Educational Measurement, Evaluation and Assessment
This domain rewards mastery of technical definitions and a small number of formulas; most marks are lost through careless use of vocabulary.
Distinctions to master
Measurement is quantitative description; assessment is the gathering and interpreting of evidence; evaluation is judging worth against criteria. Types by purpose: placement, diagnostic, formative, summative. Types by reference: norm-referenced (compares students) versus criterion-referenced (compares to a standard). Test formats: objective (supply, selection) versus subjective/essay; performance and portfolio assessment; continuous or school-based assessment.
Technical qualities
Validity (content, construct, criterion-related, face) asks whether the test measures what it claims. Reliability (stability, equivalence, internal consistency via split-half with Spearman-Brown correction, KR-20, Cronbach's alpha, inter-rater agreement) asks whether scores are consistent. Also know objectivity, usability, practicability and the standard error of measurement. A test can be reliable but invalid; it cannot be valid without being reliable.
Item analysis
Difficulty index p = number correct ÷ number tested (desirable mid-range, roughly 0.3-0.7). Discrimination index D = p(upper group) - p(lower group), with the upper and lower 27 per cent used as groups. Distractor analysis checks whether wrong options attract anyone. Use a table of specification to balance content and cognitive levels.
Scoring and reporting
Standard scores: z = (X - mean) ÷ standard deviation; T = 50 + 10z; stanines; percentile ranks. Nepal's school-leaving examinations use letter grading and GPA, with grade-point bands and non-graded continuous assessment contributing.
Traps
- Calling a low-reliability test invalid; these are separate faults.
- Ignoring the sign of D - a negative discrimination index signals a defective item.
- Confusing percentile rank with percentage score.
Worked example
Upper 27 per cent: 18 of 27 correct, p = 0.667. Lower 27 per cent: 9 of 27, p = 0.333. D = 0.667 - 0.333 = 0.334, a good discriminating item. If instead 20 of 27 lower-group pupils were correct and 10 of the upper, D = -0.37: the item is inverted or miskeyed and must be revised.