GMAT Prep Software Scoring Analysis and What If Scenarios

NEW ONLINE FORMAT

Couple of years ago we ran different scenarios with GMAT Prep Software to find out whether various myth about scoring algorithm were legit. You can find the results of that analysis in GMAT Prep Software Analysis and What If Scenarios.

On April 16, 2018, the GMAT exam shortened. The number of Quant questions were reduced by 6: from 37 to 31 and the number of Verbal questions will be reduced by 5: from 41 to 36.

Therefore, we are starting testing more or less the same scenarios with the NEW GMAT Prep Software to find out whether anything changed compared to old software.

1. What is the importance of the first 10 questions?

Experiment: We will try to disprove the myth the way OG/GMAC positions it: the first 10 questions are not critical
Methodology: we will attempt the worst case scenario and will answer the first 10 questions incorrectly (not guess but incorrectly); then we will attempt to answer the remaining 21 questions correctly (not guess)
Result: Score - Q30. Percentile - 19th (!!!). First 10 questions WRONG. The next 21 questions CORRECT
Analysis: obviously something is not right with the myth about the first 10 questions. They appear to be very important on the Quant. If you miss the first 10 (probably not a very realistic scenario for most) you have no way of climbing out of the hole - almost all questions that were offered to me were generally easy/medium difficulty.
When simulated in the previous GMAT Prep version, this scenario resulted in Score - Q38. Percentile - 48th. So, the importance of the first 10 questions seems to increase (no surprise since now 10 questions correspond to 1/3 of the questions).
Questions: 18PS and 13 DS.

TO ENSURE THAT THE RESULT ABOVE WAS NOT AN ANOMALY, WE RAN THIS SCENARIO TWO TIMES MORE:
Result: Score - Q31. Percentile - 20th (!!!). First 10 questions WRONG. The next 21 questions CORRECT. Second attempt
Questions: 18PS and 13 DS.
Total Score: every question in the verbal section was answered as "D" this time and it resulted in the final score of V6, 0 percentile. Total score: 320.

Result: Score - Q29. Percentile - 16th (!!!). First 10 questions WRONG. The next 21 questions CORRECT. Third attempt
Questions: 17PS and 14 DS.
Total Score: every question in the verbal section was answered as "C" this time and it resulted in the final score of V9, 2 percentile (Also, interesting becasie C gave a better result than D and E. Though it might be a simple luck). Total score: 320.

2. What is the importance of the last 10 questions?

Experiment: We will try to analyze the importance of the last 10 questions and if it equals the important of the first 10
Methodology: we will attempt the worst case scenario and will answer the first 21 Q questions correctly and then will answer the remaining 10 questions incorrectly (not guess but incorrectly).
Result: Score - Q49. Percentile - 75th. First 21 questions CORRECT. The final 10 questions WRONG.
Analysis: The questions were much harder this time with some new topics/types appearing on the test such as coordinate geometry/probability etc. Conclusion: the last 10 don't count as much as the first 10. This myth is True (as long as you can get the first 21 right). Please note that I have not guessed the last 10 but rather answered them incorrectly (which means I still had to solve those questions too). If you are guessing, you will most likely get a higher score.
When simulated in the previous GMAT Prep version, this scenario resulted in Score - Q50. Percentile - 92th.
Questions: 18PS and 13 DS.

3. What is the importance of the middle 10 questions?

Experiment: We will try to analyze the importance of the middle 10 questions and see how it compares to the results of the previous 2 experiments
Methodology: we will attempt the worst case scenario and will answer the first 11 Q questions correctly; then I will answer incorrectly the following 10 questions, and will answer the remaining 10 questions correctly.
Result: Score - Q49. Percentile - 75th. First 11 questions CORRECT. Next 10 INCORRECT. The final 10 questions CORRECT.
Analysis: The questions were not much easier than in the second scenario above even though I have made 10 consecutive mistakes.
When simulated in the previous GMAT Prep version, this scenario resulted in Score - Q49. Percentile - 85th.
Total Score: every question in the verbal section was answered as "E" this time and it resulted in the final score of V6, 0 percentile. Total score: 440.
Questions: 17PS and 14 DS.

