Revised GRE score range hypothesis

[resource]

I've been doing some speculating and thinking on what the revised GRE score ranges actually mean. I am going to propose a simple hypothesis I've been rolling around in my head, and I'm curious to see what flaws/improvements others can see. After all, we need something to occupy our minds until November.

Given what we know, the revised GRE has been spitting out 100 point ranges for those not bounded by 800 (i.e. 580-680) and smaller ranges for those bounded by 800 (i.e. 720-800). We also know that the highest possible score at this point is 750-800. Based on the information provided by ETS concerning the score ranges, the scoring practices on PowerPrep II, and the scoring table in ETS' book, I am going to make the assumption that these scores are raw scores based on the scoring rubric of the old GRE prior to weighting for difficulty of questions. This implies that the score is computed rather simplistically (i.e. # wrong on the quantitative section equals some single-valued score for that section). Then, ETS runs their algorithms to weight each question for the exam by the percentages of people who get certain questions right/wrong. This obviously cannot be performed consistently until enough people have taken the exam -- hence, making us wait until November for scores and dropping the price to encourage more people to take the exam. This leads me to believe that they haven't weighted the unofficial score ranges yet, whatsoever.

Further, I am going to suggest that the score ranges are centered on the single-value of the raw, unweighted score. The ranges, then, should be interpreted at the midpoint with 50 points on either side to account for the difficulty of the exam. Thus, a 650-750 on a difficult test might reflect a true score greater than 700. Conversely, a 500-600 on an easy test might reflect a true score less than 550. Additionally, the 750-800 range is more realistically 750-850, thus the lower half of the 100 point range reflects the possibility that an 800 was achieved on a relatively easier test and would be subject to deflation. I think this theory accounts for the improved precision as the range moves towards 800.

I guess the real benefit of this, if it holds any water, is that we can assign some subjective belief about how difficult our individual exam was (relative to PowerPrep as a baseline), to determine where our actual score will fall within the range we were given.

A note of caution: I am hesitant to speculate on how this will map to percentiles. The entire argument above is based on the assumption that the score ranges (out of 800) can be estimated by the raw score (number of questions right/wrong) obtained. Thus, the percentiles are determined largely independent of the raw score -- which has historically been the case. And then, depending on how ETS wants to shape the distribution of new scaled scores (out of 170), they'll assign these scores according to the percentiles they want to represent.

Thoughts? Comments?

[Timshel]

I think you have some merit here. My real question is, do you think it's a high possibility that someone could actually end up with the equivalent of the lowest score projected? So, if someone got a 600-700, they could actually end up with a 600? It seems to me that it is more likely that the person will end up with the middle number, give or take one or two on each side, so for my example, they are most likely going to get a 630-670, and that the rest of the range was just easier and gave them more cushion in case of extreme cases/discrepancies?

[resource]

I think you have some merit here. My real question is, do you think it's a high possibility that someone could actually end up with the equivalent of the lowest score projected? So, if someone got a 600-700, they could actually end up with a 600? It seems to me that it is more likely that the person will end up with the middle number, give or take one or two on each side, so for my example, they are most likely going to get a 630-670, and that the rest of the range was just easier and gave them more cushion in case of extreme cases/discrepancies?

I think that would be a fair assumption. I kind of think about it as a normal distribution around the midpoint with a standard deviation of 20 points. So, with a range of 600-700, you have a 68% chance of scoring within 630-670.

[habanero]

So you think that if someone got a 710-800 range, their likely score would be 740-780 as opposed to 735-775? That makes a lot more sense (maybe because it is exactly what I want to hear!!).

[cxxxxxx]

wow, first I have to say, to me, it seems bizarre to have thought about the scores this much!

I understand where you are coming from but wouldn't we all be better off devoting our time to revising further rather than hypothesising on possible score calculations? I bet the guys at GRE would have a good laugh at this. Although, obviously, you are going to be getting a much higher score than me!

