Hi,
I am comparing a protein level in multiple (40 to 4000 depending on the analysis) brain areas across three diagnostic groups, with a couple of covariates. I am using ANCOVA to compare the three groups.
What is the best way to account for multiple comparisons?
Where I am conducting large numbers of ANOVAS (>50), FDR appears best. I have been running this on the LSD post hoc P values from ANOVA?
But for smaller numbers of comparisons, a colleague has told me that MANCOVA is ok. For some of my analyses (in SPSS), the Wilks' Lambda for multivariate analysis is <0.05, but it does not seem to modify the post-hoc P values, so all of those that were significant if I run them as separate ancovas remain significant. This doesn't seem right. Is it OK to use MANCOVA for this? Does the Wilks' of <0.05 mean that all the subsequent <0.05 post-hocs are significant?
Alternatively, I have just run a bonferroni correction on the P values of my various ANCOVAs; it is very (expectedly) conservative. Not as 'good' as the MANCOVA.
Any help would be really appreciated,
Thank you,
Josh
I am comparing a protein level in multiple (40 to 4000 depending on the analysis) brain areas across three diagnostic groups, with a couple of covariates. I am using ANCOVA to compare the three groups.
What is the best way to account for multiple comparisons?
Where I am conducting large numbers of ANOVAS (>50), FDR appears best. I have been running this on the LSD post hoc P values from ANOVA?
But for smaller numbers of comparisons, a colleague has told me that MANCOVA is ok. For some of my analyses (in SPSS), the Wilks' Lambda for multivariate analysis is <0.05, but it does not seem to modify the post-hoc P values, so all of those that were significant if I run them as separate ancovas remain significant. This doesn't seem right. Is it OK to use MANCOVA for this? Does the Wilks' of <0.05 mean that all the subsequent <0.05 post-hocs are significant?
Alternatively, I have just run a bonferroni correction on the P values of my various ANCOVAs; it is very (expectedly) conservative. Not as 'good' as the MANCOVA.
Any help would be really appreciated,
Thank you,
Josh