From: tgordi@buffalo.edu
Subject:NMusers] Error model
Date:Thu, 10 Apr 2003 21:10:32 -0400

Dear all,

I am trying to write an error model for a system, where 2 different 
measurements were made. One of these is log-transformed while 
the other is not. I would like to test an additive model for the log-
transformed data and a proportiona error model for the other 
measurement. The two measurements are marked by a value of 1 
or 0 in a separate column called PKone, where 1 means the first 
measurement and 0 means the second. I have tried to combine 
the two models in a couple of different ways but have not been 
successful. 

Any suggestions or references that I can look up?

Toufigh Gordi
_______________________________________________________

From: Nick Holford 
Subject:Re: [NMusers] Error model
Date:Fri, 11 Apr 2003 13:28:36 +1200

Toufigh,

Its very hard to offer advice when you give no clue to the kind of problem you are having. 

I would start with something like this:

$ERROR
;assumes log transformed prediction is Y1 and none transformed is Y2
IF (PKONE.EQ.1) THEN ; log transformed DV
  Y=LOG(Y1) + ERR(1)
ENDIF

IF (PKONE.EQ.0) THEN ; none transformed DV
  Y=Y2*(1+ERR(1))
ENDIF

Nick
_______________________________________________________

From: Leonid Gibiansky 
Subject:Re: [NMusers] Error model
Date:Fri, 11 Apr 2003 10:51:15 -0400

I think there should be two different ERR():

IF (PKONE.EQ.1) THEN ; log transformed DV
   Y=LOG(Y1) + ERR(1)
ENDIF

IF (PKONE.EQ.0) THEN ; none transformed DV
   Y=Y2*(1+ERR(2))
ENDIF

Leonid
_______________________________________________________

From:Nick Holford 
Subject: Re: [NMusers] Error model
Date:Sat, 12 Apr 2003 08:11:22 +1200

Leonid,

Thanks for pointing out my typographic error. I did not intend to use the same ERR() for both
models. I agree they should certainly be different.

Nick
_______________________________________________________