I agree though, how else to pass the time... when I took the test I had no idea the scoring wasn't finalised and it completely baffled me after 3 hours to be given an estimated score. I like to imagine the GRE examiners all in lab coats in some high tech facility where everything is chrome ad polished with some giant computer spitting out printed results like confetti...

I

[habanero]

When you need to get a 770+ to be competitive, getting a ranged score is torture. If I get a 750, I will need to retake it. Being able to logically think that I will likely get 780 makes me feel much more comfortable.

I looked at your profile and you are applying for MFA programs...many of those do not even require a GRE. Those of us wishing to get into top science programs have very little wiggle room with quant scores. A similar analogy would be applying to top MFA programs with a 2 photograph portfolio. Your genius might shine through, but it'd be hard for the admissions committee to justify even giving the app a second look.

[Eigen]

I can't think of any program that relies so heavily on GRE scores that you need a 770+ to be competitive, and a 750 will require a re-take.

Unless the rest of your application is really weak, I doubt any adcom is going to throw out your application with a 750 quant, even in the physical sciences/engineering- most realize that the GRE quant just isn't that good of a measure of your ability relative to the rest of your application.

A 700 quant might require a re-take, but you still probably won't be thrown just based on your score even then.

Just for reference: MIT Computer Science doesn't even use the GRE for admissions. Berkeley CS has no minimum requirements, and only states an average of 595 Verbal and "most applicants" above the 90th percentile in Quant. That means they didn't throw out all the applicants under 90th percentile without looking, or it would be "all applicants are above the 90th percentile". When schools say no minimum score, they usually mean it, and that usually means they're looking more at the rest of your application than your standardized test scores.

Edited September 28, 2011 by Eigen

[habanero]

Yes, but I went to an undergraduate institution with no grades. Are you speaking about those people who went to small non-research colleges for undergrad and have no GPA? Additionally, 90% is an average of around 790. While my app is strong and I did well in school/ UG research, I highly doubt that admissions would overlook a 750 Q in my specific case. Unless I am totally off (which would be very fine by me)?

[Eigen]

I honestly can't comment on your specific case, but I do have a recommendation for you:

There's a poster here who's on an adcom in CS, he has a post in the NSF GRFP thread for this year- I'd suggest PMing them and asking.

I also can't directly comment on CS, but I think for my field (Chemistry), a 750 wouldn't get you thrown out, and it would come down to your research experience, SoP and letters.

And I thought a 90% was closer to around 770, but that's really neither here nor there.

Edited September 28, 2011 by Eigen

[cxxxxxx]

I'm confused, can someone explain the reasons behind the GRE exam, I thought it was a kind of qualifying exam and a higher scroe could help your application. Yes its not a requirement for an MFA but as I had in the past studied a MSc which I left to study art I thought it was worth my time to take this test hoping it may show dedication and also should I have the option of taking credits/modules in other subject areas in order to help my practice I may need to show my abilities.

Does anyone know the exact date of the 'real' results being available?

I don't mean to make light of the examination, its tough! I realise that the slight difference of 20 or so points could make all the difference in some applications. I am concerned about my verbal as an MFA requires a significant writing ability and a Q score significantly higher than my V score could raise some questions.

[resource]

I can't think of any program that relies so heavily on GRE scores that you need a 770+ to be competitive, and a 750 will require a re-take.

Unless the rest of your application is really weak, I doubt any adcom is going to throw out your application with a 750 quant, even in the physical sciences/engineering- most realize that the GRE quant just isn't that good of a measure of your ability relative to the rest of your application.

A 700 quant might require a re-take, but you still probably won't be thrown just based on your score even then.

Just for reference: MIT Computer Science doesn't even use the GRE for admissions. Berkeley CS has no minimum requirements, and only states an average of 595 Verbal and "most applicants" above the 90th percentile in Quant. That means they didn't throw out all the applicants under 90th percentile without looking, or it would be "all applicants are above the 90th percentile". When schools say no minimum score, they usually mean it, and that usually means they're looking more at the rest of your application than your standardized test scores.

