Hi all,
Im doing gene expression analysis using a ANOVA. I analyze n genes between different groups (2 controls, 2-4 treatments between 3 different days). I've read that I have to adjust my alpha according to 0,05/n. In SPSS I've run an ANOVA with a bonferroni post hoc test afterwards. Is it correct that I still have to compare the p-values to 0,05/n or is it somehow implemented in the software?
Do I need an alpha correction if I analyze n genes but only two groups (control vs treatment)?
Best regards and thanks
See the attached that I find very helpful. Best, D. BOOTH for work in gene expression the two most commonly used are Bonferroni and B-HHochberg both of which are in here. Best wishes, D. Booth
SPSS corrects the alpha values for you if you select the correct option rather than adjusting p.
The correction is traditionally done with a family of tests. The family of tests is somewhat ill-defined but often taken to be a set of tests where if any of the tests was statistically significant you would decide there is evidence of an effect that supports your research question.
It is also common just to correct for pairwise comparisons within a factor (in which case there is no correction for two means and may not be a correction required for 3 means - depending on whether the comparisons are conditional on a significant omnibus F test). My own view is that you probably want to correct for all the comparisons between conditions and separately for all the comparisons between conditions on particular days as families - which is maybe somewhat conservative).
A better approach is to specific a small set of a priori contrasts in advance to detect patterns of interest that you predicted. This has greater power if your predictions are correct and wouldn't prevent post hoc tests with more stringent correction.
As David says you'd want to use a Hochberg correction or similar rather than than Bonferroni as the latter is increasingly conservative (losing power) as the number of tests increases. These aren't built into SPSS but are relatively easily computed from p values for the set. I always use R for this as it is very easy to do just from a set of p values:
e.g.,
pvals
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA