Partial credit gives part of a question's marks for part of a right answer: three pairs matched out of five, a sound method with a wrong final figure, two rubric elements out of four. The alternative is all or nothing, where anything short of the whole answer earns zero.
Neither rule is the fair one everywhere. Part marks show what a student half knows, and they also let halves add up to a pass. This guide covers the one decision across question shapes, for a teacher, lecturer or trainer choosing the rule before a quiz.
EdProcto, which publishes this guide, keeps the choice as one switch, "Partial credit", in each quiz's settings, and its templates say how each shape is marked.
What is partial credit in each question shape?
Partial credit means something different in each shape, because the parts are different things: ticks, places, pairs, steps or rubric elements.
- Multiple answers. Some boxes are ticked rightly. The scoring rule has to say what a wrong tick costs (multiple-answer questions).
- An order. Some items stand in the right place, or the right way round relative to each other, which is not the same count (ordering questions).
- Matching and sorting. Some pairs are right, or some items sit in the right group (matching questions).
- A calculation. The method is sound and the final answer wrong. Method marks pay for the approach, and error carried forward charges a slip once (marking the working).
- A written answer. Some elements of the rubric are present (short answers and their rubric).
What does research say about part marks?
Research mostly favours part marks on questions with several right parts, though less firmly than the usual advice suggests. Where we read only an abstract, we say so.
Multiple-answer questions
Bauer et al. (2011) rescored six end-of-term examinations in internal medicine at Munich University, from 2005 to 2008: 1,255 third-year students and 180 questions with several right options. They compared all or nothing with two part-credit rules, half marks for at least half the right options and an equal fraction for each right option, neither with a deduction for wrong picks. Part credit gave better psychometric results than all or nothing, and the two rules differed little (abstract read).
The gain can be small. On a large adaptive licensing examination, Jiao et al. (2012) found ability estimates with and without part credit highly correlated and pass or fail decisions almost identical, because few questions could carry part marks (abstract read).
One calculation disagrees. Eggen and Lampe (2011), cited in our ordering guide for another result, worked reliability out from assumed levels of knowledge, not from students' answers. Where a student must select all the right options, all or nothing came out more reliable; for matching, part credit did. They recommend a plain count of right selections over rules that correct for wrong answers (abstract read).
A plain count has one cost that is arithmetic, not research: where the number of ticks is free, ticking every box collects every mark. Hence rules that set wrong ticks against right ones.
Ordering questions
For ordering there is little to lean on. Alderson et al. (2000), writing on reading tests that have a jumbled text put back in order, reported problems in scoring the responses and explored several procedures, knowing of no earlier research (abstract read; it does not say which they preferred). Eggen and Lampe (2011) found their sequencing results less conclusive than the rest. So choose by purpose.
Method marks and error carried forward
The case for marking method is that a final answer cannot tell a slip from a blank. Studies of what the marking then does are few.
Pearson and Penna (2023) built long, automatically marked calculation questions for surveying modules taken by engineering students at Newcastle University. Carried-forward marks given without further penalty when a solution was fundamentally incorrect produced marks that were not normally distributed, and the authors settled on 25% or 50% of the marks available. Questions split into too few parts behaved as all or nothing, with marks in two humps. Too many parts gave too much guidance, and marks ran unnaturally high. They suggest three or four parts (abstract read; no sample size given).
Judged part marks also cost agreement. Suto and Nádas (2008) had groups of markers mark identical samples of GCSE mathematics and physics answers. Questions calling for more complex, reflective judgement were marked less accurately than those needing a simple one (abstract read). That a method mark is such a judgement is our inference.
What students do when part marks are on offer
They use them. At a Belgian university, 683 master's students of medicine sat two real examinations, each year group taking one under a scheme that paid part marks for narrowing the options down (Vanderoost et al., 2018). Between 95.5% and 98.4% of students per sitting showed doubt between options on at least one question, and on 14.7% to 18.7% of questions on average. Scores did not differ significantly from those under a penalty for wrong answers.
The marking also teaches. In an interview study of physics lecturers, Henderson et al. (2004) found that almost all told students to show their reasoning, and about half graded in a way likely to discourage it. They were reluctant to take marks from a solution that might be correct, and tended to read sound thinking into it (abstract read).
When is all or nothing the honest rule?
All or nothing is the honest rule when part of the answer is no use: a safety procedure, a single fact or a mastery check. These are our working rules; no study chose them.
- A safety procedure. Five of six lock-out steps in order is not five sixths safe. The score should say "not yet".
- A single fact. A date, a capital, one term: there are no parts to pay for.
