Scaling School-Assessment Marks When the Cohort Is Small
A school has 11 students in Stage 2 Armenian. School assessment contributes a fixed proportion of the final result, and the school must submit a set of assessment marks that will be used alongside the external examination. The internal marks are out of 100. The distribution is: 92, 88, 85, 79, 74, 70, 66, 61, 55, 48, 40. The teacher believes the cohort is strong and wants to raise the top mark to 98 and compress the bottom to 55 so that the profile 'looks competitive'. A colleague warns that adjusting marks to shape a distribution is not the same as moderating them. The school's assessment policy says marks must reflect achievement against the performance standards, and that any adjustment must be documented and justifiable in terms of evidence.
1. Distinguish moderation from distribution-shaping and explain why the teacher's proposed adjustment is problematic under the stated policy.
Read the worked answer
Moderation is a process that checks whether marks are consistent with the performance standards and with judgments made elsewhere - for example, a second teacher reviewing a sample of work against the descriptors, or a panel comparing evidence across classes. Its purpose is accuracy: it asks 'does this mark correctly represent what the student demonstrated?'
Distribution-shaping starts from the shape of the mark set and adjusts individual marks to produce a desired profile. Its purpose is appearance: it asks 'does this look competitive?' The teacher's proposal - raising the top to 98 and lifting the bottom to 55 - is distribution-shaping. It changes the distance between students (compressing the range) without any new evidence about what any student demonstrated.
The policy problem is twofold. First, validity: if the marks already reflect achievement against the standards, altering them makes them less accurate, not more. A student on 40 who is moved to 55 now has a mark that overstates her achievement, and the external examination will expose the gap. Second, documentation: the policy requires any adjustment to be justifiable in terms of evidence. 'The cohort is strong' and 'it should look competitive' are not evidence about individual performance; they are claims about the group.
The legitimate route is to check the marks against the descriptors - re-read a sample of each student's assessed work, confirm the judgments, and adjust only where the evidence shows an error. If that process leaves the distribution as it is, the distribution stands.
2. The teacher argues that with only 11 students, a single unusually harsh or lenient judgment can distort the whole set. Explain how this concern should be addressed without resorting to distribution-shaping.
Read the worked answer
The concern is legitimate and is precisely what moderation exists to handle. In a small cohort, one mark that is out of step has a large effect on the profile, but the correct response is to test the judgment, not to reshape the set.
Steps that address the concern within policy:
- Re-mark a sample against the descriptors. Take the highest, the lowest, and two middle scripts and re-read each against the performance standards without looking at the original mark. If the re-mark agrees, the judgment is confirmed. If it diverges, the divergence is the finding, and the mark is corrected on the evidence.
- Cross-marking with another teacher. Even one other teacher who knows the standards can independently mark the same sample. Disagreement above an agreed tolerance triggers discussion, not automatic adjustment.
- Check for a systematic bias. If the teacher suspects she has been harsh across the board, the test is whether the same scripts, marked blind by another teacher, come out higher. That is evidence. 'The cohort feels strong' is not.
- Record the outcome. Whatever the moderation finds, document the sample, the judgments, and any changes. This satisfies the policy's documentation requirement and protects the teacher if the marks are later queried.
What the teacher must not do is decide the target profile first and then work backwards. In an 11-student cohort the profile will be whatever the evidence supports, and a small cohort is expected to have an uneven distribution.