%Percentile Level%

[THX]

Hello,

Is there a place where one can go and get an idea on how the 600 level question will look like vs. the one that's let's day 680 or 690 or 750?

How does GMAT (before loading them into the CAT system) determine the level of difficulty of the questions? I am just trying to figure out my own percentile level...or my ability to answer 650-700 questions with 100% accuracy. The problem is that I don’t really know if the question in front of me is of a 700+ level or it’s below 600 or in between.

Thanks in advance!

[THX]

Hello,

Is there a place where one can go and get an idea on how the 600 level question will look like vs. the one that's let's day 680 or 690 or 750?

How does GMAT (before loading them into the CAT system) determine the level of difficulty of the questions? I am just trying to figure out my own percentile level...or my ability to answer 650-700 questions with 100% accuracy. The problem is that I don’t really know if the question in front of me is of a 700+ level or it’s below 600 or in between.

Thanks in advance!

STACEY,
CAN YOU TRY TO ANSWER TO MY QUESTION ABOVE :?:

THANK YOU

[StaceyKoprince]

Approximately 25% of the questions we answer on test day do not count towards our score. These are experimental questions. When they write new questions, they first insert them into the test as experimental questions and give them to 1000 to 2000 people. They get a bunch of data from this that then allows them to set the difficulty level of that question.

It's very difficult for anyone other than an expert to distinguish between, say, a 600 and 700 level question, and for all practical purposes, you can't distinguish between a 680 and a 690. In fact, that's not even exactly how the questions themselves are defined - we just talk about it that way because it's easier to understand. Each question actually has an "S" curve of data. The X axis represents student scoring levels (1st percentile to 99th percentile) and the Y axis represents the percentage of students who go the question right.

So if you draw the above, start a line around the origin and draw it gradually going up until around, say the 50th percentile mark, when it suddenly curves up very quickly and then gradually moves back into a nearly horizontal line again, except this time that line is up towards the top of the Y axis instead of the bottom. What that means is that most people below around 45th percentile get it wrong, most people above around 55th percentile get it right, and people in the 45th to 55th percentile range sometimes get it right and sometimes get it wrong, though the closer you are to 45, the more likely you are to get it wrong, and the closer you are to 55, the more likely you are to get it right.

So the question itself isn't technically rated at a specific score.

Take a practice test that gives you difficulty level data for the questions. On our tests, we don't have experimental capability set up to determine levels beforehand, but after a question has been in the pool for several months and we have enough data, we go back and calculate its curve (as described above) and then adjust its levels in the algorithm. We also give the general difficulty level range (in 100-point increments) in the score reports, which you can see after you finish a test. So that will give you some idea.

Also, the official guide questions generally get harder as the numbers get higher, so expect the 600-700 range questions to be approximately the middle third of the questions in any given section. (Well, really, more like: if you have 100 questions, expect from about #30 or 40 to about #80 to be 600-700 level questions, mostly.)

[THX]

Approximately 25% of the questions we answer on test day do not count towards our score. These are experimental questions. When they write new questions, they first insert them into the test as experimental questions and give them to 1000 to 2000 people. They get a bunch of data from this that then allows them to set the difficulty level of that question.

It's very difficult for anyone other than an expert to distinguish between, say, a 600 and 700 level question, and for all practical purposes, you can't distinguish between a 680 and a 690. In fact, that's not even exactly how the questions themselves are defined - we just talk about it that way because it's easier to understand. Each question actually has an "S" curve of data. The X axis represents student scoring levels (1st percentile to 99th percentile) and the Y axis represents the percentage of students who go the question right.

So if you draw the above, start a line around the origin and draw it gradually going up until around, say the 50th percentile mark, when it suddenly curves up very quickly and then gradually moves back into a nearly horizontal line again, except this time that line is up towards the top of the Y axis instead of the bottom. What that means is that most people below around 45th percentile get it wrong, most people above around 55th percentile get it right, and people in the 45th to 55th percentile range sometimes get it right and sometimes get it wrong, though the closer you are to 45, the more likely you are to get it wrong, and the closer you are to 55, the more likely you are to get it right.

So the question itself isn't technically rated at a specific score.

Take a practice test that gives you difficulty level data for the questions. On our tests, we don't have experimental capability set up to determine levels beforehand, but after a question has been in the pool for several months and we have enough data, we go back and calculate its curve (as described above) and then adjust its levels in the algorithm. We also give the general difficulty level range (in 100-point increments) in the score reports, which you can see after you finish a test. So that will give you some idea.

Also, the official guide questions generally get harder as the numbers get higher, so expect the 600-700 range questions to be approximately the middle third of the questions in any given section. (Well, really, more like: if you have 100 questions, expect from about #30 or 40 to about #80 to be 600-700 level questions, mostly.)

Thanks for the detail explanation, Stacey!

[StaceyKoprince]

you're welcome!