Tag Archives: GMAT scoring algorithm

yes, GMAT scoring is weird

Let’s play a little GMAT guessing game. Take a look at the two GMATPrep screenshots below, and see if you can guess the quant score for each of these tests. (Sorry, the screenshots aren’t beautiful, but I did my best to make them legible.)  Keep in mind that the GMATPrep software uses exactly the same algorithm as the real GMAT test.

On GMATPrep test #1, the student missed a total of 7 quant questions out of 37 (81% correct):

And on GMATPrep test #2, the test-taker missed a total of 21 questions out of 37 (43% correct):

Go ahead, take a guess. What quant score do you think these two students received on their GMATPrep tests? The lucky student who got 81% correct probably got a solid but imperfect score, right? And the poor schmuck who got 43% correct must have been vaguely suicidal after that GMAT quant disaster, yes?

Well, the first student got a perfect 51 on her quant section, and the second student earned a 44.  (For what it’s worth, he paired the 44 quant with a 41 verbal, for a grand total of 710, and he earned a very similar score on the real GMAT. He ended up at Harvard. Things work out.)

Surprised? It turns out that a “perfect” GMAT quant score doesn’t necessarily require complete perfection, and you can miss a ton of questions on the GMAT quant section without endangering your chance at a 700.

I discussed the GMAT scoring system in both a recent GMAT blog post (which profiled a mediocre math student who still earned a 720) and in an ancient, crusty GMAT blog post (which explored the fundamentals of the GMAT scoring algorithm), but I’ll say it here again: your GMAT quant score doesn’t really depend on how many questions you miss.  It depends on which questions you miss.

As you probably already know, the GMAT exam “adapts” to your performance, selecting each question based on your answers to previous questions.  Your final score is based primarily on the difficulty level of the questions you see on the test. If you see tons of hard questions, your score will be higher than if you see nothing but GMAT cream puffs.

By the time you reach, say, question #29, the GMAT scoring system already has 28 data points about your skills.  If you get 27 of the first 28 questions correct–as student #1 did–you’ll convince the GMAT scoring algorithm that you’re pretty awesome. And it will spit out the hardest question it can find.  Even if you miss that question, you’ll still have missed only two out of the first 29, and the computer will spit out another really tough question.

So as you look at the student who earned a quant score of 51 while missing seven questions, you shouldn’t be too shocked:  six of her seven mistakes came at the end of the test, once the GMAT scoring system had “already made up its mind about her.” All of the seven questions that she missed were unbelievably difficult, and the GMAT algorithm doesn’t really punish test-takers for missing hard questions.  Just ask Ms. A from this GMAT blog post.

Student #2 is admittedly an even more extreme case.  He missed 21 questions and still scored above 700; that’s not normal, but it’s obviously possible under the right circumstances.  In his case, he had a reasonably strong start to the test, missed only the toughest questions that the GMATPrep software threw at him, and managed to get just enough questions right to prevent a score meltdown.  Again, this is a great illustration of the GMAT scoring system:  you can miss piles of questions and still do well on the GMAT.  You just can’t afford to miss the easier questions, since those errors will send your GMAT score into a tailspin.

The bottom line is that a 51 isn’t necessarily “perfect” on the GMAT, and a 700 doesn’t necessarily require a high rate of accuracy.  You can miss tons of hard questions and still do incredibly well on the GMAT, as long as you don’t miss the questions that are relatively easy.   And when you see impossibly difficult questions on your GMAT exam, just smile, and accept the fact that you can miss them without torpedoing your GMAT score.

 

 

deciphering adaptive GMAT scores

Math-phobic students have always been a part of my life as a GMAT tutor, and I currently work with several New Yorkers who used to sweat profusely whenever they heard the words “slope” or “equation.” One of my favorite current students is a 34-year-old actor who has taken exactly no math since high school, and he is raising all sorts of interesting questions as he starts to ascend toward a decent GMAT score.