Economics is pretty brutal in terms of quantitative GRE scores. It is thought that admissions committees sort applications based on GRE scores, and then pore through them from top to bottom. A 750 puts you on the bottom (and top programs typically get 500+ applications).

In my situation in particular, I scored a 760Q on the old GRE and my new 750-800Q is unsettling only because of the possibility (albeit, a small one) that I didn't exceed my previous score.

[process chemist]

This is a very productive discussion. I retook my GRE. I bombed the old one, but I did much better on the revised version, and I even bumped my quant score up (I don't know by how much thought). I got a 650-750Q, what does this mean, I have no idea, but I hope it means something good. My question is more general in how will the adcoms at grad programs, and fellowship orgs. view the new scores. I heard this rumor of a conversion chart, but will they look more at hard numbers or percentages. I have a feeling that the percentage will count more than your actual scores using the new scale. I just submitted part I of my fellowship app, and I hope all turns out well.

[patientpatient]

Interesting discussion. Just to let the topic initiator (is it called an OP? Not hip on the discussion board lingo) know, some 800-bound scores had a 100 point spread. I had a 700-800 on both sections. Could this be a recent development? I don't think so, but if it were, wouldn't that show a lack of confidence in previous score ranges. My guess is that that particular range just hasn't been that popular (what with a 750-800 corresponding with a raw score of 0-4 or 5 missed questions of 50 and a 700-800 corresponding to only (I think) 8 missed questions).

Is it safe to assume that a 700-800 means that I will likely not have a score at the top of that range? Are my chances for a mid 700 on either section (my goal before the test) any better than if my score band was, say, 670-770? This has me freaked out.

And, as a hopefully comforting aside to those concerned with overinflated score ranges, I was set to take my test last month (when it was half off) with a bunch of friends. I had to reschedule, but of my friends that took it that day, most scored in the 500-600s -- and they're pretty bright. One or two even scored in the 400 range. So I think this forum represents a pretty self-selecting group. Plus, let's be real, most people slightly inflate their scores.

[resource]

Interesting discussion. Just to let the topic initiator (is it called an OP? Not hip on the discussion board lingo) know, some 800-bound scores had a 100 point spread. I had a 700-800 on both sections. Could this be a recent development? I don't think so, but if it were, wouldn't that show a lack of confidence in previous score ranges. My guess is that that particular range just hasn't been that popular (what with a 750-800 corresponding with a raw score of 0-4 or 5 missed questions of 50 and a 700-800 corresponding to only (I think) 8 missed questions).

Is it safe to assume that a 700-800 means that I will likely not have a score at the top of that range? Are my chances for a mid 700 on either section (my goal before the test) any better than if my score band was, say, 670-770? This has me freaked out.

And, as a hopefully comforting aside to those concerned with overinflated score ranges, I was set to take my test last month (when it was half off) with a bunch of friends. I had to reschedule, but of my friends that took it that day, most scored in the 500-600s -- and they're pretty bright. One or two even scored in the 400 range. So I think this forum represents a pretty self-selecting group. Plus, let's be real, most people slightly inflate their scores.

OP is indeed correct ("original poster"). The 700-800 range is not a recent development -- it's been consistent with powerprep scores, ETS practice book scoring approximations, and actual reported scores. I don't think it represents a lack of confidence in the point spread. I think the ranges allow for ETS to say with >99% certainty that your score will fall within your range -- this shouldn't change for narrow ranges since, if my hypothesis holds any water, a 750-800 is more accurately a 750-850 with the top half of the distribution truncated.

