i spend about 5-10 minutes during my final preparation for MC exams determining a guessing strategy.
some people finish 1 hour early and get upset they cant leave… im the guy that has 5-15% of answers with blanks and spend the final minutes of the exam praying.
ive found there is little value in “eliminating” an answer. ill leave it blank for the end to guess… ive found that often enough the one option that i quickly discard is actually right… so i guess between all the options. (*unless im sure its not one)
i have considered using the same letter option for all the blanks and hoped for a + deviation.
i have generated a random number on my calculater and determined the guess based on the percentile of the number.
i have guessed depending what the answer before it was and picked the same letter option.
i have alternated between a-b-c for all blanks
i have counted the distribution of my selected responsed and guessed based on an anlysis of the results… (either normalizing or going with the mode)
this question is actually a very intense behaivioral economics/statistics exercise. unlike a roulette wheel-the distribution of correct answers on an exam is unknown… would you become suspicious if the first 20 answers were A? depending on the accuracy of purposeful choices, the result of guesses may or may not matter…(leaving 15 blank and still pass).
Aren’t there only three answer choices instead of four now? Does that mean that B is the new C? If I were you dvictr, I would just pick B on all of the questions – you’ll be way ahead of the pack for sure.
@ bromion: the answer sheet has ovals from A to F I think… standard form regardless of the format of the question. always fill in C, even if there are only two possible answer choices. C will never be replaced by inferior options such as B.
I was proctoring an undergrad final exam when I was in grad school and this guy just pulled out a coin and started flipping it to decide what answer to put. I recommend you follow his strategy.
Well if you are truly stuck and can’t choose at all between the 3 possibilities, then just look at your answer key and see if there is an answer choice you haven’t used for a while.
seriously however, you need to follow the principle of maximum entropy: (http://en.wikipedia.org/wiki/Principle_of_maximum_entropy) - most physical systems evolve in way that maximizes entropy over time, so absent any other information you must choose the probability distribution of the answers that has maximum entropy.
for a discrete distribution supported on a finite set (A, B, C, D), the one that maximizes the entropy is the uniform distribution. in other words, keep a running count/histogram of your answers up to the current time, and for your random guess pick an answer which will bring the distribution of your answers closer to uniform (i.e. the least represented answer so far). this is just a scientific proof of iteracom’s intuition from above - you’re welcome
That’s true, but I’m not sure how that informs your guessing strategy. If the idea is that the test-designers try to make the test answer distribution close to maximum entropy, I’m not sure if the test-taker gains any advantage between making their guesses also have maximum entropy as well, or if they just guess A on everything.
I suppose what the entropy principle does is take information from what you have already answered and allows you to figure out whether adding a column of As, Bs, or Cs produces maximum entropy in your own answer set, on the assumption that if you have more As in your “known” answers, you probably shouldn’t be choosing A as your guess.
And please make sure to do this before time is up!
Informational entropy is a measure of unpredictability (disorder in the thermodynamic sense). So you can hypothesize that the maximum entropy probability distribution of the answers is a natural outcome of the test design rather than an explicitly formulated goal. Based on that, and given the information from what you have already answered - as you state in your second paragraph - you make a guess that maximizes the entropy/unpredictability of the ‘running’ probability distribution.
Of course with no other constraints imposed, the ‘target’ distribution is trivially uniform. But on a case-by-case basis, you can maximize the entropy under a set of appropriate constraints, which are likely to improve your guess. You’ll just need to solve a numerical system using the method of Lagrange multipliers, but lucklily the exam allows you to use a scientific calculator, so you’re all good.