See this old Statalist post:

https://www.stata.com/statalist/archive/2009-09/msg00159.html

You could also compute it yourself fairly easily using the r(porder) stored result from the -ranksum- command. Here is an example showing both methods.

// Example from Wikipedia page for Rank Correlation:

// https://en.wikipedia.org/wiki/Rank_correlation#Rank-biserial_correlation

// Read in the data

clear

input byte(group y)

1 1

1 2

1 3

1 4

1 6

2 5

2 7

2 8

2 9

end

// Use the -ranksum- command with porder option

ranksum y, by(group) porder

// return list // show the stored results from -ranksum-

local f = max(r(porder), 1-r(porder))

local u = 1-`f'

display _newline ///

"Rank-biserial correlation = " `f'-`u'

// Now use Newson's -somersd- command, as shown here:

// https://www.stata.com/statalist/archive/2009-09/msg00159.html

somersd group y, transf(z) tdist

----------------------------------------------

The first method shows:

Rank-biserial correlation = .9

The second method shows:

Asymmetric 95% CI for untransformed Somers' D

Somers_D Minimum Maximum

y .9 -.34470506 .99730383

HTH.

Bruce Weaver 's method seems to be correct. However you can get much more information with about the same amount of work with an ordinal logistic regression. Details and an example follow. David Booth

https://youtu.be/Zh4wN7xQ3_I

See the ordinal regression information I promised in these attached course notes. Best wishes David Booth

Bruce Weaver

Thank you so much for the detailed explanation. I'm confused what's the next step after "ranksum y, by(group) porder"? . I used

local f = max(r(porder), 1-r(porder))

local u = 1-`f'

but nothing was displayed.

Meanwhile, I can't use the second method "Newson's -somersd- " as it needs version 16 STATA and I'm working with 14.2.

David Eugene Booth

I was planning to do Rank-biserial correlation, and if I find correlation I will then do ordinal logistic regression. I think regression should be proceeded by correlation test. Am I wrong?

You omitted the -display- command that displays the result:

display "Rank-biserial correlation = " `f'-`u'

Re your other question, if you want to estimate an ordinal logit model, go ahead. I can think of no good reason to make it contingent on finding a significant rank-biserial correlation.

Bruce Weaver

It worked, thank you so much.

By the way, Manar Serageldin, I see nothing in the help file for Newson's -somersd- saying that Stata 16 or later is required. Furthermore, I just tried it in Stata 15.1, and it gave exactly the same result I get using Stata 16. So I would be quite surprised if it does not work with Stata 14. (You could try emailing Newson to ask him.)

And one advantage of using -somersd- is that you get a 95% CI along with the point estimate.