- A mastery check. The question is whether the student can do the whole thing, and the remedy for "no" is another attempt, not a fraction.
On mastery the evidence is mixed. In spring 2015, Sangelkar et al. (2017) divided 117 engineering students in four sections of one course into a group given marks only for essentially correct solutions, with up to three tries, and a group given part marks. On a common final examination the totals did not differ significantly. Across six calculus classes over two years, Harsy and Hoofnagle (2020) reported higher course grades under mastery testing, with fewer hours of study (abstracts read).
When all or nothing hides what a student knows
All or nothing hides knowledge whenever a question has many parts and the score is read as "how much". Swap two neighbouring steps in an order of six and the mark equals a blank page. The longer the set, the more it throws away.
How can a part mark hide a gap?
A part mark hides a gap when the same share can be earned by knowing half or by doing something cheap, and when halves are left to add up. Reflecting on a discrete mathematics course, Linhart (2020) wrote of students "putting together enough partial credit to pass the class" without having mastered its content (abstract read).
- The student who always gets half. Half on every question is 50% with no question secure, and the total looks like half the unit mastered.
- The cheap half. Drop every item of an evenly split two-group sort into one group and half are right. Tick every box where the right options outnumber the wrong ones and the net is still positive.
- The average. A question that half the class got wholly right and a question that everyone half answered both show 50%.
The remedy is to read part marks by student and by question, and to keep an all-or-nothing question on anything essential.
Which rule fits which shape and purpose?
Fit the rule to what the score is for: part marks to see what a student knows, all or nothing to certify that a student can do the whole thing.
| Shape | Practice or a lesson check | A test that counts | Mastery or safety |
|---|---|---|---|
| Multiple answers | Part marks, wrong ticks charged | The same | All or nothing |
| An order | Part marks | Part marks, rule stated | All or nothing |
| Matching, sorting | A share for each pair or item | The same | All or nothing, short sets |
| A calculation | Steps marked, a slip charged once | Carried marks capped | Exact final answer, with a retake |
| A written answer | Rubric, a mark an element | The same | Every essential element present |
| Rows one to three in EdProcto | "Partial credit" on | "Partial credit" on | "Partial credit" off |
The last two rows are marked by a reader.
What happens when one class is marked both ways?
The same answers give two rankings, and each rule hides something. Four questions on the heart, 5 points each, were marked under EdProcto's rules, every mark run through its marking code:
- Multiple answers: which of five vessels carry deoxygenated blood (two do).
- An order: six places blood passes after returning from the body.
- Matching: five parts of the heart to their jobs.
- Sorting: eight features into "artery" or "vein".
| Student | The answers | All or nothing | Partial credit |
|---|---|---|---|
| A | Everything right | 20 | 20 |
| B | Question 1 right; two neighbouring steps swapped; two partners swapped; one feature misplaced | 5 | 16.71 |
| C | Questions 1 and 2 right; 3 and 4 not reached | 10 | 10 |
| D | One right tick of two; the lungs put first and two steps swapped; two pairs of five; every feature in "artery" | 0 | 10 |
Under all or nothing, B, who slipped once on each of three questions, sits below C, who left half the quiz. At a 60% pass mark only A passes.
Under partial credit B passes on 84%, which is fairer to B. But C and D are now level on 10, and D has no question right: 2.5 of D's points came from putting all eight features in one group. On the results page the sorting question's "Got it" would read 25% under the first rule and 59% under the second.
Neither column is wrong. Open a student's attempt before trusting a middling score.
How do you tell students the rule?
Tell them before the quiz, in one sentence a question, in the words of the rule and not its name. "Partial credit applies" says nothing. These do:
- "Each wrong tick cancels a right one."
- "Marks are for steps in the right order relative to one another."
- "One mark for each correct pair."
- "A final answer with no working earns the last mark only."
Put the sentence in the question's prompt, and keep one rule across a course.
How does EdProcto give partial credit?
EdProcto is the best tool for this decision, because the rule is one switch a quiz, the part marks that need no switch are known in advance, and any mark stays yours to change. The worked example shows what rides on it: the same answers earned 5 or 16.71.
"Partial credit" is in the Grading group of a quiz's settings, with the note "On multiple-answer, ordering, matching, sorting and best-and-worst questions. Gaps, lists and a number’s half margin earn their part marks either way." It starts off, and the Practice, Standard quiz and Exam presets leave it as it stands. With it on:
- Multiple answers: right ticks less wrong ticks, over the number of right options, never below nothing.
- An order: every pair of items the right way round counts for the student and every pair the wrong way round against. The net, over all the pairs, is the share, never below nothing; an item left out counts neither way.
- Matching: the share of pairs right. Sorting: the share of items in the right group.
