LJHSCE cheat sheet

The exam outline for LJHSCE was not supplied, so this sheet covers only general exam-authoring and verification facts; treat every item as provisional until the official blueprint is confirmed.

A domain weighting is the percentage of scored items drawn from one content area; weights across all domains must sum to 100%, and any sheet section that claims a weight without a source is unverified.

A cut score (passing standard) is set independently of item difficulty statistics; a common defensible method is the modified Angoff procedure, in which subject-matter experts estimate the probability that a minimally competent candidate answers each item correctly.

Item difficulty is reported as a p-value equal to the proportion of candidates answering correctly; values near 1.0 indicate easy items and near 0.0 indicate hard items, with 0.30–0.80 often targeted for scored items.

Item discrimination is commonly reported as a point-biserial correlation between item score and total score; negative values flag items where stronger candidates did worse, a signal to review the key or wording.

Classical test theory expresses observed score as X = T + E, so reliability is the ratio of true-score variance to observed-score variance; reliability rises with item count and with inter-item consistency.

Cronbach's alpha estimates internal consistency from item variances and total variance; alpha is a lower bound on reliability when items are not essentially tau-equivalent, so it can understate true reliability.

Standard error of measurement equals the standard deviation of observed scores times the square root of (1 minus reliability); a 95% confidence band around a score is roughly 1.96 times that standard error.

Criterion-referenced exams interpret scores against a defined performance standard, while norm-referenced exams interpret scores against the performance of a comparison group; mixing the two interpretations in one report is a common error.

A single-best-answer item has one defensible key and three or more distractors; every distractor should be plausible to a candidate who has partial knowledge, and no distractor should be defensible as also correct.

Negative stems and double negatives increase misreading rates; when a negative must appear, capitalize or bold the negating word so it cannot be skimmed past.

Absolute qualifiers such as always, never, and only make a stem or option easier to eliminate on test-wiseness grounds rather than content knowledge; reserve them for cases where the fact is genuinely absolute.

Each item should test one primary point; compound stems joined by and require the candidate to be right about two things, which confounds the difficulty estimate.

Options should be grammatically parallel and similar in length and specificity; a longest-and-most-detailed option is frequently the key by construction, an unintended cue.

Numbers in options should be ordered consistently (ascending or descending) and drawn from plausible error paths, such as unit-conversion slips or off-by-one mistakes.

A distractor analysis reviews the proportion choosing each option; a distractor chosen by almost no one carries no information and should be revised or replaced.

Fairness review screens items for construct-irrelevant content, including references to local institutions, idioms, or cultural knowledge not required by the domain being measured.

Every item needs an independent second review for key accuracy before use; a wrong key is the most damaging defect because it penalizes the best-prepared candidates.

Raw score is the count of correct responses; a scaled score is a linear or nonlinear transformation of the raw score used to equate forms and report on a stable metric.

Equating adjusts for differences in difficulty between test forms so scores are interchangeable; common designs are common-item (anchor) equating and random-groups equating.

The standard error of a difference between two scores is the standard error of measurement times the square root of 2; score differences smaller than that band are not meaningfully interpretable.

A confidence interval around a reported score should accompany any high-stakes result; reporting a point score alone overstates precision.

Percentile rank is the percentage of the norm group scoring at or below a given score; percentile ranks are ordinal, so averaging or subtracting them is not valid.

A z-score is (X minus mean) divided by standard deviation; T-scores set mean 50 and standard deviation 10, and stanines divide the distribution into nine bands with mean 5.

Reliability of a speeded test is not well estimated by internal consistency, because unanswered items at the end reflect time limits rather than knowledge; alternate-form or test-retest designs are preferred.

Inter-rater reliability for constructed responses is often estimated with Cohen's kappa for two raters or a generalizability coefficient when multiple facets (rater, task, occasion) contribute error.

A test security plan covers item exposure, secure transmission, access logging, and a documented response to suspected compromise; exposure of live items is the primary threat to bank value.

Accommodations change the conditions of administration (for example extended time or a separate room) without changing what is measured; a modification changes the construct and must be reported as such.

Standardization requires identical instructions, timing, and materials; any deviation must be logged with the reason so score interpretation can account for it.

Irregularity reports document events such as late starts, power loss, or suspected cheating; they are adjudicated before scores are released, not after.

Chain of custody for secure materials requires signed transfer records at each handoff; an unbroken log is the evidence that a compromise did not occur.

Pretesting embeds unscored items in live forms to gather difficulty and discrimination data; pretest items must be excluded from candidate scores and identified in the answer key.

A retake policy should specify the waiting period and whether the prior attempt is visible to the candidate; unlimited immediate retakes inflate exposure and depress the value of a pass.

Score cancellation for irregularity requires a defined appeal path and written notice; cancellation without documented findings is vulnerable to challenge.

A valid credentialing program documents a job or practice analysis that defines the content domain, and the exam blueprint is derived from that analysis rather than from available items.

A passing standard should be reviewed on a fixed cycle and after any material blueprint change; carrying an old cut score onto a new blueprint is not defensible.

Recertification requirements (continuing education, periodic assessment, or both) exist to keep the credential tied to current practice; the interval and units must be published.

Conflict-of-interest disclosure for item writers and standard-setters is a governance control; subject-matter experts who train candidates for the exam should be screened out of key-writing roles.

Score reporting policy must state what is released (pass/fail versus numeric), to whom, and for how long records are retained.

An item bank inventory tracks live, pretest, and retired items; a bank with too few live items per objective cannot assemble parallel forms.

Transparency statements should describe the blueprint, the passing standard method, and the retake policy, while withholding live item content to protect security.

Version control on the blueprint matters: every scored form should be traceable to the blueprint version in force on its administration date.