EGATSU-PI cheat sheet

EGATSU-PI - Unverified Content Outline: Blueprint Verification and Score Interpretation

• Blueprint weight check: expected item count per area = total scored items × that area's stated weight; the weights must sum to 1.00 (100%) before any published outline can be treated as complete or authoritative.

• Standard error of measurement: SEM = SD × √(1 − r). A 95% confidence band around an observed score is approximately observed ± 1.96 × SEM, and the band narrows as reliability r increases.

• Classical item difficulty is p = (number correct)/(number attempting), ranging 0 to 1. For a multiple-choice item with k options, discrimination is maximized near p = (1 + 1/k)/2, i.e. about 0.625 when k = 4.

• Corrected-for-guessing formula score = R − W/(k − 1), where R = right, W = wrong, and k = number of response options; omitted items are neither credited nor penalized.

• Standard-score transforms: z = (X − mean)/SD and T = 10z + 50. Both are linear, so they change the scale but never the rank order of candidates.

• Internal consistency for right/wrong items: KR-20 = [k/(k − 1)] × [1 − (Σpq)/σ²total]; with dichotomous scoring KR-20 equals Cronbach's alpha computed on the same data.

• Rasch (one-parameter IRT) model: ln[p/(1 − p)] = θ − b, where θ is person ability and b is item difficulty, both expressed on a shared logit scale; the model assumes equal item discrimination.

• Angoff cut score = Σ p̂ᵢ over all items, where p̂ᵢ is the judged probability that a minimally competent candidate answers item i correctly; the resulting raw cut is then placed on the reporting scale by equating.

• Percentile ranks are ordinal: they may not be averaged, added, or subtracted. Convert to z-scores first when any arithmetic combination of candidate results is required.