GMAT Score Penalties

Regarding the penalty for omitting questions, it's proportional...but the proportion is taken off of a score (your "theta estimate") that only the algorithm sees before it converts it to your scaled score. So essentially if you leave 3 of the 37 quant questions unanswered, your quant theta score will get multiplied by 34/37, and then that adjusted theta will be converted to your scaled score.

So that's the explanation. Then for how you should approach it...

1) GMAC's psychometricians released a blog post on the subject: https://www.mba.com/us/the-gmat-blog-hub/the-official-gmat-blog/2017/july/tactics-and-guessing.aspx

2) I did an interview with Dr. Rudner, the former chief psychometrician at GMAC, on the subject: https://www.veritasprep.com/gmat/talking-with-the-testmaker/#Rudner_Lesson_2

3) And *really* good strategy rewards you for thinking about when to guess before you ever have to rush to either guess or omit: https://www.veritasprep.com/blog/2014/04/gmat-tip-of-the-week-your-3-step-pacing-plan/

Then for your second question, here's Dr. Rudner and I explaining that one, too: https://www.veritasprep.com/gmat/talking-with-the-testmaker/#Rudner_Interview_17

I hope that helps!

Hope you've been doing well Brian!

This is a technical discussion that won't be of interest to test takers:


I've always been curious about this. There's confirmation from a couple of sources from years ago that the penalty is proportional, so some "score" is being reduced by k/37 when you leave k questions unanswered at the end of the Quant section. But it's never been clear just what "score" is being penalized - the scaled score, the percentile, the theta score, or some other interim score? I just discovered your Lawrence Rudner interview videos (which are great, thanks for those!), but before seeing those interviews, the only numerical data I could find about the penalty was in the OG, in one of those "Myth vs Fact" bubbles (on pg 13 in OG2018). There, the book explained that five unanswered questions in Verbal would reduce a Verbal score from the 91st to the 77th percentile. If you go back to the percentile tables when that blurb was first published (they haven't updated the numbers in that fact bubble since, so in later OG editions those percentiles often don't match any specific Verbal score), those percentiles corresponded to the scores V41 and V36. So I've always assumed, without being certain, that the penalty was applied to the scaled score, because in the OG example, with five unanswered questions, (36/41)*41 is indeed equal to 36.

After watching your second interview video, I'm still not sure what to think! Rudner first says that the penalty is applied to an "interim score", but later talks about applying the penalty to the scaled score ("your scaled score will go from 39 to 88% of 39, which is 34"). I don't think it's possible that the penalty is applied directly to the Theta value, because an average Theta value is 0 (and a below average value is negative), and applying any proportional penalty to a score of zero wouldn't change the score at all. My hypothesis now is that the "interim score" Rudner refers to is the scaled score, but before it is rounded off to the nearest integer. So someone might leave two Verbal questions unanswered, and have an unrounded scaled score of, say, 33.47 in Verbal, and the penalty would produce a score of (39/41)*33.47 = 31.84, which would become a V32. But I still can't be sure that's right.

Thank you. This is news to me. I have followed the old principle stated by GMAC that they take 3 percentile points from the scaled score for each question that was left unanswered - I have done some googling and it seems that is no longer the official policy/approach they are publishing and some of the docs have actually been removed... interesting. GMAC are not really saying what approach they are taking though but it was great to see the video touching on that here: https://www.gmatclub.com/forum/veritas-prep-resource-links-no-longer-available-399979.html#/talkin ... r_Lesson_2. (Choose Video 2)

Doing the math in the video, it seems the old rule of 3 percentile point reduction seems to hold (I know I am going to get a bunch of hate comments on my imprecise calculation but if one to had to invent one, what would you use?). In the example, leaving 5 Verbal questions unanswered took the score from V39 to V34, which is a reduction from 89th percentile to 72nd (17 percentile points)

To confuse the whole thing, I found this paper on GMAC's site that almost implied it made no difference in guessing vs. leaving questions blank (circa 2009): https://www.gmac.com/~/media/Files/gmac/ ... ssWhat.pdf

