Store the results of a regression in a matrix

[Steven Karan]

Hi!

I'm trying to run a regression and separating the results by a binary variable, this is easy to do it with the use of bysort. However, in the following steps I'm trying to construct a matrix with the values of each regression so first I need to store the results of the regression with a certain name. The problem is that when I use matrix list to see the stored results I only get the results of the last regression and not of both

sysuse auto.dta


bysort foreign: regress price mpg weight headroom
matrix list r(table)

To be a little more clear I want to achieve something like this, but preferably using a loop or something that can be replicated:

reg price mpg weight headroom if foreign == 0
matrix res1=r(table)

reg price mpg weight headroom if foreign == 1
matrix res2=r(table)

matrix list res1
matrix list res2

[Ken Chui]

If or a foreach loop may be more suitable for this case. You can also try statsby. For example:

sysuse auto, clear
statsby cons = _b[_cons] mpg = _b[mpg] weight = _b[weight] headroom = _b[headroom], by(foreign): reg price mpg weight headroom
list

Results:

     +---------------------------------------------------------+
     |  foreign        cons         mpg     weight    headroom |
     |---------------------------------------------------------|
  1. | Domestic   -10845.59    192.2954   4.451391   -526.4644 |
  2. |  Foreign    -5283.24   -21.69082   5.121311    131.9128 |
     +---------------------------------------------------------+

[Steven Karan]

Thank you, I was trying to do it as in #1 because I need to store results of a particular value in a matrix. That's why I was trying to use a loop and then use the return results to build the matrix. But I think it could be useful your code, so thanks

[Ken Chui]

webuse nhanes2, clear

* Extract levels
levelsof region, local(lv)

* Loop
foreach x of local lv{
    quietly reg bmi i.sex i.race age if region == `x'
    matrix res`x' = r(table)
}

mat list res1
mat list res2
mat list res3
mat list res4

[Eric Makela]

Hi Steven. Ken's code is perfect if the 'levelsof' variable categorizes your data exactly. If you're interested in extracting a specific result from the regression results (for example the b coefficients), you might further consider something like:

mat result = J(4,2,.)
mat result[1,1] = res1[1,4]
mat result[2,1] = res1[1,3]
...
mat result[1,2] = res0[1,4]
...
mat result[4,2] = res0[1,1]
mat coln result = "foreign" "domestic"
mat rown result.........

[Steven Karan]

Thank you both, its been really helpful. Now, I wanted to follow up a little bit on Erics' code. I want to extract different coefficient and build a new matrix, but is there any way I can automate this that can work for different data sets? For example, following #4, now that I have the results in res1, res2, ... I know the position of the betas and the pvalues, however I would like to put these in a matrix that looks like this:

	beta	p-value
region 1
region 2

Is there any way to do this? I'm not sure if I'm being clear in my explanation. Thanks again for further help!

[Ken Chui]

Thank you both, its been really helpful. Now, I wanted to follow up a little bit on Erics' code. I want to extract different coefficient and build a new matrix, but is there any way I can automate this that can work for different data sets? For example, following #4, now that I have the results in res1, res2, ... I know the position of the betas and the pvalues, however I would like to put these in a matrix that looks like this:

	beta	p-value
region 1
region 2

Is there any way to do this? I'm not sure if I'm being clear in my explanation. Thanks again for further help!

Not Eric, but I've recently answered a question similar to this: https://www.statalist.org/forums/for...s-and-p-values