How to narrow down to specific contrasts with lsmeans

Question

I would like to use contrast() with an lsmeans object to perform some specific planned comparisons, but I cannot find a method to perform the comparisons I want. I want to compare whether one factor has an effect at either of two factor levels. For example, in the below I would like to compare whether A,C differs to B,C, and whether A,D is different to B,D. I do not want to compare whether A, C is different to B, D or different to A, D, etc.

fac_one <- c(rep("A", 200), rep ("B", 200))
fac_two <- rep(c("C", "D"), 200)

dats <- data.frame(fac_one= c(rep("A", 200), rep ("B", 200)),
                   fac_two= rep(c("C", "D"), 200))

dats$y <- NA

dats$y[dats$fac_one=="A" & dats$fac_two=="C"] <- 
  rnorm(100, mean=0.9, sd=1)
dats$y[dats$fac_one=="B" & dats$fac_two=="C"] <- 
  rnorm(100, mean=0.9, sd=1)

dats$y[dats$fac_one=="A" & dats$fac_two=="D"] <- 
  rnorm(100, mean=0.6, sd=1)
dats$y[dats$fac_one=="B" & dats$fac_two=="D"] <- 
  rnorm(100, mean=1.4, sd=1)

mod <- lm(y ~ fac_one*fac_two, data = dats)
Anova(mod)

lsmns <- lsmeans(mod, ~fac_one*fac_two)
#currently does many contrasts that I do not want to do
contrast(lsmns)

Thanks!

Russ Lenth · Accepted Answer · 2019-10-04T20:51:26.403

2

contrast(lsmns, “pairwise”, by = “fac_two”)

It’s in the documentation

edited Oct 04 '19 at 20:51

answered Oct 04 '19 at 20:42

Russ Lenth

5,922
2
13
21

`fac_two` needs to be in quotes, I believe. – Aaron left Stack Overflow Oct 04 '19 at 20:50
Wait a minute, though, you're not just any @rvl, you're the package author, yes? It's a great package, I love it, thanks so much! – Aaron left Stack Overflow Oct 04 '19 at 20:51
Yup, I am. Glad you like the package. – Russ Lenth Oct 04 '19 at 20:53

score 1 · Answer 2 · answered Oct 04 '19 at 20:49

You can use | instead of * to tell emmeans to separate them into groups based on that factor. You can then use pairs to get pairwise comparisons within each group. Also note that both the package and the function have been renamed in the latest versions to emmeans from lsmeans, though this functionality is in the older versions as well.

> library(emmeans)
> lsmns <- emmeans(mod, ~fac_one|fac_two)
> pairs(lsmns)
fac_two = C:
 contrast estimate    SE  df t.ratio p.value
 A - B      -0.044 0.142 396 -0.309  0.7574 

fac_two = D:
 contrast estimate    SE  df t.ratio p.value
 A - B      -0.974 0.142 396 -6.848  <.0001

PS. Nice reproducible example. :)

How to narrow down to specific contrasts with lsmeans

2 Answers2