Forcing nlme to give the results for the levels of a treatment as "absolute" values instead of contrasts I am using a nonlinear model to fit an equation to data using nlme in R.
Several values (one by treatment) are estimated for a fixed effect. 
As usual, by default of nlme, the estimated values of this fixed effect
are given relative to the smallest value as contrasts. 
I would like, instead, to have the results as "absolute" values of the 
fixed effect for each treatment. 
My question is the following: how  can I specify to nlme  that contrasts
should not be used to give the estimates of the  values of this fixed effect by  treatment.
 A: This is just about how categorical variables are coded in R and how the forumlas work.  It doesn't really have anything to do with whether the model is nonlinear or if you use a mixed model.  The easiest way to get your model returned fitted with level means coding instead of reference level coding is to suppress the intercept when you have a factor variable.  In R, this can be done with either +0 or -1.  Consider:  
set.seed(123)                 # this makes the example exactly reproducible
x = runif(30, min=0, max=10)  # x is a continuous variable
g = as.factor(sample(rep(c("C","T"), 15), 30, replace=F))  # g is categorical
y = 3 + .3*x + 2*ifelse(g=="T",1,0) + rnorm(30)            

## this is the typical way, which uses reference level coding:  
mod.refL = lm(y~x+g)
summary(mod.refL)
# ...
# Coefficients:
#             Estimate Std. Error t value Pr(>|t|)    
# (Intercept)  3.18841    0.41014   7.774 2.33e-08 ***
# x            0.28444    0.05557   5.118 2.22e-05 ***
# gT           2.15804    0.31838   6.778 2.80e-07 ***
# ...

## here I used '0' to suppress the intercept:  
mod.Lmeans = lm(y~0+x+g)
summary(mod.Lmeans)
# ...
# Coefficients:
#    Estimate Std. Error t value Pr(>|t|)    
# x   0.28444    0.05557   5.118 2.22e-05 ***
# gC  3.18841    0.41014   7.774 2.33e-08 ***
# gT  5.34644    0.36615  14.602 2.46e-14 ***
# ...

A: What you really want are predictions from the model at each treatment level. If it is just a one-factor experiment, with one covariate, do something like this:
grid <- data.frame(treat = c("t1", "t2", "t3"), covariate = c(rep(5.5,3))
predict(my.model, newdata = grid)

Of course, you should use the actual variable names and treatment levels you used. Typically, people would want to make the predictions at the mean value of each covariate.
[Earlier stuff with gung's example removed because it is linear. Also, note that his results with the second approach are predictions at x = 0, not x equal to its mean.].
