EasyReg: 1997 versions

Difference between Version 1.08 (October 1997) and Version 1.07 (September 1997):
- The Fisher information matrix of the Poisson regression was wrong.
- The check on near-singularity in the Gauss-Jordan matrix inversion subroutine was too conservative. The current version aborts the Gauss-Jordan algorithm only if an overflow error is encountered.
- An artificial cross-section data generator has now been build in.
Difference between Version 1.09 (November 1997) and Version 1.08 (October 1997):
- The Tobit likelihood gradient was wrong.
- The stopping rules for the Newton iteration of the discrete variables maximum likelihood models (Logit, Probit, Poisson, Binomial Logit/Probit, Multinomial Logit) and the Tobit model have been made explicit and adjustable.
- The identifiability of the parameters in the multinomial logit model is now checked automatically.
Difference between Version 1.10 (December 1997) and Version 1.09 (November 1997):
- The KPSS tests of the (linear trend) stationarity hypothesis against the unit root (with drift) hypothesis have been included.
- A non-essential bug in the VAR innovation response module has been repaired.
- Output text that is appended during the execution now scrolls down automatically.
- Some minor cosmetic upgrades.
Difference between Version 1.11 (December 1997) Version 1.10 (December 1997):
- It is now possible to make the EasyReg version (i.e., student version with choice of levels versus research version) permanent, by clicking the new 'Versions' button in the start-up window. The version can still be changed, though.
- The window for the choice of models for discrete dependent variables has been re-arranged in order to distinguish more clearly between qualitative response models and count data models.
- A bug in the module for deleting variables from the input file has been repaired.
- The procedure for computing the suggested number of cointegrating vectors in the Johansen cointegration module has been improved.
- The procedure for saving the three-dimensional kernel regression plot in the nonparametric regression module as a bitmap file did not work properly. It does now.
- Some minor cosmetic upgrades.
Difference between Version 1.12 (December 1997) and Version 1.11 (December 1997):
- When upgrading the multinomial logit module (Version 1.08) I made a trivial error by which the multinomial logit was always aborted. This has been repaired.
- A minor cosmetic upgrade in the module for chosing a subset of observations.
Version 1.12 has been reviewed by Peter Sephton.

Acknowledgement: Most of the bugs in versions 1.07-1.12 have been detected by students in my ECON 510 Econometrics class during the Fall 1997 semester.