Methodological issues in the calculation of mean length of utterance.

Abstract

Mean length of utterance (MLU) is widely used as a diagnostic, monitoring and group matching measure. This study investigated methodological issues regarding the calculation of MLU. The aim was to establish whether different calculation procedures render different MLUs, and whether there is a high correlation between MLU measured in words (MLU-w) and in morphemes (MLU-m). Language samples from 15 Afrikaans-speaking 6-year-olds with and 15 with ot specific language impairment were analyzed. MLU was calculated eight times for each participant, varying sample size (50 or 100 utterances), unit counted (words or morphemes) and calculation method (traditional or alternate). Significant differences in resultant MLUs were due to the calculation method used, rather than sample size or unit counted. A high positive correlation (>0.96) between MLU-w and MLU-m was found. The results imply that researchers and clinicians should clearly state their MLU calculation procedures, otherwise reliable comparisons between MLU scores from different sources cannot be made. The results furthermore imply that, in order to generalize research results and make diagnostic decisions based on MLU, consistent procedures should be used, not only with regard to language sampling, but also to MLU calculation.