4. What is the penalty if I miss every third question?

Experiment: I will try to test a scenario in which a test takers answers every third question incorrectly (e.g. 3, 6, 9, etc. Thus giving themselves an extra 2 mins on quant). This scenario will simulate someone who takes 3 mins instead of 2 per question and then guesses every third question. Please note that this is the worst case scenario (we do not get any guesses right)
Methodology: Every 3rd questions answered incorrectly. So, 10 incorrect and 21 correct answers.
Result: Score - Q48. Percentile - 69th. Every third questions WRONG.
Analysis: Slightly lower score than in the case of missing the last 10 questions.
When simulated in the old Desktop GMAT Prep version, this scenario resulted in Score - Q49. Percentile - 85th. More details here
Questions: 18PS and 13 DS.

5. Not Answering Last 10 Questions (ran out of time) vs. 10 Wrong Answers

Experiment: We will try to analyze the importance of answering questions in time.
Methodology: we will attempt answering the first 21 questions correctly and then wait till the time run out leaving the last 10 questions unanswered.
Result: Score - Q39. Percentile - 37th vs. Q49. Percentile - 75th. First 21 questions CORRECT. Next 10 NOT ATTEMPTED.
Analysis: I got lower score than in scenario 2 (First 21 questions CORRECT. The final 10 questions WRONG: Score - Q49. Percentile - 75th). This might indicate that answering all questions, finishing a test is very important. So, it's better finish the test and answer all the final questions, even incorrectly than not finish the test and leave final questions unanswered. Only scenario 1 (missing first 10 questions) gave a lower score than this case.
Total Score: every question in the verbal section was answered as "A" this time and it resulted in the final score of V6, 0 percentile. Total score: 370 (8th percentile).
Questions: 12PS and 9DS.

6. Importance of DS questions

Experiment: We will try to analyze the importance DS vs PS.
Methodology: we will attempt answering the first 10 questions correctly and then miss all DS questions, while answering PS correctly.
Result: Score - Q50. Percentile - 86th. First 10 questions CORRECT and then missed 10 DS while answering 11 PS questions correctly.
Total Score: every question in the verbal section was answered as "B" this time and it resulted in the final score of V6, 0 percentile. Total score: 460 (19th percentile).
Questions: 18PS and 13DS.

7. Importance of PS questions

Experiment: We will try to analyze the importance DS vs PS.
Methodology: we will attempt answering the first 10 questions correctly and then miss all PS questions, while answering DS correctly.
Result: Score - Q49. Percentile - 75th. First 10 questions CORRECT and then missed 11 PS while answering 10 DS questions correctly.
Total Score: every question in the verbal section was answered as "D" this time and it resulted in the final score of V8, 1 percentile. Total score: 460 (19th percentile).
Questions: 18PS and 13DS.



_________________

New Format - Verbal What-If Scenarios


Last year when Vercules ran the what-if scenarios in this thread we found some very interesting results. Not that the GMAT has changed (or just reduced in number and minutes), I decided to make some changes. I thought of doing things in quarters (9 out of 36 questions form a quarter).

1. Importance of the first 9 questions

In the first part of the test, I have tried to see what happens if we miss the first quarter entirely (i.e. the first 9 out of the 36 questions answered wrongly while answering correctly the others). To check for experimental questions and other question level based biases I ran the scenario thrice.

Experiment - GMAC says the first 10 questions are as important as the other questions. We test that theory. We have changed it to 9 because we want to take into account the reduced number of questions in the test.

Methodology - Run 3 instances of A) missing the first 9 questions . B) answering the next 27 questions correctly.

Result

- V28, V29, V30

Analysis - To make this analysis more effective, let's first check the scenarios:

First Scenario - Missed the first 9 questions - 4 CRs, 4 SCs and 1 RC - V29 - 57th Percentile
Second Scenario - Missed the first 9 questions - 5 RCs, 3 SCs and 1 CR - V28 - 52nd Percentile
Third Scenario - Missed the first 9 questions - 2 RC, 3 SC and 4 CRs - V30 -59th Percentile

It appears that there's no pattern to when the first RC will appear. In the second test, the RC was as early as the third question and as a result it made me miss the entire RC, which, I think, contributed to the score decline more than the other two tests. Also, while the average difficulty of the test was pretty easy/medium, I did see a few questions with 35%-55% difficulty on GMAT Club which will easily include them in the 700-800 level category.

Last year missing the first 11 questions resulted in the V22 (27th percentile)

Importance of the last 9 questions

Experiment In the second part, I decided to see what happens when you mark every question in the test, save the last 9 questions, correctly. Not to my surprise, I got a V40 or above every single time.

Methodology Run 3 instances of missing the last 9 questions on purpose but everything else marked correctly.

