Multilevel models, covariates, and controlled factors in experimental speech research: Unified analyses of highly structured data

• DOI: 10.1121/1.4790331
• Publication Date: 2013-03-18T22:42:21Z
• Authors: Noah Silbert; Jared A. Linck; Mark VanDam
• Keywords: 03 medical and health sciences; 0305 other medical science

Experimental speech research often makes use of complex experimental designs, but even when multiple factors are manipulated, measured outcomes may be influenced by non-controlled and incompletely controlled factors. Multilevel models (of which mixed-effect a special case) enable unified analysis the relationships between, on one hand, trial-level data and, other, potentially important variables. Fitted multilevel allow us to draw inferences simultaneously about group-level effects covariates (the typical focus work) as well individual subject item properties (both can in applied research). The utility will illustrated with analyses from number studies. We present phonological structure, gender differences, within-gender variability acoustics spoken English consonants; simultaneous modeling factors, variability, second language proficiency bilingual lexical processing; age, hearing loss, phonological/lexical properties, vocabulary-related early development.