Debunking the “First 7 Questions” Myth


I don’t even need to say what the myth is! Everyone already knows—that’s how pervasive it is. Ever since the GMAT and GRE CATs launched in the 1990s people have believed that the earlier questions are worth more, that if we could get the first 7 (or 5, or 10) questions in a row right, we’d be guaranteed a really high score.

And you’ve likely also heard that this is a myth: from me, from other teachers, from Dr. Lawrence (Larry) Rudner, Chief Psychometrician of GMAC (the organization that makes the GMAT). And yet so many people still talk about it and believe it—so who should we believe?

Let’s talk about this and, hopefully, lay the myth to rest once and for all.


What is the Myth?

Different people talk about different details: if we get the first 5, or 7, or 10 questions right, then we’ll get a high score no matter what else happens, or a higher score than we would otherwise get. (And, conversely, if we do poorly on the first 5, or 7, or 10 questions, then our score will be terrible no matter what, or lower than we would otherwise get.)

How Did the Myth Get Started?

Item Response Theory (IRT), the concept on which the GMAT is based, has been around for more than 50 years. In the 1990s, the GMAT decided to switch from the old-fashioned paper-based format to a new computer-adaptive testing (CAT) format based upon IRT.

During that timeframe, the Educational Testing Service was responsible for producing test items for both the GRE and the GMAT (now, the GMAT test items are produced by another organization, ACT). As ETS developed its new GRE CATs, it did initially have a format that emphasized the earlier questions. The test prep world soon figured that out and ETS redesigned the format so that this would no longer happen. That was the start of the myth: it wasn’t actually a myth at first!

But why did it persist? Everyone knew that the early form had been cracked and that ETS responded accordingly by changing things.

In 1999, ETS researchers presented a white paper at the annual meeting of the National Council on Measurement in Education. The researchers, Manfred Steffen and Walter D. Way, presented the results of a number of scenarios they ran to test the IRT-based algorithm and show how it worked. I’m going to give you some of that data below, but I want to point something out first: this paper is why the rumor persisted.

The paper itself is completely correct and accurate; in fact, it’s an example of quite good research. But a lot of that research was misinterpreted by people in the test prep industry—misinterpreted to mean that the earlier questions still were worth more and that students should spend a lot more time on those earlier questions. The paper, however, shows just the opposite.

One thing the paper tells us is that, for a true examination level of 650, answering the first two questions correctly vs. incorrectly results in a 31 point score difference (658 vs. 627). Note, though, that this assumes all the other items are identically answered…in other words, the examinee who answered the first two questions correctly didn’t have to guess on any questions at the end. How does that happen? The examinee doesn’t run out of time because the examinee didn’t take extra time (or not much, anyway) in the first place. (Also: no, you can’t really score 658 or 627 on the test; these are the average results of many simulations.)

One scenario they ran was what happened to the true scoring level if someone got the first, first two, or first three in a row right, or wrong, vs. the final question, the final two, or the final three. The results are very interesting.

For someone with a true scoring level of 750, getting the first 3 questions right (and leaving everything else the same) results in a 10-point lift to 760. Great! We should spend more time on the first three questions, right?

Not so fast. Getting the last three questions in a row wrong results in a 20-point drop to 730. What do these two data points really mean? If you can answer those first three questions correctly without sacrificing later questions, then great. Do it. But chances are pretty good that you’d have to spend extra time…and then your score will drop at the end. (And this is how the myth was perpetuated. People forgot that there are consequences for trying to get the first X number of questions right. You have to take extra time!)

At the 650 true-scoring level, getting the first 3 right results in a 16-point jump, to 666. Getting the last three wrong, however, results in a 15-point drop, to 635. Six of one, half a dozen of the other it doesn’t really matter.

I will concede that, at lower scoring levels (sub-550), a strategy that involves getting the first 3, or 5, or 7 in a row right works in theory because there isn’t much of a drop at the end for getting a similar number of questions wrong. There’s only one problem with this strategy. What are the chances that someone at a true scoring level of 500 is going to get the first 5 (or even 3) questions in a row right? Think about what happens each time you get a question right.

Next, the study talks about scenarios in which someone gets the first ten questions right or the last ten questions right. Here’s an interesting statistic: if a test-taker with a true scoring level of 670 gets the first 10 questions right, the study showed that the resulting score would be 728. Clearly, we should spend that extra time on the first 10 questions!

