The Soft Drink Excitation: minitab 6, analysing data 4 (transformations)

Описание: See the overview of the project and analysis (Part 1) here:    • The Soft Drink Excitation (Visualisation) ...  

This goes with Part 2 here:    • The Soft Drink Excitation (Visualisation) ...   which has some visualisations of how anova works, why equal variances and normal residuals matter, and problems often associated with residuals like skewness (left or right) and kurtosis (platykutic, mesokurtic, leptokurtic).

This video shows how to transform data, rerunning the glm and looking at which hypotheses we expect to be affected by the unequal variances (heteroscedastic).

Transformations are not always included in introductory data analysis courses (depending on the math background required) so if this material is too difficult check with your lecturer about the requirements for your project/course.

Common transformations are ln(X) and sqrt(X) , particularly for right-skewed data; I ended up using 1/X which worked better in this case with the kurtosis. Left skewed data is more unusual (but does happen), you can try positive power transforms, like X squared etc.

The formula for the arctan transform
input - mean
arctan( c ------------ )
stand. dev.

is from here if you want to chase up the references: http://www.faqs.org/faqs/ai-faq/neura...