The LSAT: A Treatise on Guessing Forum

[dabossman]

I just took the June LSAT, and was in a mad rush to finish the final game in the LG section. I pretty much had to make semi-guesses on the first three questions (i.e., guessing after eliminating some options), and make total guesses on the last two. The conventional wisdom on TLS, based on my lurking, seems to be that D is the default guess option. And I guessed D on my last two for LG, but my obsession with my score has led me to see whether this assumption that D is the unqualified best guess is true, so I looked at the frequency of certain answers for the last two questions of LSATs 54-65 (minus 64, since I don't have it)--a sample of 11 recent LSATs. This is what I found:

Since I used 11 tests and only looked at the last two answers for all four respective sections, I had 88 total answers. A and E turned out to be the worst guesses overall, as A was the correct answer only 12 times and E was the correct answer only 14 times. This seems aligned with conventional wisdom: LSAT might be punishing novice test takers for making guesses by making the most likely guesses by novice test takers (A and E) the wrong answer. D only came in third at 18, C came in second at 21 and B came in at first with 23. So, from the looks of things, B and C, with relatively the same amount converting to the correct answer choice, are the best guesses you can give (at the tail-end of the test). Now, obviously this should be taken with a grain of salt, as 18 is not too far from 23, and I might be toying with a too small sample here, but when you're totally guessing you might as well go with what might give you a slight edge. There's little to lose. And at the very least never outright guess A or E; in fact, B almost shows up twice as frequently as A. That seems like a significant insight. I also did a breakdown per section:

LG: A3 B7 C5 D4 E3
RC: A1 B5 C6 D6 E4
LR: A8 B11 C10 D8 E7

From the looks of things, A shows the most consistent and, most pertinently to us, predictable patterns. A, in general, is not a good guess, but for RC it is an absolutely awful guess at the very end of the LSAT. Only once, out of 22 possibilities from the last 11 tests (excluding test 64 from my study), has it been the correct answer, which is in the neighborhood of a 4.5% chance of being correct, when random guess should give you a 20% chance at being correct. And since we know E is also chosen relatively infrequently (although the distribution is not as striking as A), we can effectively raise our chances at random guess to approximately 25% if we choose between B, C and D for RC, and it looks like we might be able to similarly raise our odds if we choose between B and C for LG (though I am less confident about the latter). We can also have some measure of confidence of B, C and D raising our odds, as it seems like the relative infrequency of A and E at the ends of sections is to be due to the intentions of the LSAT makers, rather than random variation.

So, in conclusion, it seems that you're best off choosing between B and C, and possibly even D, if you're in a mad scramble to guess at the end of a section (and when else would you be in a mad scramble to guess?). Again, I am aware that this sample may not be representative of future patterns, but it doesn't seem crazy to say that it might be. And if you're wildly guessing, you've honestly got nothing to lose by following the previous pattern. It would be interesting if someone did this with all available prep tests... that would be time consuming though.

Anyways, good luck with your LSAT, TLS, and good night,
dabossman

[RodionRaskolnikov]

I just took the June LSAT, and was in a mad rush to finish the final game in the LG section. I pretty much had to make semi-guesses on the first three questions (i.e., guessing after eliminating some options), and make total guesses on the last two. The conventional wisdom on TLS, based on my lurking, seems to be that D is the default guess option. And I guessed D on my last two for LG, but my obsession with my score has led me to see whether this assumption that D is the unqualified best guess is true, so I looked at the frequency of certain answers for the last two questions of LSATs 54-65 (minus 64, since I don't have it)--a sample of 11 recent LSATs. This is what I found:

