st: Re: Difference between -spearman- and -pwcorr- with ranked variables

["Joseph Coveney" <stajc2@gmail.com>]

Owen Gallupe wrote:

My understanding is that Spearman correlations are simply Pearson
correlations using ranked data. However, when I run the spearman
command I get different results from when I convert the variables to
ranks and then run the pwcorr command with the ranks.

. spearman var1 var2

rho = -0.0112, p = 0.5642


. egen rankvar1 = rank(var1)
. egen rankvar2 = rank(var2)

. pwcorr rankvar1 rankvar2, sig

r = -0.0100, p = 0.6058

I suspect it has something to do with the way I created the ranked
variables, but I'm not sure.

If someone could point me in the right direction, I would really appreciate it!

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

You rank the pooled variables.  See below.

Joseph Coveney


. sysuse auto, clear
(1978 Automobile Data)

. rename price var1

. rename gear_ratio var2

. quietly reshape long var, i(make) j(dummy)

. egen double rank = rank(var)

. quietly reshape wide var rank, i(make) j(dummy)

. spearman var1 var2

 Number of obs =      74
Spearman's rho =      -0.2525

Test of Ho: var1 and var2 are independent
    Prob > |t| =       0.0299

. pwcorr rank1 rank2, sig

             |    rank1    rank2
-------------+------------------
       rank1 |   1.0000 
             |
             |
       rank2 |  -0.2525   1.0000 
             |   0.0299
             |

.

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/