Sign up free

Should students be able to go back and change quiz answers?

What studies say about letting students go back and change quiz answers, what locking answers does to scores, time and anxiety, and when to switch it off.

The EdProcto editorial teamWritten and checked against the research it cites. Last checked . How the guides are made

Guides · Tools, compared · · 13 min read

Should an online quiz let students go back and change their answers?

Students should be able to go back and change answers in most quizzes: studies over nearly a century find changes go from wrong to right more often than the reverse. Switch going back off only where answers could be passed around during the sitting, or where a later question gives an earlier one away, and tell students first.

Whether students can go back and change an answer is one switch on a quiz, and an old argument. One side holds that a first answer should stand; the other, that review is part of knowing something.

This guide is for a teacher, lecturer or trainer choosing the setting. Each source was read on 10 October 2026; "abstract read" marks one where only the abstract or an index record was open.

EdProcto, which publishes this guide, carries the switch as "Allow going back to change an answer". The three presets covers the rest of the settings page.

Should students be able to go back and change answers?

Yes, by default. Switch going back off in two cases: where answers could be passed around during the sitting, and where a later question gives an earlier answer away. The table is our recommendation; the evidence follows.

What the quiz is for Going back Why In EdProcto
Practice On Correcting a slip is the learning The Practice preset
A weekly check On Nothing to protect The Standard quiz preset
A unit test in one room On Changes gain marks more often than they lose them Standard quiz, in EdProcto
An exam sat apart, where answers could be passed around Off: one question a page, everyone at once An answer closes before it can be compared The Exam preset, or the one switch
A case that unfolds Off The next question states the last answer The case study template holds it off

What does research say about changing answers?

Changed answers go from wrong to right more often than the reverse, in study after study since the 1920s.

Wrong to right, more often than right to wrong

Benjamin et al. (1984) reviewed 33 studies: most test takers change some answers, most changes run from incorrect to correct, and most students who change gain marks (abstract read).

Larger samples agree. Liu et al. (2015) analysed 8,538 and 9,140 candidates on the quantitative and verbal sections of the GRE, in 12 countries: 82.8% and 67.8% scored higher for their changes, about 4% and 15% lower. Ouyang et al. (2019) followed 27,830 candidates through one block of 40 to 45 questions in an American medical licensing examination. A candidate revisited 16 questions on average and changed 1.4: 0.59 from wrong to right and 0.40 from right to wrong. Of those who changed anything, about 45% gained and 28% lost.

One dissent did not stand. Bridgeman (2012) writes that a 2011 paper reporting losses from changing rested on flawed data, and that a plain count on the same data showed an overwhelming majority scoring higher after review (abstract read).

Why does the "first instinct" belief persist?

Because a change that goes wrong is remembered. Kruger et al. (2005) argue that switching away from a right answer brings more regret than sticking with a wrong one, so those occasions come to mind more easily and pass for evidence (abstract read).

Teachers hold the belief too. Among the lecturers Merry et al. (2021) surveyed, the commonest view, held by 36%, was that changes usually harm grades. Eraser marks on that study's exam papers showed, across two exams, 2.8 changes a student from wrong to right and 1.0 the other way (abstract read).

Who gains from changing?

Every group studied, and the strongest most. On the GRE the benefit rose with ability (Liu et al., 2015). In the licensing data, strong candidates made 1.58 wrong-to-right changes for each right-to-wrong one and candidates who failed made 1.28: both above one (Ouyang et al., 2019).

Sticking is not foolish, though. Couchman et al. (2016) found that revising and sticking both ended right more often than wrong, and that confidence rated question by question predicted which was wise (abstract read).

What happens when going back is switched off?

Average scores barely move, exams finish no later, and the students asked disliked it. The comparisons are of computer-based exams in health programmes, each cohort against the year before, none randomised.

Scores and time

  • Cochran et al. (2020). Six pharmacy exams at one American university, with backward navigation in 2018 and without in 2019. On questions identical in both years the mean moved between +3.3% and −2.6%, never significantly; time a question fell on two exams.
  • Caetano and Pawasauskas (2020). 576 unchanged questions over two years of a pharmacy programme. Difficulty and discrimination held, as did the lowest-scoring 27%; the highest scorers lost 0.95% (abstract read).
  • Giruzzi et al. (2022). Calculation exams in four pharmacy courses: first attempts significantly worse in two, retakes better in the other two, time down or level.

The size fits the answer-changing figures: in Ouyang et al. (2019) the net gain was 0.19 of a question in a block of 40 to 45, under half a percentage point by our arithmetic.

Anxiety

Giruzzi et al. (2022) surveyed 400 of those students: 84% agreed that a single chance at each question added to their test anxiety, and about 90% of those asked had changed how they sat a test, reading more carefully and no longer second-guessing.

A brake on collusion, and its cost

The case for the lock is that an answer cannot be amended after it has been compared. Giruzzi's college adopted it over concern that neighbours were lining up on the same question despite shuffling.

Li et al. (2021) built an online final on it for 78 undergraduates at one American university in April 2020: 40 questions, one at a time in fixed slots with no revisiting, the sequences arranged so that a likely copier usually met a question before a likely helper did. Their model put the average gain from collusion at 0.0073% of the score, against 19.23% for one paper seen by all at once: a calculation, not an observation.

It is a brake, not a seal. Half the students Giruzzi surveyed did not think it reduced dishonesty, and the lecturer's midterm describes an exam with no backtracking in which information still flowed from earlier rounds to later ones. No study we read measured copying with and without the lock. Honest students pay the half point above, the strongest most, and the anxiety.

What do question order and a review screen add?

Order matters more once nobody can return, and a review screen turns doubt into a short list.

