Sample size effects on temporal reliability of language sample measures of preschool children

The present study examined the temporal reliability of four quantitative measurements of operationally defined linguistic behaviors observed in a naturalistic setting. The measures of language production were computed using the Systematic Analysis of Language Transcripts (SALT) software from two 20-minute language samples obtained 3 to 14 days apart for each of 20 preschool-aged children. Samples were edited to different sizes based either on duration (12 or 20 minutes) or on limits of the total number of complete and intelligible utterances (from 25 to 175 in increments of 25). Inadequate reliability was found for the language sample measure, total number of words; hence, the validity of this measure is questionable. In contrast, very high temporal reliability coefficients (r > .92) were obtained for the language sample measures of number of different words, mean length of utterance in morphemes, and mean sentence length in morphemes when derived from a large number (> or = 175) of complete and intelligible utterances. The temporal reliability of these measures reached acceptable levels, not only for research purposes, but for diagnostic purposes as well.