Result V40, V42, V42

Analysis In the first scenario the 9 questions I missed were - 1 entire RC (3 questions), 4 SC and 2 CR. In the second scenario, I missed 6 RC questions each. Clearly, even though RC questions are more on the test, missing SC and CR questions wrongly makes a bigger dent in the score.

Sidenote Last year, missing the 11 questions at the end resulted in a score of V38. So it is definitely a good news for test takers

3. Guessing SC but marking everything else correctly

This time we did a sectional test to see what happens if I guess all SC questions but mark everything else correctly.

Methodology: Mark CR and RC questions correctly and mark C for all SC questions.

Result: V37 with 11 questions answered wrongly.

Guessing RC and marking everything else correctly

In similar fashion, I wanted to check what happens when we guess RC but keep the SC and CR answers correct. Surprising results follow!

Methodology: Mark CR and SC correctly but mark all RC questions C.

Result - SHOCKING V34

Analysis: It is pretty clear that RC matters a whole lot more now than it used to. I would advise with caution against using this strategy. Practice your RCs and go well prepared.
_________________

We can test outcome when one performs exceptionally in some sections but fails abysmally in other section.
The cases such as :
1. SC, CR > 80+, RC<30
2. RC, CR > 80+, SC<30
3. SC, RC> 80+, CR<30

It will be useful for the test takers to know GMAT better.
_________________

Hi,please share the distribution of question types in verbal.Like in quant,I see there are 17-18 PS and 13-14 DS,similarly would love to know number of ques from RC,SC AND CR
Thanks

Posted from my mobile device

Great stuff, Bunuel! Cheers to all your valuable contributions to the Quant forums.


I ran a question-type count on the new (online) Verbal section of Exam 1, and here is what I found: 15 SC, 13 RC, 8 CR.

Make of that what you will! Also note that this balance/frequency of question Verbal types on the new GMATPrep tests does NOT appear to reflect their true proportions on the actual GMAT, based on the debriefs we've read so far, which suggest that RC is now the most common question type:



My best guess for the actual number of counted questions on the GMAT is this: 15 PS, 13 DS, 12 RC, 10 SC, 8 CR.

Keep in mind that I only ran one simulation. It is entirely possible that every GMATPrep test now has a slightly different balance of question types, much like the real GMAT itself, due to the random distribution of experimental questions by type.

These analyses also demonstrate that although the GMAT Official Practice Exams are the obviously most realistic in terms of raw material, the scoring on the real GMAT is clearly different. For example, 2 of your demonstrations earned scores of Q50 despite answering 10 questions wrong, which would never happen on the real GMAT: according to 5 years of ESR evidence, for example, we have never seen more than 6 (counted) questions wrong correlate to a Q50. Similarly, you can answer up to 9 questions wrong (in a row!) on GMATPrep and yet still earn a V42. This scenario would never occur on the real GMAT, either: a V42 score correlates to only 2-5 counted questions incorrect.



For more info: https://gmatclub.com/forum/psa-the-scor ... 67046.html

For this reason, to be safe, I would subtract 20-30 points from the average of your GMAT Official Practice Tests in order to estimate your likely score on the real GMAT.

-Brian
_________________

Any attempts on the myth of skipping questions on verbal? I read that skipping the last questions (up to 4) in verbal insead of guessing might be a better idea (what if you get 3 very easy questions wrong because of guessing vs just skipping them alltogether?)

I think thats a what if worth doing as there are many people strrugling to finish verbal (i have difficulties to finish the section and usually rush or guess the last 4)

I also heard that if you answer the last X questions under Y minutes, the software detects that you are guessing and penalizes you for not managing your time. Is that true?

Thanks for the great content!

Posted from my mobile device

Interesting interpretation. I always thought skipping questions was equivalent to answering them incorrectly. Can someone verify?

You can't "skip" a question on the GMAT. You either get it right or wrong (you have to answer a question in order to move on to the next one), or you run out of time and are forced to omit questions at the end. Despite your suspicions, it appears from previous GMATPrep studies that the opposite is actually true: leaving questions unanswered is even worse than getting them wrong!

The GMAC is very clear that there is a "significant" penalty for not finishing all questions in a section, so if we take GMAC at its word, then it's actually better to answer (some/most/all?) questions at the end of the GMAT incorrectly, than it is to leave them unanswered.


Source: OG 2019

That being said, I don't necessarily always trust the GMAC at its word, so I welcome seeing this hypothesis tested experimentally using the new online tests.

