What means "out of estimation"?

Price estimations are based on a stepwise linear regression with ANOVA.
That method demands the respect of mathematical assumptions that are tested (a posteriori) with a normality test for residuals.
These tests (D'Agostino and Anderson-Darling) have only become successfull after the elimination of some wines, adverse to mathematics, they are the ones classified as "Out of estimation".

Why two types of estimation?

One of my main goals is to offer as much information as possible and to do so, I prefer the classifications involving as many wines as possible.
The first estimation is about the average score (in fact the score and its two, three and four powers), and it takes into account 405 wines in 2005.
A second one is about the whole panel of tasters,and makes it possible to assess the influence of each taster on the price. It gives a better "R2" but has two flaws:
- it is about a smaller number of wines (188 wines in 2005) i.e. those tasted by ALL significant tasters
- apart from year 2005, they do not satisfy the mathematical assumptions of a linear regression.


This estimation is based on one sample of wine (or tasters), results may change with another sample.
For those who are brave enough
Regression:1982 1990 2003 2004 2005 2006 2007 2008 2009 2009 2010 2011 2012