Dear R users,
Does anyone know of a way to obtain approximate 95% confidence intervals
for predicted values for factor levels of fixed effects from lme? Our
goal is to use these intervals to interpret patterns across our
predicted values for certain factor levels.
Our mixed model has the following form with 7 levels of mtDNA, 2 levels
of autosome, 2 levels of brood and 2 levels of block,
> lme(fitness ~ mtDNA*autosome + brood, random = ~1 | block)
We have used the predict.lme function to obtain predicted values, but
are unsure how to obtain appropriate standard errors on these predicted
values.
Using predict.lme to predict "fitness" across a subset of our factor
levels (2 mtDNA, 2 autosome) generates the following output,
autosome mtDNA brood block predict.fixed predict.block
1 ore ore A A 0.4977047 0.5016255
2 ore simw501 A A 0.4278287 0.4317495
3 ore ore B A 0.5042857 0.5082065
4 ore simw501 B A 0.4344098 0.4383306
5 ore ore A B 0.4977047 0.4937839
6 ore simw501 A B 0.4278287 0.4239079
7 ore ore B B 0.5042857 0.5003649
8 ore simw501 B B 0.4344098 0.4304890
9 aut ore A A 0.5321071 0.5360279
10 aut simw501 A A 0.4866497 0.4905705
11 aut ore B A 0.5386882 0.5426090
12 aut simw501 B A 0.4932308 0.4971516
13 aut ore A B 0.5321071 0.5281863
14 aut simw501 A B 0.4866497 0.4827289
15 aut ore B B 0.5386882 0.5347674
16 aut simw501 B B 0.4932308 0.4893099
We would like to calculate, for example, the appropriate 95% confidence
intervals for the predicted values of autosome=ore + mtDNA=ore,
autosome=ore + mtDNA=simw501, etc.
Sincerely,
Kristi Montooth and Colin Meiklejohn
Ecology and Evolutionary Biology
Brown University