Guidelines for statistical analysis of percentage of syllables stuttered data.

Research paper by Mark M Jones, Mark M Onslow, Ann A Packman, Val V Gebski

Indexed on: 16 Aug '06Published on: 16 Aug '06Published in: Journal of speech, language, and hearing research : JSLHR


The purpose of this study was to develop guidelines for the statistical analysis of percentage of syllables stuttered (%SS) data in stuttering research.Data on %SS from various independent sources were used to develop a statistical model to describe this type of data. On the basis of this model, %SS data were simulated with varying means, standard deviations, and sample sizes. Four methods for analyzing %SS were compared.Results suggested that %SS data can be adequately modeled with a gamma distribution. Simulations based on a gamma distribution showed that all 4 analysis techniques performed favorably with respect to Type I error except for F. E. Satterthwaite's (1946) t test, which had increased Type I error on two occasions. Power was generally lower for the Wilcoxon-Mann-Whitney test compared with the other methods. Analysis of variance (ANOVA) performed on square-root-transformed data performed adequately under all scenarios, but ANOVA performed on nontransformed data and Satterthwaite's t test performed poorly when sample sizes were small or when sample sizes and variances of the groups were markedly different.Standard techniques such as t test and ANOVA are appropriate for most analysis scenarios with %SS data. Two occasions when this is not the case are when sample size is small, with fewer than 20 in each group, or when sample sizes and variances of the groups are markedly different. Under these circumstances, analyses should be based on standard methods, with a suitable transformation performed prior to analysis.