This particular guy (let’s call him Mr. L… that’s short for Law & Order, since he once appeared on an episode as the main victim) took a Princeton Review GMAT course, got a horrendous score on the math section (21, which is probably on the wrong side of the 10th percentile), demanded his money back from Princeton Review, hired a GMAT tutor in India (he was volunteering there at the time), and then managed to get a 33 on the math–a huge, huge improvement. (If anybody needs a GMAT math tutor in Ahmedabad, India, I can recommend a great one. Oddly enough, I also know a great auditor who lives there, too. And no, I’ve never set foot in Ahmedabad or anywhere else in India.)

Upon his return to NYC, Mr. L contacted me for GMAT tutoring, and we’ve been working together for about a month. His patience for the GMAT started to wane recently, and I suggested that he take a few practice math tests–not because I thought that he was almost ready to take the GMAT, but because I thought that online practice tests would keep him a little bit more engaged than paper-based practice problems.

And Mr. L pretty much crapped himself when he saw his first score: on a Manhattan GMAT test, he scored 40 on the math. I figured that it wasn’t a complete fluke–Mr. L had improved by quite a bit. But he was highly skeptical, so I asked him to take another MGMAT math test. And he scored 40 again. Clearly, the 40 wasn’t a fluke, unless you’re skeptical of MGMAT’s scoring (that’s another topic entirely, but I usually find that the math scores are fairly accurate for students who haven’t taken their course).

I haven’t quite succeeded in convincing Mr. L that he deserved the 40, and here’s why: on both tests, he got 20 questions right, and missed 17. In the American educational paradigm, getting just over half right usually means that you barely passed, which means that you suck. Mr. L couldn’t really get his head around this: he missed enough GMAT math questions to suck pretty badly, but his score was higher than he ever dreamed possible.

So here’s the deal: adaptive tests such as the GMAT and GRE are designed to make you miss lots of questions. (That’s one of many reasons why taking these tests can be such a painful experience.) Each GMAT question is essentially assigned a difficulty level–if it helps, you can think of each GMAT question as, say, a “700-level” or a “520-level” question. The test basically tries to figure out the level of question at which you get 50% right. It seems logical that you might be able to get 55-60% of the questions right, and still get a decent score–your score is based on which questions you miss, not necessarily on how many you miss.

If you don’t believe me, check out a recent entry in the official GMAT blog that addresses this issue.  Or keep reading.  Whichever you prefer.

If it helps, imagine that you’re destined to earn the equivalent of a 650 on the math section of the GMAT. (For the sake of simplicity, I’m going to ignore some of the complexities of GMAT scoring. Forgive me.) The first question of the test will be a 550-level question (roughly), and let’s suppose that you get it right. You’ll get a harder question next, and you probably won’t start to screw up consistently until you see a few 650- or 700-level questions. But it won’t take too long to get to that level–if you get the first three questions right, the fourth question of the GMAT quant section will probably make you sweat.

Now, imagine that the fourth question is a 650-level question… seems reasonable enough, right? If you’re a 650-level test-taker, you’re likely to miss about half of the remaining 34 questions. I’m obviously making some gross oversimplifications here, but it isn’t hard to imagine that you could get a 650 on the GMAT, or perhaps something even higher… without getting more than about 20 questions right.

My point is this: in all likelihood, you have a huge margin for error on the GMAT. As long as you don’t fall apart at the beginning of the test, you can miss tons of questions, and still get a fantastic score. So when you see some crazy, indecipherable GMAT combinatorics problem, there’s really not much harm in guessing and moving on–one missed question won’t hurt your composite score by much, and you’ll have plenty of chances to recover.

So if you’re reading this, Mr. L… have I convinced you yet? You actually deserved those 40s, and we’re not even warmed up yet. Crazy as it sounds, getting 60% of the questions right might be enough to get you wherever you want to go on the GMAT math section.