问题
This works:
testmodel=glm(breaks~wool,data=warpbreaks)
emmeans::emmeans(testmodel,"wool")
This works:
warpbreaks %>%
group_by(tension) %>%
do(models=glm(breaks~wool,data=.)) %>%
ungroup() %>%
mutate(means=map(models,~emmeans::emmeans(.x,"wool")))
This doesn't:
warpbreaks %>%
group_by(tension) %>% nest() %>%
mutate(models=map(data,~glm(breaks~wool,data=.x))) %>%
mutate(means=map(models,~emmeans::emmeans(.x,"wool")))
Error in is.data.frame(data) : object '.x' not found
Error in mutate_impl(.data, dots) :
Evaluation error: Perhaps a 'data' or 'params' argument is needed.
Any idea what's causing this?
回答1:
I figured it out. The issue is the way emmeans tries to recover data from lm/glm objects: it tries to run the call stored in the object, which fails if emmeans() is called in a different environment than the original glm() call:
emmeans:::recover_data.lm
Here's an easy example:
wb=warpbreaks
model=glm(breaks~wool,data=wb)
emmeans(model,"wool")
rm(wb)
emmeans(model,"wool")
Here's the way to make emmeans() work with map():
warpbreaks %>%
group_by(tension) %>% nest() %>%
mutate(models=map(data,~glm(breaks~wool,data=.x))) %>%
mutate(means=map(models,~emmeans::emmeans(.x,"wool",data=.x$data)))
It seems strange that recover_data() doesn't just automatically use the data attribute of the lm/glm objects and instead assumes that the call will function in the current environment...
回答2:
We can do it in a two-step process
df1 <- warpbreaks %>%
group_by(tension) %>%
nest() %>%
mutate(models = map(data,~glm(breaks~wool,data=.x)))
warpbreaks %>%
split(.$tension) %>%
map( ~glm(breaks ~ wool, data = .x) %>%
emmeans(., "wool")) %>%
mutate(df1, Means = .)
# A tibble: 3 x 4
# tension data models Means
# <fctr> <list> <list> <list>
#1 L <tibble [18 x 2]> <S3: glm> <S4: emmGrid>
#2 M <tibble [18 x 2]> <S3: glm> <S4: emmGrid>
#3 H <tibble [18 x 2]> <S3: glm> <S4: emmGrid>
来源:https://stackoverflow.com/questions/48576160/cant-use-emmeans-inside-map