Since I used 11 tests and only looked at the last two answers for all four respective sections, I had 88 total answers. A and E turned out to be the worst guesses overall, as A was the correct answer only 12 times and E was the correct answer only 14 times. This seems aligned with conventional wisdom: LSAT might be punishing novice test takers for making guesses by making the most likely guesses by novice test takers (A and E) the wrong answer. D only came in third at 18, C came in second at 21 and B came in at first with 23. So, from the looks of things, B and C, with relatively the same amount converting to the correct answer choice, are the best guesses you can give (at the tail-end of the test). Now, obviously this should be taken with a grain of salt, as 18 is not too far from 23, and I might be toying with a too small sample here, but when you're totally guessing you might as well go with what might give you a slight edge. There's little to lose. And at the very least never outright guess A or E; in fact, B almost shows up twice as frequently as A. That seems like a significant insight. I also did a breakdown per section:

LG: A3 B7 C5 D4 E3
RC: A1 B5 C6 D6 E4
LR: A8 B11 C10 D8 E7

From the looks of things, A shows the most consistent and, most pertinently to us, predictable patterns. A, in general, is not a good guess, but for RC it is an absolutely awful guess at the very end of the LSAT. Only once, out of 22 possibilities from the last 11 tests (excluding test 64 from my study), has it been the correct answer, which is in the neighborhood of a 4.5% chance of being correct, when random guess should give you a 20% chance at being correct. And since we know E is also chosen relatively infrequently (although the distribution is not as striking as A), we can effectively raise our chances at random guess to approximately 25% if we choose between B, C and D for RC, and it looks like we might be able to similarly raise our odds if we choose between B and C for LG (though I am less confident about the latter). We can also have some measure of confidence of B, C and D raising our odds, as it seems like the relative infrequency of A and E at the ends of sections is to be due to the intentions of the LSAT makers, rather than random variation.

So, in conclusion, it seems that you're best off choosing between B and C, and possibly even D, if you're in a mad scramble to guess at the end of a section (and when else would you be in a mad scramble to guess?). Again, I am aware that this sample may not be representative of future patterns, but it doesn't seem crazy to say that it might be. And if you're wildly guessing, you've honestly got nothing to lose by following the previous pattern. It would be interesting if someone did this with all available prep tests... that would be time consuming though.

Anyways, good luck with your LSAT, TLS, and good night,
dabossman

That was way too much for me to read because all I kept thinking about, and what I'll tell you now, is: Get the LG powerscore book, which will give you statistics on all the released tests' last 5 or so questions and their answers and it will specifically tell you which letter to guess with (on all three different sections too).

I usually go with B because I've always done it that way. Luckily, I don't need to guess much but definitely look at Powerscore.

[phillykid]

From the looks of things, A shows the most consistent and, most pertinently to us, predictable patterns. A, in general, is not a good guess, but for RC it is an absolutely awful guess at the very end of the LSAT. Only once, out of 22 possibilities from the last 11 tests (excluding test 64 from my study), has it been the correct answer, which is in the neighborhood of a 4.5% chance of being correct, when random guess should give you a 20% chance at being correct. And since we know E is also chosen relatively infrequently (although the distribution is not as striking as A), we can effectively raise our chances at random guess to approximately 25% if we choose between B, C and D for RC, and it looks like we might be able to similarly raise our odds if we choose between B and C for LG (though I am less confident about the latter). We can also have some measure of confidence of B, C and D raising our odds, as it seems like the relative infrequency of A and E at the ends of sections is to be due to the intentions of the LSAT makers, rather than random variation.

Your conclusions are based on way too small of a sample size. 11 tests is peanuts. Even 65 tests means nothing. Get back to us in the year 3000 when there are thousands of tests and a reliable sample size.

[Jeffort]

OP,

If you submitted that for an assignment or on an exam in a basic college level statistics class it would probably be graded D by the professor. It is also terrible advice to people that are going to take the LSAT, plus your various conclusions/recommendations are all over the place in conflicting and confusing ways.

The distribution of which of the five answer choices is correct holds very close to 20 percent for each letter across all administered tests in aggregate. However, it does not necessarily hold strictly true per section, nor per particular test form.

The LSAT test form development/assembly process and computer programs that produce the test forms from the item pool are very complicated with processes designed to insure randomness of credited answer choice letters across all section types and within question number ranges.

