The test is marked and the scores are lower than you planned. A student asks whether you will grade on a curve. The useful first answer is a question: which curve?
Four adjustments share the name. One ranks students against each other, and research and published rules are hardest on it. None mends a faulty question.
EdProcto, which publishes this guide, reports every score as it was marked and sets the evidence for adjusting it alongside, in a quiz's results. In short: adjust for a fault the questions show, by a rule you announced, never to fill a quota.
What does grading on a curve mean?
Grading on a curve means one of four things, and only the first sets students against each other.
| Called a curve | What it does | A result depends on |
|---|---|---|
| Fixed shares | A set share of the class gets each grade, or the class mean is held to a figure | The other students |
| Scaling to a target | Every score is raised until the top score, or the mean, reaches a chosen figure | One figure taken from the class |
| Adding points | Everyone gains the same points | The teacher's choice of how many |
| Dropping the lowest quiz | Each student's weakest quiz leaves the term's total | The student alone |
Scaling and adding points keep the students in order, so neither ranks them. Both leave every mark on a faulty question as it was.
Grading against a standard or against the class
Grading against a standard, or criterion-referenced grading, asks what a student can do. Grading against the class, or norm-referenced grading, asks where the student stands among the others.
Under a standard everyone may pass, or nobody. Under fixed shares someone comes last however much the class knows. A pass mark is a standard: see setting a pass mark you can defend.
What does grading against the class do to students?
Grading against the class gives students a reason not to help each other, and its low grades weigh on who stays. The evidence is suggestive, and one experiment found no effect on effort.
Cooperation between students
- Scenarios. Burleigh and Meegan (2018) ran two studies with hypothetical scenarios, varying the grading (curved or absolute) and a classmate's standing (near the participant's, distant or unknown). Under curved grading participants were less willing to share lecture notes with a classmate of similar or unknown standing. We read the abstract on the first author's page; it gives no sample sizes, and a scenario is not a classroom.
- A review. Schinske and Tanner (2014), read in full, appear in our distribution guide for the curve's history. On its effects they cite Seymour and Hewitt's work of 1997 and Tobias's of 1990: the competition a curve breeds is one reason able, often under-represented students leave science.
- A field experiment. Czibor et al. (2020) compared absolute and relative grading at a Dutch university and found no significant effect on preparation or exam scores (abstract read; it gives no figures).
Who stays in the course
- A curved introductory course. Harris et al. (2020), read in full, followed 25,768 students through general chemistry at the University of Washington from 2001 to 2016, where final grades were curved to a mean of 2.6 on a four-point scale. Women, under-represented minority students and students of low socioeconomic status who got below a C− were less likely to take the next course than peers with the same grade and preparation; with a C or better, more likely. The study looks back at one university and did not test the curve itself.
- A cap on course averages. Butcher et al. (2014) studied Wellesley College's cap on most course averages at a B+. In the departments it bound, gaps in grades between racial groups widened, enrolments and majors fell, and students rated professors lower (abstract read).
- A curve nobody announced. Calsamiglia and Loviglio (2016), in records from public schools in Catalonia, found that teachers' grades behaved as a curve: stronger classmates lowered the grade a pupil received (abstract read).
Where did the bell curve of marks come from?
The bell curve of marks came from early intelligence testing, by an argument that needed a class to be a sample of the whole population, which a class of thirty is not. Schinske and Tanner (2014), citing Finkelstein's book of 1913, give the reasoning: if a classroom held a representative sample of the population, its grades should fall on a normal curve, as aptitude seemed to. Even supporters, they add, saw the catch: a classroom might not be such a sample, and teachers mark what a student has done, not aptitude.
Guskey (2011), read in full, objects: the normal curve describes "the distribution of randomly occurring events when nothing intervenes". Teaching intervenes. If marks after a unit still form a bell, he argues, the teaching made no difference.
A class of thirty is chosen, by options, sets and prerequisites; it has been taught; and it is small. In the worked example below, the top scorer's absence would be worth fifteen points to every classmate.
What do published rules say about curving grades?
Published rules disagree: one fixes the distribution, one declines to, and an examining body moves the line for the paper, not the candidates. All three were read on 10 October 2026.
- A mandatory curve. Northwestern Pritzker School of Law (n.d.) sets the mean of first-year required doctrinal courses at 3.45, with at most 5% of A+ grades and at least 7% of B− and below. Upper-level doctrinal courses are held to a mean only where 13 or more are enrolled.
- No required distribution. The University of Melbourne (2026) requires criterion-referenced grading and holds no dean to a specific distribution of grades. A board of examiners may still standardise provisional results for a marking error, a disproportionate cohort or an irregular distribution. Any adjustment must be documented and transparent, and students given raw scores that may be scaled should be told.
