Retesting in selection: A meta-analysis of coaching and practice effects for tests of cognitive ability.

50 studies, 107 samples, N = 134,436

Cognitive-ability tests in educational and employment selection; mean sample age 19.3, and the sample is not adult-only: 28% K-8, 26% high school, 46% college

By far the largest synthesis of the retest question, and the age breakdown is what makes it usable here - more than half the samples are school-age. Graded C as a meta-analysis of mixed designs, with the additional caveat that reliability had to be imputed for 67 samples. Most of the underlying contexts are selection settings rather than classroom achievement testing.

The complete decomposition of a retest gain, on 134,436 people. Simply taking the test again raises the score by an adjusted d = 0.26 first-to-second administration, with further cumulative gain by the third. Adding COACHING between administrations raises it to delta = 0.70 against about 0.24 without - so coaching roughly triples the retest gain. And the form matters exactly as it should if the gain is partly test-specific: identical form delta = 0.46 (k = 38) versus alternate form delta = 0.24 (k = 64). Coaching contact time was a significant positive predictor in the weighted regression, which independently supports the dose-response pattern that Messick & Jungeblut proposed and the Kulik group could not replicate. The authors themselves flag that retest gains may not carry validity - the question Lievens et al. answered directly in the same year.

Low for the retest effect itself (the same people are compared to themselves), medium for the coaching contrast, since who obtains coaching between sittings is not random.

Reproduces the Kulik group's 1984 identical-versus-parallel form gap on a sample three orders of magnitude larger, twenty-three years later, in different settings.