Grading practice
How to Award Partial Credit Consistently
Use independent criteria, defensible mark increments, and moderation to recognise partial understanding consistently.
- By
- MarkingEase Editorial Team
- Published
- Reading time
- 9 minute read
All-or-nothing scoring treats a nearly complete chain of reasoning like a blank answer. Partial credit can represent what a student demonstrated, but only when the route to those marks is planned and applied consistently.
The aim is not to reward effort that is invisible in the answer. It is to separate independent evidence and reserve teacher judgment for cases the scoring plan could not sensibly anticipate.
Break the answer into independent evidence
Consider: ‘Calculate the mean of 6, 8, 10, and 12 and explain what it represents.’ A student might correctly sum to 36, divide by four incorrectly, or calculate 9 without explaining it. Calculation and interpretation are separable evidence.
- One mark: correct sum or equivalent method.
- One mark: divide by four and obtain 9.
- One mark: explain that 9 represents the equal-share or central value for this set.
Separate partial understanding from a lucky output
A correct final number with no work may satisfy an answer-only item, but it does not demonstrate a required method. Conversely, a clear method with one arithmetic slip may show most of the intended understanding. State whether method evidence is required and how consequential errors are handled.
Avoid overlapping criteria
If a software-design response is scored for ‘appropriate design pattern’ and ‘justification’, naming the pattern belongs to the first criterion; explaining why it fits the constraints belongs to the second. Rewarding the name twice double-counts one piece of evidence.
- Underline the distinct evidence phrase in each criterion.
- Merge criteria that cannot be scored independently.
- Remove presentation marks unless presentation is an explicit outcome.
Use meaningful mark increments
The smallest increment should correspond to a distinction a grader can reliably observe. A ten-point criterion with no descriptors invites invented differences between 6 and 7. Describe those boundaries, use fewer levels, or split the criterion.
| Award | Evidence |
|---|---|
| 4 | Correct claim, mechanism, and application |
| 3 | Correct claim and mechanism; minor omission in application |
| 2 | Core idea correct but explanation or application incomplete |
| 1 | Relevant fact without an adequate connection |
| 0 | No relevant evidence |
Handle unexpected valid answers
A rubric is a guide, not permission to reject sound reasoning. Pause when an answer reaches the learning outcome by an unanticipated route. Check subject accuracy, document the decision, and apply the same interpretation to comparable responses.
- Do not quietly change the rule for one student.
- Record accepted alternatives for the marking session.
- If the question caused ambiguity, consider a cohort-wide response.
Moderate the boundaries
Choose samples near important boundaries rather than only obvious high and low answers. Ask graders to cite the exact evidence supporting each mark. Disagreement about evidence suggests a marking issue; disagreement about the task may reveal a design issue.
- Score the same sample independently.
- Compare criterion-level awards.
- Resolve the reason for each difference.
- Update anchor examples and recheck earlier scripts if guidance changes.
Practical checklist
- The mark scheme separates distinct evidence.
- Every increment has an observable meaning.
- Method expectations are explicit.
- Consequential errors are handled consistently.
- Criteria do not double-count evidence.
- Unexpected valid reasoning has a moderation path.
- Boundary samples have been compared.