Calculate a covariance/ correlation between two variables within a group in stata.

[Samira Bakkali]

Hello,

I want to calculate a covariance/ correlation between two variables within a group in stata.

May data is the following:

prior_date	c_pairs	prox
1980	BE-NL	0.576022
1980	BE-IT	0.711139
1980	BE-GB	0.255543
1980	BE-FR	0.257338
1980	BE-US	0.07187
1980	AT-BE	0.799441
1980	BE-OTHERS	0.69717
1980	BE-DK	0.585538
1980	BE-DE	0.076019
1980	BE-SE	0.686572

For each c_pairs, I want to calculate a covariance/ corralation between prior_date and prox.

Could someone help me to do that?

Thanks

[Dave Airey]

Here's an example using the postfile command.

clear
sysuse auto
tempname memhold
tempfile results
postfile `memhold' foreign cor cov using "`results'"
levelsof foreign, local(levels)
foreach foreign of local levels {
    correlate weight length if foreign == `foreign'
    local cor = r(rho)
    correlate weight length if foreign == `foreign', covariance
    local cov = r(cov_12)
    post `memhold' (`foreign') (`cor') (`cov')
}
postclose `memhold'
use "`results'", clear
list

. list

     +-------------------------------+
     | foreign        cor        cov |
     |-------------------------------|
  1. |       0   .9210244   12838.44 |
  2. |       1   .9105639   5394.719 |
     +-------------------------------+

[Nick Cox]

Your example isn't promising as prior_date is constant and there is only one observation for each pair. But guessing at what you want, consider this example using rangestat (SSC).

webuse grunfeld , clear
rangestat (corr) invest mvalue, int(year 0 0)
tabdisp year, c(corr_x) format(%4.3f)

--------------------------------
     year |               corr_x
----------+---------------------
     1935 |                0.930
     1936 |                0.834
     1937 |                0.808
     1938 |                0.797
     1939 |                0.887
     1940 |                0.906
     1941 |                0.919
     1942 |                0.924
     1943 |                0.914
     1944 |                0.934
     1945 |                0.951
     1946 |                0.946
     1947 |                0.944
     1948 |                0.887
     1949 |                0.928
     1950 |                0.926
     1951 |                0.929
     1952 |                0.918
     1953 |                0.941
     1954 |                0.925
--------------------------------

That leads to guess that you want something like

ssc install rangestat
gen foo = 42
rangestat (corr) prior_date prox, int(foo 0 0) by(c_pairs)

Here foo is just to keep rangestat happy.You don't give a decent data example using dataex (do please read and act on FAQ Advice #12) so I can't tell whether c_pairs is string or numeric with value labels.

You can get covariances too: just read the help file.

[Dave Airey]

I had to go read the rangestat help file after this example. In addition to correlation, one can do a user defined Mata function. Pretty neat!