Best Guess on New Gre Percentiles

[Daisy Fields]

I've been trying to dig around to find some information which might give me an idea of the percentiles on the new test. This is the only piece of information that I could find

http://www.ets.org/Media/Tests/GRE/pdf/14521_0ETSS036_supplement.pdf

They state on the second page that ets expects that the top 25% will be spread over 15 points.

So, please correct these numbers if I've made a mistake but, assuming a normal curve this gives a standard deviation of 7.42 and the following percentiles (score, percentile):

155 75th

156 79th

157 83rd

158 86th

159 89th

160 91st

161 93rd

162 95th

163 96th

164 97th

164 98th

166 98th

167 98th

168 99th

169 99th

170 99th

Thoughts?

[Lox26]

The sample score report indicates that 156 is roughly 89th percentile for Verbal and that 157 is about 63rd percentile for Quant. Did you just average the two to calculate your percentiles? Correct me if I'm wrong: don't the math/verbal scores have two different distributions? Therefore, might it not be more meaningful to calculate the percentiles by section?

Thank you for posting this supplement.

Edited August 19, 2011 by Lox26

[Kitkat]

I can understand why Daisy Fields posted it in this way. It says in the supplement says that they are trying to get it so that the 25% will be in the top 15 points in their new point system. And based on that, Daisy Fields I think that your setup with the numbers are correct.

But to Lox26, it does say that on the sample. It might be that it is because they were bringing that from the previous exam and that was the sort of spread that you got from that exam. They expect something different from the new exam, but won't put it into actual percentages until they have more data to work with. It might very well fall into the same pattern, but who knows. From the looks of it, the quant section does not seem that much different, but to me the verbal seems easier in the sense that there is less dependence on knowing large words that are used infrequently, and more contextual, and therefore will most likely bring the percentiles closer if not make them completely equal.

[Lox26]

Ah, my apologies, Daisy.

[chak]

Ah so now the interesting part is how the model that the original poster put up would translate to the two sections. I'm not reiterating Lox's comment entirely, but pointing out the part where there is a difference between how weighted scores translates to the percentiles is wholly different in the two sections. Using the old gre as the case in point:

Quant:800~94th percentile and up

Verbal:640-94th percentile and up

http://en.wikipedia....rd_Examinations

Not saying what Daisy did is wrong; not at all. I was just curious if the percentile could be accurately extrapolated from the wide range score that the new gre test is currently handing out (e.g. 650-750, 750-800, etc) to the 130-170 range, other than taking the easy route and just averaging the range and then figuring out the percentage.

Edited August 19, 2011 by chak

[chak]

Or maybe ETS will do already translate the discrepancy into the weighted score, which would then be translated into the percentile range? E.g., using the new gre as the example, 30/40 correct would give a raw score of 160, but being how verbal is in the old gre and assuming the same would happen with the new gre, the weighted score would be bumped up a couple points to 162 or so. This example is purely used as an example, and not backed with actual math.

[Daisy Fields]

Sorry, my bad! I have only been doing the quant since I got 98th on the first verbal but not so good on the quant. Grrr. So, I live in a world in my mind where everyone is doing only quant. Ugh, I have to pull my head out of this occasionally. Anyway, the article that I referred to was discussion the spread for the top 25% in the QUANTITATIVE section. I don't know what that tells you, except that if I were ets, I'd strive to structure both tests that way so that scores were related to percentiles in the same manner for both verbal and quant. In other words, both would be beautiful normal curves with the same standard deviation and mean. At least that's what I'd aim for. I don't know if they can do that as percentiles will change as people get used to the tests and I don't supposed they'd change in the same manner. But at first it seems like they would use different formulas to get from the raw score to the 130/170 scale score which would have the same percentiles. But that's a huge guess on my part!!!!

As for the sample scores, I didn't really trust them. I just assumed they were made up, but I could be wrong. The article I linked is the only information I found that was real information from ets. Again, just shooting for a best guess here. I know as much as you do at this point.

Caffeinated: As I see it, the problem with moving from a range in the previous scale to a guess in the new scale is that in the quant the top 6% are all bunched into an 800 score. Hmm. I've been getting ranges like 750 - 800 as well, but no smaller range (on practice tests only). For instance, nothing like 790-800. Has anybody else received a tight range near the top? My impression is that near the top things get very hazy because a 750 is only about the 82nd percentile. That's not a very tight guess. Personally, I think they're just being very careful in giving estimates, but I've only taken practice tests. And of course if you're hoping for 97th or 98th in the new test, you can't translate that from the old test. You can only know that you're somewhere above 94th percentile.

Btw, Princeton Review is giving their online practice test results in the new scale, but no percentiles which is what got me searching in the first place. I hope their scores and percentiles are close to what I'm estimating. I'm really losing patience with these tests that don't give feedback on how you'd do on the real test!

[Daisy Fields]

Decaf: I did not look at the sample report at all. I calculated using the 15 point spread. So,

Assuming a normal distribution and a mean of 150,

((170-15) - 150)/(standard deviation) = .674 (z score for 75th percentile)

Which works out to sd = 7.42

Then I just I used this formula:

(score-150)/7.42 to find the z score for each score possible in the top 25 percent and used a standard z score table to find the percentiles.

[crater21]

Thanks everyone for this really useful analysis! I too have been really frustrated not knowing how my practice tests will translate into percentiles. This begins to give us some picture of what might be in store.

