The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Topics » Wealth Index » Not positive definite correlation matrix
Not positive definite correlation matrix [message #24647] Tue, 14 June 2022 03:32 Go to next message
kptanalgo is currently offline  kptanalgo
Messages: 2
Registered: January 2018
Member
Hi!

In the Nepal Further Analysis of their 2006 DHS, technical note #3 indicates removing problematic variables from the PCR because the correlation matrix turns out to be "not positive definite". May we get some guidance as to how said problematic variables are identified?

Thank you in advance.

Klaren

Reference: https://dhsprogram.com/pubs/pdf/FA53/FA53.pdf

[Updated on: Tue, 14 June 2022 03:35]

Report message to a moderator

Re: Not positive definite correlation matrix [message #24656 is a reply to message #24647] Wed, 15 June 2022 09:01 Go to previous message
Janet-DHS is currently offline  Janet-DHS
Messages: 666
Registered: April 2022
Senior Member
Following is a response from DHS Research & Data Analysis Director, Tom Pullum:

"Not positive definite" means that a matrix cannot be inverted because the components are too highly correlated with one another. It's equivalent to "multi-collinearity". When this happens during the construction of the wealth index with principal components analysis, I believe it is basically a judgment call how to reduce the number of variables. I don't believe there is any formal guidance on what to keep and what to drop--and I don't think it would make much difference, because multi-collinearity just means that the variables are duplicating each other.
Previous Topic: Problem with recreating Asset quintiles
Next Topic: Comparative wealth Index
Goto Forum:
  


Current Time: Thu Mar 28 17:06:31 Coordinated Universal Time 2024