Statistical Analyses

John Kingston, Harald Baayen and Cynthia G. Clopper

in The Oxford Handbook of Laboratory Phonology

Published in print December 2011 | ISBN: 9780199575039
Published online September 2012 | | DOI:

Series: Oxford Handbooks in Linguistics

Statistical Analyses


This article provides an overview on the statistical techniques that are appropriate for speech research. A common method that is employed when a categorical predictor has more than two values involves determining its significance overall by means of an analysis of variance, and then to run post-hoc tests comparing pairs of predictor values. Another method is to recode the original predictor with a set of contrasts that embody the required comparisons. Treatment recoding is identical to simply comparing each non-default treatment to the default treatment. Mixed-effects models provide the researcher with a more sophisticated tool for analyzing repeated measures data that is more flexible, more powerful, and more insightful. Fixed-effect factors are coded numerically using dummy coding, such that a factor with n levels contributes n-1 predictors to the model. Treatment coding is most efficient and easy to interpret, especially in the case of analysis of covariance. One level of the factor is selected as the default or reference level. The adjustments for any given random-effect factors are implemented to allow precise predictions for the individual units sampled, such as the individual speakers in an experiment or corpus. These adjustments are assumed to follow a normal distribution with mean zero and some unknown standard deviation.

Keywords: speech research; treatment coding; collinearity; Finnish vowels; cross-validation

Article.  16799 words. 

Subjects: Linguistics ; Phonetics and Phonology

Full text: subscription required

How to subscribeRecommend to my Librarian

Buy this work at Oxford University Press »