EI: A Program for Ecological Inference
                              Gary King
			   King@Harvard.Edu
                       http://GKing.Harvard.Edu

	    (c) Copyright 1995-2001, All rights Reserved.
            
What's new?  What's planned?  How do I make suggestions?  See below.
For automatic notification via email of all program updates, see
http://GKing.Harvard.Edu/netmind.shtml.

***************************************
* What's Planned for Future Versions? *
***************************************

-- Automating procedures to whittle down the number of problem cases
   of nonfitting
-- An automatic procedure for RxC tables

*****************************
* Any ideas or suggestions? * 
*****************************

Please email King@Harvard.Edu (for EI) or KBenoit@Harvard.Edu (for EzI)

*************************************
* What's new in this version of EI? *
*************************************

6/28/01 fixed bug in bounds2

5/17/01 added bounds2, abounds2, double, _t, and _x to eiread

12/29/00 Fixed _Ebounds error

9/18/00 Added functions R, Ri, and loglikS to eiread.

4/13/99 switched to new gauss cdfni function, a bit faster and more
accurate, added cdfbvng function and _Ecdfbvn=6 option which is much more 
accurate but slower.

12/28/98 Improved eigraph profile and profileR functions.  Fixed rare
problem with infinities showing up in likelihood function.

12/25/98 Added eiread 'checkr' option.  The library command to
intialize ei should now be "library ei;" rather than the old version,
"library ei,cml,pgraph;".  Added some programming efficiencies
(switched to Gauss loess procedure).

12/15/98 Added eigraph options "profile" and "profileR" (and
corresponding eigraph global _eigraph_pro).  These are useful for
exploring the likelihood function, or for checking for the effect of
potential cdfbvn inaccuracies.

12/4/98 Fixed a bug with _Erho

10/24/98 Increased speed and numerical stability slightly by switching to
new Gauss procs for lncdfn, lncdfn2, etc.  Changed PhiSims option in eiread

10/12/98 increased the speed of the importance sampling

9/22/98 new global for _EcdfTol.  This fixes most of the numerically
induced discontinuities in the likelihood function, while keeping cdfbvn
as precise as possible.

8/6/98 changed _EvTol's default to 1e-322 (previously it was
  1e-15). This makes EI more accurate.

7/28/98 Added new EI global _EvTol, fixed several buglets.  More
problems fit well automatically.  _EcdfTol now works for all values of
_Ecdfbvn (although _Ecdfbvn should probably not be changed from the
default any longer).

7/6/98  Added eiread option "ebounds"

6/16/98 Added new grid search option (see _Estval and the FAQ), which
can be useful for problem cases (it is also nice if you don't like the
uncertainty of how long the iterations may take; the grid search
procedure tells you at the start).  Added new EI global _EcdfTol, to
tune the new lncdfbvn function.  Added new eiread options: beta
(combination of betaB and betaW), and GEbw, GEbwa, GEwb, GEwba (gives
estimates if you know that betaB>betaW or betaB<betaW).  Added new
eigraph option "results".  Please also see the FAQ at the end of the
documentation.

5/19/98 uses Gauss' new cdfbvn2 function that makes EI run about 1.6
times faster and more accurately. _Ecdfbvn=5 is now the default
(previously it was 2). Note that this requires a version of Gauss that
includes the new procedure (if you use the HP version, it should be
3.2.39, or later).

5/4/98 Fixed HP core dump problem

3/31/98 Fixed bug in EI2

2/23/98 Confidence intervals (which are conditional on X) on FIT graph
    are no longer drawn when Z's are specified.

2/20/98 Added EIGRAPH GLOBALS _eigraph_Xhi, _eigraph_Xlo,
    _eigraph_Thi, _eigraph_Tlo, removed _eigraphlo, _eigraphhi, and
    made _eigraph_BBlo, _eigraph_BBhi, _eigraph_BWlo, eigraph_BWhi
    more precisely adjustable (especially useful for data sets, like
    mortality, that have only very small values of T).

2/18/98 Added eigraph option movieD

2/1/98 Added a much improved method of computing the variance matrix
    in troublesome cases (see _ei_vc) and added it to the default list
    as the second option.  For backwards compatibility and
    replication, the old default value is _ei_vc={1 0, 2 0.1, 2 0.05,
    3 0.1,1 0.1, 1 0.2}. Included workaround to avoid Gauss core
    dumps.  
   
1/13/97 Unified DOS/Unix version to cover differences in Gauss across
     platforms 

11/12/97 A bug was fixed that will improve variance matrix
     calculations in a larger fraction of very difficult cases.

5/22/97 EI is now faster (due to using new Gauss looping procedures)

4/8/97 Bug fixed in Palmquist's Inflation Factor computation

11/1/96 added _EselRnd global to allow randomly selected
       observations to be deleted from the estimation stage.  This is
       useful for increasing the speed of the program when operating
       with very large datasets. estimates will still be available for
       all precincts.

10/12/96 new global _EisT to allow the multivariate t density to
       substitute for the multivariate normal in importance
       sampling; also fixed a bug that affected the variance matrix
       calculation in rare situations.

9/25/96 fixed a bug that affected homogeneous precincts rarely

9/15/96 fixed a small buglet for unanimous precincts

8/20/96 eigraph's "movie" option is now much faster on DOS systems

8/11/96 Globals _GhFall, _GhDelta, _GhStart, and _GhQuad have been
        replaced by _ei_vc.  This provides the user more flexibility
        in choosing methods to compute the variance matrix of phi.
        The defaults now rarely need to be changed.

8/6/96  Fixed bug in eiread/ei2 that occured when reading betaW from
        ei2 data buffers (thanks to Mike Alvarez for finding this
        one!).  Also fixed bug that occasionally made it seem like the
        variance matrix computation was stuck.

7/23/96 Added _Eres support for ei2, and fixed a likelihood function
        buglet for homogeneous precincts

7/9/96  Added facilities to do 2xC tables for C>=2.  See ei2() and new
        options, and newly modified options, under eiread and eigraph.
        These methods appear in versions of the manuscript only after
        7/1/96.

7/9/96  changed the graphs included in tomogS

7/5/96  Added a more robust version of the proc called by _Ecdfbvn=2.
        _Ecdfbvn=1 will often be faster now, although often less
        accurate.