Use results store in matrices

[Charlie Joyez]

Dear Statalister,
many commands automatically store results in r(), either scalar, local or matrices. I focus in the latter case.

I'd like to access directly to some values in the matrix stored. See the example below :

sysuse auto.dta,clear

corr mpg price trunk
return list

The matrix of the correlation coefficient is store under the name r(C).
However any direct use of r(C) doesn't work. See :

matrix list r(C)
local b1=r(C)[1,2]
di `b1'

However, I've get this information using an intermediate matrix m set equal to r(C). And now the commands that didn't work directly on r(C) work on m

I've done :

matrix m=r(C)
matrix list m
local c1=m[1,2]
di `c1'
local c2=m[1,3]
di `c2'

Is there a way to get this info without creating m?

Thanks,
Charlie

Ps : I know that concerning correlation coefficient I could have simply done two by two correlations, which directly store in the local r(rho) avoiding a matrix, but this was for illustrational purposes. Indeed I use corrci command, which returns only matrixes.

[Clyde Schechter]

As far as I know, in order to retain a returned matrix, you have to create a new matrix as you have done in #1. This is also true for matrices in -ereturn-.

[Charlie Joyez]

Ok, thanks Clyde, this is all I wanted to know.
It seemed strange to me, quite unusual compared to scalar stored (local b=r(rho) would work) and quite long, but I guess there is some good reason behind.

Charlie

[Nick Cox]

You can copy a single number to a local macro, but how would you expect a matrix to be stored in a local macro? You could store the name of a matrix in a local macro, clearly a different matter.

Remember, a local macro holds a string. The string can have numeric characters only, but all local macro definitions are variants on

local foo "bar"

even when they are assignments.

[Charlie Joyez]

Dear Nick,
I didn't want to store a matrix in a local macro, but only one element of the matrix (i.e. here the [1,2]), which I do (see my code local c1=m[1,2]), but only after copying the matrix named r(C) into the matrix named m (or rather, after creating a new matrix m equal to r(C).)

What I didn't get is why I could store the element [1,2] of my matrix m into a local macro, but couldn't do it for the matrix r(C), whereas they are supposed to be the same (same type : matrix, and same values). But I guess a matrix automatically generated as is r(C) is not exactly the same as a matrix I create using matrix, this is the one thing I didn't know, and made this issue look strange to me.

Best,
Charlie

[Nick Cox]

Fair point. I suspect that it just is a complication too far for the syntax. There would have to be a check of whether an r(C) exists that can take subscripts.. Implementing that without creating other problems sounds like a medium-size nightmare compared with the small benefit you are missing now.

[William Lisowski]

Using the el() matrix function enables the program to store the value without creating an intermediate matrix.

local b1=el(r(C),1,2)

[Charlie Joyez]

Thanks William,
This is what I was looking for !

Best,
Charlie

[William Lisowski]

For those who search finds this topic, with Stata 16 it is now possible to use the syntaxCharlie Joyez hoped to use in his code in post #1.

. sysuse auto.dta,clear
(1978 Automobile Data)

. quietly corr mpg price trunk

. matrix list r(C)

symmetric r(C)[3,3]
              mpg       price       trunk
  mpg           1
price  -.46859669           1
trunk  -.58158503   .31433161           1

. local b1=r(C)[1,2]

. di `b1'
-.46859669

[Charlie Joyez]

Great news!
Thanks William for the update.