The t test and beyond: Recommendations for testing the central tendencies of two independent samples in research on speech, language and hearing pathology

Abstract

The main messages are (a) that researchers should present more relevant features of their data (means, medians, SD, skewness, tailedness, outliers etc.), (b) not routinely use conventional non-parametric tests like Wilcoxon-Mann-Whitney test in case one or more of the assumptions of t tests are not met, and (c) should consider using less conventional, but robust statistics which have been developed and tested in the last decades.

Shapes and timing in charismatic speech - Evidence from sounds and melodies

“…We used the z-score proportion tests for counts and the Welch t-test for acoustic data. Compared to the regular Student's t-tests for independent samples, the Welch t-test performs better for heterogeneous variances and for unequal samples sizes [34], both of which applied to parts of our data.…”

Treatment intensity and childhood apraxia of speech

“…Several statistical assumptions must be met before t ‐tests can be applied (Rietveld and Van Hout , Wilcox ). These include assumption of normality of the difference scores, assumption of equal variances between the groups (Pitman's test; Pitman ), non‐significant skewness and kurtosis of difference scores and assumption of non‐significant interaction between participants and treatment (Tukey's Test of Additivity).…”

A Role for Maternal Factors in Suppressing Cytoplasmic Incompatibility

“…The data were analyzed for normality using the Shapiro–Wilk test and for homogeneity of variance by Levene’s test ( Shapiro and Wilk, 1965 ; Lim and Loh, 1996 ; Mohd Razali and Bee, 2011 ). For the data with normal distribution, mean differences were evaluated using a t -test if variance was homogeneous, and Welch’s t -test if it was not ( McDonald, 2014 ; Rietveld and van Hout, 2015 ). If the data did not fit the normal distribution, the Mann–Whitney U- test was performed for data showing homogeneity of variance, and an independent t -test was performed with bootstrapping as an approximation when variance was uneven ( LaFlair et al, 2015 ; Rietveld and van Hout, 2015 ).…”