Clustering of Human Sleep Recordings Using a Quantile Representation of Stage Bout Durations - Biomedical Engineering Systems and Technologies - page 353

Fig. 5. Mean values of clustering variables, k =2
Fig. 6. Mean values of clustering variables, k =3
Fig. 7. Mean values of clustering variables, k =4
Multiway Cluster Comparisons. Statistical significance of the observed differences
in the means among clusters was assessed by ANOVA and Kruskal-Wallis tests for mul-
tiway comparisons, and by t , and Wilcoxon rank sum tests for pairwise comparisons. A
nonparametric two-sample Kolmogorov-Smirnov test was also used to determine differ-
ences between pairs of clusters in the overall distributions of the bout duration quartile
variables. All p -values were corrected for multiple comparisons using the Benjamini-
Hochberg method on a per clustering basis, so that reported p -values are upper bounds
on the false discovery rate relative to all findings over the given family of k clusters.
Stage Bout Duration Quartile Means. The results are as follows. For all clustering
families, k =2 , 3 , 4 , the mean values of the wake stage and stage NREM1 duration
quartile variables differ significantly among clusters in a multiway comparison using the
Kruskal-Wallis test ( p< 0 . 05 ). ANOVA results are in agreement with Kruskal-Wallis,
with the exception that the inter-cluster difference in the quartile variable REM.Q1 is