Table 2 

Analysis of variance of variance 

Source of variation 
ave.kME.green, total proportion of variance explained = 15.8% 
logPvalue Stouffer, total proportion of variance explained = 6.7% 




Source 
Degrees of freedom 
Sums of Sq 
Proportion of total variance 
F statistic 
Pvalue (Ftest) 
Sums of Sq 
Proportion of total variance 
Pvalue (Ftest) 


PCG (Suz12+Eed+H3K27me3) OccupancyCount 
1 
49.35 
0.071 
2013.0 
< 2.2E16 
82,530 
0.050 
< 2.2E16 
CPG_Island 
2 
50.78 
0.073 
1035.7 
< 2.2E16 
24,867 
0.015 
< 2.2E16 
X chromosome 
1 
9.74 
0.014 
397.4 
< 2.2E16 
764 
0.000 
5.5E04 
Distance to TSS 
1 
3.50E01 
0.001 
14.1 
1.7E04 
1,827 
0.001 
9.3E08 
Residual error 
23,903 
586.02 
0.842 
1,529,855 
0.933 

Total variation 
696.24 
1,639,843 



The first six columns detail the sources of variation on ave.kME with respect to the green (aging) module. Polycomb group occupancy count explains the highest proportion of the total variance followed by CpG island status (inside, outside, and shore). Comparing the results from ANOVA applied to average kME with those from applying it to the logPvalue (significance measure resulting from Stouffer's metaanalysis) reveals marked differences in the proportion of variance explained (15.8% versus 6.7%), which suggests again that the average kME measure is biologically more meaningful than the metaanalysis Pvalue. TSS, transcription start site. 

Horvath et al. Genome Biology 2012 13:R97 doi:10.1186/gb20121310r97 