- A line set for each paper. At AQA, an examining body in England, Pointer (2022) explains that papers vary slightly in difficulty and a harder one should not cost a student a grade. So grade boundaries are set for each exam after marking, by senior examiners, with statistics keeping the boards in line.
Where your institution requires a distribution, apply it as written. Where it does not, copy the examining body: judge the paper, then move the line for everyone.
When is adjusting scores defensible, and how is it done without ranking?
An adjustment is defensible when the paper turned out harder than intended and the questions show it. Low scores alone do not: the paper, the teaching and the class can each produce them.
Schinske and Tanner (2014) advise finding out why students did poorly, and excluding a confusing question before thinking of a curve.
- Hold the scores. In EdProcto, "Show the score" offers "At once", "After submitting", "When released" and "Never": "When released" keeps every score from students until you press "Release the scores" in the results.
- Find the questions most missed. "Got it", under "Each question", is the share who got each right: see item difficulty.
- Answer each one cold and name its fault: a wrong key, no right option or two, content not taught, or none. The question with the wrong key lists the signs.
- Repair the question. Correct a wrong key and mark again. Drop a question with no defensible answer, or credit everyone: when to throw a question out compares the two.
- Only then consider the pass mark. If sound questions asked more than the course taught, move it by a method stated beforehand, as in setting a pass mark you can defend, and record the reason.
- Apply the change to everyone, then release.
No step asks where a student stands among the others.
How do you tell students before the test?
Tell students in writing, before the first test, how a mark becomes a grade and what could change it. Students often assume a curve where there is none: Schinske and Tanner (2014) report this from Tobias, and advise saying outright that nobody competes with a classmate. Put four things in the course outline:
- the pass mark, and that everyone can reach it;
- what happens to a faulty question: re-marked, dropped or credited, for all, before scores go out;
- whether any scaling is possible, by what method, and who approves it;
- any course rule, such as dropping the lowest quiz.
Which adjustment fits which finding?
Our advice: match the adjustment to what the questions show, not to how low the scores are.
| What you find | What to do | In EdProcto |
|---|---|---|
| A wrong key | Correct it and mark again | "Re-mark this quiz", in the results |
| No right option, or content never taught | Drop the question, or credit everyone | Delete it and re-mark, or "Add a note, or set the points by hand" |
| Sound questions pitched above the course | Move the pass mark by the stated method | "Pass mark" |
| A sound paper and low scores | No adjustment: reteach and test again | The scores stand |
| A rule requiring shares, a mean or scaling | Apply it exactly, in the open | "Download the marks", then your own sheet |
A worked example: one class under three rules
This class is invented. Its marks and class figures are what EdProcto's marking and analysis code returned for its answers; Rules 1 and 2 are our arithmetic on those scores. Thirty students sat twenty four-option questions, one point each, with a pass mark of 60%.
As marked, the page reads "Average score" 55.7%, "Median" 55% and "Passed" 43.3%: 13 of 30. Beside "Score distribution" stands "30 submitted, mean 55.7%, spread 17.4, lowest 25%, highest 95%". The next score down is 80%.
On Q8, "Got it" is 13%: four students chose the keyed option and averaged 46.3% on the paper; 24 chose one other and averaged 59%. The full analysis prints "Check the key on Q8: the stronger students got it wrong more often than the weaker ones." Worked cold, the 24 were right: the key was a slip. On Q15, at 20%, the answers spread 6, 7, 6 and 11 over the four options, and rereading finds that none is right.
| Rule | Passed, of 30 | What decides a result |
|---|---|---|
| 1. Add points until the top score is 100% | 16 | One student's 95% |
| 2. Fixed shares of 10, 20, 40, 20 and 10% for A to F | 27, with D a pass | Rank in the class |
| 3. Re-mark Q8, drop Q15, keep the pass mark | 15 | Nineteen sound questions |
Rule 1 adds five points to everyone. Had the top scorer been absent, 80% would have set it: twenty points each, and 24 of 29 passing. The 24 who answered Q8 correctly stay marked wrong.
Rule 2 fails the bottom three, on 30%, 30% and 25%, and passes two students on 35%. Four are tied on 75%: one takes the last A and three a B, by a tie-break the rule never states.
Rule 3 changes marks, not the scale. With Q8 keyed correctly and Q15 deleted, the paper is out of 19: 26 scores rise and four fall, those of the students who chose the old key. The mean is 61% and 15 pass. Two students on 55% who answered Q8 correctly now pass on 63.2%. A third on 55%, who chose the old key, has 52.6%; Rule 1 would have passed all three.
Half the class is still under 60% on nineteen sound questions: a finding about the teaching or the paper's level, for step 5. A curve would have buried it with the fault.
What does EdProcto show and do?
