Algarythem gurus

[tomslawsky]

Assuming a perfect score on a particular sub section, at which question is a test taker getting 50% right and 50% wrong. For example, if someone were EXACTLY a 50 quant scorer, and the exam worked 100% correct, by which question would they get their 1'st miss and subsequently 50% right/wrong?

[RonPurewal]

Assuming a perfect score on a particular sub section, at which question is a test taker getting 50% right and 50% wrong. For example, if someone were EXACTLY a 50 quant scorer, and the exam worked 100% correct, by which question would they get their 1'st miss and subsequently 50% right/wrong?

heh. i would wager that none of us will really have any sort of concrete answer to this question.

aside from answering "i don't know" to the principal question here, i will also point out that the highest score in each subsection is 51, not 50. (actually, in some of the gmac tables this score is indicated as "51-60", as if there are basically unused spots on the scale.)
again, i have no idea for sure, especially regarding the official test, but (if i remember correctly) i don't think the gmat prep software will actually award you a score of 800 unless you actually get all of the questions right. of course, this could be quite different from the official test -- in particular, the software may contain a much smaller complement of extremely hard questions than the official test, thus justifying the necessity of actually getting all of them right. but -- to my best memory, as i haven't actually taken gmat prep in about three years -- i seem to remember running through it with 1-2 questions incorrect on the whole thing and being given a score of 790, not 800.

more out of curiosity than anything else, i'm wondering what would make you wonder something so incredibly specific-- it's certainly an interesting thought.

[StaceyKoprince]

Everything Ron said. Also, though I don't actually know, I would wager that there isn't one spot (one particular question number) where it has to happen every time.

Also, just technically - it's actually the case that people get closer to about 60% right and 40% wrong (at most scoring levels), not 50/50. At the "perfect score" level, though, I would expect the percentage correct to be a lot higher, but probably not perfect. (If nothing else, I think it would be very hard to go for 3.5 hours without making a single careless mistake!)

Here's another (philosophical) question:

If someone gets all of the operational questions right but misses some of the nonoperational (experimental) questions, does that person have 100% correct? Or do we count all questions given in terms of percentage correct, even though some of those questions don't count toward the score? After all, when we're taking the test, we don't know which ones count and which ones don't...

[tomslawsky]

Ron- that's just the way my brain is hardwired. When I decide to conquer something of intellectual challenge, I try to dissect it from every possible angle. In this case, knowing the answer to my question may serve no pragmatic function, but it would still fill a spot in my brain for wanting to know EVERYTHING i'm capable of knowing about the GMAT (and the GRE and IRT in general). I tend to (sometimes to a fault) approach difficult problems like this- when something gets me, it's hard to sate the itch!

Stacey- 60% sounds right. I've thought about this when taking paper practice exams (for the GRE and SAT) where they publish the percent correct for the questions. I was getting frustrated when I was hitting a "35-ish" % question wrong. Then, I did a "dirty" calculation (not sure of its validity) but I said, if 20% get the question correct, that leaves 80% who guess. Of the 80%, 20% will guess correct, or about 80(.2)= 16% of the test pool guess correctly. That means that if I correctly chose an answer that is at the "top 20%", then 16% will guess with me and thus an top 20 %question is really a 36% hit rate. Doing the math for the top 10% means that 10 + 9(.2)= 28%, 5%: 5+9(.2)= 23%. So, I estimate that a "51 level" question on the quant, should have about a 20-25% correct answer rate. So, for your 50% bench mark, that would be 50 + 50(.2)= 60% correct. Sounds sound to me.

As far as the experimental questions go, I would think it unethical to count them in your score if they hadn't been fully validated, so I would hope to god that the GMAC wouldn't do so. In other words, if I answered all "valid" questions correctly and all "experimental" questions wrong, it would be incumbent on the exam scorers to grade me as "perfect"...

thanks for your responses.

[tomslawsky]

Actually, as I think about it, there is a reason I went down this path of thought. I used to not be a big fan of the computer adaptive format. I'm not sure why, but I think I felt that there were too few questions to actually gauge one's skill level. However, I have ALWAYS had a problem with a standardized exam being multiple choice. Here is the problem: in a standard, non "CAT" GMAT, there is no leeway given for a teat taker who actuallt knows how to solve a problem correctly, goes ahead and solves the problem correctly and makes a mistake by checking "B" when the answer is "A". Also, if that same test taker understands a question, sets up the solution correctly and then makes a silly mistake, his wrong answer is weighted exactly the same as the test taker who gets addled by the question, blindly guesses and moves on.

Enter the CAT format. Now, if someone makes a careless mistake or 2, there is actually (in theory) room for redemption. Based on my experience with the GRE (admittedly not the GMAT) POWERPREP, one can miss 4 (possibly more?) questions in quant and still get an 800. On the paper exam, this would be a low to mid 700s. Being someone who is prone to making silly mistakes and especially in math, has always been a B/C student who came away with much more from the class than the tippy top A students (and on many occasions, actually tutoring some of said students before exams), the computer exam started top strike me as a better format. I assume that both the GRE and GMAT share the same basic statistical and mechanisms, albeit tailor tweaked. Thus, now I'm changing my tune and thinking the computer exam MIGHT be a more fair assessment of one's skill level. This got me to thinking about exactly how the algorithm worked. The concept is fascinating to me.

Who knows, maybe I'll invent a "Better" test someday....haha~

Just my thoughts on the subject.

[StaceyKoprince]

Interesting thoughts! Yes, in terms of assessing one's ability to take these kinds of exams, the CAT format does tend to produce more accurate results than the paper format, and the CAT can do so in fewer questions.

They don't count the experimentals at all UNLESS you leave one blank. (eg, if you run out of time, and one of those questions WOULD have been an experimental if you'd gotten to it, that no longer matters - they give you the same penalty as for the ones that would have counted.)