-Brian
_________________

PS vs. DS Question Importance - First Post Updated

Scenario 6: Weight & Importance of DS questions

Experiment: We will try to analyze the weight and importance of DS vs PS in terms of final Quant Score
Methodology: Answer the first 10 questions correctly and then miss all DS questions, while answering PS correctly.

Result: Score - Q50. Percentile - 86th. First 10 questions CORRECT and then missed 10 DS while answering 11 PS questions correctly.
Total Score: every question in the verbal section was answered as "B" this time and it resulted in the final score of V6, 0 percentile. Total score: 460 (19th percentile).
Questions: 18PS and 13DS.

Scenario 7: Weight & Importance of PS questions

Experiment: We will try to analyze the weight and importance of DS vs PS.
Methodology: Answer the first 10 questions correctly and then miss all PS questions, while answering DS correctly.

Result: Score - Q49. Percentile - 75th. First 10 questions CORRECT and then missed 11 PS while answering 10 DS questions correctly.
Total Score: every question in the verbal section was answered as "D" this time and it resulted in the final score of V8, 1 percentile. Total score: 460 (19th percentile).
Questions: 18PS and 13DS.

Conclusion: It seems there is no or little difference in terms of weight that GMAC places on the DS vs. PS questions. Both Scenarios resulted into a very close score: Q50 with 10 DS wrong and Q49 with 11 PS wrong.

_________________

bb OmerPelman

In the official GMAT website under the FAQ section of "What are GMAT® Official Practice Exams and Practice Questions?" it is stated that -
"Practice with real GMAT questions from past exams and full-length practice exams that use the same scoring algorithm as the actual exam."

The link - https://www.mba.com/frequently-asked-qu ... -questions

Thank you for the link. Yes. That is correct - that’s the update that GMAC has made after upgrading to the online platform.

Posted from my mobile device
_________________

Hi ,
can somebody analyse the below mock test result for me because i am clueless how i got V28 , that is too low. I though miss one question in verbal.
Please provide your feedback as how can i improve my verbal score.

Brutal. GMAT, in my opinion, doesn't give you enough chances to redeem yourself after you miss an "easy" question, which is likely what happened to you.

It would be great to look more closely at some of these 3 question CCW strings. You no doubt know the test is adaptive. What may be happening to you, because you get two questions correct in a row, is that you may be tempting the test to deliver the next question at a higher level of difficulty.

Unfortunately, you regularly got that third question wrong, and so have sent the algorithm a clear signal.

You probably know that their algorithm is top-secret, so that any attempt to assign a difficulty level to these questions is purely subjective. However, this would be at least a decent place to start your investigation.

You may also like to read up about what to do next in Verbal, in terms of Return on Time Invested. I go into it at more length: https://www.gmat.amsterdam/gmat_secrets_of_success/gmat-english

Kind Regards,
Omer
_________________

Thank you for this awesome effort.
Is it possible to test something similar to that second scenario?
I'm looking for the result of something like this:
The difference between the effect of having around 10 incorrect answers distributed evenly over questions after number 10 and the effect of having 10 final questions incorrect.

Thanks

Hey, here I am to confirm.
Got 3 RC passages in the first 10 questions. I'm weak at RC so I got 6/10 wrong. Then I got 2 CRs wrong, then 17 Qs correct in a row, one mistake in SC and boom, there it is. 27/36 correct and a result of V33.
This is ridiculous. I am glad I'm reading this post now, thank you OP.

mcelroytutoring,

I hope you're wrong. I literally destroy SC and my accuracy for RC questions is ~30%. That's so sad!!!!

This analysis is GOLD!!!
Was continuously scoring V35-V37 in GMAT Prep, Manhattan and Veritas Tests.
Took the GMAT late December and scored a V40.
And I think this was mainly because of my time-management. Did not leave any question, Left a couple of questions in the middle to balance the time.

Can't thank souvik101990 and Bunuel enough.

So interesting...i feel more confident right now

What about a scenario for verbal in that you miss every 3rd question (Get 2 right, miss 1, get 2 right, miss one etc)? (Like Bunnel's Quant Test)

So I would say that the conclusions are the following:

Quant:

10 first questions are really important
10 last questions are not so important
10 mid questions are not so important
PS is a little more important than DS

Verbal:
RC is the most important section
9 last questions are not so important

Overall Conclusions:

ANSWER ALL QUESTIONS

If I am wrong, please tell me.
_________________