Dear List,
Is there a way to specify more than one error term in the formula for
a mixed/random effects ANOVA? For example, given this data and that
both 'order' and 'admin' are random factors, my understanding based on
the construction of expected mean squares is that the 'order:admin'
interaction is the appropriate error term for the main effects of
'order' and 'admin', but that the usual residuals are the appropriate
error term for the 'order:admin' interaction. Although it is possible
to do this using two separate models, I am hoping they can be combined
somehow.
Thanks for your time,
Josh
dat <- structure(list(nresponse = c(26, 15, 26, 24, 33, 27, 36, 28,
30, 33, 25, 33, 26, 32, 37, 42, 25, 23, 27, 17, 30, 24, 37, 33,
28, 30, 27, 26, 31, 26, 39, 25), admin = structure(c(1L, 1L,
1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 3L, 3L,
3L, 3L, 3L, 3L, 3L, 3L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L), .Label = c("1",
"2", "3", "4"), class = "factor"), order = structure(c(1L, 1L,
2L, 2L, 3L, 3L, 4L, 4L, 1L, 1L, 2L, 2L, 3L, 3L, 4L, 4L, 1L, 1L,
2L, 2L, 3L, 3L, 4L, 4L, 1L, 1L, 2L, 2L, 3L, 3L, 4L, 4L), .Label = c("I",
"II", "III", "IV"), class = "factor")), .Names = c("nresponse",
"admin", "order"), row.names = c(NA, -32L), class = "data.frame")
summary(aov(nresponse ~ order + admin + Error(order:admin), dat))
summary(aov(nresponse ~ order * admin, dat))