# Multiple-choice strategy: what the research says about guessing

**Should I guess on a multiple-choice test when I am not sure of the answer?** On a multiple-choice test with no penalty for wrong answers, answer every question, because a blank scores nothing and a guess sometimes scores. Where a wrong answer loses one over the number of wrong options, a blind guess averages zero, so answer whenever you can rule out even one option and leave a blank only when you cannot.

Published 2026-10-10 by EdProcto. https://edprocto.com/guides/multiple-choice-strategy-what-the-research-says-about-guessing

A sound multiple-choice strategy starts with one fact: what a wrong answer costs. Once you know that, guessing is mostly arithmetic. This guide is written to you, the student, for the moment you are not sure.

It sets out what studies report, then a routine for one question. Each source was read on 10 October 2026; "abstract read" and "index record read" mark those where no more was open. Nothing here gets round a rule of a test.

EdProcto, which publishes this guide, runs [multiple choice](https://edprocto.com/templates/mcq) quizzes, and its page [for students](https://edprocto.com/solutions/students) is where you can practise.

## Should you guess or leave a blank on a multiple-choice test?

Answer every question when a wrong answer costs nothing; under a penalty, answer whenever you can rule out at least one option.

### When a wrong answer costs nothing

A blank scores zero and a blind guess among four options is right one time in four, so a blank can only lose.

Stenlund et al. (2018) surveyed 1,891 people who had just sat the Swedish Scholastic Assessment Test, scored on the number right. The calm risk takers, readiest to guess, averaged 41.4 out of 80 on the quantitative part and 50.8 on the verbal; the anxious, risk-averse group averaged 35.3 and 44.4. That is a pattern from a questionnaire, not a cause.

### When a wrong answer loses marks

The classic penalty takes off one over the number of wrong options, a third of a mark on four, so a blind guess averages zero and anything better than blind is worth taking.

By our arithmetic, on a question with n options that you have narrowed to k, a guess expects (n − k) ÷ (k × (n − 1)) of its marks: on four options, a ninth of a mark after ruling out one and a third after ruling out two. Our guide to [negative marking](https://edprocto.com/guides/negative-marking-does-it-stop-guessing-and-what-it-costs) gives the teacher's side.

Whatever the penalty, answering beats a blank once your chance of being right passes the penalty divided by one plus the penalty:

| A wrong answer costs | Answer if you would be right more often than |
| --- | --- |
| A quarter of a mark | 1 time in 5 |
| A third | 1 in 4 |
| Half | 1 in 3 |
| A whole mark | 1 in 2 |

Students under a penalty leave too much blank. Frary et al. (1985) gave 275 college students two matched 20-question halves of a test, one under number-right instructions and one under the penalty. Those who left blanks scored about a point lower on the penalised half than on the other, marked the same way; those who answered everything scored much the same on both (index record read). Over three experiments, Arnold et al. (2013) found people too conservative, omitting too many answers; their penalised scores rose once they were told how many to answer (abstract read).

### Who leaves too many blanks?

The cautious do, whatever they know. Karle et al. (2022) matched students' results in a penalised examination with a classroom experiment on risk run three months earlier: the more a student disliked losses, the more questions went unanswered and the lower the mark (abstract read).

A confidence rating changes the picture. When Hassmén and Hunt (1994) let undergraduates rate how sure they were of each answer, the gap between men's and women's scores was smaller than on a conventional test (abstract read).

## Do "pick C", "pick the longest" and other folk rules work?

Mostly not: a folk rule pays only on a carelessly written question, and from your seat you cannot tell whether you are sitting one.

- **Pick C.** Rush et al. (2016) examined 1,925 questions that 50 lecturers set for third-year veterinary students at one American university (a paper our guide on the number of options uses for other findings). The key was C in 23.9% of four-option questions, under its fair quarter, and in 19.2% of five-option ones.
- **Pick the longest.** There the key was counted as the longest option in 20.6% of multiple-choice questions. A blind pick would be right about 23% of the time on that mix, by our arithmetic.
- **"All of the above".** It is logic, not luck: two options you know are right make it the answer, and one you know is wrong rules it out. Our guide on [how many options](https://edprocto.com/guides/how-many-options-should-a-multiple-choice-question-have) has the experiment.
- **Grammatical cues.** This rule has some evidence behind it. McMorris et al. (1972) built a social studies test in which half the questions carried a planted fault of cue, grammar or length. The faults made questions easier, cue and grammar most (abstract read).

