CHANGES IN TSFA VERSION 2006.x-x

Changes (Possibly) Affecting Users

    o  small changes related to changes in tframe were necessary.
    
    o  demos  SOM-2005.R  and WP-2006-3.R were added.

    o  the number of parameters in the calculation of fit statistics for the
       saturated model was fix for the possibility that the Lederman bound may
       not be an integer.
 
    o  object structure was rationalized. TSFestModel is now gone, and the
       estimation sub-object is an optional $stats element in the FAmodel 
       object. One important aspect of this is that $model$estConverged and
       $model$rotationtConverged are now $stats$estConverged and 
       $stats$rotationtConverged. FAmodel does not have factors, but fFAmodel 
       extends it by including factors, and TSFmodel extends fFAmodel by 
       including time associated pieces (tframe and diff). The fFAmodel is 
       experimental and not extensively used yet, but there is a fair amount
       code already committed to having factors in the TSFmodel as an
       extension of an FAmodel.

    o  argument methodArgs changed to rotationArgs in several functions, for
       improved clarity.

    o  added estimation function estFAmodel using Sigma rather than data (and
       so does not return factor predictions). This also provides estimation 
       for non-time-series factor models (FAmodel objects).

    o  estTSF.ML  re-organized to call estFAmodel using Sigma then add 
       factor predictions.
     
    o  changed estTSF.ML to call rotation by name rather than GPFoblq, so
       othogonal rotations are now possible. (See rotation 
       names in GPArotations.)

    o  changed permusign to have default Phi of NULL, rather than I, which is 
       more consistent with GPForth for orthogonal rotations.

    o eps and maxit were not being passed from estTSF.ML to GPFoblq.
 
    o The Bartlett prediction calculation in estTSF.ML was converted to use a 
      more robust computation, which works better when there are more indicators
      and possible problems of near singular computations.

    o Methods for tfplot were changed so that adding the Title 
      can be turned off using a global option setting PlotTitles=FALSE.
 
    o is.R() and some code related to possibly running in Splus was
      removed. (The package has never been tested in Splus.)

    o fixed a problem with tsfa tfplot method for monte carlo class.