inations of sociodemographic variables were identified and students were then randomly selected as fitting given categories, e.g. North Central, male, fourth grade, black, parent a college graduate, or North Central, male, fourth grade, white, parent a college graduate). Students selected by this process constituted the normative sample.
The PIAT-R manual (Markwardt, 1989) reports four tests of reliability defined as the consistency of scores across repeated testings of a subject or a similar test under comparable conditions. The first reported reliability estimates are split-half reliabilities calculated to determine the consistency of a subject's performance on content in a single administration. For all grade levels, total test estimates varied from .97 to .98 indicating extremely high reliability. Indeed, the lowest estimates (kindergarten level in math and spelling) were in the .80s.
Kuder-Richardson reliability estimates were also high with only a few exceptions. This was said to indicate a strong degree of content homogeneity. Further, test-retest reliabilities (computed to determine the consistency of scores from one test period to another) ranged from the low to upper .90s---values that were said to be much higher than in most similar tests.
Finally, reliability was estimated using the item-response theory model which utilizes a different mathematical assumption relative to true and error variance of scores. These coefficients were also very high, once again reflecting strong consistency of scores.
Regarding the samples used, split-half reliabilities, Kuder-Richardson coefficients, and item response reliabilities were calculated using the entire normative sample, cross-tabulated by both grade and age level. However, test-retest reliabilities were computed using 50 randomly selected subjects from the normative sample at each of the following grade levels: Kindergarten, Grade 2, Grade 4, Grade 6, Grade 8, ...