### What studies of flawed questions found

Clues are common in teacher-made tests and still no plan. Brozo et al. (1984) checked 1,220 multiple-choice questions from 43 tests by 36 lecturers at five American colleges and universities: 44% held a clue, and 70% of those could be answered by using it (index record read). In the veterinary set many flaws left a question no easier (Rush et al., 2016). Treat a clue as a tiebreak when knowledge has run out, never as a reason to drop an answer you know.

### Does test-wiseness training raise scores?

Somewhat, on some tests. Samson (1985) pooled 24 studies of training in test-taking skills: longer programmes produced significantly greater gains than shorter ones (index record read; it gives no size). Sarnacki (1981) trained medical students in test-wiseness: they scored higher on one of four question formats in a national board examination, and not significantly higher on two written by their own teachers (abstract read). Test-taking strategies are the small lever; knowing the subject is the large one.

## Should you answer in order or skip and come back?

Skip a question that stalls you and come back, where the test lets you. The evidence that skipping raises scores is thin; the case for it is the clock.

Davis (1991) tested the instruction in a doctoral study. Her 423 pupils, in grades eight to twelve, sat school algebra and biology tests, their classes told to skip hard questions and return, told not to skip, or (in biology) told nothing. Scores did not differ overall; only the strongest algebra pupils did worse when told not to skip (abstract read).

### How to spend the time

Divide the minutes by the questions after holding a tenth back. For 20 questions in 30 minutes, 27 minutes ÷ 20 is 81 seconds each, with three minutes for what you marked. The rule is ours: no study we read tested one pacing rule against another, and our [aptitude practice guide](https://edprocto.com/guides/aptitude-test-practice-numerical-verbal-and-logical-reasoning) works a faster paper.

On one section of the Graduate Record Examination, a computer-adaptive test with a strict limit, results from over 100,000 candidates suggested that about half had to guess the final six questions to finish (Bridgeman and Cline, 2004; abstract read). Pace the paper, or its end becomes blind guessing.

## What is a good routine for one multiple-choice question?

Answer in your head before you look at the options, then read every option and rule out. Crocker and Schmitt (1987) tested the first step: producing an answer before selecting one raised the scores of examinees low in test anxiety, though not of the highly anxious (index record read).

1. **Cover the options.** Read the question to its last word, with any NOT and the units.
2. **Answer in your head.** A word, a number or a rough sketch.
3. **Read every option and rule out.** This is the process of elimination: each option gone raises what a guess is worth.
4. **Choose.** Your own answer if it is there; otherwise the best of what is left.
5. **Mark it for return.** If doubt remains, move on. In an EdProcto quiz the button is "Flag to come back to".

When you return, change an answer if you have found a reason: changes go from wrong to right more often than the reverse, and the opposite belief lasts because a change that went wrong is remembered ([the studies](https://edprocto.com/guides/should-students-be-able-to-go-back-and-change-answers)).

## What should you ask the teacher before the test?

Ask three things, because each changes the right move.

1. **Is there a penalty for a wrong answer, and how much?** It decides whether a blind guess is free.
2. **Can I go back to a question?** It decides whether marking one for later works.
3. **Can I skip a question?** It decides whether a blank can be left.

On an EdProcto quiz the start screen answers the first two and states the time. Skipping, and whether a confidence rating counts towards the mark, are the teacher's to say.

