Empirical p-value after sampling matching distribution in R

Question

[User 7754]

I am trying to understand how to compute an empirical p-value after a sampling a procedure, to test whether a variable "var" we observe in a "case" dataset appears more than expected from a "control" dataset, after having matched for a variable "length".

cases = data.frame(id = 1:10, length = sample(1:1000,10), var = sample(c(TRUE,FALSE), 10, TRUE)) 
control = data.frame(id = 1:100, length = sample(1:1000,100), var = sample(c(TRUE,FALSE), 100, TRUE)) 

res = data.frame()
nperm = 10
for (perm in 1:nperm) {
    control_random = control[sample(nrow(control),nrow(control),replace=FALSE),] # draw the control into a random order
    for (i in 1:nrow(cases)) { # loop through cases to find a match for each
    for (j in 1:nrow(control)) { # for each case, loop through control looking for a match
        if (abs(control_random$length[j] - cases$length[i]) < 100) { # match if length is within 100
            break
        }
    }

 res = rbind.data.frame(res, data.frame(perm, cases$id[i], control_random$id[j], control_random$var[j] ))
 # and remove it so we don't use it again
 control_random = control_random[-j,]
 }
 }

# pvalue:
# count how many times var is TRUE in control set compared to cases
library(dplyr)
num_controls = res %>%
    group_by(perm) %>%
    summarise(n = sum(control_random.var.j.))
# Is it the number of original true values compared to random controls?
num_cases = sum(cases$var)

pval = (sum(num_controls$n > sum(cases$var))) / nrow(res)

Is the p-value correct? What if we had bins instead of matched random samples, would we compare within each bin? Any help would be very appreciated!