finding the closest rivals in the group - r

Question

i have a unbalaned panel company data as followed：
time comid group sales closeone
1988m1 tw1701 1 2.45 tw1410
1988m1 tw1213 1 1.98 tw1701
1988m1 tw1707 1 2.67
1988m1 tw2702 1 9.45
1988m1 tw9902 1 4.16
1988m1 tw1410 1 2.57
1988m2 tw2601 3 27.44 tw2505
1988m2 tw2505 3 9.49
1988m2 tw1413 3 1.46
1988m2 tw2901 3 3.74
1988m2 tw1417 4 1.87 tw1506
1988m2 tw1506 4 3.24
1988m2 tw1215 4 3.58
my aim is to find out the closest rival within the same group and time.
just as column colseone show. for example,
in the closeone first row ,tw1410 is under group=1,time=1988m1 condition,
abs(all sales - sales of tw1701) and find the min value and exclude zero(i.e. can't be itself).

Answer 1

I'm not sure why you were downvoted, I didn't think this was trivial. This is how I solved it. There might be an easier way. I couldn't get data.table operations to set the final value so had to use a for loop.
basically it sorts the data by groups, calculated the difference between the rows above and below within the same group, finds the min of those two values, and then sets the value of the rival by that reference.
library(data.table)
setDT(dat)
setorder(dat,time,group,-sales)
dat[ , "Diff" := c(NA, diff(sales)), by = .(time,group)]
dat[ , "Diff2" := c(diff((sales)),NA), by = .(time,group)]
dat[ ,"Min" := ifelse(abs(Diff) < abs(Diff2), 1, 2)]
dat[ ,"Min" := ifelse(is.na(Diff),2,Min)]
dat[ ,"Min" := ifelse(is.na(Diff2),1,Min)]
dat[, "Rival" := NA]
for(i in 1:nrow(dat)){
if(dat$Min[i] == 2){
dat$Rival[i] = as.character(dat[i+1,comid])
}else{
dat$Rival[i] = as.character(dat[i-1,comid])
}
}
> dat
time comid group sales Diff Diff2 Min Rival
1: 1988m1 tw2702 1 9.45 NA -5.29 2 tw9902
2: 1988m1 tw9902 1 4.16 -5.29 -1.49 2 tw1707
3: 1988m1 tw1707 1 2.67 -1.49 -0.10 2 tw1410
4: 1988m1 tw1410 1 2.57 -0.10 -0.12 1 tw1707
5: 1988m1 tw1701 1 2.45 -0.12 -0.47 1 tw1410
6: 1988m1 tw1213 1 1.98 -0.47 NA 1 tw1701
7: 1988m2 tw2601 3 27.44 NA -17.95 2 tw2505
8: 1988m2 tw2505 3 9.49 -17.95 -5.75 2 tw2901
9: 1988m2 tw2901 3 3.74 -5.75 -2.28 2 tw1413
10: 1988m2 tw1413 3 1.46 -2.28 NA 1 tw2901
11: 1988m2 tw1215 4 3.58 NA -0.34 2 tw1506
12: 1988m2 tw1506 4 3.24 -0.34 -1.37 1 tw1215
13: 1988m2 tw1417 4 1.87 -1.37 NA 1 tw1506
If anyone has a better solution I'd love to see it.
EDIT
The reason I couldn't get this into vector format apparently was because comid was a factor. I have no idea why that would break the function, but when I changed it to character it worked.
Replace the for loop with this:
dat$comid = as.character(dat$comid)
dat[, "Rival" := ifelse(Min == 2, shift(comid, type = "lead"), shift(comid, type = "lag"))]
> dat
time comid group sales Diff Diff2 Min Rival
1: 1988m1 tw2702 1 9.45 NA -5.29 2 tw9902
2: 1988m1 tw9902 1 4.16 -5.29 -1.49 2 tw1707
3: 1988m1 tw1707 1 2.67 -1.49 -0.10 2 tw1410
4: 1988m1 tw1410 1 2.57 -0.10 -0.12 1 tw1707
5: 1988m1 tw1701 1 2.45 -0.12 -0.47 1 tw1410
6: 1988m1 tw1213 1 1.98 -0.47 NA 1 tw1701
7: 1988m2 tw2601 3 27.44 NA -17.95 2 tw2505
8: 1988m2 tw2505 3 9.49 -17.95 -5.75 2 tw2901
9: 1988m2 tw2901 3 3.74 -5.75 -2.28 2 tw1413
10: 1988m2 tw1413 3 1.46 -2.28 NA 1 tw2901
11: 1988m2 tw1215 4 3.58 NA -0.34 2 tw1506
12: 1988m2 tw1506 4 3.24 -0.34 -1.37 1 tw1215
13: 1988m2 tw1417 4 1.87 -1.37 NA 1 tw1506
That should run a lot faster.