Basically, it is designed so that no particular blind guess answer choice letter will consistently produce extra points in any section type or overall per test form.

The logical and statistically sound method of blind guessing on questions one is not able to read, analyze and answer in the allotted time per section is to stick with one letter for all of them and hope you are lucky with the letter you decide to use.

[Noitaraperp]

Furthermore, you might expect things to regress back to the mean, in which case you would do to the opposite of what you have suggested. This seems more plausible, since the LSAC often seems aware of such strategies and wants to ensure a level of fairness in its tests.

[Mal Reynolds]

OP,

If you submitted that for an assignment or on an exam in a basic college level statistics class it would probably be graded D by the professor. It is also terrible advice to people that are going to take the LSAT, plus your various conclusions/recommendations are all over the place in conflicting and confusing ways.

The distribution of which of the five answer choices is correct holds very close to 20 percent for each letter across all administered tests in aggregate. However, it does not necessarily hold strictly true per section, nor per particular test form.

The LSAT test form development/assembly process and computer programs that produce the test forms from the item pool are very complicated with processes designed to insure randomness of credited answer choice letters across all section types and within question number ranges.

Basically, it is designed so that no particular blind guess answer choice letter will consistently produce extra points in any section type or overall per test form.

The logical and statistically sound method of blind guessing on questions one is not able to read, analyze and answer in the allotted time per section is to stick with one letter for all of them and hope you are lucky with the letter you decide to use.

Does he owe you a formal apology too?

[Mal Reynolds]

Also OP, that was an incredible waste of time. Go spend some time outside please.

[oaken]

I just took the June LSAT, and was in a mad rush to finish the final game in the LG section. I pretty much had to make semi-guesses on the first three questions (i.e., guessing after eliminating some options), and make total guesses on the last two. The conventional wisdom on TLS, based on my lurking, seems to be that D is the default guess option. And I guessed D on my last two for LG, but my obsession with my score has led me to see whether this assumption that D is the unqualified best guess is true, so I looked at the frequency of certain answers for the last two questions of LSATs 54-65 (minus 64, since I don't have it)--a sample of 11 recent LSATs. This is what I found:

Since I used 11 tests and only looked at the last two answers for all four respective sections, I had 88 total answers. A and E turned out to be the worst guesses overall, as A was the correct answer only 12 times and E was the correct answer only 14 times. This seems aligned with conventional wisdom: LSAT might be punishing novice test takers for making guesses by making the most likely guesses by novice test takers (A and E) the wrong answer. D only came in third at 18, C came in second at 21 and B came in at first with 23. So, from the looks of things, B and C, with relatively the same amount converting to the correct answer choice, are the best guesses you can give (at the tail-end of the test). Now, obviously this should be taken with a grain of salt, as 18 is not too far from 23, and I might be toying with a too small sample here, but when you're totally guessing you might as well go with what might give you a slight edge. There's little to lose. And at the very least never outright guess A or E; in fact, B almost shows up twice as frequently as A. That seems like a significant insight. I also did a breakdown per section:

LG: A3 B7 C5 D4 E3
RC: A1 B5 C6 D6 E4
LR: A8 B11 C10 D8 E7

From the looks of things, A shows the most consistent and, most pertinently to us, predictable patterns. A, in general, is not a good guess, but for RC it is an absolutely awful guess at the very end of the LSAT. Only once, out of 22 possibilities from the last 11 tests (excluding test 64 from my study), has it been the correct answer, which is in the neighborhood of a 4.5% chance of being correct, when random guess should give you a 20% chance at being correct. And since we know E is also chosen relatively infrequently (although the distribution is not as striking as A), we can effectively raise our chances at random guess to approximately 25% if we choose between B, C and D for RC, and it looks like we might be able to similarly raise our odds if we choose between B and C for LG (though I am less confident about the latter). We can also have some measure of confidence of B, C and D raising our odds, as it seems like the relative infrequency of A and E at the ends of sections is to be due to the intentions of the LSAT makers, rather than random variation.

