Finding the observation number of the smallest/largest value in a data set.

[Jeff Wooldridge]

Suppose I have data on a variable x, and I want to find the observation number associated with the minimum and maximum value in the data. I know I can get the min and max easily, and then I can write a loop. Checking whether there's an easier solution.

[Leonardo Guizzetti]

The trick here is to use an intermediate variable to identify those which are equal to the min/max. However, you'll have to be mindful of what you want to do if you have multiple observations at the min/max value. The code below just lists out all observation numbers, so you could use that potentially.

sysuse auto
keep mpg
summ mpg, meanonly

gen byte which_min = cond(mpg == r(min), _n, .)
gen byte which_max = cond(mpg == r(max), _n, .)

* min and max summary statistics now indicate first and last obs number for the min and max values
summ which_min
summ which_max

[Nick Cox]

Well, holding observation number in a byte will undoubtedly bite (*) you hard with a large dataset.

In general,

gen long obsno = _n 

summarize foo, meanonly 

levelsof obsno if foo == r(min)

and so forth.


(*) Worst joke on Statalist since 1994.

[Leonardo Guizzetti]

Well, holding observation number in a byte will undoubtedly bite (*) you hard with a large dataset.

(*) Worst joke on Statalist since 1994.

True on both counts.
Indeed long (or c(obs_t)) is the more appropriate type.

[Nick Cox]

This problem was also discussed in the Stata Journal in 2006

https://journals.sagepub.com/doi/pdf...867X0600600313

https://journals.sagepub.com/doi/pdf...867X0600600414