Except for one little detail. That part of the study assumed that the test-taker didn’t have to rush on any other questions. In other words, the study assumed that the test-taker didn’t need any significant extra time in order to get those first 10 questions right. (And, again, we missed this in the 1990s when interpreting this data.)

By the way, what can you do if you simply don’t know how to do a problem? Will spending an extra minute or two help? The vast majority of the time, no. If you can’t do it in the normal time (or perhaps about 30 seconds above normal time), then this just means that you don’t actually know what you’re doing, since there is a “normal-time” solution. Spending even more time, then, is not going to do much (except blow time).

And finally we get to the portion of the report that mimics real-life conditions the best: what the report calls the early care / late guessing condition. In this scenario, the test-taker takes additional time on some early number of questions and then has to guess on questions towards the end in order to finish the test on time. There’s one more not-so-minor detail. This scenario assumed that, for the early care (extra time) situation, the tester would answer every single question correctly. That is: you spend more time, you automatically get it right. I don’t need to point out that it doesn’t usually work that way, right? : )

If a test-taker with a true scoring level of 500 answers the first 5 questions correctly, then that test-taker is likely to end up with a higher-than-500 score as long as he doesn’t guess on more than 6 questions at the end. That sounds pretty good—until you remember that this requires a 500-level tester to answer the first 5 questions in a row correctly. Again, remember what happens when you answer questions correctly.

What about a tester at the 780 level? This tester has a pretty good shot at answering the first 5 in a row correctly. In this scenario, however, the tester cannot guess on more than a single question at the end without the score dropping below 780. Only 1 question!! If this tester answers the first 10 questions in a row correctly, he can guess on no more than 3 questions at the end before the score level drops below 780.

What does all of this mean?

If you’re going for a 750+ score, then the strategy actually boils down to this: get the first 5, or 7, or 10 in a row right while spending barely any extra time so that you have to guess on zero or very few questions at the end—otherwise, your score will actually go down. (By the way, if you can actually get the first 5+ in a row right without spending extra time, then you don’t need to worry about any of this. You just do all the questions normally for you.)

What about a 500-ish score? We’re allowed to guess more at the end but we’d still have to get 3+ questions in a row right at the beginning of the test, and we all know how the test works. I’m going to start the test with a medium-level problem. If my level really is around 500, there’s no way I’m going to get 3 in a row right, because that third question is going to be way too hard for me no matter how much time I spend (and possibly the second one as well).

What about in the 600 range or right at 700? The final scenario in the study (early care / late guessing) didn’t provide data for these specific scores. But look at all of the data given collectively in the paper. We haven’t found one case in which spending more time on the early ones actually works. Basically, we’ve got a tug of war between how many questions we really could get right in a row and how many times we’d have to guess at the end as a result—and the data shows that they’ve figured out how to balance this in such a way that gaming the test is just going to backfire in the end.


1. It really is a myth. Don’t spend lots of extra time on any one problem anywhere in the test. It’s not worth it.

2. The real strategy that derives from the research is: get everything right that you can without spending a bunch of extra time. (This does not mean that you can’t go 30 to 45 seconds over the average time when you think some extra time is warranted! Beyond that, though, the extra time is likely just indicating that you don’t know how to do the problem anyway.)

3. Read these two articles: In It To Win It and Time Management.

All data points cited from Test-Taking Strategies in Computerized Adaptive Testing, Steffen, M. and Way, W. D., Educational Testing Service, April 1999. Presented at the annual meeting of the National Council on Measurement in Education.

  1. Pete November 10, 2016 at 2:00 am

    Hi Stacy,

    It seems to me that it would be beneficial to get the first several questions wrong. If you zipped through the first 7 questions and put C on all of them, then you would have an extra 15 minutes to solve the remaining 30 questions AND assuming you didn’t guess may right, you would get easy questions (that take less time) for the first several questions you try.

    Wouldn’t this leave you with roughly 20 extra minutes to slowly work your way up to the difficult questions and finish the test strong? This seem easier than answering question at the brink of your intelligence level the entire test. Do you have any info on if that is a failed strategy?