So, in conclusion, it seems that you're best off choosing between B and C, and possibly even D, if you're in a mad scramble to guess at the end of a section (and when else would you be in a mad scramble to guess?). Again, I am aware that this sample may not be representative of future patterns, but it doesn't seem crazy to say that it might be. And if you're wildly guessing, you've honestly got nothing to lose by following the previous pattern. It would be interesting if someone did this with all available prep tests... that would be time consuming though.

Anyways, good luck with your LSAT, TLS, and good night,
dabossman

This is the dumbest thing I've read in a while.

Anyway, I'm pretty sure Steve Schwartz or somebody did do this with all available prep tests, right?

[Glaucon]

PowerScore has compiled statistics for this very subject from a very comprehensive set of answer choices. If my memory serves me right, D and B are def the most frequent right answers for LG but not for LR or RC. I think for LR and RC, it's C and D.

[Jeffort]

OP,

If you submitted that for an assignment or on an exam in a basic college level statistics class it would probably be graded D by the professor. It is also terrible advice to people that are going to take the LSAT, plus your various conclusions/recommendations are all over the place in conflicting and confusing ways.

The distribution of which of the five answer choices is correct holds very close to 20 percent for each letter across all administered tests in aggregate. However, it does not necessarily hold strictly true per section, nor per particular test form.

The LSAT test form development/assembly process and computer programs that produce the test forms from the item pool are very complicated with processes designed to insure randomness of credited answer choice letters across all section types and within question number ranges.

Basically, it is designed so that no particular blind guess answer choice letter will consistently produce extra points in any section type or overall per test form.

The logical and statistically sound method of blind guessing on questions one is not able to read, analyze and answer in the allotted time per section is to stick with one letter for all of them and hope you are lucky with the letter you decide to use.

Does he owe you a formal apology too?

No, (s)he didn't accuse me of anything or attack me. However, I do think the OP should apologize to his/her statistics professor for not paying attention in class and not doing the homework.

[Nova]

If you have to guess, you're doing it wrong.

[LexLeon]

dabossman,

Your contribution is not without merit. I think your post provides us with valuable information that even 99th percentile takers can at times use to good effect.

Thank you.

[soj]

A treatise on guessing should really focus on different strategies for quickly eliminating wrong answers so that even when you can't eliminate every wrong answer, you at least give yourself better than 20% chance of getting the right answer.

[uchi12]

.

[dabossman]

Not sure where all the vitriol is coming from...

Um yeah, 11 prep tests is definitely a really small sample size, I'm aware of that, but it seemed noteworthy how infrequently A showed up relative to other answer choices, as well as E. And according to powerscore, which I just looked up based on someone' recommendation, A and E, from all 65 preptests, are the most infrequent answer choices. But powerscore has definitely made my post pretty obsolete in terms of usefulness

no, guessing is not ideal but not a mark of failure either

And yes, I go outside--this didn't take that long to do

Maybe calling it a 'treatise' was misleading for some?

I'm just obsessed with my score, this passed the time

[Nova]

Not sure where all the vitriol is coming from...

Welcome to TLS

[Helicio]

Not sure where all the vitriol is coming from...

Welcome to TLS

Truth.

[laxbrah420]

Not sure where all the vitriol is coming from...

Um yeah, 11 prep tests is definitely a really small sample size, I'm aware of that, but it seemed noteworthy how infrequently A showed up relative to other answer choices, as well as E. And according to powerscore, which I just looked up based on someone' recommendation, A and E, from all 65 preptests, are the most infrequent answer choices. But powerscore has definitely made my post pretty obsolete in terms of usefulness

no, guessing is not ideal but not a mark of failure either

And yes, I go outside--this didn't take that long to do

Maybe calling it a 'treatise' was misleading for some?

I'm just obsessed with my score, this passed the time

If something seems noteworthy, and is a statistic, do statistics.

[HarlandBassett]

Also OP, that was an incredible waste of time. Go spend some time outside please.

lol