Number of observations when calculating mean with --by-- command

[Thomas Mitterling]

Hey,

I'd like to calculate averages of different groups and get information on the number of observations used, like in the following example:

webuse union, clear
by id: egen mean_age = mean(age)

In this case there are no missing values, hence the number of observations is always the same. In my case this is not the case and I would like to know how many observations are non-missing for each group. The --mean-- command stores the number of observations in the results. Is there a way to get the results even if I use --by--?

Thanks in advance!

[Nick Cox]

webuse union, clear by id: egen mean_age = mean(age)
by id: egen n_age = count(age)

The count() function of egen counts how often its argument is not missing, which is what you want here.

[Thomas Mitterling]

 webuse union, clear by id: egen mean_age = mean(age) by id: egen n_age = count(age)

The count() function of egen counts how often its argument is not missing, which is what you want here.

Ok, so I need two commands. Thanks!

[Thomas Mitterling]

Additional question. My panel is sorted by cross-section and time (obviously). Is there a possibility to calculate the means by time without resorting the whole dataset?

Using the current sorting Stata gives me "not sorted".

[Nick Cox]

No; I don't think so with this method. There is no operation by: unless observations are in appropriate sort order.

[Thomas Mitterling]

Ok, that means I have to use ifs. I guess there is no possibility to use mean combined with replace?

My current approach would be to loop through all dates in my panel and calculate the mean for each group. Since mean and replace can't be used together the only way is to use a scalar as temporary variable? Isn't there a more straightforward approach?

edit:
Problem solved. While code like

by id: egen n_age = count(age)

doesn't work, this works:

egen n_age = count(age), by(id)

[Nick Cox]

That is a distinction without a difference. Here's the code behind count() from which it can be seen that Stata has to sort, regardless of which syntax you choose to call it with.

*! version 3.1.3  22feb2015
program define _gcount
    version 6, missing

    gettoken type 0 : 0
    gettoken g    0 : 0
    gettoken eqs  0 : 0

    syntax anything(name=anythin) [if] [in] [, BY(varlist)]

    tempvar touse
    quietly {
        gen byte `touse'=1 `if' `in'
        sort `touse' `by'
        by `touse' `by': gen `type' `g' = /*
            */ sum(!missing(`anythin')) /*
            */ if `touse'==1
        by `touse' `by': replace `g' = `g'[_N]
    }

end

[Thomas Mitterling]

Ah, good to know. I'll try the code anyway and look whether it works or not.