GMAT Scoring Algorithm

[AlexKuijper]

Hey there,

First post for my self on this board. I searched and couldn't really find an explanation on this board and was just wondering the prized 3-digit number comes to be. For instance, I just scored the following on a MGMAT practice exam:

690 (91% percentile)
Quant: 45 (77% percentile)
Verb: 38 (85% percentile)

Does this mean that the quantitative section had a ton more 700-800 level questions? Just kinda confused by these numbers and if they make sense. If someone could help explain or direct me to a better source of information that would be much appreciated.

Thanks!

AK

[Ian Stewart]

I'm not precisely sure what it is about the numbers that you find unusual, but if you let me know, I can try to answer in more detail - is it that your scaled score is higher, but percentile rank is lower, in Quant than in Verbal? The Quant and Verbal scaled scores are not directly comparable. They were fifty years ago when the test was first developed, but over time, as test takers have devoted increasing amounts of time to preparation, test takers have become much stronger in Quant; Verbal scores improve less with preparation. So as you saw on your test, you can have a much lower scaled score in Verbal than in Quant, but actually be better in Verbal, in percentile terms, just because Verbal scores tend to be so much lower than Quant scores in general.

It is your percentile rankings in each section, and not your raw scaled scores, that really matter in determining your score out of 800, so you should pay greater attention to those. In your case, you're very strong in both sections already, but you actually stand to make greater improvement in Quant - that's not a bad position to be in, because you're much more likely to improve in Quant with preparation than you would be in Verbal.

As for whether you would have had a lot of 700-800 questions in your Quant section, there's no way to know based on your score what difficulty level of question you saw. In theory, the test can give you mostly easy questions, but then you need to be almost perfect to get a respectable score, or the test can give you mostly hard questions, and then you could still get a respectable score despite many, many mistakes. The test tends to be most accurate, and most efficient, when most questions are around the ability level of the test taker, and while that's normally what happens for most test takers, that doesn't need to be true (and sometimes isn't when a test taker has an unusual response pattern, e.g. because of a string of lucky guesses). Further, you shouldn't draw any conclusions at all about the real scoring algorithm from any prep company tests. Prep companies are not, in general, using a scoring algorithm that is all that similar to the real one.

[AlexKuijper]

Thanks for the response, helps quite a bit.

For instance, I complete one of the MGMAT exams the other day and got a scaled score of 50 in the quant(95%). When I look at the actual questions, I got 15 wrong out of 37. Maybe I'm just not used to the way this test is designed but I thought that getting 22/37 correct and still scoring in the 95% percentile was strange.

Alex

[Ian Stewart]

Thanks for the response, helps quite a bit.

For instance, I complete one of the MGMAT exams the other day and got a scaled score of 50 in the quant(95%). When I look at the actual questions, I got 15 wrong out of 37. Maybe I'm just not used to the way this test is designed but I thought that getting 22/37 correct and still scoring in the 95% percentile was strange.

Alex

Almost everyone who takes the GMAT gets about one third of his or her questions wrong in Quant. Even people who get a Q50 score usually have 10-13 mistakes. I've often seen people draw an analogy between the GMAT scoring algorithm and athletic competitions like the high jump or the pole vault. I know nothing about pole vaulting, so I'll use numbers like the ones on the GMAT score scale. Suppose I wanted to test how good a pole vaulter you are. Say I set the bar at 35 feet, 38 feet, 41 feet, 44 feet, 46 feet, 47 feet and 48 feet, and you clear the bar every time. At that point, I'd be pretty certain you can handle heights up to 48 feet quite comfortably. Then suppose I give you fifteen tests at 49 feet and you clear them every time, and fifteen tests at 51 feet and you miss every time. I'd then think you can handle the 49 level but not the 51 level, so you're probably a 50 level pole vaulter, even though you missed 15 out of 37 attempts.