[chak]

Daisy: who is decaf? lol

On another note, yes 750-800 is the highest range anybody can get. I myself dislike the ranged score that they give us because it means a huge difference once it translates to percentiles so I usually just use the smaller number in the range (e.g. 650 from 650-730) to find my percentile; at least in this case, I won't get any strange surprises

[Daisy Fields]

chak: I know, another goof!

[chak]

haha, I'm just really interested in how this will work out when it's said and done. It's just strange because of the percentile discrepancies that I get from the various study books. =/

[lollipop]

Daisy: who is decaf? lol

lol

the title above avatars is really confusing here and I made the same mistake!

[orst11]

I have seen on Delaware's Poli Sci program they have already figured they are setting a minimum score of 153 on each of the sections, whereas on the old format it was 1100. Here is the link:http://www.udel.edu/poscir/grad/prospectiveStudentsFAQ.shtml

[Kitkat]

It's nice to know that the programs are already thinking about this, but until we (and adcomms) see the actually outcomes from the first test, I don't know how much that will mean.

I think that a lot of science programs will be depending on percentiles still. I think that they are going on the idea of the fact that it is still a comparison of how you do compare to other people.

[H24]

Thanks for all the thoughts on this, and especially to Daisy for working out some of the possible percentiles--what a mess in trying to understand what to make of your score.

As if there wasn't already enough pressure around these issues. It'd be great to hear some actual scores people are getting too (or the ranges that is, since it sounds like that's all you get until November or later this year)...

[H24]

By the way, i was just going over some of my numbers for the little pile of practice tests i've done, and Daisy's numbers seem to jive with what i was seeing there.

E.g., in the math section i got 26 out of 40 (& 39 for one) right in 3 out of the 5 practice tests i've done so far (or 65% correct for 2 and 66.7% correct for the 3rd)--one was ETS' online PowerPrep test, and 2 were Princeton Review's online tests.

For the P.Prep one it estimated my score, using the old scoring system, as being b/w 680-780, in the 66-89 percentile range; that was one that i got 65% correct (26 of 40).

The 2 Pr. Rvw. tests, like Daisy mentioned above, somehow are ahead (if they're accurate-ish) of ETS and give you your score estimate in the new scoring format. So i got a 154 for both of these 2 tests--that's 26 out of 40 for one (65%) and 26 out of 39 for the other (66.7%).

SO, all to say that if i got 26 out of around 40 on all of them/around 65% of the Q's correct, giving me a 154 by Pr Rvw's estimate, and a 680-780 by ETS' estimate--and more importantly falling in the 66-89 percentile in ETS' estimate--that all seems to confirm Daisy's estimates above (where she guessed a 155 would be in the 75th percentile).

That seem reasonably well reasoned to everyone? Hoping it bears out, both in Daisy's estimate being right and in my eventual percentile after i take it tomorrow!!

All best to everyone as you traverse this madness they are calling education (which i think is only half-accurate)....

[George2248]

Nice analisys...

Does anyone knows then what are the percentiles requierements for tops universities like Berkeley Engineering, Columbia and University of Illinoi.

Im also getting in PowerPrepp around 750 and in Princeton Tests around 159 in math (maybe that can help you with your data)

Thanks... nice work

[Daisy Fields]

Thanks everyone for your feedback. Very good comments!

H24 and George - thank you for posting your practice test results. That's really good information. George, it looks like you're getting around the 82nd-83rd percentile on the PowerPrep and the 89th percentile on the Princeton tests (if my percentile guesses are correct). That's pretty close.

Here is what my practice tests scores have been, just to add to our little heap of knowledge:

Manhattan 800 (94th percentile)

Kaplan 750-800 (80th - 94th percentile)

Princeton Review 164 (97th percentile)

I haven't taken the powerprep yet, but I'll post when I do.

[Daisy Fields]

Okay, just took the PowerPrep and got 750-800. Which really tells me nothing at all since it spans 18 percentile points.

[Kitkat]

On the PowerPrep I got a 730-800 on the quant, and 670-770 on the verbal. Have yet to take the two online Princeton Review and Kaplan practice exams, but I will update later with those and with my final scores later.

[George2248]

Ok Daisy... so if I am correct, according to your data, the relationship between old and new scores is not proportional (15 points Old = 1 point new); that is good to know because I was afraid the 159 translated into 635 which is no that good. I still have 2 more princeton practices and one kaplan, I will post results then.

What about the percentiles required for top schools? do you know where can I find that info? I am interested in Engineering for Berkeley, Columbia and Illinoi.

[crater21]

Just took the Princeton Review practice:

Quant: 157 (29/40)

Verbal: 161 (31/40)

Kaplan, on the other hand, gave me:

Quant: 720 - 800 (24/40)

Verbal: 720 - 800 (34/40)

[crater21]

P.S. Did anyone else notice huge typographical errors on the Princeton Review test?

One of the quantitative comparison questions lists the following equation: a + b + 1 = 3

So, I solved the question based on this. But, in the answers, the equation is given as ab + 1 = 3. Obviously, that made a difference as to whether you got the right or the wrong answer! I hope this doesn't happen in the actual test.

[bluejay16]

A friend of mine just took the revised GRE and her immediate score was not given to her with the new scores. Instead the comp showed her an approximate score range based on the old GRE scores, for example, 450-520 for the Verbal section. Hearing this confused me because I was assuming that ETS would present the score with the new score scale?? Has anyone taken the revised GRE yet and can you give some more info on this?