What about changing the missing values to zero after you do the pca. I
would use pca based on the covariance matrix.
To change the missing values to zero, use
mvencode variablename ,mv(0)
where variablename is the name of the variable having missing values. I
am assuming you have no missing values other than due to the
periods when the person was not in office.
David J Svendsgaard, PhD
Biostatistician
EPA/ORD/NCEA/RTP, Mail Drop B-243-01
Research Triangle Park, NC 27711
Phone (919) 541-4186
Fax (919) 541-1818
From: Nick Eubank <nickeubank@gmail.com>
To: statalist@hsphsun2.harvard.edu
Date: 07/03/2011 12:17 AM
Subject: st: PCA with Missing Values (or other factor analysis)
Sent by: owner-statalist@hsphsun2.harvard.edu
Hello All,
I'm trying to do a PCA on legislator votes (1 or 0), where different
legislators have overlapping but distinct terms. In other words,
legislator A is in office periods 1 and 2, legislator B is in office 2
and 3, so A has no votes in period 3 and B has no votes in period A. I
want to map their "ideal points" (i.e. pca predicted value from votes)
into the same space. But PCA obviously dislikes missing values.
Suggestion? I'm really open to any factor analysis algorithm.
Thanks!
Nick
--
PhD Student, Political Economics
Stanford University GSB
nickeubank@gmail.com
Cell: +1 (303) 918.6528
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
