Assessment and Quiz Design Template

For a module quiz (formative), 5-10 questions covering the key concepts is sufficient. For a final exam (summative), 30-50 questions provides enough coverage to reliably assess all learning objectives. Use the formula: 3-5 questions per learning objective being tested. Too few questions and a single lucky guess skews the score. Too many and learner fatigue affects validity.

Item discrimination measures how well a single question differentiates between learners who know the material and those who do not. A discrimination index above 0.3 is considered good. Below 0.2 means the question is not useful for assessment. Items where low-performing learners score higher than high-performing learners should be revised or removed. Run this analysis after your first 50-100 learners complete the assessment.

Start with a standard like 70% for formative assessments and 75-80% for summative ones. After your first cohort, review the score distribution. If fewer than 40% of learners pass, either the content preparation is insufficient or the assessment is too difficult. If more than 95% pass, the assessment may be too easy to differentiate skill levels. Adjust based on data, not assumptions. The [NPS Calculator](/tools/nps-calculator) can help you measure learner satisfaction alongside pass rates.

For formative assessments (practice quizzes), yes. Unlimited retakes with new question variants encourage mastery learning. For summative assessments (exams, certifications), allow 1-2 retakes with a waiting period (e.g., 48 hours) to prevent brute-force memorization. For high-stakes certifications, a single retake after a mandatory study period is standard practice.

Match your integrity measures to the stakes. For low-stakes quizzes, question and answer randomization is sufficient. For medium-stakes exams, add time limits and draw questions from a larger pool. For high-stakes certifications, consider proctoring software. The most effective anti-cheating measure is designing questions that test application and analysis rather than recall. It is hard to cheat on "analyze this dataset and recommend a strategy." ---