How many attempts a quiz should allow depends on what its score is for. A second attempt can be more learning or a second throw of the dice, and the setting cannot tell which.
This guide, for a teacher, lecturer or trainer choosing that setting, reports what studies found and says where our advice is only a working rule.
EdProcto, which publishes this guide, holds the choice in two settings, "Whole-quiz retakes" and "Hours between retakes"; the rest are in the settings guide.
How many attempts should a quiz allow?
A quiz should allow as many attempts as its score can bear: no limit where nothing rides on it, two where the first is the record, one where the score decides something. The table is our recommendation.
| Purpose | Attempts | Wait | Show between attempts | In EdProcto |
|---|---|---|---|---|
| Practice | No limit | Days, by habit | Answers and explanations | A study quiz, or the Practice preset |
| Weekly check | Two: the first is the score | 24 hours | Answers and explanations | "Whole-quiz retakes" at "1", "Hours between retakes" at 24 |
| Unit test | One, and a resit under the pass mark | A week, with reteaching | Answers, if the resit is a new paper | Retakes at "None"; the resit is a second quiz |
| Certificate | Two | Three days or more | The score, not the answers | Retakes at "1", the wait at 72 |
| Hiring round | One | No retake | Nothing | Retakes at "None" |
The attempts follow the studies below; the waits are working rules, since no study we read fixes a figure. We read Kortemeyer's preprint and the publishers' pages in full, every other study in abstract only.
What do retakes do to later exam scores?
In the two course studies we read, the chance to try again went with higher scores.
- Marden et al. (2013) tried four models of online multiple-choice quiz with immediate feedback in an introductory physiology course. Only the model allowing multiple unsupervised, untimed attempts was associated with a significant rise in mean examination marks over the year before.
- Morphew et al. (2020) compared two cohorts of an undergraduate engineering course. The cohort tested more often, with second-chance exams, scored seven percentage points higher and had half as many failing grades.
Neither assigned students at random, and the engineering study changed frequency and second chances together. Retrieval practice covers the classroom experiments on quizzing, and the settings guide the pooled studies of mastery learning.
What do students do with unlimited attempts?
Many stop solving and start trying answers, and the more attempts allowed, the less each one achieves.
Guess-and-check and fast re-submission
Kortemeyer (2015) analysed four semesters of an introductory physics course of typically more than 500 students, with over 200 homework problems a semester, most answered with a typed number. Feedback was immediate, extra tries cost nothing, and instructors allowed 10, 12, 20 or 99. With 10 tries students solved 98.0% of problems in the end; with 99, 91.8%, giving up on 8.2%.
After the first try or two, the chance of solving a problem on any one try stayed constant: students were not learning from earlier tries, whatever their exam scores. From his earlier study he reports tries typically following one another within a few seconds. His preprint calls unlimited tries "definitely a mistake" and suggests about ten; the published abstract's model predicts five. With four options, a third try is elimination: our point, not his.
The first attempt against the last
The first attempt is the one that says what a student knew. Kibble (2007) gave a large medical physiology class unsupervised online quizzes. Course credit of 0.5% to 2% a quiz raised participation and brought evidence of widespread inappropriate use. Scores predicted the summative examinations when the first of several attempts was compared.
How the number allowed changes effort on the first
A second attempt in hand seems to cheapen the first. In Kortemeyer's data the success rate on each try fell as the tries allowed rose. Orchard (2016) allowed up to two attempts at online assessments in an operations management course: students who attempted once did better, even beside students of similar ability, though a student content with a first score need not try again.
Which score should count: first, best, last or an average?
Count the first when the score is a record of what a student knew unaided, and the best or a weighted average when the aim is mastery.
| Score that counts | Rewards | Risks |
|---|---|---|
| First | Preparing before the first sitting | Nothing gained by learning afterwards |
| Best | Persistence | A careless first attempt, and luck |
| Last | Improvement | One bad day undoes a good one |
| Average | Both sittings | A sum students must be shown |
The table is our reasoning; two studies bear on it. Studying admission to medical studies in Belgium, Lievens et al. (2005) found that at the same test score, candidates who passed on the first attempt later performed better than those who passed on the second.
Herman et al. (2020) compared two rules for a second-chance exam in one course: the second score replacing the first, capped at 90%, or 90% of the higher score plus 10% of the lower. Under the weighted rule marginal students were significantly more likely to take the second chance, and the data suggested they learned more.
How long should students wait between attempts?
Long enough to study and to forget the paper: a day for a weekly check, days or weeks for a test that counts. Rawson and Dunlosky (2011), after three experiments with 533 students, recommend recalling an idea correctly three times at first, then relearning it three times at widely spaced intervals. Re-tests that come due turns that into a timetable.
Waiting alone corrects little. Feinberg et al. (2015) followed the questions that repeat examinees on a credentialing examination had answered wrongly: 68% of the time they chose the same wrong option again. The wait needs reteaching inside it.
