Skip to content

factor analysis - constraints

2 messages · Robert Ruser, Peter Dalgaard

#
Dear R Users,
I'm wondering how is it possible to get |factor loadings| <1 in factor
analysis model (factanal) performing maximum-likelihood estimation. Is
it caused by some constraints? If so, what kind of constraints are
placed and when (during the estimation or after)?

Robert
#
On Apr 19, 2011, at 13:58 , Robert Ruser wrote:

            
factanal() works with the correlation matrix, so you cannot have a valid factor model otherwise. Non-negative uniquenesses (aka specific variances) are ensured by constraints in the estimation algorithm. You do not infrequently see cases where the uniqueness is being set to its limit of 0.005 (Heywood cases)