Awesome discussion, guys! And good point, Ian - I know what we do on the Veritas Prep tests is take the theta score (which like you said is on a -3 to 3 scale), add 3, then apply the x/37 proportional deduction, then subtract 3 to get it back on the scale. That may not be what GMAC does though (although Dr. Rudner signed off on our methods when we had him out consulting on our IRT system, so it's at least really close). That idea that the interim score may just be the not-yet-rounded scaled score is pretty interesting (and probably has an eerily similar effect, but would make a lot of sense, too).

Then as far as the net effect, yeah that "take off approximately 3 percentile points per omission" seems like a pretty decent rule of thumb at least for students to think about. Kind of cool how that still seems to mostly hold up even though that's not the actual calculation that the scoring system uses.

Since we're invoking Dr. Rudner's name a lot on this thread, I'd be remiss if I didn't honor this point he likes to make: "please don't call it a penalty!" Pretty soon after I met him he mentioned how he really liked the Veritas Prep "Think Like the Testmaker" philosophy but wished that we (and all test prep companies) would stop using terms like "trap answer," "scoring penalty," and "the GMAT punishes..." so often, because in his words "we're not trying to trap, punish, or hurt anyone...we're trying to ensure a fair assessment of candidate abilities." So we agreed that we'd talk every bit as much about the thought processes that the test *rewards* as we do about those that lead to incorrect answers. So call it a penalty if you like, but I think I've promised at this point to call it a "proportional adjustment!"

It shouldn't be too surprising that you get similar results whether you apply the penalty to scaled scores, or to theta values, since historically, the scales were designed to be perfectly interchangeable. On the 0-60 scale, 30 was originally average, and 10 points corresponded to one standard deviation, and on the theta scale, 0 is average and 1 point corresponds to a standard deviation. So sixty-odd years ago, when the test was first developed, you'd get identical penalties whether you adjusted the scaled score proportionally, or the theta score (after adding 3). Of course, back then the penalty didn't exist, but if they had CATs back then, it wouldn't have mattered which score you adjusted.

Verbal scores still hew fairly closely to that original distribution, which would explain why the results we're getting are all quite similar. Quant scores have drifted quite far from that distribution though, which is why I wish I had even one data point that told me how Quant scores are adjusted when questions are unanswered - with that info it would probably be clear exactly what the adjustment is. But I think the method you're using on your tests is either exactly what the GMAT uses (and it makes sense, considering the design of the scoring system), or is a very close approximation.

This was a good discussion. I am really grateful to you guys to share your views on my queries. This is much of a technical discussion, so what would you suggest to a student - to attempt all the questions (though make few guesses if possible) and not leave any unanswered questions to be on a safe side both in verbal n quant? If I am stuck on any question, should I simply guess my best option and move forward so that I am not left with any unanswered questions at the end? Since I am average in verbal, I am finding it difficult to attempt all the questions. I am left with 5-6 unanswered at the end.

Sent from my Pixel using GMAT Club Forum mobile app

According to a GMAC research report, test takers who are above average should always try to finish the test, even if they need to guess randomly. If you're around an average level (low Q30s or high V20s level) it usually won't matter too much what you do, but in that case I'd also recommend answering every question, even if you need to guess. If you are below average in a section, then it's usually better not to guess randomly just to finish (in that case, you're often guessing randomly at easy questions, and that can lower your score more than the penalty would lower your score).

Ideally though you wouldn't need to guess at the end of the test. You definitely should be moving on fairly quickly, in the middle of the test, if you get "stuck". On an adaptive test, that will probably happen several times, because an adaptive test will give you a lot of questions that are meant to be challenging for you. So it's completely normal to get stuck a few times, and you shouldn't be concerned at all if that happens. But you do want to be sure you don't invest a lot of time in those questions, since it usually won't help. Save that time for later questions that you will know how to solve, just as long as you have the time available.