pspp-dev
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: GLM factorial analysis


From: John Darrington
Subject: Re: GLM factorial analysis
Date: Sat, 13 Aug 2011 17:28:48 +0000
User-agent: Mutt/1.5.18 (2008-05-17)

On Fri, Aug 12, 2011 at 05:37:20PM -0400, Jason Stover wrote:
     
     I looked into this problem.  Briefly stated, this happens because the
     meaning of "type 3 sums of squares" changes in the presence of
     interactions.  The fix should be simple enough, but requires get_ssq
     to know which variables, if any, went into an interaction. Is there an
     easy way to do this?

Yes.

struct interaction which is defined in src/math/interaction.h is a transparent
struct defined as follows:

struct interaction
{
  size_t n_vars;
  const struct variable **vars;
};


So it's easy to add a function which checks if a variable is a member of an 
interaction. 

Perhaps in the future, we should change the implementation of interactions to 
use a hash
table, so that looking up a variable can be done in constant time.  However for 
now, a 
linear search should do.

J'
     
     -Jason
     
     > 
     > J'
     > 
     > On Mon, Jul 25, 2011 at 02:29:07PM -0400, Jason Stover wrote:
     >      On Mon, Jul 25, 2011 at 09:11:21AM +0000, John Darrington wrote:
     >      > The sum of squares for factors with interactions are correct, and 
all the degrees of
     >      > freedome are correct.  Unfortunately the ssq for factors without 
interactions are 
     >      > wrong (my experiments showed 3.0 and 3.5 times too high) if the 
analysis also 
     >      > includes interactions between those factors.
     >      > 
     >      > However, if the analysis is run on the individual variables, 
without the interactions,
     >      > then the results are correct.  
     >      > 
     >      > So I think the get_ssq function needs some more work.
     >      
     >      Can you send syntax and output that shows the problems?
     > 
     > -- 
     > PGP Public key ID: 1024D/2DE827B3 
     > fingerprint = 8797 A26D 0854 2EAB 0285  A290 8A67 719C 2DE8 27B3
     > See http://pgp.mit.edu or any PGP keyserver for public key.
     > 
     
     
     

-- 
PGP Public key ID: 1024D/2DE827B3 
fingerprint = 8797 A26D 0854 2EAB 0285  A290 8A67 719C 2DE8 27B3
See http://pgp.mit.edu or any PGP keyserver for public key.

Attachment: signature.asc
Description: Digital signature


reply via email to

[Prev in Thread] Current Thread [Next in Thread]