在R中构造重复测量方差分析并提取ls均值。

Question

I am trying to properly construct a repeated measures ANOVA in R and extract the associated lsmeans. My data consists of a dependent variable (rSWC) and a predictor (Geno). The full dataset is as below:

> str(mydata)
'data.frame':   153 obs. of  5 variables:
 $ Geno          : Factor w/ 5 levels "8306","8307",.. 
 $ BioRepeat     : int  1 1 1 1 1 1 1 1 1 2 ...
 $ Geno_BioRepeat: Factor w/ 17 levels "8306_1","8306_2",..
 $ Day           : Factor w/ 9 levels "1","2","3","4",.. 
 $ rSWC          : num  104.5 92.5 81.8 65.6 61 ...

I am constructing my repeated measures anova as:

rmaModel <- aov(rSWC ~ Geno + Error(Day/Geno), data=mydata)

I wish to extract the lsmeans (and associated variance terms) for Geno for every repeated measure (Day). At the moment if I try to extract the lsmean, I just get one lsmean for each Geno and a warning message that I cannot interpret:

> library(lsmeans)
> lsmeans(rmaModel, specs = "Geno")
 Geno      lsmean        SE    df lower.CL  upper.CL
 8306    59.43538  8.905658  8.00 38.89890  79.97187
 8307    58.06825  9.988820 12.45 35.03399  81.10251
 8417    71.16686 10.158125 13.24 47.74219  94.59154
 Control 86.97797 10.488538 14.84 62.79136 111.16459
 WT      45.76538  9.988820 12.45 22.73112  68.79964

Confidence level used: 0.95 
Warning message:
In lsm.basis.aovlist(object, trms, xlev, grid, ...) :
  Some predictors are correlated with the intercept - results are biased.
 May help to re-fit with different contrasts, e.g. 'contr.sum' 

Any help to understand whether my model is constructed appropriately, how to extract the lsmean for each repeated measure, and how to interpret the warning message would be very much appreciated. Thanks!

Answer 1

The default contrasts for statistical tests in R regression functions are treatment contrasts, so the "Intercept" level is generally the first of each of the factor variables. The warning message is suggesting you might redefine the contrasts for your Geno variable. So that might be done with:

 ?contrasts  # to get the background theory and example code
 contrasts(mydata$Geno) <- contr.sum(5) 
 rmaModel <- aov(rSWC ~ Geno + Error(Day/Geno), data=mydata) # refit
 lsmeans(rmaModel, specs = "Geno")