Date:         Mon, 25 Jun 2007 16:30:07 -0400
Reply-To:     John James <free562james@YAHOO.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         John James <free562james@YAHOO.COM>
Subject:      Re: Correlation at the country level

I had a look at the autoreg procedure but it seems to me that this proc
takes care of serial correlation etc and should be preferred over proc reg
depending on the data. However, i would just like to compute a correlation
coefficient at this stage and nothing more but because of the issues that
David noted I would like somehow to compute it at the country level but get
a single correlation coefficient at the end for the total sample.

Help please...

On Sun, 24 Jun 2007 23:44:08 -0700, David L Cassell <davidlcassell@MSN.COM>
wrote:

>free562james@YAHOO.COM wrote:
>>
>>I have a country-year dataset including data on a number of countries for
a
>>number of years. I want to compute the correlation coefficient between two
>>variables at the country level. Is it possible to do that use proc corr in
>>sas?
>
>You have a time series problem.  (Actually, you have a panel data
>problem.)  So don't do ordinary correlations, because the results
>out of PROC CORR will be assuming that you have independent
>observations.  You know that's not the case.  You can get an r,
>but the tests that use that r will be misleading, if not seriously biased.
>
>Look into time series approaches that use the temporal structure of
>your variables.  PROC AUTOREG might be a good starting point,
>depending on what you plan to do with your data.
>
>HTH,
>David
>--
>David L. Cassell
>mathematical statistician
>Design Pathways
>3115 NW Norwood Pl.
>Corvallis OR 97330
>
>_________________________________________________________________
>Like puzzles? Play free games & earn great prizes. Play Clink now.
>http://club.live.com/clink.aspx?icid=clink_hotmailtextlink2