A student who can go back makes an order of their own: sure answers first, hard ones last. Lock the answers and that choice passes to you, or to the shuffle; easy, hard, easy has the studies of order by difficulty. Cochran et al. (2020) add that with no return a later question may state an earlier answer, which is how a case that unfolds is built.

Review mostly confirms: 16 questions revisited, 1.4 changed (Ouyang et al., 2019). On the Medical Council of Canada's qualifying examination, McConnell et al. (2012) found that candidates more often flagged, and more often changed, the questions they had first answered wrongly (abstract read).

What does accessibility and exam-anxiety guidance say about locking answers?

It leans towards review.

  • The web accessibility standard. WCAG 2.2, criterion 3.3.4 at level AA, covers pages that submit a user's test responses: a submission must be reversible, or checked for input errors, or preceded by a way to review, confirm and correct. The reason given: people with reading disabilities may transpose characters, and those with motor disabilities may hit keys by mistake (World Wide Web Consortium, 2025).
  • The cognitive pattern. A supplementary W3C pattern, not required for conformance, says that being able to return to what they have entered and check it matters greatly to people with anxiety or memory difficulties (World Wide Web Consortium, 2021).
  • Exam-stress advice. The NHS tells students not to panic when stuck on a question and to leave time at the end to come back to it (National Health Service, 2023). A quiz with no return takes that advice away.

So a lock needs a reason, a warning, and another route for a student whose arrangement calls for review. Who has one is for your institution's policy; this is not legal advice, and nothing here claims conformance for any tool.

How do you tell students that answers are final?

Tell them before the day, give the reason, and let them practise under the rule. A script:

This exam shows one question at a time. When you press Next, that answer is final and you cannot return to it. The rule is there so that nobody can compare answers and then amend their own. Read the whole question, answer it, and check it before you move on. Nothing can be saved for the end, so spread your time evenly.

EdProcto's start screen repeats it: "No going back: each answer is final." Where going back is on, the brief is shorter: flag what you doubt, return to the flags, and change an answer when you have a reason.

What does the decision look like across one module?

Two quizzes with going back on and one with it off. This example is invented: a pharmacology module of 84 students.

  • Weekly practice, ten questions. On: a corrected slip is the point.
  • Mid-module test, 30 questions, in a computer room. On: an invigilator is present, and review gains marks.
  • Final, 40 multiple-choice questions, sat at home at 10:00. Off: by the licensing figures the lock costs an average student a fifth of a question, and no answer can be amended after a friend's message.

For the final, in EdProcto:

  1. Switch "Allow going back to change an answer" off, or press the Exam preset, which switches skipping off too.
  2. Leave "Layout" at "One question a page" and "A review screen before the final submit" on.
  3. Keep "Allow skipping a question" off: "Next" then waits for an answer, and no question is passed blank by accident.
  4. Leave "Move on as soon as a question is answered" off. Its hint reads "For speed drills.", and with no return a first click would stand.
  5. For a student whose arrangement calls for review, "Duplicate this quiz" makes a copy with a link of its own: switch going back on there.

How does EdProcto handle going back?

EdProcto is the best tool for this decision because the rule is held on its server and said to the student before the first question. Both matter: a lock that a reload undid would be no brake, and a rule first met mid-exam is the kind students fear.

The switches

Under "Flow and pacing" are "Layout" ("One question a page", "All on one page" or "Grouped by section"), "Allow going back to change an answer", "Allow skipping a question" and "A review screen before the final submit". The three switches start on, and Layout at one question a page. Practice and Standard quiz switch going back and skipping on; Exam switches both off; all three keep the review screen.

With going back on

The start screen says "You can go back and change an answer before you submit." "Previous" works from the second question, and each question carries "Flag to come back to". "Review and submit", on the last question, opens the review screen. Headed "Before you submit", it counts the answers ("18 of 20 answered, 2 flagged."), marks each question "answered" or "blank" and says "Open any question to change it." "Back to the questions" returns to the paper; "Submit the quiz" hands it in. After a reload the attempt reopens on the first question still unanswered.

With going back off

The start screen says "No going back: each answer is final." With one question a page, "Previous" is greyed and the server closes each question behind the student: once they have moved past it, anything sent for it later is set aside for the answer already held. The furthest question reached only moves forward, so after a reload the same link in the same browser reopens the attempt there, never on a question already passed.

The questions themselves arrive together, without their keys, when the attempt begins (why it is built this way); the page shows one at a time, and what the server closes is the answer. With the other two layouts the questions share a page and stay open while the student is on it, and the start screen says so in place of "No going back". The review screen reads "Answers cannot be changed."

Templates that hold it off

A quiz built on Case study, Scaffolded sequence, Listen once, Memory recall or the four other memory shapes keeps going back off whichever preset is pressed; the settings page greys the switch with the reason "the template has no going back".

These switches are on every plan. Free takes 30 students a quiz, Tutor 100, and Educator 500 with the full analysis; the Exam preset as a whole reaches Medium integrity, saved from Tutor up (pricing). No setting makes copying impossible, a logged event is a record and not proof, and every score is yours to release.

Quick questions

Should you change your answer or trust your first instinct?

Change it when you have a reason: across 33 studies, most changes went from wrong to right (Benjamin et al., 1984).

Does switching off going back lower scores?

Hardly, on average: across six exams, Cochran et al. (2020) found no significant fall.

Does one question at a time with no return stop cheating?

It slows the passing of answers and does not end it; no study we read has measured by how much.

Which EdProcto quizzes never allow going back?

Those built on the case study, scaffolded sequence, listen once and five memory templates.

References

The settings

Time, attempts, feedback, grading, access, identity, security, tools, accessibility: every rule of a quiz, set by itself.

On every plan

See the settings

Read next