Why We Shouldn't Be Viewing New GRE Quant Scores in Context of the Old Ones

[anonecon]

I think a lot of people are viewing 166Q on the new GRE to be an equivalent signal to an 800Q on the old GRE. And, while the percentile for 166 is, in fact, the same as the percentile for 800, the signal is not the same. This is for a number of reasons:

(1) A score of 166 reveals your "type" as at the lower end of what used to be 800. This is obviously not good given that before the 800 GRE constraint was pretty binding (e.g. MIT's admissions site states that they are primarily interested in students who received the top possible score on the quant section). This indicates that admissions committees at top programs would like to raise their GRE requirements, but could not do so (until now) because there didn't exist a signal better than "800."

(2) All sorts of heuristics/biases kick in. We seem susceptible to anchoring 166~800; however, I don't find it unlikely that admissions committees who are used to just burning all applications without top scores are going to be biased towards 170 ~ 800.

Anyways, that's just my two cents. Anyone agree? Have counter arguments?


Also, I have a question (and I know this doesn't technically belong here, so thank you to whoever, if anyone, answers):
Approximate (I know it varies by test) the number of questions one can miss in the quantitative sections and still receive:
-170Q
-169Q
-168Q
-167Q
-166Q

Thanks!

[Ondrej]

Approximate (I know it varies by test) the number of questions one can miss in the quantitative sections and still receive:

How do you even know how many you've missed?

Anywho, there is one more consideration - did you miss any in the first quant test? If so, I think it's much worse than missing in the second part (as it is adaptive).

[OneArmedEcon]

FWIW, if I were an applicant with the new GRE, I wouldn't lose much sleep over getting a 166 versus a 168 or whatever. My understanding is that most department just use it as a filter to weed out people that have weak quantitative reasoning skills (and yes, I fully agree that the GRE Quant is a less than ideal instrument...). If you've scored in the >95th percentile or whatever, then that indicates that you have sufficient mastery of high school math. Whether you happen to score 95th or 99th is really irrelevant because:
a) the test itself is noisy and, on a single administration, I doubt that the difference is significant
b) even setting aside the sampling issues, I'm not aware of any evidence that suggests that a 99th percentile on high school math indicates greater likelihood to succeed in a PhD program than a 95th; mastery of math at the high school level is a necessary--but far from sufficient--condition.

Again setting aside sampling concerns, if an adcom compares an applicant with a 75th percentile and 95th percentile, then it will have questions about the math capabilities of the former relative to the latter. The same issue doesn't exist in comparing 95th and 99th, IMHO.

As for how many questions you can get wrong... because its a computer adaptive test (CAT), it depends on where you make the errors and the difficulty rating of the question (at least when I took it, each question had a difficulty rating of 1-5, with 5 being most difficult). If you answer a 5 wrong, then it hurts your score considerably less than if you answer a 1 wrong (and it's actually double-whammy, because answering a 1 wrong increases the likelihood of getting 1 and 2's at the expense of 4 and 5's). To score well on a CAT, you need to do really well on the first half dozen or so questions and not miss easy questions. I seem to recall reading a claim that if you answered the first 6 questions correctly and got the remainder of the questions wrong then you would score better than if you answered the first 6 questions wrong and answered the remainder right.

For me, taking my time and being sure about the early questions was the single most important piece of GRE test taking strategy that I learned when I was preparing for the exam. The first time I took it, I tried to pace myself and got too many early questions wrong and wound up with a 740Q. The next time I took it, I spent more time on the early questions and was willing to not finish the test (although I wound up doing so) and I got a 800Q.

[OneArmedEcon]

How do you even know how many you've missed?

At least when I took it, after you got the official score in the mail, you could go online and review a detailed score report. It broke down the question by subject type (algebra, geometry, etc), rated difficulty (1-5), and question number.

[Ondrej]

At least when I took it, after you got the official score in the mail, you could go online and review a detailed score report. It broke down the question by subject type (algebra, geometry, etc), rated difficulty (1-5), and question number.

Oh, interesting. Now you first get it online (by mail a couple weeks later) and there is no further information whatsoever. Just your score and percentile, that's it.

Therefore I am not sure, but I believe I either did not make a mistake or had one answer wrong... and got a 170. A friend of mine a week later got one wrong and already had 169. Most of my friends (say 8 out of 10) got 168-169, all of these reaching the 99th percentile, same as me. So if a school is more concerned about percentiles (I think Boulder does), getting a perfect score does not matter that much.

[OneArmedEcon]

Oh, interesting. Now you first get it online (by mail a couple weeks later) and there is no further information whatsoever. Just your score and percentile, that's it.

Therefore I am not sure, but I believe I either did not make a mistake or had one answer wrong... and got a 170. A friend of mine a week later got one wrong and already had 169. Most of my friends (say 8 out of 10) got 168-169, all of these reaching the 99th percentile, same as me. So if a school is more concerned about percentiles (I think Boulder does), getting a perfect score does not matter that much.

Fortunately I had still had the bookmark saved. Looks like it's still available for free.

https://grediagnostic.ets.org/GREDWeb/gred/signIn.jsp

[Zeno]

Not a contribution to the present discussion, but I find it hilarious that 700V was a higher percentile than 800Q.

[Ondrej]

Fortunately I had still had the bookmark saved. Looks like it's still available for free.

https://grediagnostic.ets.org/GREDWeb/gred/signIn.jsp

Cool, but it doesn't work for me. It's probably only for pre-August '11. Can someone with the newer format double check?

Edit: Oh, here's the catch: "and it will remain available for six months following your test administration date." Too bad, could have helped.

[Catrina]

I tried it and it worked for me. I took the test most recently in July.

[anonecon]

Does anyone have more precise information (ideally based on real data, e.g. themselves)?