I would interpret a 700-800 at the midpoint of the range. I think it's equally as likely you got an 800 as you did a 700 (which are both very small probabilities). That said, a 700-800 gives a better signal for the mid 700's than a 670-770 would. Try to imagine a normal distribution between 700 & 800 with a mean 750 and a standard deviation of 15. That should be your expectation. For more generalizability, imagine a normal distribution with (mean=lower bound of range + 50, sd=15).

And yes, the small group of folks who are avid posters on these online forums are definitely a self-selecting group. I don't, however, think people embellish their scores to impress anonymous internet folk. At least, I'd hope not.

[jpo645]

So here's the thing. Page 117 of the GRE Prep practice booklet from ETS lists predicted score ranges based entirely on your raw score from the practice test. There's really little else to go on except a table on the preceding page which shows what percentage of test takers correctly answered each question. With that said, the range probably matches the best and worse case scenarios based on your raw score. The low-end means every question answered had the easiest weighting, the high-end every question a hard weighting. I reckon the computer-based exam is the same thing or near similar.

[Grunty DaGnome]

Nice catch on the page 117 chart! I think you settled the question.

[contretemps]

So here's the thing. Page 117 of the GRE Prep practice booklet from ETS lists predicted score ranges based entirely on your raw score from the practice test. There's really little else to go on except a table on the preceding page which shows what percentage of test takers correctly answered each question. With that said, the range probably matches the best and worse case scenarios based on your raw score. The low-end means every question answered had the easiest weighting, the high-end every question a hard weighting. I reckon the computer-based exam is the same thing or near similar.

But that's for the Paper-based test. Correct me if I'm wrong, but from what I've heard, the scoring for that is slightly different to make up for the adaptive nature of the computer-based test.

[Kitkat]

But that's for the Paper-based test. Correct me if I'm wrong, but from what I've heard, the scoring for that is slightly different to make up for the adaptive nature of the computer-based test.

It is for the paper based test, but I think it might not be that different from the computer based version. I am assuming that at least the first section for the computer based version is similar to the paper based version, as that is the base point for everyone on that test. Where you go on the second section is dependent on the first and where I think the variation between the paper and computer based versions begin to change. I would assume that with the computer based version there is a bit more variation that you can have on the amount that you can get wrong, depending on the difficulty of your second section, and what kind of grade you might get. I think that the paper based might be close to say a middle of the road version of the computer based version.

Although I think that the amount of questions you get right and wrong shouldn't matter, rather how many you get right or wrong in comparison to other people. After all, when you take the test, how do you know how many you got right or wrong, and how much control do you have over that other then knowing the material and how to get through it?

[eomentl]

So here's the thing. Page 117 of the GRE Prep practice booklet from ETS lists predicted score ranges based entirely on your raw score from the practice test. There's really little else to go on except a table on the preceding page which shows what percentage of test takers correctly answered each question. With that said, the range probably matches the best and worse case scenarios based on your raw score. The low-end means every question answered had the easiest weighting, the high-end every question a hard weighting. I reckon the computer-based exam is the same thing or near similar.

I don't think this helps very much. I don't know about you guys, but the Revised GRE I took only had 40 questions per category. The list in the pdf is scaled for 50 questions per category.

[jpo645]

Ultimately, how the computerized version is graded may be different but that's not what's important here. The question is: how are the gre score ranges estimated? Remember, the paper based test hasn't been adjusted for new test takers yet, either. And it does not make sense for ETS to come up with two different methodologies for score estimation.

[jpo645]

Sorry for the annoying formatting, but I'm just trying to stress that the score estimate is simply a range to pacify us while we wait. It does not require any basis in how the actual test is scored. It's even possible (although unlikely) that we might score outside the ranges provided to us. Such a scenario would still fit with ETS's description of their score ranges just "giving us an idea" of how we performed.

[habanero]

Actually, 159 Q was a 750 on the prior scale and is an 82nd percentile. http://www.ets.org/s/gre/pdf/concordance_information.pdf

[jadler]

Oops, you're right. The numbers I was looking up were wrong. Thanks.