GMAT FORMAT IS CHANGING APRIL 16, 2018 - it will be 30 minutes shorter

I think what we sometimes forget is that the GMAT 200-800 composite score is a reflection of your RELATIVE PERCENTILE scores on the Quant and Verbal sections of the test (percent of test-takers who you outperformed during that particular 3-year period), not your ABSOLUTE PERCENTILE scores (your actual performance on the test, including percent correct, question difficulty level, and scoring algorithm management).

Confused? Think of your ABSOLUTE PERCENTILE as the percent of questions you answer correctly, and your RELATIVE PERCENTILE as the percent of students who you outperform.

On an easier test, you might earn a high ABSOLUTE PERCENTILE, but still earn a low RELATIVE PERCENTILE due to the unchallenging nature of the test. ("scaled down")
Conversely, on a harder test, you might earn a low ABSOLUTE PERCENTILE, but still earn a high RELATIVE percentile due to the challenging nature of the test. ("scaled up")

In other words, even if the GMAT does get objectively easier as a result of a shortened test with fewer experimentals, it should not get relatively easier: composite scores and percentiles should stay stable, though section scores, which unlike the composite score are absolute scores, may indeed rise. We have seen that to be the case over the last 10-20 years, when composite percentiles have stayed mostly the same (the only real change at the top end is that a 750 used to be 99%) while section percentiles have lowered drastically, on Quant in particular. (See examples below of Quant percentiles from 2000 and 2017, as well as composite percentiles from 2000 and 2017.)

That being said, since the GMAC's numbers take into account the last 3 years of test-taker information, there might indeed be a short window of time where one can take advantage of a slightly easier GMAT that the GMAC's curve hasn't yet compensated for (if it is in fact any easier), since the data is averaged out over the last 1,000 or so days, and the percentiles are only calculated once annually in July.

-Brian

I'm not sure what "absolute percentiles" means, but the GMAT's percentile calculation is done only once a year (not the last ~1000 days from any given day). In fact, it's going to be done very soon (end June). So for the full year after that (and not just the period shortly after), the percentiles will be locked at whatever values we get in July. So, if the exam does get "easier", the scores test takers get from now on are going to be compared with scores on the old format. The proportion of old scores used in the calculation of GMAT percentiles will start falling only from July 2019.

Not sure whether the GMAT is going to be any "easier" or "tougher", but it would have been good if they had managed to get the new GMATPrep out at least a month before making the changes they've announced.

I understand that the term "absolute percentile" must in some ways seem absurd, because a percentile score is by definition relative.

To clarify, I would define an "absolute percentile" as a percentile that does not vary from year to year in relation to scaled scores. GMAT composite percentiles can be considered absolute percentiles because they have barely budged within the last 20 years (see above tables) in relation to the actual composite scores--especially at the top end. The cutoff for 99%, for example, has always been either 750 or 760--it's remarkably consistent over time.

On the other hand, section percentiles are relative, not absolute, because they tend vary greatly over time--especially on Quant. For example, a score of Q45 was 82% overall in the year 2000, but now a Q45 only earns you a score of 57%! The scaled section score is absolute--a Q45 in 2000 is the same as a Q45 in 2018--but a 750 composite from 2000 cannot be compared to a 750 from 2018, because composite scores are relative to overall test-taker performance, and the test has become significantly more competitive since then.

Good point about the percentiles only being calculated once per year by GMAC, at the beginning of July. Hence, as I have already noted, if the the shorter, updated GMAT does indeed prove to be easier, which we have every reason to expect, then it might be particularly advantageous to take during the first year or so of its implementation, before the scoring curve and percentiles have been fully adjusted to reflect test-taker performance on the new exam. This advantage will persist in a diminished way during the 2nd year of the new exam, and will disappear completely after the updated GMAT has been around for 3 full years.

To summarize:

1) GMAT composite scores are RELATIVE to test-taker performance, but GMAT composite percentiles are ABSOLUTE--they do not change much over time, and they have a high degree of correlation with the overall composite score. In fact, you could argue that the composite score and the percentile are roughly the same thing, since composite score conversions change over time, along with percentiles.

2) GMAT scaled section scores are ABSOLUTE, but GMAT section percentiles are RELATIVE--they reflect the changing demographics of the test-taker population.