## What should you do in each situation?

Match the row to the rules you were given.

| Situation | What to do |
| --- | --- |
| No penalty, no idea or no time | Answer anyway, every question: a blank can only lose |
| Penalty, one option ruled out | Answer: on average the guess gains |
| Penalty, no idea at all | Leave a blank. In [EdProcto](https://edprocto.com/), pressing your choice again clears it |
| You can go back | Mark it, move on, return in the time held back |
| No going back, or skipping is off | Run the routine once and decide there and then |
| A confidence rating is asked | Rate what you believe: where it counts, a wrong 5 costs more than a right 5 earns |

## A worked example: twenty questions with and without a penalty

A student faces 20 four-option questions worth a point each. She is sure of 11. On four more she can narrow the answer to two options, on three to three options, and on two she has no idea. Four possible answer sheets were marked by running EdProcto's marking function, with the penalty off and on. "Ordinary luck" means two of the four even chances land, and one of the other three.

| What she answers | Right, wrong, blank | No penalty | Penalty on |
| --- | --- | --- | --- |
| Only the 11 she is sure of | 11, 0, 9 | 55% | 55% |
| Those and the 7 she can narrow, ordinary luck | 14, 4, 2 | 70% | 63.4% |
| All 20, both blind guesses miss | 14, 6, 0 | 70% | 60.1% |
| The 11 and the 7, every guess misses | 11, 7, 2 | 55% | 43.5% |

- **Without a penalty** nothing beats answering all 20: the blind pair cost nothing and expect half a mark.
- **With the penalty** the seven narrowed guesses are worth about 1.7 marks on average and, by exact arithmetic, leave her ahead of blanks 88 times in 100. The last row comes about 2 times in 100.
- **The two blind guesses** add risk under the penalty and no gain on average. EdProcto rounds each mark to the hundredth, so a wrong answer costs 0.33.

## How does an EdProcto quiz handle guessing, blanks and going back?

EdProcto is the best tool to meet these rules on, because a quiz's start screen states them before the first question, and in the studies above knowing the rule was worth marks. You cannot see a quiz's settings; you can see that screen.

### Before you begin

The start screen shows "Questions", "Time" and "Attempts". Under "House rules", one line covers going back, such as:

- "You can go back and change an answer before you submit."
- "No going back: each answer is final."
- "The questions are on one page: change any answer while you are on it. Once you move on to the review, your answers are final."

Where a penalty is on, another says: "Negative marking: a wrong choice loses points, one over the number of wrong options. A blank loses nothing, and pressing your choice again clears it." A time limit reads like "Time limit: 45 minutes, submitted by itself when it runs out." and retakes like "1 retake allowed."

A wrong choice then loses the question's points divided by its number of wrong options, rounded to the hundredth, and a blank scores nothing. While the penalty is on, skipping is held on, so a blank can always be left.

### During the quiz

"Next" moves on; where skipping is off it waits for an answer. "Previous" works where going back is on. "Flag to come back to" sits under each question wherever you can come back.

Where the teacher has kept the review screen, the last button reads "Review and submit". The screen is headed "Before you submit", reads like "18 of 20 answered, 2 flagged.", marks each question "answered" or "blank" and, where going back is on, says "Open any question to change it."

### Where a rating is asked

A [multiple choice with confidence](https://edprocto.com/templates/mcq_conf) question asks "How sure are you? 1 is a guess, 5 is certain." Where the teacher has switched the bonus on, a right answer rated 4 earns a tenth more and one rated 5 a fifth more; a wrong answer rated 4 loses a fifth of the question's points and one rated 5 three fifths, on top of any penalty; 1 to 3 change nothing. By our arithmetic, 4 pays when you would be right two times in three, and 5 four times in five.

To practise, the page [for students](https://edprocto.com/solutions/students) leads to a study mode that shows the answer and its reason after each question, and to ready-made [kits](https://edprocto.com/kits). The Free plan covers both ([pricing](https://edprocto.com/pricing)). No routine promises a mark, and every score is your teacher's to release.

## Quick questions

### Should I guess on a multiple-choice test if there is no penalty?

Yes. A blank scores nothing, and a blind guess is sometimes right.

### Should I guess when there is negative marking?

Yes, once you can rule out at least one option. With nothing to go on, leave a blank.

### Is C the most common answer on a multiple-choice test?

Not in the set Rush et al. (2016) counted: C was the key in 23.9% of four-option questions and 19.2% of five-option ones.