EdProcto is the best tool for deciding whether to adjust a set of scores because it keeps apart the two things a curve blurs: what each student earned, and whether the paper was sound. In the example, one screen showed 24 students marked wrong on a question they answered correctly.
What it shows. A score is the points earned over the points on the paper, as a percentage, whoever else sat the quiz. At the top of the results stand "Average score" with "spread (sd)", "Median", "Passed" and "Score distribution": ten bands of ten points, the count over each bar, a line at the pass mark, and the lowest and highest score beside the mean.
What it does when you ask.
- "Re-mark this quiz" opens a card: "Every attempt handed in is marked again from the questions and keys as they stand now. A mark you or Procto AI gave is kept." Press "Yes, re-mark" and it reports how many attempts have new marks. A question deleted in the builder leaves the totals at that re-mark.
- "Add a note, or set the points by hand", under a marked answer on a student's attempt, opens "Points", "Note for the student" and "Save my grade". The points run from 0 to the question's own.
- "Pass mark", under "Grading" in the settings, starts at 60% and takes any figure from 0 to 100. "Passed" is counted against it.
- "Download the marks" saves a spreadsheet file with a row for each attempt handed in: "Name", "Email", "Sitting", "Started (your time)", "Handed in (your time)", "Score %", "Points", "Out of", "Reduced by %", "To grade", "Minutes", "Security events", "Notes on the record" and the points on each question.
Where a course rule requires scaling, fixed shares or a dropped quiz, that sum is done in your own sheet from that file. EdProcto's record stays the scores as marked; what they become is your decision.
The plans. All of this is on every plan: Free takes thirty students a quiz, Tutor 100 and Educator 500. The "Check the key" sentence and the takers of each option are in the full analysis, on Educator. See pricing, then start on the home page in multiple choice.
Quick questions
Is grading on a curve fair?
Fixed shares are not: a grade then depends on who else is in the room. Adding the same points to everyone treats students alike and leaves a faulty question marked wrongly.
What is the difference between curving and scaling?
Curving, strictly, gives set shares of the class each grade. Scaling moves every score by one sum and keeps the students in order.
Should I curve a test that most of the class failed?
Read the questions first. Re-mark a wrong key, drop a question with no right answer, and only then ask whether the pass mark should move by a stated method.
Does EdProcto grade on a curve?
EdProcto reports each score as marked, re-marks from a corrected key when you press "Re-mark this quiz" in the results, and gives the marks as a file for any rule your course requires.
References
- Burleigh, T. J., & Meegan, D. V. (2018). Risky prospects and risk aversion tendencies: Does competition in the classroom depend on grading practices and knowledge of peer-status? Social Psychology of Education, 21(2), 323–335. https://tylerburleigh.com/research/articles/burleigh-meegan-2017
- Butcher, K. F., McEwan, P. J., & Weerapana, A. (2014). The effects of an anti-grade-inflation policy at Wellesley College. Journal of Economic Perspectives, 28(3), 189–204. https://doi.org/10.1257/jep.28.3.189
- Calsamiglia, C., & Loviglio, A. (2016). Grading on a curve: When having good peers is not good (BSE Working Paper No. 940). Barcelona School of Economics. https://bse.eu/research/working-papers/grading-curve-when-having-good-peers-not-good
- Czibor, E., Onderstal, S., Sloof, R., & van Praag, C. M. (2020). Does relative grading help male students? Evidence from a field experiment in the classroom. Economics of Education Review, 75, Article 101953. https://research.vu.nl/en/publications/does-relative-grading-help-male-students-evidence-from-a-field-ex/
- Guskey, T. R. (2011). Five obstacles to grading reform. Educational Leadership, 69(3). https://www.ascd.org/el/articles/five-obstacles-to-grading-reform
- Harris, R. B., Mack, M. R., Bryant, J., Theobald, E. J., & Freeman, S. (2020). Reducing achievement gaps in undergraduate general chemistry could lift underrepresented students into a "hyperpersistent zone". Science Advances, 6(24), Article eaaz5687. https://pmc.ncbi.nlm.nih.gov/articles/PMC7286681/
- Northwestern Pritzker School of Law. (n.d.). Grading policy. Northwestern University. https://www.law.northwestern.edu/registrar/gradingpolicy/
- Pointer, W. (2022, June 14). Making the grades – How does an exam turn into a qualification? AQA. https://www.aqa.org.uk/aqi/making-the-grades-how-does-an-exam-turn-into-a-qualification
- Schinske, J., & Tanner, K. (2014). Teaching more by grading less (or differently). CBE Life Sciences Education, 13(2), 159–166. https://pmc.ncbi.nlm.nih.gov/articles/PMC4041495/
- University of Melbourne. (2026). Assessment and results policy (MPF1326) (Version 29). https://policy.unimelb.edu.au/MPF1326/