Plotting a histogram with multiple correlations

[Arthur Carvalho Brito]

Hi, I have a dataset with columns named i, j, product, v, value.


How can I calculate the correlation coefficient between v and value for each triple of i-j-product, and then plot what I get with a histogram?

I know how to get the correlation matrix for each group, but not how to use that to get the plot. This is what I tried:

bysort i j product: pwcorr v value
gen correlation = r(rho)

Thanks!

[Nick Cox]

To get a histogram, you need the correlations in a variable, but that approach can find only one correlation left in memory after pwcorr, which is why you are asking the question.

Here is a silly example to show some technique. I use rangestat from SSC, which must be installed before you can use it. Here the histogram is not interesting enough to show, but I am optimistic that this is what you're seeking.

If you have many correlations, and some are based on very small samples, I would keep a close eye on the sample sizes used, which are automatically generated too.

webuse nlswork, clear 

egen group = group(msp nev_mar collgrad)
rangestat (corr) ln_wage age, int(group 0 0)
tabdisp group, c(msp nev_mar collgrad corr_*)

egen tag = tag(group) 

* options to suit 
histogram corr_x if tag

So, although you don't give a data example, something like this should help:

egen group = group(i j product)
ssc install rangestat 
rangestat (corr) v value, int(group 0 0)
tabdisp group, c(i j product corr_*)

* select just one observation from each group, to avoid multiple counting
egen tag = tag(group) 

* add options to suit 
histogram corr_x if tag

[Arthur Carvalho Brito]

Nick Cox, thank you so much!

[Nick Cox]

The fix is to ignore it. You could fire up list to show you the results, but you wouldn't want (I guess) really to look through thousands of them.

[Arthur Carvalho Brito]

Nick Cox thank you, you are right in that I don't need to look at the individual values