- Best and worst: half for each end.
With it off, each of those is all or nothing. Three kinds of answer earn part marks whatever the switch says: gaps in a passage, gap by gap; a typed list, item by item; and a number inside its wider margin, which earns half. Give a number's two margins the same figure and it is all or nothing as well. A single choice, one word and a click on an image are right or wrong.
"Negative marking" is a separate switch for single-choice questions: "A wrong choice loses points: one over the number of wrong options."
Written answers and show your working wait for a reader. On the grading page each has "Points", in half marks, "Note for the student" and "Save my grade", so the part mark follows your rubric: write error carried forward into it. In a scaffolded sequence each step is marked against its own key, so open each step with the right result of the one before. Any answer the rules marked has "Add a note, or set the points by hand", for the answer the key did not foresee.
The grading page labels each answer "right", "partly", "wrong" or "to grade", with the points beside a marked one. The results page's "Got it" is the average share of a question's points earned. Every score is yours to release.
Free takes thirty students a quiz, Tutor 100, and Educator 500 with the full analysis; see pricing.
Quick questions
Is partial credit fairer than all or nothing?
It is fairer to the student who knows most of a many-part question, and less able to certify that anyone knows all of it.
Should a wrong tick lose marks on a multiple-answer question?
Yes, when the number of ticks is free; otherwise ticking every box earns full marks. EdProcto's multiple answers sets wrong ticks against right ones and never goes below zero.
Can students pass on partial credit without mastering anything?
They can, when halves add up to the pass mark. Keep an all-or-nothing question on each essential and read each student's marks question by question.
Is negative marking the same as partial credit?
No. Negative marking takes points for a wrong single choice; partial credit pays for the right part of a many-part answer. Our guide to negative marking covers it.
References
- Alderson, J. C., Percsich, R., & Szabo, G. (2000). Sequencing as an item type. Language Testing, 17(4), 423–447. https://doi.org/10.1177/026553220001700403
- Bauer, D., Holzer, M., Kopp, V., & Fischer, M. R. (2011). Pick-N multiple choice-exams: A comparison of scoring algorithms. Advances in Health Sciences Education, 16(2), 211–221. https://doi.org/10.1007/s10459-010-9256-1
- Eggen, T. J. H. M., & Lampe, T. T. M. (2011). Comparison of the reliability of scoring methods of multiple-response items, matching items, and sequencing items. CADMO, 19(2), 85–104. https://doi.org/10.3280/CAD2011-002008
- Harsy, A., & Hoofnagle, A. (2020). Comparing mastery-based testing with traditional testing in Calculus II. International Journal for the Scholarship of Teaching and Learning, 14(2), Article 10. https://doi.org/10.20429/ijsotl.2020.140210
- Henderson, C., Yerushalmi, E., Kuo, V. H., Heller, P., & Heller, K. (2004). Grading student problem solutions: The challenge of sending a consistent message. American Journal of Physics, 72(2), 164–169. https://doi.org/10.1119/1.1634963
- Jiao, H., Liu, J., Haynie, K., Woo, A., & Gorham, J. (2012). Comparison between dichotomous and polytomous scoring of innovative items in a large-scale computerized adaptive test. Educational and Psychological Measurement, 72(3), 493–509. https://doi.org/10.1177/0013164411422903
- Linhart, J. M. (2020). Mastery-based testing to promote learning: Experiences with discrete mathematics. PRIMUS, 30(8–10), 1087–1109. https://doi.org/10.1080/10511970.2019.1695236
- Pearson, C., & Penna, N. (2023). Automated marking of longer computational questions in engineering subjects. Assessment & Evaluation in Higher Education, 48(7), 915–925. https://doi.org/10.1080/02602938.2022.2144801
- Sangelkar, S., Lasher, W., Warley, R. L., & Hollinger, A. S. (2017). Comparison of partial credit and mastery assessment on student learning and retention. In 2017 IEEE Frontiers in Education Conference (FIE) (pp. 1–6). IEEE. https://doi.org/10.1109/FIE.2017.8190603
- Suto, W. M. I., & Nádas, R. (2008). What determines GCSE marking accuracy? An exploration of expertise among maths and physics markers. Research Papers in Education, 23(4), 477–497. https://doi.org/10.1080/02671520701755499
- Vanderoost, J., Janssen, R., Eggermont, J., Callens, R., & De Laet, T. (2018). Elimination testing with adapted scoring reduces guessing and anxiety in multiple-choice assessments, but does not increase grade average in comparison with negative marking. PLoS ONE, 13(10), Article e0203931. https://doi.org/10.1371/journal.pone.0203931