Skip to content

summary.lm for post-hoc tests

2 messages · Alex Bokov, Dimitri Liakhovitski

Original
Alex Bokov
# 
Let's say I've run Anova(lm(y~a*b)) and found the a:b interaction to be 
significant. Now I'm interested in which specific level combinations of 
a and b significantly differ from the control group. Can I use the 
t-tests from summary(lm(y~a*b)) to answer that question?

I saw no mention of multiple comparison in the documentation for 
summary.lm, so am I right in assuming I need to run p.adjust on the 
p-values column?

Thanks.

PS: I am aware of TukeyHSD and that works fine if a and b are both 
factors. However, if 'a' is numeric and I'm interested in which levels 
of 'b' have a different slope from the control group, TukeyHSD cannot do 
that, and I'm hoping the above can.
Dimitri Liakhovitski
# 
Check this out:

http://www.rforge.net/doc/packages/NCStats/compSlopes.html

However, it looks like one need to install NCStats
(http://www.rforge.net/NCStats/files/).
Or is there a more "mainstream" method of running multiples
comparisons among slopes?

Dimitri
On Sat, Oct 30, 2010 at 1:20 PM, Alex Bokov <bokov at uthscsa.edu> wrote: