Reliability

Imagine stepping on a bathroom scale three times in a row and getting 150, 172, and 141. You'd toss that scale out — not because it's inaccurate, but because it's inconsistent. That's reliability in a nutshell: the consistency of a measurement, and examiners love it because it's the floor every good test must stand on.

Reliability asks whether a test yields stable, repeatable results. Four flavors get tested. Test-retest reliability gives the same test twice and correlates the scores — it captures stability over time (watch for practice effects if the interval is short). Parallel forms (a.k.a. alternate/equivalent forms) uses two different-but-equivalent versions to check equivalence and dodge memory effects. Internal consistency asks whether items hang together within a single administration — estimated by Cronbach's alpha, split-half (splitting the test in two, often odd vs. even, then applying the Spearman-Brown correction), and KR-20 for dichotomous items. Inter-rater reliability checks agreement between scorers — crucial for subjective ratings and observations (Cohen's kappa is the classic index).

Reliability coefficients range from 0 to 1.0; .80+ is a common rule of thumb, and .90+ is expected for high-stakes clinical decisions. Every observed score is true score + error, and the standard error of measurement (SEM) quantifies that error band around a score — it's how you build a confidence interval. The key relationship examiners drill: as reliability rises, SEM shrinks. A perfectly reliable test would have an SEM of zero; a worthless one would have a huge SEM.

Clinical Pearl: Reliability is consistency, not accuracy — and SEM moves opposite to reliability: the more reliable the test, the smaller the SEM, and the tighter your confidence band around a client's score.