FAQ:

Q: Can I compare my 750 composite in 2018 to my (_______)'s 750 in 2000?
A: No, you can't. A 750 in 2018 is far more impressive than a 750 in 2018--even though a 750 in 2018 is "only" 98% and a 750 in 2000 was 99%.

Q: But you just finished explaining why composite percentiles are "absolute," which implies that they don't change over time.
A: Yes, that's true. They are absolute in the sense that they don't vary much relative to composite scores.

Q: So why is a 750 in 2018 more impressive than a 750 in 2000?
A: Because although they are (roughly) the same percentile scores in both years, a percentile score is by definition relative to the strength of the test-taking population, and the GMAT is a much more competitive test now, which can be seen in the significant rise in Quant scores over the last 2 decades.

Q: Wait, you just said that GMAT composite percentiles are absolute, and now you are saying that they are relative?
A: Yes, I can understand why that might be confusing. Both are true: although GMAT composite scores have a high degree of correlation with overall percentiles, you cannot compare percentiles from different years because they are relative to how prepared test-takers are for the GMAT during that particular period of time.

Q: So what CAN I compare over time?
A: You can compare scaled section scores, which--unlike section percentiles, composite percentiles or scaled composite scores--are consistent over time, due to the fact that the test content has not changed (much). A Q50 from 2000 is as impressive as a Q50 from 2018, for example. And anything above V40 (90% in 1998-2000 and 91% in 2015-2017) is excellent, no matter when you took the test. However, was it easier to score 99% overall way back then? Definitely. For one thing, there was no GMAT Club helping raise everyone's Quant scores. On the other hand, Verbal scores and percentiles have stayed about the same over the last 20 years--probably because the same international test-takers who are "ruining the curve" on Quant tend to be English as a second language students who on average, struggle more on Verbal.

You can even see these effects in the short term. I was lucky enough to earn my 770 (99%) in January 2012 with a Q47 / V48, though I suspect that particular combination of section scores would convert closer to a 760 or even 750 on today's GMAT.

-Brian

Hi, bb

Have you got a chance to attend a GMAC seminar to confirm that the number of SC question is indeed be reduced to 8-9? It is really hard for me to believe that and there is no way to confirm the rumors discussed above. Thanks in advance.

I have not seen a US webinar not have I received an invitation to it. I also have not seen announcements from MGMAT or Kaplan or Veritas Prep about the test structure which makes me believe there was not one yet for the US. Which is weird. (Of course i may not have been invited to one... you know invitation lost in the mail )

So the best info we have is what was shared in the webinar that Arjesh Arun attended.

Posted from my mobile device

Of course there's always the possibility (as others have already mentioned) that GMAC made a mistake in this webinar, and mixed up its figures for Critical Reasoning and Sentence Correction--otherwise the Verbal section has changed significantly in its overall composition--namely, the frequency/proportion of each question type (CR, SC, RC). For now, the question remains: has Sentence Correction truly gone from the most common question type on Verbal, to the least common question type, or was that just a type-O?

Timing issues would also have to be addressed, as most GMAT test-takers take far longer per CR question than per SC question. I know that I certainly do! Usually I take about 2 min per CR, 1.5 min per RC, and 1 min per SC. I'm excited to see the format changes in the ESR format as well--according to GMAC, these changes should apply to older ESRs as well, if you choose to re-download them.

GMAC has assured us (or at least implied--see below) that the GMAT itself and its balance of questions has not changed, so this appears to be conflicting information. If the balance/distribution of question categories on Verbal really did change so drastically, then you would think that the GMAC would at least share this news with us, despite its reputation for obfuscation.


source: GMAC website

For comparison's sake, here is the current proportion of question types on the Verbal section of GMATPrep: 17 SC, 13 RC, 11 CR. If SC has truly gone from the most common question type to the least common question type, then this is a fact that test-takers deserve to know before the new exam debuts on April 16th.

Then again, I love teaching the Critical Reasoning sub-section/question type, so I am OK with that news if it does turn out to be true. SC grammar (and the rote memorization entailed) can be mastered easily, but it gets a bit dull and unchallenging over time. On the other hand, there are so many logical layers to most CR questions--and at the most difficult level, they can be VERY challenging.