That's more or less what the algorithm does - it tries to find what level of question you can answer comfortably, and what level you struggle with. Your % of right and wrong answers is not a very important factor in determining your score; instead it's the *difficulty level* of the questions you answer correctly which matters. That's important to understand for a few reasons. For one thing, you should not be overly concerned with your 'hit rate' on problem sets or on tests, since the difficulty level of the questions is what really matters - on a very hard problem set, even quite a low hit rate might be very good. Even more importantly, on test day, you should never panic if you can't answer a question. If you get the wrong answer on a really hard question, that barely hurts you at all - if you get a 51-level question wrong, that only tells the algorithm that you might not be a 51-level test taker (and almost no one is) but you can still be a 50-level test taker. So you should not be concerned if you need to guess at some hard-seeming questions. What you don't want to do is rush through easy questions and make careless mistakes, since getting easy questions wrong can hurt you a lot.

[AlexKuijper]

That makes a lot of sense. Thanks for the response Ian.

This morning I wrote another MGMAT practice exam and I let my emotions grab a hold of me as I couldn't answer a 700-800 level question. I worked on it for 3:54 and ended up getting it right. Spent over 3 minutes on another few question and got them wrong. As a result, I panicked as my remaining time was less than 2 minutes a question and I ended up botching some 600-700 level questions. Pretty much exactly what you are talking about above. Ended up with a 43 on the Quant. Thanks for your help in showing me how this works.

[chieftang]

Very informative!

I'm just getting started in studying, and took a random free Quant test online yesterday. It was said to emulate the GMAC CAT algorithm.

Well, the questions got really crazy, really quickly and it never let up. About 22 questions in I got frustrated and realized I needed to go back and rememorize some things before taking another one, and at that point it was clear I wasn't going finish on time. So I literally randomly answer the remaining 15 questions without reading them. I did have about 15 minutes left, but the level of difficulty was much higher than 1 minute per question, so I quit.

I still got a Q44. Strange.

[Nina1987]

I have seen with my own experience and from experts' views on various forums that one can get a score of Q50 while still missing 10-15 questions. However, with regards to Q51 I have not seen such more or less conclusive opinion. I do remember reading a post here (I think by a chicago based tutor) that one can't hope to score Q51 while answering more than 5 questions wrong.
Do the experts agree yet on the following?
"How many question one can't afford to get wrong while still hoping to score Q51?"
Thanks!

[DavidG@VeritasPrep]

I have seen with my own experience and from expert's views on various forums that one can get a score of Q50 while still missing 10-15 questions. However, with regards to Q51 I have not seen such more or less conclusive opinion. I do remember reading a post here (I think by a chicago based tutor) that one can't hope to score Q51 while answering more than 5 questions wrong.
Do the experts agree yet on the following?
"How many question one can't afford to get wrong while still hoping to score Q51?"
Thanks!

There's no magic number. I can tell you that, from years of evaluating students' GMATPRep tests, I've seen someone get a Q51 with as many as 6 incorrect answers. But as Ian mentioned in an earlier post, the algorithm isn't just looking at how many questions you missed, but the difficulty level of those question. Mix in the fact you'll see some experimental questions that won't count towards your final score at all, and it's impossible to say definitively how many questions you can miss and still get that 51. (And it's important to note that while one could, in theory, get a 50 with 10 incorrect answers, one could also get a substantially lower score with 10 incorrect answers, depending on how the algorithm assesses the difficulty level of those questions. This is all to say that you don't want to attach too much importance to the number of questions you miss.)

[[email protected]]

Hi Nina1987,

CAT review is a big part of the study process, since that review can help to pinpoint any issues you might be having when you take a FULL-LENGTH CAT. Since your question is ultimately about the Quant section, here's something to consider when analyzing YOUR performances:

When you get questions wrong in the Quant section, how many are because of...
1) A silly/little mistake?
2) You didn't know the 'rules' involved?
3) The question was just too hard?
4) You were low on time and had to guess?

To score at the highest levels on Test Day, you really have to take responsibility for all of the 'gettable' questions (regardless of what 'level' you think they are). The more you end up saying "I could/should have gotten that question correct....but didn't...." the more likely you're letting easy points slip away.

GMAT assassins aren't born, they're made,
Rich