Styjun

I am conducting some time series modelling and want to remove terms from my time series based on removing the lowest rank p-values. The criteria I have is to only rank p-values that exclude "ar", "ma", "intercept", "price.diff" and rank only if p>0.2.

Here is an example:

term pval rank
ar1 0.001 NA
ar2 0.292 NA 
ar3 0.000 NA
ma1 0.000 NA
intercept 0.000 NA
Price.Diff 0.859 NA
School 0.818 2
Easter 0.149 NA
Christmas 0.049 NA
High.Week 0.000 NA
Low.Week 0.000 NA

This is the function I have written:

rank_p<-function(x) 
 x["rank"]<-NA
 x$rank<-ifelse(test = substr(x$term,1,2) != "ar" & 
 substr(x$term,1,2) != "ma" & 
 substr(x$term,1,stop = nchar(x$term)) != "intercept" &
 substr(x$term,1, stop = nchar(x$term)) != "Price.Diff" & 
 x$pval > 0.2,
 yes = rank(-x$pval, na.last = NA),
 no = NA)
 return(x)
 

My issue is that the rank for this example begins at 2.This would be the second highest p-value however because I am excluding price.diff, this should be ranked 1.

Is the issue in ordering of the conditions?

The reason your ranking is not as desired is because your yes term includes the whole column. So you need to subset the column based on your condition.
What is happening in your current function is that you are ranking the column without a filter then changing to NA everything that doesn't fit the condition.

I have just added a slightly changed version of your code to do this.
All I changed was to define your condition then subset the data frame accordingly. EDIT: this will work on the displayed data but not if there are more than one rows which meet the condition.

rank_p <- function(x) 
 x["rank"] <- NA
 cond <- substr(df$term,1,2) != "ar" & 
 substr(df$term,1,2) != "ma" & 
 substr(df$term,1,stop = nchar(df$term)) != "intercept" &
 substr(df$term,1, stop = nchar(df$term)) != "Price.Diff" & 
 df$pval > 0.2

 x$rank <- ifelse(test = cond,
 yes = rank(-x[cond, ]$pval, na.last = NA),
 no = NA)
 return(x)

What this does is only rank the rows that meet your condition( x[cond, ] ).

This can be further simplified because you have already set them to NA just change the ones that meet the condition. It doesn't seem like you need an ifelse at all! This will also assign a vector of ranking to a subset of the same length which should work for larger sets which contain more than 1 row which meets the condition.

rank1_p <- function(x) 
 x["rank"] <- NA
 cond <- substr(df$term,1,2) != "ar" & 
 substr(df$term,1,2) != "ma" & 
 substr(df$term,1,stop = nchar(df$term)) != "intercept" &
 substr(df$term,1, stop = nchar(df$term)) != "Price.Diff" & 
 df$pval > 0.2
 x[cond, "rank"] <- rank(-x[cond,]$pval, na.last = NA)

 return(x)