What happens to scores on a retake of a test that counts?
They rise, and part of the rise is not learning. Among those repeating the Belgian admission tests, second scores were a third of a standard deviation higher on the knowledge and situational judgement tests and half on the cognitive ability test (Lievens et al., 2005).
The same paper against a parallel one
Here the studies disagree. Pooling 40 studies of practice on aptitude and achievement tests, Kulik et al. (1984) found larger gains when the identical form was sat again than with a parallel form, and larger gains the more practice tests were taken.
Raymond et al. (2007) randomly assigned repeat examinees to the same form or a parallel one: gains were 0.79 standard deviations for 71 examinees in computed tomography and 0.48 for 765 in radiography, nearly the same on either form. Feinberg et al. (2015) found likewise.
Our reading, not a finding: where only a score comes back there is little to memorise; a quiz that shows its answers is a different case.
What test publishers publish about waiting periods
They publish a waiting period, sometimes with a yearly limit, and cite no study for either. Two pages, read on 10 October 2026:
- The UK car theory test: at least 3 working days, with a letter saying which parts fell short (GOV.UK, n.d.).
- Microsoft certification exams: 24 hours after a first failure, 14 days between later attempts, five attempts in 12 months (Microsoft, n.d.).
Why does a retake of the same questions measure memory of the paper?
Because the second sitting can be answered from the first. Lievens et al. (2007) studied 941 candidates on a cognitive ability test that counted: retest scores measured general ability less, went with memory more, and no longer predicted what the test was used to predict. The abstract does not say whether the form was the same.
What to change, in our order of effect:
- The questions. New numbers, a new context, the same idea: a parallel paper.
- What is shown. If the paper must be the same, show the score alone until the retake is in.
- The rule that deals them. A set made by rule gives each sitting its own questions, as EdProcto's rule-made kits can.
- Not only the order. Shuffling moves a question; it does not change what it asks.
How do you tell students the rule?
Put the retake policy in writing before the first sitting, in five lines: how many attempts, which score counts, the wait, what they will see in between, and what changes on the retake. For a weekly check:
You may sit this quiz twice. Your first score is the one recorded. The second attempt opens 24 hours after you hand in. You see the answers after each attempt. Use the second to check what you have learned since.
What does this look like for a unit test with a resit?
One test, one reteaching lesson and one second paper. The class and every figure are invented: 28 students, 20 questions on forces, a pass mark of 60%.
- In EdProcto, apply Standard quiz (on Free, switch "Full screen" off). "Whole-quiz retakes" stays at "None", and answers show at hand-in.
- On Thursday 28 sit. Nine score under 60%.
- Read the results question by question and reteach the two questions most missed.
- Press "Duplicate this quiz" at the foot of the builder and rewrite each question in the copy with new numbers. The copy is a new quiz with its own code.
- The next Thursday the nine sit the second paper. A student with 45% then 70% is recorded at 67.5%: nine-tenths of the better score plus a tenth of the other, the weighted rule above, announced with the test.
How does EdProcto count attempts and retakes?
EdProcto is the best tool for fitting attempts to the stakes, because the server keeps the limit and the wait, and the class is read on first sittings: the attempt that predicted examination scores in Kibble's (2007) course.
The settings
On a quiz's settings page, under "Attempts and retakes":
- "Whole-quiz retakes" takes "None", "1", "3" or "No limit" and starts at "None": "1" allows two sittings, "3" four.
- "Hours between retakes", shown once retakes are allowed, runs from 0 to 720: "A retake is refused until this long after the last hand-in."
Practice sets retakes to "No limit"; Standard quiz and Exam set "None". No preset touches the hours. A reasoning kit from the hiring page opens with "None", whatever its preset.
How the server counts
The server counts, on a quiz of the exam kind or a quiz night. A student is the email address typed where "An account is needed to take it" is on, as it is from the start, and otherwise the name typed, capitals ignored. A sitting counts once handed in. After the last, a start is refused: "You have used every attempt at this quiz." Inside the wait the student reads a line like "Your next attempt opens on Friday, October 16 at 09:40 AM, by this device’s clock."
The address is typed, not checked: the count is a record, not proof of who sat. Leave the email on wherever attempts are limited, so that a sitting still open under an address must be handed in before another begins. Your own previews, signed in, use no attempt and stay out of the results.
A study quiz is practice and is not counted. On your own study screen its result offers "Retry the 3 you missed", a new attempt at those questions alone (study mode).
What the student sees
The start screen shows "Attempts" as "1", "2 in all", "4 in all" or "No limit": the allowance, the same at every sitting. A study quiz, which is not counted, shows "No limit". The rules say "1 retake allowed.", "3 retakes allowed." or "You may retake the whole quiz." To sit again a student opens the quiz's link and begins as before; on the same device, "See your result" stands under the button.
