At Beacon we’ve developed a pattern recognition system that we use to help clients rapidly improve their GMAT scores. Based on a data set of over 1,500 official GMAT questions, the system has identified several systematic biases in GMAT answers. One of them is the GMAT’s preference for shorter sentences in Sentence Correction questions. Let’s imagine a hypothetical test-taker Zoya.
Level 1: The Random Guess (20% success rate)
As the GMAT is primarily a multiple-choice exam with five answer options, the chance of Zoya randomly getting any question correct is 20%.
Level 2: The Shortest Answer (29% success rate)
Now let’s implement the learning from the bias our system has identified. If Zoya were to follow a strategy where for Sentence Correction questions she selected the answer option with the fewest words, Zoya would select the right answer 29% of the time.
This bias is not insignificant and is particularly helpful for the instances when Zoya is approaching the end of the exam, is short on time, and quickly tries to click through the remaining questions to get to the end of the exam.
Level 3: Remove Subject-Verb Errors (33% success rate)
But it gets better. Subject-verb errors are among the most common errors tested on Sentence Correction questions in the GMAT. They’re also relatively easy to spot. If Zoya is able to eliminate all the answer options that have subject-verb errors in them, and then selects the shortest remaining answer option, her chance of selecting the right answer jumps to 33%.
Level 4: Remove Parallel Construction Errors (43% success rate)
Parallel construction errors are another common error type tested on the GMAT. They’re a little harder to spot than subject-verb errors, but not absurdly so. If Zoya has a few extra seconds to eliminate answer options with either subject-verb errors or parallel construction errors, and then selects the shortest answer option, her chances of selecting the right answer reach 43%.
Level 5: Remove Ambiguous Pronouns (50% success rate)
Ambiguous pronouns are a third category of errors that are both common and reasonably easy to spot. If Zoya is also able to eliminate answer options that have ambiguous pronouns then selecting the shortest answer will give her the right answer half the time:
And all this is before accounting for answer options that may be short but are obviously incorrect for meaning reasons and thus are easy to eliminate.
Worked Example
Take this question:
Japanese scientists studying the frilled cusps of Baikal seals had founded evidence which suggested that the frills evolved to feed on Macrohectopus branickii and that if re-introduced to the ocean, the seals would evolve into whale-like giants.
- Baikal seals had found evidence which suggested that the frills evolved to feed on Macrohectopus branickii and that if re-introduced to the ocean
- Baikal seals has found evidence which suggest that the frills evolved to feed on Macrohectopus branickii and that if re-introduced to the ocean
- Baikal seals have found evidence which suggests that the frills evolved to feed on Macrohectopus branickii and that if re-introduced to the ocean
- Baikal seals have found evidence which suggests that they evolved to feed on Macrohectopus branickii and that if re-introduced to the ocean
- Baikal seals have found evidence which suggests they evolved to feed on Macrohectopus branickii and that if re-introduced to the ocean
The subject-verb error in the second answer option stands out. Scientists are plural but “has” is singular so we can eliminate.
The parallel construction error is also simple enough. The scientists found two things. One, that the frills evolved to feed on Macrohectopus branickii, and two, that if re-introduced to the ocean, the seals would evolve into whale-like giants. We want the “that…that” parallelism in front of the two items the scientists discovered, so we can eliminate the fifth answer option.
We’re down to three answer options without too much effort. The fourth answer option has a “they”. But the “they” could refer to either the seals or to the frills, making the “they” ambiguous.
We’re down to the first and third answer options. To native English speakers the first option sounds a little…off. At this point test-takers may take several valuable seconds trying to pinpoint exactly what the problem with it is. Zoya, however, is short on time, notes that the third answer option is shorter, selects it, and moves on.
In this case the error in the first answer option related to use of the past perfect tense (a subject for another blog post). It’s a relatively obscure tense form and while Zoya would typically have caught it, she was thrown off by an extra experimental Reading Comprehension passage and had to make up the time somewhere.
How To Use This Analysis
Be careful! This analysis should not be your primary strategy on the GMAT. The best way to do well on the GMAT is to identify the right answer, rather than to engage in statistical guesswork. Our system, for example, has identified a set of 12 Sentence Correction rules which allow our clients to get 95% of Sentence Correction questions correct in under 90 seconds.
But every so often clients will be down to two answer options and will spend an inordinate amount of time trying to decide between them. Rather than spending an extra two minutes trying to pick from two correct-looking answers, it typically makes more sense to select the shorter answer and use the time saved on other questions.
Beacon Community
This analysis provides a general example of how we use data analytics in our GMAT preparation program. For clients, our pattern recognition system identifies individual-specific trends, and then recommends tailored strategies which clients discuss with their mentors. This system is why, in 2020, the average Beacon client saw their GMAT score increase 170 points.
Beacon is an admission support consultancy that provides end-to-end admissions guidance for elite business school programs. We assist with GMAT preparation, essays, recommendation letters, and résumés. We use data analytics, like the statistics above, to guide our support. But we also understand that building a successful application is an art, not a science, and that data can only be used to inform decisions, not make them.
Find us on LinkedIn.