  2. Manhattan Prep March 14, 2016 at 4:07 pm

    Hi, Sravya. Here’s what Stacey had to say in response to your question:

    “No, the early questions are not worth more than the other ones. The main message of the article is that it is a *myth* that the earlier questions are more important.

    The test is not scored based on # correct, so looking at that data alone is misleading (and may lead someone to mistakenly draw the conclusion that the earlier questions are worth more).

    Think about a weighted average. (Note: I am using this as an analogy: the scoring is not a weighted average either.) If you had five quizzes in a class that were each weighted at 10% each and then one final exam weighted at 50% – that would be a very different grade than if you just took all 6 tests and calculated a straight average. Right now, you’re doing the equivalent of trying to calculate a straight average while the test is calculating a weighted average – in other words, you’re doing the wrong calculation. (Again, the test is not a weighted average either – I’m just using that as an analogy.)

    Your score is based not just on getting questions right vs. wrong but on the difficulty levels of the specific questions – think of those difficulty levels as applying “weightings” to calculate your score. But the test report does not tell you the difficulty levels of the various questions, so there’s no way for you to “backwards engineer” your score. You’re missing the necessary data.

    Your score can be adversely affected by strings of wrong answers – usually 4+ wrong answers in a row or a cluster of, say, 5 wrong out of 6. You missed the first 3 questions in a row, so that did hurt a bit.

    (Your next question is probably: how do I make sure I don’t get 3 or 4 wrong in a row? There’s no way to guarantee that you’ll get certain questions right. You either know how to do a question, or have a good idea of how to make a guess, or you don’t. Just do your best and move on when you’re not sure or when something is taking too much time.)

    The short answer: don’t think about ways to “game” the test. There aren’t any. Study hard and do your best to make good decisions. Know what you do not know, so that you can guess on those questions and move on to things that you do know how to do. Good luck!”

    Let us know if that makes sense to you!

    Manhattan Prep

  3. Sravya March 13, 2016 at 3:30 am


    I got 9 questions wrong in the verbal section of the purchased GMAT PREP Test 4. But, I got a score of 29. In GMAT PREP Test 3, the number of wrong questions were the same but they were not present in the first 10 and I got a 39; however, in GMAT PREP test 3 I had the four mistakes in the first 10 questions: I got the first three questions wrong and the 10 question wrong. Does this mean, one has to get the first 10 right always?

  4. Stacey Koprince November 24, 2014 at 6:22 pm

    @thegmatguru: I understand what you were trying to do, but it’s just these kinds of experiments that perpetuate the myth.

    First, the algorithm is quite complex; you can’t conclude anything based on running 1 scenario (even running the same scenario a few times). You would need to build a Monte Carlo simulation that would run the same scenario thousands of times to generate valid data. The data that I discuss in this article was produced in this way. (And, by the way, I could point you to plenty of posts / one-off experiments that try a similar test and come up with a result that does support what GMAC says.)

    Second, people who are already good at taking the test (as I’m guessing you are, both because of your handle and because you’re actually taking the time to play around with this algorithm – most “regular” students aren’t going to take an entire practice test just to do this!) are unfortunately creating an artificial environment in your experiments by “gaming” the test in a way that a regular student wouldn’t be able to do. It isn’t actually the case that spending more time on a question in general produces a greater likelihood of reaching a correct answer – not after a certain point, anyway.

    If I remember the number correctly, studies have shown that after about 2m42s on quant, your chances of answering the question correctly go *down* the more time you spend. So spending that much extra time doesn’t actually help a “normal” test-taker (vs. us weirdos who like these tests and generally do perform well on them).

  5. Ted November 22, 2014 at 9:40 pm

    @thegmatguru. A 49?

  6. thegmatguru February 12, 2013 at 1:06 am

    So I would be in really bad shape if I “took care” on the first 27 questions of the math section and then put “A” down 10 times in a row to finish the test? I actually did this on my last test as a timing/scoring experiment. Anyone want to guess what my raw score was?

  7. brand December 23, 2012 at 4:54 am

    @Starcraft 2

  8. brand December 23, 2012 at 4:53 am

    Great post. here are some recommended website for you : Whistler

  9. Wonderlic May 2, 2012 at 4:16 pm

    I think people who start off well and get their confidence up are more likely to do better because they feel better about the test and how they are doing on it. This leads to better results. I have to imagine the questions are randomized to protect from manipulation.