top observations in a variable

[Joseph Chav]

I have a variable, let's say X, with 100+ categories. Then, I want to see the distribution of Y in these categories. However, I only want the top 10/ or 5 percent in my graph. Is there any way to specify it in the below command

graph hbar Y, over( X, sort(1) ) by(year)

Thank you

[Jared Greathouse]

h egen rank

[Nick Cox]

The top 10 categories will have more than 5% of the total!

Your question could be read in various ways and here is one in the absence of a data example from you.

In this dataset there are 13 categories and we suppose that we want to show just the 5 most frequent.

webuse nlswork, clear

bysort occ_code : gen freq = _N
egen tag = tag(occ_code)
egen rank = rank(-freq) if tag, unique 
sort rank occ_code
l rank freq occ_code if rank < .

graph hbar (asis) freq if rank <= 5, over(occ_code, sort(rank)) l1title(Occupation code) ytitle(Frequency) blabel(bar) ysc(alt r(0 12000))

And here is some more technique:

gen cu_freq = sum(freq) if rank < . 
su cu_freq, meanonly

gen cu_prob = 1 - cu_freq / r(max)

list rank cu_* occ_code if rank < . 


graph hbar (asis) freq if cu_prob >= 0.2 & cu_prob < ., over(occ_code, sort(rank)) l1title(Occupation code) ytitle(Frequency) blabel(bar) ysc(alt r(0 12000))

See also

https://journals.sagepub.com/doi/pdf...6867X221106436

https://www.stata-journal.com/articl...article=st0496

[Joseph Chav]

Thanks Jared! Thank you so much Nick, for your detailed comment and the resouces.