Correlation by gender in a Panel Data

[Daniel Lerner]

Hi, I have an artificial panel data in which the individual is the Household ID and time is defined as the individual's gender, meaning that for each household I have one couple with one individual male and other female. I have several explanatory variables such as years of education, age, nationality, religiosity and so on. I want to create a matrix with the correlation between males and females for each of my explanaotry variables. I tryed to use:

correlate varlist, by (gender)

But this option is not allowed. Any suggestion?

Thank you,
Daniel

[Carlo Lazzaro]

Daniel:
welcome to this forum.
You may want to test something like:

bysort gender: correlate <varlist>

[Daniel Lerner]

Hi, when I use bysort it just repeat the command for males and females. What I wanna test is the correlation between both groups for each explanatory variable.
Let's say:

correlation between male's years of education and female's years of education

[Carlo Lazzaro]

Daniel:
correct, as this is what you asked for.
That said, if you want to -correlate- your variables as you reported in your #3:

. use "C:\Program Files\Stata17\ado\base\a\auto.dta"
(1978 automobile data)

. g price_1=price if foreign==0
(22 missing values generated)

. g price_2=price if foreign==1
(52 missing values generated)

. corr price_1 price_2
no observations
r(2000);

Conversely, if you're interested in the correlation of regression coefficients, things are easier:

use "C:\Program Files\Stata17\ado\base\a\auto.dta"
. regress price i.foreign##c.trunk

      Source |       SS           df       MS      Number of obs   =        74
-------------+----------------------------------   F(3, 70)        =      3.45
       Model |  81822937.2         3  27274312.4   Prob > F        =    0.0211
    Residual |   553242459        70   7903463.7   R-squared       =    0.1288
-------------+----------------------------------   Adj R-squared   =    0.0915
       Total |   635065396        73  8699525.97   Root MSE        =    2811.3

---------------------------------------------------------------------------------
          price | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
----------------+----------------------------------------------------------------
        foreign |
       Foreign  |   1114.855    2657.69     0.42   0.676    -4185.741    6415.451
          trunk |   261.6024   91.41565     2.86   0.006     79.27966    443.9252
                |
foreign#c.trunk |
       Foreign  |   6.257639   211.4828     0.03   0.976    -415.5316    428.0468
                |
          _cons |   2213.787    1403.61     1.58   0.119    -585.6244    5013.199
---------------------------------------------------------------------------------

. estat vce, corr

Correlation matrix of coefficients of regress model

             |        1.           1.fore~n#          
        e(V) |  foreign     trunk   c.trunk     _cons 
-------------+----------------------------------------
   1.foreign |   1.0000                               
       trunk |   0.5074    1.0000                     
   1.foreign#|                                        
     c.trunk |  -0.9575   -0.4323    1.0000           
       _cons |  -0.5281   -0.9607    0.4153    1.0000 

.

[Nick Cox]

Let's suppose gender is coded 0 for male and 1 for female. (By the way, a really good convention that should be a rule for researchers is to name an indicator for the condition coded 1, so that 0 for male and 1 for female and a variable name female then would be more transparent.)

Then what you want is then a generalisation of

bysort id (gender) : gen whatever_female = whatever[2] if _n == 1 & _N == 2 

corr whatever whatever_female

[Nick Cox]

Second thoughts are that this might, unusually, be a situation where a reshape wide is a really good idea. Given a data example that is real or realistic I would be happy to show token code.