拟泊松GLM的阶跃函数

Question

I'm trying to fit quasi-poisson GLMs and GAMs using the step function to perform backwards selection. I used the following code but R returned an error as it couldn't calculate the AIC. The same model using a poisson distribution works but not quasi-poisson, is there any way to alter the function so it works?

full.mod03q<-glm(Total.Positivo_2003~X.baix03 + Tot_PrevDesnut04
                 + Tot_HosDesid04 + Despesa_06 + prpnurb03 + prpnkid03
                 + dev03 + watertoilet03 + elec03 + dollpercap03
                 + abspov03 + gini03 + Importado.Outro.Municipio.Da.Uf_2003 
                 + Importado.Uf.Pais_2003 + elevation + floodplain + BIO1 
                 + BIO6 + BIO12 + deforestation_prop,
                 family="quasipoisson", offset=log(pop_2003), 
                 data=ama1@data)
step(full.mod03q, direction="backward")

Answer 1

AIC cannot be calculated at quasi-poisson models. Function "step" uses AIC for choosing more relevant variables to include in model. There is a "not purist" solution calculating "quasi-AIC".

I am fighting against same problem. It appears can be solved at:

https://cran.r-project.org/web/packages/bbmle/vignettes/quasi.pdf