从glmer（family = inverse.gaussian（link =“ log”））进行反向转换

Question

I am currently running a glmer with family=inverse.gaussian(link="log"). The "top model" I have is as follows:

 full_mod2=glmer(cpueplus1 ~ assnage * logcobb + (1|fyear) + (1|flocation),
 data=yc,family=inverse.gaussian(link = "log"))

with the output for the coefficients being:

  Fixed effects:
                   Estimate Std. Error t value Pr(>|z|)    
  (Intercept)      1.53670    0.16126   9.529  < 2e-16 ***
  assnage         -0.30168    0.04909  -6.146 7.96e-10 ***
  logcobb          0.42032    0.06155   6.829 8.54e-12 ***
  assnage:logcobb -0.10132    0.02395  -4.231 2.33e-05 ***

I am hoping to have an equation which I will be able to hold one of the variables constant (for example assnage) and determine the effect of the other variable across the observed values (for example logcobb). With gmler, you are able to use "invlogit()" easily for binomial dristributions, is there something similar to inverse.gaussian? For example, when assnage equals the mean (2), max(4), or min(1):

 mean_age=FUNCTION(1.53670 + -0.30168*(mean(assnage)) +
 0.42032*observedvalues(logcobb) + -0.10132*(mean(assnage)*observedvalues(logcobb))

Answer 1

You can get the inverse function from the inverse.gaussian object.

inv.gaus <- inverse.gaussian(link = "log")

inv.gaus$linkfun(10)
inv.gaus$linkinv(inv.gaus$linkfun(10))

You can also look at the definition of these functions directly.

inv.gaus$linkfun

 function (mu) log(mu) <environment: namespace:stats> 

inv.gaus$linkinv

 function (eta) pmax(exp(eta), .Machine$double.eps) <environment: namespace:stats> 

you can see more of the attributes in the link object by looking at attributes(inv.gaus)

to complete your back transformation, you may use

inv.gaus$linkinv(1.53670 + -0.30168*(mean(assnage)) +
 0.42032*observedvalues(logcobb) + -0.10132*(mean(assnage)*observedvalues(logcobb))