Of course, sometimes "very challenging" just means "poorly written," but I digress. I'm officially signing off on this issue for a while until we start to get some official test-taker information in about a week: until then it's mostly conjecture anyway.

-Brian

Yeah, I'm about the same, but like you I'm a fast-reading native English speaker. My non-native study buddies take longest on RC, because they take a long time to read the passage, and have to read it more times.

This change moves the verbal section away from a reliance on rote learning and towards a deeper understanding of concepts and applying those concepts to different scenarios, which I think is good. B Schools probably aren't interested in how many random grammar rules a candidate can memorize.

tinyinthedesert



I totally agree with the text in blue since the test is called Verbal Reasoning
and is not about knowing grammar rules or memorizing tons of idioms. If at all SC questions are indeed reduced ,
the only consolidation for the non-natives (read Asia Pacific and India in particular) would be to acknowledge the
struggle of natives (read US, in particular) in quant, accept the same and face test heads on.

That said, CR and RC are indeed lot difficult to improve for non-natives (leave along getting
them right under timed mock) than improving in quant for natives.

Hi,

Can someone please quote the official source for the break up of questions:
(a) RC: 13-14 (4 passages)
(b) SC: 8-9
(c) CR: 13-14
(d) PS: 17-18
(e) DS: 13-14

I do not find this information on the official website.

Thanks,
Rahul

Hi rahulchougule,

It's not completely official yet, but the source of information appears to be legit. That being said, the results are very surprising (SC is going from the most common question type to the least common question type??), so I have asked GMAC to confirm to make sure no mistakes were made. I believe that the original source is a GMAC webinar for India attended by AjiteshArun.

These numbers obviously conflict with the reassurances we were given by GMAC that the overall balance of questions on the GMAT has not changed, so something has got to give.

-Brian

Now that the weekend is (almost) over, I think we can expect more information from the GMAC.

They were answering all the other questions, but I had to ask about the number of SC questions four times to get a response, or even an acknowledgement that such a question had been raised. Oddly enough, only one other instructor seemed to be concerned about this (no one else was asking).

Here is actual slide that they shared:

I'll repeat a few points that I think might have been lost in all the conversation:

1. This is the distribution of questions in the GMATPrep, not the actual GMAT. Depending on what they need from a question development perspective, the distribution could change as a result of the distribution of experimental questions.

2. They were just relaying information the psychometrics team had sent them. They said they'd confirm that data and get back to us.

3. At the beginning, they had said that they'd tried to keep the "proportion" (not the word they used) roughly the same, which leads to me believe that they could have mixed up SC and CR.

4. This was a GMAC webinar, not a GMAC India webinar

Your updates came at 3:58 and 4:02 AM Pacific Time so I assumed that GMAC would not hold a webinar at such exotic time and it was a webinar for a non-US audience. Usually I also get invites to these events but not this one. I may have assumed wrong.

Posted from my mobile device

well, It also sounds astonishing to me that BB havent been invited to such a event. But I am still curious about the exact number of SC in the real exam. Since in the past record, it seems that test-takers will actually get more SC when they score 700 and higher, and less SC questions when they score below 700. Is that past record true?

This was for other comments in the thread, not yours. You were absolutely right that the webinar was only for India (and maybe some other countries in the region).

I'm just addressing some of the concerns in the thread that perhaps this information is not directly from the GMAC.

No, the exam cannot change the "ratio" on the basis of a test taker's performance. The exam would not be fair, and anyone who is very quick (at doing SC questions) would have a big advantage.

In other words, the GMAT algorithm takes care of content balancing. You don't have to worry about getting a significantly different distribution of questions.

Just in: Confirmed that the two (SC and CR) had been mixed up. Some other changes:

RC: 13-14
SC: 12-13
CR: 9-10
PS: 17-18
DS: 13-14

Thank you for clarifying and making me feel better about not being invited

At the same time, are they planning a US webinar? I have not heard any indication of that. From the the outside, seems a very haphazard approach to the whole thing...

PS. No, I don’t doubt the info. I am certain it is from GMAC (and we have another source confirming it too - experts global)

Posted from my mobile device

Posted from my mobile device

REALLY? GMAC? REALLY???

I am just going to stop at this.
Posted from my mobile device

Posted from my mobile device