MGMAT CAT Workings

[kelvin.s.kwong]

Hi there,

I have a quick question relating to how the MGMAT CAT works and how that relates to the actual GMAT. In taking a few CATs, I have seen a few instances where I will get a string of '700-800' level questions correct, then the computer will give me a '600-700' level question. That seems counterintuitive since my understanding was that a difficult question usually means that the next question will be increasingly difficult.

Along those same lines - I have seen cases where I will again get a string of '700-800' level questions correct and be given a reading comp passage that has one '700-800' level question, one '600-700' level question and one '300-500' level question. From what I understand, the questions from passage to passage are not adaptive. But assuming that the test-taker is at the 700-800 level, I would think that would mean that all three questions would be 700-800.

I ask these two questions because it seems a little worrisome - in the event that I am on an upward progression (i.e. getting 700-800 level questions correct), but then I get one of these curve ball lower level questions and - god forbid - get that lower level question wrong, will I get penalized from that lower level?

Thanks for all your help,

Kelvin

[ilakshmir]

I also share Kevin's concerns. I observe that I get 4-5 questions @700-800 level right and then go wrong on a 600-700 level question. But the next question is still a 700-800 level one on the MCAT. This is rather generous, as it allows one to compete at the higher level but is this how the real GMAT works?

[StaceyKoprince]

The "get one right, get a harder one" rule is only ONE of the multiple rules the algorithm has to follow in giving you a question. It also has to factor in the right number of problems from the different question categories, and the right number of problems from all of the sub-categories (eg, the algorithm might be set to give you at least 3 but no more than 6 geometry problems, of which at least one has to be a triangle / quad problem, one a circle problem, and one a coordinate plane problem). This is just an example, of course.

Further, on our tests, we don't allow the algorithm to give you a problem you've already seen. So, if you're on your third test, and the test needs to give you a DS geometry coordinate plane problem that you haven't seen before, and it has a choice between a 99th percentile level question and a 92nd percentile level, and it wants to find something as close as possible to the 95th percentile level, then it's going to give you the 92nd percentile level problem. (The ratings you see on your score reports aren't exactly how the test rates problems - we just use those categories to make it easier for students to understand relative difficulty level buckets.)

You don't have the above constraint to the same degree on the real test, of course - that's just one of the built-in drawbacks of a practice test database. (And part of the reason why the real test won't let you take the test twice in a 31-day period.)

Finally, on RC, there are typically only 6 to 8 questions written for a passage, one or two of which are general (main idea) questions. You're supposed to be given a main idea question, so if that question is a lower-ranked question, you're still going to get it even if you've been scoring higher.

Make more sense?