Statalist The Stata Listserver


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: Re: orthogonal regression


From   "Nick Cox" <[email protected]>
To   <[email protected]>
Subject   st: RE: Re: orthogonal regression
Date   Thu, 14 Dec 2006 15:06:45 -0000

I think Kit is right. As some people may recognise, the 
name -sdline- stems from the term "SD line" used in 
a popular introductory text by David Freedman and friends
from Berkeley, but many other names exist in the literature,
including uglier ones such as "reduced major axis", and 
the beast has been re-invented many times. 

It boils down to a slope that is 

	sign(corr(y,x)) * sd(y) / sd(x) 

and an intercept that is determined by a constraint that the 
line passes through the intersection of the means. That's 
all there is to it; thus there are no handles for tweaking 
estimation according to various extra assumptions. 

Incidentally, the original question did ask for programs
that fitted lines, and it wasn't originally evident that
the real problem is not line-fitting at all, but higher
dimensional. 

Finally, some readers may be interested to know that
Wiley have just re-issued Wayne Fuller's "Measurement 
error models" in paperback. It is still not cheap, but 
waiting for the movie will not work, and it is a very 
solid treatment. Note, as said, a re-issue, and not
a revision. 

Nick 
[email protected] 

Kit Baum
 
> Instrumental variables procedures are often used to deal with  
> problems of errors-in-variables, and a common instrument in 
> that case  
> is the rank of the regressor. Of course it is not 'independent' of  
> the regressor itself--if it was it would make a lousy instrument.
> 
> If you have a model in which y1 and y2 are jointly determined, you  
> need some variables not in the equation to estimate the relationship  
> in which they both appear. That could be instrumental variables or a  
> k-class estimator such as LIML. IV is just a special case of the k- 
> class where k==1. But I do not think that consistent estimation of  
> the "orthogonal regression" (k ~= 1) model can be performed (by,  
> e.g., Nick Cox's -sdline-) if the regressor is correlated with the  
> error.

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index