Correlation between variables

[juliana pinto]

Hello Statalists,
Please, could anyone help with the correct command?
I have two datasets measuring the weight of a product using two different methods. I want to know the correlation between this two variables called 'grams1' and 'grams2'.
The two data sets have the same id variable, the same time var and the grams var differ between both data. The two data sets samples and my codes are below. Is this the correct way to assess the correlation between this two variables? What I expect is that the correlation is very high indicating that both weighting methods are similar and yield close weight in grams.
Thank you!

Grams1.dta

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(id grams1 time)
11 13 1
11 12 2
11 19 3
22 21 1
22 10 2
22 15 3
33 35 1
33 29 2
33 20 3
end

Grams2.dta

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(id grams2 time)
11   21 1
11   12 2
11   18 3
22 21.5 1
22   11 2
22 14.7 3
33   36 1
33   29 2
33 20.6 3
end

I then tried to do the following:

use grams1,clear
sort id time
merge 1:1 id time using grams2

I get the following merged dataset:

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(id grams1 time grams2) byte _merge
11 13 1   21 3
11 12 2   12 3
11 19 3   18 3
22 21 1 21.5 3
22 10 2   11 3
22 15 3 14.7 3
33 35 1   36 3
33 29 2   29 3
33 20 3 20.6 3
end
label values _merge _merge
label def _merge 3 "matched (3)", modify

And finally I did:

pwcorr grams1 grams2, star(0.01)

[Clyde Schechter]

Yes, that will do it.

Since you are only working with 2 variables, you can also use -corr- instead of -pwcorr-; everything will come out the same.

[juliana pinto]

Many thanks Clyde!

[Nick Cox]

the correlation is very high indicating that both weighting methods are similar and yield close weight in grams

This is not what correlation measures. It measures linearity of relationship, not agreement. Consider that y = a + b x and x are correlated at 1 so long as b is any positive value whatsoever. But y is not equal to x unless a = 0 and b = 1 and y is not close to x without a being comparatively small and b being close to 1.


Concordance correlation is a measure of agreement. In your case it is high too.

 . concord grams1 grams2 

Concordance correlation coefficient (Lin, 1989, 2000):

 rho_c   SE(rho_c)   Obs    [   95% CI   ]     P        CI type
---------------------------------------------------------------
 0.937     0.046       9     0.847  1.026    0.000   asymptotic
                             0.753  0.985    0.000  z-transform

Pearson's r =  0.946  Pr(r = 0) = 0.000  C_b = rho_c/r =  0.990
Reduced major axis:   Slope =     1.025   Intercept =    -1.595

Difference = grams1 - grams2

        Difference                 95% Limits Of Agreement
   Average     Std Dev.             (Bland & Altman, 1986)
---------------------------------------------------------------
    -1.089       2.670                 -6.322      4.144

Correlation between difference and mean = 0.076

Bradley-Blackwood F = 0.679 (P = 0.53773)

I used concord from the Stata Journal.

[juliana pinto]

Many thanks Nick Cox!