swap columns and rows of e(b) and e(V)

[Bo Chen]

Hi all,

I am new to mata, but I think this issue can be solved by mata.

I am wondering how to swap rows (and columns) based on the names of the columns (and rows) in stata. For example, the e(b) matrix I get is shown in Impage 3.png. What I would like to achieve is to swap columns so that all the columns associated with equation 1 (equ1) precede all columns associated with equation 2 (equ2) etc. And for columns of a single equation, the variables are also in a particular order-cons, lnmp lnp1 lnp2 lnp3 lnp4 lnp5 lnp6.

I have already saved the result in result.ster (https://1drv.ms/u/s!An6r3b9xWElTl0ETiaVzoRzSJST2).

Thank you all in advance

[Mike Lacy]

I don't see Mata as relevant to the solution. I think this is a tedious problem, and I'm not up to providing a complete solution, but I can suggest how I would approach it, at least the part that involves moving columns of e(b). Also, your description of what you want is not clear to me, so I'm *guessing* you to mean that you want a matrix with all the equ1 columns first, then all the equ2 columns, and then all the equN columns. Within each block of (e.g.) equ1, I'm *guessing* you want columns ordered as they were in e(b).

First, I'd use the extended macro function -colnames- to obtain a local containing the column names (see - help extended_fcn-).

mat B = e(b)
local cnames: colnames B

Then, I would loop over the column names, and concatenate the columns associated with a particular equation (e.g., equ1) into a Stata matrix. For example:

local foundfirst = 0
local j = 0
foreach cn of local cnames {
     local ++j
     if (strpos("`cn'", "equ1") & (`foundfirst' == 0)) { 
          mat EQ1 = B[1..`R',`j']
          local foundfirst == 1
     }
     else if (strpos("`cn'", "equ1") & (`foundfirst' == 1)) {
          mat EQ1 = (EQ1, B[1..`R',`j']
     }
}

Presuming you do this with all of equ1, ..., equN, you could then put them together as you like:

mat EQ = (EQ1, EQ2, ..., EQN)

Since I didn't have anything on which to test this code, there are likely errors, but it should at least give the idea of my approach.

[Bo Chen]

I don't see Mata as relevant to the solution. I think this is a tedious problem, and I'm not up to providing a complete solution, but I can suggest how I would approach it, at least the part that involves moving columns of e(b). Also, your description of what you want is not clear to me, so I'm *guessing* you to mean that you want a matrix with all the equ1 columns first, then all the equ2 columns, and then all the equN columns. Within each block of (e.g.) equ1, I'm *guessing* you want columns ordered as they were in e(b).

First, I'd use the extended macro function -colnames- to obtain a local containing the column names (see - help extended_fcn-).

mat B = e(b)
local cnames: colnames B

Then, I would loop over the column names, and concatenate the columns associated with a particular equation (e.g., equ1) into a Stata matrix. For example:

local foundfirst = 0
local j = 0
foreach cn of local cnames {
local ++j
if (strpos("`cn'", "equ1") & (`foundfirst' == 0)) {
mat EQ1 = B[1..`R',`j']
local foundfirst == 1
}
else if (strpos("`cn'", "equ1") & (`foundfirst' == 1)) {
mat EQ1 = (EQ1, B[1..`R',`j']
}
}

Presuming you do this with all of equ1, ..., equN, you could then put them together as you like:

mat EQ = (EQ1, EQ2, ..., EQN)

Since I didn't have anything on which to test this code, there are likely errors, but it should at least give the idea of my approach.

Thanks, Mike. This strategy works well.

[John Mullahy]

Bo: While Mike's solution works for your purposes, if you wanted to do this work in Mata I think it's possible with careful use of st_matrixcolstripe(.) and sort(.). Here is the essence of the idea.

mata

b=st_matrix("e(b)")
v=st_matrix("e(V)")
sb=st_matrixcolstripe("e(b)")
sba=sb,strofreal(1::rows(sb))
sbas=sort(sba,(1,2))
sind=strtoreal(sbas[.,cols(sbas)])
bsort=b[sind]
vsort=v[sind,sind]

end

In your example I think

sb=st_matrixcolstripe("e(b)")

will return a kx2 string matrix with nonmissing elements in the first column (check this). If so then the key issue is whether

sbas=sort(sba,(1,2))

orders the names as you desire. You might have to play around with the sort order to get the results you want, but I think it's doable.