I'm trying to predict the impact of avg grasshopper density(ghavg) on plant biomass. Both the response and predictor are continuous data. I had to log transform biomass (logmass) for a normal distribution, giving the following dataset:
logmass = c(8.032393925,7.439531107,7.307924891,7.036315375,6.679316231,6.545784839,6.414481385,6.39297518,6.209312602,6.209312602,5.698486978,4.862609605,4.367692388,3.608137836)
ghavg = c(30.4,30.4,7.7,124.8,7.7,7.7,123.2,30.4,21.1,21.1,21.1,123.2,47.9,124.8)
In R I ran a glm:
biomass<-glm(logmass~ghavg,data.txt).
To predict the effect of grasshoppers on plant biomass (up to 300 m2) I ran:
nd<-data.frame(ghavg=0:300)
pred_mass<-predict.glm(biomass,type="response",newdata=nd)
Then back transformed the output with:
trans_pred<-(exp(1)^(pred_mass))
I think this works well but I would now like to add confidence intervals to the prediciton. I tried:
pred_massSE<-predict.glm(biomass,type="response",se.fit=TRUE,newdata=nd)
but the standard errors are clearly off when back transformed.
Any assistance on how to properly add CI's to predicted values from transformed data would be greatly appreciated
Best Answer
@ogrisel: bootstrap seems overkill here! rather:
should work (if confidence intervals for predictions are actually what you want)
cheers