Version 1.1  Sept 2009

- The  box-plot has been simplified : no picture whose size does not change with a a scale change.
- Limitation to 100 of the number of iterations for the robust method. 
When the robust method does not converge in less than 100 iterations, the last calculated values are given.
The non-convergence occurs when the lambda of box-cox is >10.
- Non alphabetical symbols are now taken into account in the graphics names.
- the sheets names are "report for variable_name"
	In Excel 2003 a sheet name should not 
		- contain a forbidden symbols as / \ : ?
		- have a length greater than 31.
  These restrictions were not taken into acount in Version 1.0 leading to output in the middle of the data sheet.
  Version V1.1 replaces the forbidden symbols by _ and troncates the sheet name to 31 characters.
- Cells of data columns containing string of characters are now neglected and do no longer cause a crash of RefValAdv

19/09/2009
- The tolerance in the back transformation of Box-Cox was set to 10^(-4) leading to limits equal to 0 or equal to 9999.
  It is now set to 10^(-7). This problem illustrates the need for a generalised Box-Cox transformation that will be considered in version 2.0
- The inverse Box-Cox transformation of negative results were set to 0.0. They are now calculated.
  The same problem occured for confidence intervals of the limits. This problem has been fixed.
- In some cases, the computation of lambda in box-cox transformation did not stop: the program got stuck and displayed the "hourglass" symbol.
  This happened mainly when lambda was very negative. This convergence problem has been fixed. 
  However, such values of lambda does not allow other computations to be completed.
  A solution would be to implement a generalised BC transformation.

Version 1.2  Oct 2009
17/10/2009
- When an entire column containing empty cells was selected, the data selected for the computation were data contained between the first numeric 
cell and the cell just before the  first empty cell. This problem has been fixed.
- The generalised Box-Cox transformation is now implemented. See the Guide for RefValAdv for more information on it.
This transformation should be numerically more stable than the previous one.

Version 1.3  Jan 2010
10/01/2010
- Comments explicitly cite the IFCC-CLSI C28-A3 guideline on reference intervals.
- Comments have been simplified. 
	No reference is now made to the normality test (Anderson-Darling)
	When the nonparametric method is not applicable, the robust (Horn) method with Box-Cox transformation is the default method.
	This method requires symmetry of the data distribution that is not tested in the current version.
	A new warning appears : when the max of the length of the 90% CI > 0.2 * (length of RI) the following message is printed "The 90% CI of one (or more) limit is larger than recommanded in IFCC-CLSI C28-A3."

Version 1.4  May 2010
23/05/2010
- The empirical cumulative distribution function has been changed. It was defined as 1/n \sum_{i=1}^n 1_{[X_i <= x]}. 
It is now defined as 1/(n+1) \sum_{i=1}^n 1_{[X_i <= x]} to be in accordance with  the IFCC-CLSI C28-A3 guideline.

TODO
- Implementation of a symmetry test
- Implementation of partitionning (test+method)

Version 2.0  November 2011
2/11/2011

- All graphs are now Excel 2010 compatible
- the reference interval given in the comments was a bad rounded version of the one given in the table. The rounding has been redesigned.
- A symmetry test dor the data distribution has been implemented according to (http://www.jstor.org/pss/2289611). 
The P-value of this test is computed for the robust method.
It appears in the row "p-value Anderson-Darling/ symmetry test for Robust"

6/12/2011
- The confidence interval is now available for the nonparametric method as soon as n>40
- The Anderson Darling AND the symmetry test are available for the robust method
- Graph titles are now written in 10 pts

Version 2.1  february 2012
- the format file is no longer used
- all graphical routines have been rewritten to be Excel 2010 compatible
- multiple parametric regressions are performed
	- weighted residuals are computed
	- test for lambda box-cox
- four nonparametric smoothing methods are performed (for large sample size)
- a new and simpler installation procedure has been implemented for office 2010.

to do
- implement Harris & Boyd method for categorical covariates
- add comments on the usefulness of the covariates by comparing the reference band to the reference interval obtained without covariate.
- implement specific methods individual reference intervals built from a for follow-up (mixed models)