|
Activity Number:
|
390
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Social Statistics Section
|
| Abstract - #309866 |
|
Title:
|
How Many Factors? A Strategy for Identifying Latent Structure in Factor Analysis
|
|
Author(s):
|
Mitchell Harris*+ and Michael K. Lauritzen and Landon Poppleton and Robert Bubb and Bruce Brown
|
|
Companies:
|
Brigham Young University and Brigham Young University and Brigham Young University and Brigham Young University and Brigham Young University
|
|
Address:
|
1001 SWKT, Provo, UT, 84602,
|
|
Keywords:
|
factor analysis ; dimensionality ; latent structure
|
|
Abstract:
|
Factor analysis has now been used for over a century. One of the persistent questions in using factor analysis has been the problem of determining when to quit factoring--that is, how does one know how many underlying factors should be used to capture the essence of the observed surface variables. A number of solutions to the number-of-factors problem have been proposed: the eigenvalues criterion, the skree plot method, etc. A Monte Carlo simulation method is used to create of a variety of surface variable structures from systematically varying underlying factor structures. This simulation has the advantage of evaluating the many and varied strategies for identifying the number of factors against the actual known number of factors in each simulation. A new strategy, based upon the central limit theorem, is shown to be most effective in identifying the actual number of factors.
|
