[R] extending lines for prediction intervals

[Andras Farkas]

Dear All,

would you have some thoughts on how to extend the prediction interval lines to beyond the "range of data" modeled?



example:


y <-c(0.4316092,0.4156757,0.3517915,0.3669508,0.3899471,0.3964143,

0.4001074,0.3851003,0.4222451,0.375324,0.3652045,0.3376978,0.383012,

0.3763665,0.3550609,0.2958678,0.3726571,0.3442298 

#,0.3403275,0.2973978 

)*100 

x <-seq(1,length(y),1) 


z<-c("07/01/2015","08/01/2015","09/01/2015","10/01/2015","11/01/2015",

"12/01/2015","01/01/2016","02/01/2016","03/01/2016","04/01/2016","05/01/2016",

"06/01/2016","07/01/2016","08/01/2016","09/01/2016","10/01/2016","11/01/2016",

"12/01/2016","01/01/2017","02/01/2017") 


fit <-lm(y~x) 


temp_var <- predict(fit, interval="prediction") 


new_df <- data.frame(cbind(x,y, temp_var)) 

#new_df$x<-factor(new_df$x, ordered = T) 


library(ggplot2) 

ggplot(new_df, aes(x,y))+ 

geom_point() + 

theme(panel.background = element_rect(fill = 'white', colour = 'black'))+ 

geom_line(aes(y=lwr), color = "black", linetype = "dashed",size=0.75)+ 

geom_line(aes(y=upr), color = "black", linetype = "dashed",size=0.75)+ 

scale_x_discrete(limits=z)+ 

theme(axis.text.x = element_text(angle = 45, hjust = 1))+ 

theme(panel.grid.major=element_line(colour = "grey"))+ 

lims(y=c(0,50))+ 

geom_smooth(method=lm, se=TRUE,fullrange=TRUE,fill="darkgrey",col="black")+labs(title = paste("Adj R2 = ",signif(summary(fit)$adj.r.squared, 4), 

"Intercept =",signif(fit$coef[[1]],4 ), 

" Slope =",signif(fit$coef[[2]], 4) 

# " P =",signif(summary(fit)$coef[2,4], 3) 

))+ 

ggtitle("Consumption Over Time") + 

theme(plot.title = element_text(hjust = 0.5))+ 

labs(y="y",x="x")+ 

geom_point(shape=15,aes(x=c(7),y=new_df[,2][7]), color="black",cex=4)+ 

geom_point(shape=15,aes(x=c(8),y=new_df[,2][8]), color="black",cex=4)+ 

geom_point(shape=17,aes(x=c(19),y=0.3403275*100), color="black",cex=4)+ 

geom_point(shape=17,aes(x=c(20),y=0.2973978*100), color="black",cex=4) 



as you  will see the regresssion line and confidence interval is extended, but would also want to extend the prediction interval lines to the "same length"... Wonder if you have any insights to this question...



appreciate the help,



Andras Farkas