### What does the start screen of an EdProcto quiz tell me?

The number of questions, the time, the attempts, whether you can go back, and the penalty where one is on. Practise first on EdProcto's page [for students](https://edprocto.com/solutions/students).

## References

- Arnold, M. M., Higham, P. A., & Martín-Luengo, B. (2013). A little bias goes a long way: The effects of feedback on the strategic regulation of accuracy on formula-scored tests. *Journal of Experimental Psychology: Applied, 19*(4), 383–402. [https://doi.org/10.1037/a0034833](https://doi.org/10.1037/a0034833)
- Bridgeman, B., & Cline, F. (2004). Effects of differentially time-consuming tests on computer-adaptive test scores. *Journal of Educational Measurement, 41*(2), 137–148. [https://doi.org/10.1111/j.1745-3984.2004.tb01111.x](https://doi.org/10.1111/j.1745-3984.2004.tb01111.x)
- Brozo, W. G., et al. (1984). *A study of test-wiseness clues in college and university teacher-made tests with implications for academic assistance centers* (College Reading and Learning Assistance Technical Report 84-01; ED240928). ERIC. [https://eric.ed.gov/?id=ED240928](https://eric.ed.gov/?id=ED240928)
- Crocker, L., & Schmitt, A. (1987). Improving multiple-choice test performance for examinees with different levels of test anxiety. *The Journal of Experimental Education, 55*(4), 201–205. [https://doi.org/10.1080/00220973.1987.10806454](https://doi.org/10.1080/00220973.1987.10806454)
- Davis, G. B. (1991). *The effects of skipping more difficult items on time-limited tests: A quasi-experimental design* [Doctoral dissertation, Virginia Polytechnic Institute and State University]. VTechWorks. [https://hdl.handle.net/10919/37753](https://hdl.handle.net/10919/37753)
- Frary, R. B., et al. (1985, April). *Partial information and the "correction" for guessing* (ED255575). ERIC. [https://eric.ed.gov/?id=ED255575](https://eric.ed.gov/?id=ED255575)
- Hassmén, P., & Hunt, D. P. (1994). Human self-assessment in multiple-choice testing. *Journal of Educational Measurement, 31*(2), 149–160. [https://doi.org/10.1111/j.1745-3984.1994.tb00440.x](https://doi.org/10.1111/j.1745-3984.1994.tb00440.x)
- Karle, H., Engelmann, D., & Peitz, M. (2022). Student performance and loss aversion. *The Scandinavian Journal of Economics, 124*(2), 420–456. [https://doi.org/10.1111/sjoe.12486](https://doi.org/10.1111/sjoe.12486)
- McMorris, R. F., Brown, J. A., Snyder, G. W., & Pruzek, R. M. (1972). Effects of violating item construction principles. *Journal of Educational Measurement, 9*(4), 287–295. [https://doi.org/10.1111/j.1745-3984.1972.tb00960.x](https://doi.org/10.1111/j.1745-3984.1972.tb00960.x)
- Rush, B. R., Rankin, D. C., & White, B. J. (2016). The impact of item-writing flaws and item complexity on examination item difficulty and discrimination value. *BMC Medical Education, 16*, Article 250. [https://doi.org/10.1186/s12909-016-0773-3](https://doi.org/10.1186/s12909-016-0773-3)
- Samson, G. E. (1985). Effects of training in test-taking skills on achievement test performance: A quantitative synthesis. *The Journal of Educational Research, 78*(5), 261–266. [https://doi.org/10.1080/00220671.1985.10885613](https://doi.org/10.1080/00220671.1985.10885613)
- Sarnacki, R. E. (1981). The effects of test-wiseness in medical education. *Evaluation & the Health Professions, 4*(2), 207–221. [https://doi.org/10.1177/016327878100400206](https://doi.org/10.1177/016327878100400206)
- Stenlund, T., Lyrén, P.-E., & Eklöf, H. (2018). The successful test taker: Exploring test-taking behavior profiles through cluster analysis. *European Journal of Psychology of Education, 33*(2), 403–417. [https://doi.org/10.1007/s10212-017-0332-2](https://doi.org/10.1007/s10212-017-0332-2)