The Assessment performance report shows the psychometric performance of every assessment on the platform, with the ability to drill into a single assessment, then into individual items. Use this article to find the report, read the headline statistics, and understand when Cirrus recalculates them.
Open the report
Open Reports > Assessment performance.
The list lets you:
- Search by assessment title.
- See a headline overview: last sat, pass rate, reliability, times delivered, number of items.
- Drill into any assessment for detail.
- Export the data.
Per-assessment detail
Select an assessment title to drill in.
The detail view shows:
- Average chance score for the assessment.
- Average RIR value.
- Average RIT value.
- Reliability for the assessment.
- Average Pc value (corrected P value).
- Average P value (difficulty).
See Statistics: RIT and RIR values for the discrimination metrics, and Item statistics for P and Pc values.
Reliability (Cronbach's alpha)
Cirrus uses Cronbach's alpha for the reliability number:
Reliability = (N / (N − 1)) × (1 − Σ(Si²) / Sx²)
Where:
| Symbol | Meaning |
|---|---|
| N | Number of items |
| Σ Si² | Sum of the item variances |
| Sx² | Assessment variance = Σ(score − average score)² / number of candidates |
See Cronbach's alpha on Wikipedia for the wider context.
Per-item view
Select the assessment title a second time (or follow the item link) to see the items in the assessment and their per-item statistics in the context of this assessment.
The per-item view is in beta. Feedback is welcome through the Customer Portal.
Where psychometric data appears
Statistics show up in several places:
- Library > Collection > Statistics tab for per-item statistics in the library. See Item statistics.
- Assessment dashboard for assessment-level statistics.
- Reports > Assessment performance for the report described in this article.
When statistics recalculate
Statistics are calculated when:
- The schedule has ended, and
- Every candidate's results have been submitted (including auto-submitted "not handed in" results).
Scenario 1: auto-scored exam
- The assessment uses Do not assign assessors at publish time.
- Schedule: 4 April to 14 April, 50 candidates.
Cirrus calculates statistics on 14 April when the schedule ends, or sooner if every candidate has submitted. If one candidate did not start, Cirrus waits for the schedule to end so it can auto-submit the missing attempt.
Removing the candidate from the schedule does not trigger a new calculation.
If every candidate has submitted and statistics have been calculated, adding a new candidate later within the window triggers a recalculation as soon as that candidate's attempt is in.
Scenario 2: manually marked exam
- The assessment uses Simple or Extended marking workflow.
- Schedule: 4 April to 14 April, 50 candidates, multiple assessors in the pool.
Cirrus calculates statistics once every assessor has submitted their scores for every candidate. The schedule's end date does not trigger it on its own; marking submission does.
Rescoring
When rescoring runs after marking is complete, statistics are recalculated as part of the rescore.
When rescoring runs during marking (before publication), statistics are not recalculated yet; they wait for marking submission and then reflect the rescored values.
Per-version statistics
The version history of each assessment also tracks per-version pass rate and candidate count. Use the version history (three-dot menu in the top right of the assessment) to compare versions side by side. See Assessment versioning.