backward_gevreg: Backward Selection on GEV Parameter
In pengyuwei94/evreg: Extreme value regression modelling

Description Usage Arguments Details Value See Also Examples

Significance controlled variable selection selects variables in either mu, sigma, and xi with backward direction based on likelihood-ratio-test, AIC, or p value from Wald test.

1 2	backward_gevreg(fit, criterion = "pvalue", alpha = 0.05, do_mu = TRUE, do_sigma = FALSE, do_xi = FALSE)

`fit`	An object of class `c("gev", "evreg")` returned from `gevreg` summarising the current model fit.
`criterion`	Either based `LRT`(Likelihood ratio test), `AIC`, or `pvalue`(by Wald test). Default criterion is `pvalue`.
`alpha`	Significance level if criterion equals pvalue or LRT. Default value is 0.05.
`do_mu`	do backward selection on mu if `do_mu` equals TRUE. Default is TRUE.
`do_sigma`	do backward selection on sigma if `do_sigma` equals TRUE. Default is FALSE.
`do_xi`	do backward selection on xi if `do_xi` equals TRUE. Default is FALSE.

The function performs backward elimination for an object of class c("gev", "evreg"). When do_mu, do_sigma, and do_xi all equal TRUE, the function performs backward selection on xi first, then on sigma, and finally on mu.

Non-zero components of inital value of xi may mean that the likelihood is zero at the starting values. This is because for xi not equal to zero, there is a constraint on the parameter space. To avoid this set all components of initial value of xi to 0, i.e. the Gumbel case.

An object (a list) of class c("gev", "evreg") summarising the new model fit (which may be the same as fit) and containing the following additional components

`Input_fit`	The input object of the class `c("gev", "evreg")`.
`Note`	A message that tells if a covariate has been dropped or not.
`Output_fit`	A list that contains formulae for the parameter, and the output object of the class `c("gev", "evreg")` if the output fit is different from the input fit.
`dropped_covariate`	A character vector shows dropped covariate
`criterion_value`	criterion value for if both input model and output model are different.

gevreg

### Annual Maximum and Minimum Temperature

P6 <- gevreg(y = TMX1, data = PORTw[, -1], mu = ~MTMAX + AOindex + STDTMAX + STDMIN + MDTR)
backward_gevreg(P6)