Accessing Variance-Covariance matrix elements by variable names.

[Alfonso Sánchez-Peñalver]

Hi consider the following

clear all
set more off

sysuse auto
quiet reg price mpg foreign weight
mat vc = e(V)
mat list vc
di vc[mpg,foreign]
di vc[1,2]

This produces the following output:

. clear all
. set more off
. 
. sysuse auto
(1978 Automobile Data)
. quiet reg price mpg foreign weight
. mat vc = e(V)
. mat list vc

symmetric vc[4,4]
                mpg     foreign      weight       _cons
    mpg   5508.7774
foreign   9089.6431   467826.27
 weight     36.2912     218.835   .39784425
  _cons   -229604.2  -993431.72  -2039.2381    11404044

. di vc[mpg,foreign]
.
. di vc[1,2]
9089.6431

Why is it that, even though we can clearly see that the variance-covariance matrix has row names and column names, we can't access its elements using them? Clearly this is not a problem for the diagonal elements, i.e. the variances, since we can always access them through the system vector _se, but what about the covariances? I find this very surprising. Is there any way to access them via the variables' names, or _cons for the intercept? I have searched the forum and Google, and have found no solution for this. I apologize if I have missed it.

[Clyde Schechter]

You have to do it indirectly:

. local r = rownumb(vc, "mpg")

. local c = colnumb(vc, "foreign")

. display vc[`r', `c']
9089.6431

When working with commands that begin with -matrix-, you don't have to intermediate with rownumb() and colnumb(), you can access them directly with the row and column names, but only in those commands. So another approach would be:

. matrix one = vc["mpg", "foreign"]

. matrix list one

symmetric one[1,1]
       foreign
mpg  9089.6431

. display one[1, 1]
9089.6431

[Alfonso Sánchez-Peñalver]

Hi Clyde, yes I was aware of what I call the rownumb and colnumb workaround. I find that cumbersome.

I wasn't aware of the second option you give me, which I think is slightly better, but still cumbersome.

I believe the downfall is that the operator vc["mpg", "foreign"] is returning a matrix instead of a scalar. In a matrix if you access one of its elements, whether by index or by name, it should return a scalar. Otherwise you have to create new macros or matrices that load the memory more. Clearly if you access only a column or a row, also by index or by name, you should return a matrix, but when both row and column in a matrix are specified it should return a scalar.

[Clyde Schechter]

I believe the downfall is that the operator vc["mpg", "foreign"] is returning a matrix instead of a scalar. In a matrix if you access one of its elements, whether by index or by name, it should return a scalar. Otherwise you have to create new macros or matrices that load the memory more. Clearly if you access only a column or a row, also by index or by name, you should return a matrix, but when both row and column in a matrix are specified it should return a scalar.

Yes. Actually, this might be something to add to the Wishlist.