Faq what are reduced T^2 and Q Statistics: Difference between revisions

From Eigenvector Research Documentation Wiki
Jump to navigation Jump to search
imported>Lyle
imported>Lyle
Line 7: Line 7:
Hotelling's T<sup>2</sup> (T-squared) and Q residuals (sum squared residuals) are often reported in units which are sensitive to the total number of variables in the data, and the number of components and particular preprocessing used for a model. As a result, comparison of values reported by different models, or setting a standard alarm level for all models requires normalizing the statistics.  
Hotelling's T<sup>2</sup> (T-squared) and Q residuals (sum squared residuals) are often reported in units which are sensitive to the total number of variables in the data, and the number of components and particular preprocessing used for a model. As a result, comparison of values reported by different models, or setting a standard alarm level for all models requires normalizing the statistics.  


A common way to normalize the statistics is to divide by a confidence limit calculated from each model's particular data and parameters (In PLS_Toolbox and Solo, the default confidence limit is at the 95% level.) The '''Reduced Q''' and '''Reduced T<sup>2</sup>''' are simply the statistic divided by the corresponding confidence limit.  
A common way to normalize the statistics is to divide by a confidence limit calculated from each model's particular data and parameters (In PLS_Toolbox and Solo, the ''default'' confidence limit is at the 95% level.) The '''Reduced Q''' and '''Reduced T<sup>2</sup>''' are simply the statistic divided by the corresponding confidence limit.  


'''PLS_Toolbox'''
'''PLS_Toolbox'''

Revision as of 13:45, 8 January 2019

Issue:

What are the "Reduced" T2 and Q Statistics?

Possible Solutions:

Hotelling's T2 (T-squared) and Q residuals (sum squared residuals) are often reported in units which are sensitive to the total number of variables in the data, and the number of components and particular preprocessing used for a model. As a result, comparison of values reported by different models, or setting a standard alarm level for all models requires normalizing the statistics.

A common way to normalize the statistics is to divide by a confidence limit calculated from each model's particular data and parameters (In PLS_Toolbox and Solo, the default confidence limit is at the 95% level.) The Reduced Q and Reduced T2 are simply the statistic divided by the corresponding confidence limit.

PLS_Toolbox

In PLS_Toolbox and Matlab, you can view the exact value used to normalize by viewing the following fields in a model:

>> q_limit = model.detail.reslim{1} 
>> t2_limit = model.detail.tsqlim{1}


Still having problems? Please contact our helpdesk at helpdesk@eigenvector.com