0

I used the MuMIN package to do a model averaging based on information criterion, following this question.

options(na.action = "na.fail")

Create a global model with all variables and two way interaction:

global.model<-lmer(yld.res ~ rain + brk + 
  onset + wid + (1|state),data=data1,REML="FALSE")

Standardise the glboal model since the variables are on different scale

stdz.model <- standardize(global.model,standardize.y = TRUE)

Create all possible combinations of the model

model.set <- dredge(stdz.model)

Get the best model based on deltaAICc<2 criteria

top.models <- get.models(model.set, subset= delta<2)

Average the models to calculate the effect size (standardised slopes of the input variables)

s<-model.avg(top.models) 
summary(s);confint(s)

The effect size of the variables are as follows: 

Variable           slope estimate
brk                 -0.28
rain                 0.13
wid                  0.10
onset                0.09

As you can see, I had standardize my model in step 3 so I can compare these slope estimates i.e. I can say slope estimate of brk is greater (in negative direction) than rain. However, since these slope estimates where standardised, I wanted to know if there is any way I can get the unstandardised slopes?

Please let me know if my question is not clear.

Thanks

www
  • 38,575
  • 12
  • 48
  • 84
user53020
  • 889
  • 2
  • 10
  • 33
  • related: http://stackoverflow.com/questions/23642111/how-to-unscale-the-coefficients-from-an-lmer-model-fitted-with-a-scaled-respon/23643740#23643740 – Ben Bolker Nov 22 '14 at 21:31
  • Many thanks for this. I referred to the link you directed and it is exactly what I was looking for you. But what does the symbol like `Y0`,`mu0`,`b0` mean in these equations `(Y0-mu0)/s0 = b0' + (b1'*(1/s1*(x1-mu1))) + b2'*(1/s2*(x2-mu2))+ ` . Is there any text book that I can refer to? Thanks a lot again – user53020 Nov 25 '14 at 16:41

0 Answers0