On a kit's rule-made quiz left as the kit made it, "A fresh set for each student" deals each sitting its own questions by the same rule. The reasoning kits on the hiring page open with it on.
What you see in the results
"Students" lists every sitting handed in. A later sitting under the same email address carries "sat again" and is left out of the class's figures. Where the quiz asks for no address, a later sitting under a name already used carries "same name as an earlier sitting" and stays in the figures, since it may be another student of that name. "Download the marks" gives a row for each sitting, with "Sitting", "Handed in (your time)" and "Score %", so first, best, last or an average is yours to work out.
These settings are on every plan. Free takes thirty students a quiz, Tutor 100 and Educator 500 with the full analysis; a retake takes no second place. See pricing. No setting makes cheating impossible, and every score is yours to release.
Quick questions
Should a graded quiz allow unlimited attempts?
No. With 99 tries allowed, students in Kortemeyer (2015) solved fewer problems than with 10.
Which attempt should count, the first or the best?
The first, for a record of what was known unaided; the best or a weighted average, for mastery. EdProcto reads the class on first sittings and lists every later one.
How long should students wait before a retake?
A day for a weekly check and longer for a test that counts. In EdProcto, "Hours between retakes" refuses an early start and tells the student when the next attempt opens.
Can a retake use the same questions?
For practice, yes. For a score that counts, change the questions or show the score alone in between: Lievens et al. (2007) found retest scores went with memory.
References
- Feinberg, R. A., Raymond, M. R., & Haist, S. A. (2015). Repeat testing effects on credentialing exams: Are repeaters misinformed or uninformed? Educational Measurement: Issues and Practice, 34(1), 34–39. https://doi.org/10.1111/emip.12059
- GOV.UK. (n.d.). Theory test: cars: Pass mark and test result. Retrieved October 10, 2026, from https://www.gov.uk/theory-test/pass-mark-and-result
- Herman, G. L., Cai, Z., Bretl, T., Zilles, C., & West, M. (2020). Comparison of grade replacement and weighted averages for second-chance exams. In Proceedings of the 2020 ACM Conference on International Computing Education Research (pp. 56–66). Association for Computing Machinery. https://doi.org/10.1145/3372782.3406260
- Kibble, J. (2007). Use of unsupervised online quizzes as formative assessment in a medical physiology course: Effects of incentives on student participation and performance. Advances in Physiology Education, 31(3), 253–260. https://doi.org/10.1152/advan.00027.2007
- Kortemeyer, G. (2015). An empirical study of the effect of granting multiple tries for online homework. American Journal of Physics, 83(7), 646–653. https://doi.org/10.1119/1.4922256
- Kulik, J. A., Kulik, C.-L. C., & Bangert, R. L. (1984). Effects of practice on aptitude and achievement test scores. American Educational Research Journal, 21(2), 435–447. https://doi.org/10.3102/00028312021002435
- Lievens, F., Buyse, T., & Sackett, P. R. (2005). Retest effects in operational selection settings: Development and test of a framework. Personnel Psychology, 58(4), 981–1007. https://doi.org/10.1111/j.1744-6570.2005.00713.x
- Lievens, F., Reeve, C. L., & Heggestad, E. D. (2007). An examination of psychometric bias due to retesting on cognitive ability tests in selection settings. Journal of Applied Psychology, 92(6), 1672–1682. https://doi.org/10.1037/0021-9010.92.6.1672
- Marden, N. Y., Ulman, L. G., Wilson, F. S., & Velan, G. M. (2013). Online feedback assessments in physiology: Effects on students' learning experiences and outcomes. Advances in Physiology Education, 37(2), 192–200. https://doi.org/10.1152/advan.00092.2012
- Microsoft. (n.d.). Exam and Applied Skills assessment lab retake policies. Microsoft Learn. Retrieved October 10, 2026, from https://learn.microsoft.com/en-us/credentials/support/retake-policy
- Morphew, J. W., Silva, M., Herman, G., & West, M. (2020). Frequent mastery testing with second-chance exams leads to enhanced student learning in undergraduate engineering. Applied Cognitive Psychology, 34(1), 168–181. https://doi.org/10.1002/acp.3605
- Orchard, R. K. (2016). Multiple attempts for online assessments in an operations management course: An exploration. Journal of Education for Business, 91(8), 427–433. https://doi.org/10.1080/08832323.2016.1256262
- Rawson, K. A., & Dunlosky, J. (2011). Optimizing schedules of retrieval practice for durable and efficient learning: How much is enough? Journal of Experimental Psychology: General, 140(3), 283–302. https://doi.org/10.1037/a0023956
- Raymond, M. R., Neustel, S., & Anderson, D. (2007). Retest effects on identical and parallel forms in certification and licensure testing. Personnel Psychology, 60(2), 367–396. https://doi.org/10.1111/j.1744-6570.2007.00077.x