Statalist The Stata Listserver

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

st: Re: positive definite matrices

From   "Rodrigo A. Alfaro" <[email protected]>
To   <[email protected]>
Subject   st: Re: positive definite matrices
Date   Wed, 20 Sep 2006 15:10:48 -0400


A is positive definite if for any vector z then z'Az>0... quadratic form.
Liberal translation: a positive definite refers in general to the variance 
should be positive. Therefore, you have a negative variance somewhere.

In your case, the command tries to get the correlation using all the
available information... because you have missing something the
correlations that you get do not meet the condition that the var-cov
is positive definite. 

Solutions: (1) use casewise, from the help file "Specifying casewise 
ensures that the estimated covariance matrix will be of full rank and 
be positive definite." (2) fill some missing data with -ipolate- or 
-impute-, (3) drop the too-much missings variables, (4) work with 
multiple-imputation datasets... using -ice- or some other package.


----- Original Message ----- 
From: "Jason Yackee" <[email protected]>
To: <[email protected]>
Sent: Wednesday, September 20, 2006 2:46 PM
Subject: st: positive definite matrices

I am trying to run -xtpcse, pairwise- on unbalanced pooled cross
sectional time series data, with no single period common to all panels.

Depending on the model I can occasionally get the routine to work by not
including panel and/or time dummies.  But usually the routine spits out
error message r(506), which in long form is explained thus:

[P]     error . . . . . . . . . . . . . . . . . . . . . . . .  Return
code 506
        matrix not positive definite;
        You have issued a matrix command that can only be performed on a
        positive definite matrix and your matrix is not positive

I know very little about matrix algebra.  Would someone be willing to
substantively "translate" the error message?  I would love to have a
more intuitive sense of what my problem is, and how I might go about
fixing it. 

Jason Webb Yackee, PhD Candidate; J.D.
Fellow, Gould School of Law
University of Southern California
[email protected]
Cell: 919-358-3040

*   For searches and help try:

*   For searches and help try:

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