and should count on, at least, some reward ...? One may spend time answering a bunch of experimental questions correctly and missing not experimental questions, thus having the low scaled scores at the end.

How testing system balances this problem?