Interpreting Partial Least Square Results

E

Estelita

Hello everyone,

I run a Partial Least Square analysis with 15 predictors in Minitab 15 and would appreciate some help on how to interpret the results.
Under "Model Selection and Validation", the Minitab output window gives a list of 10 "Components" that the PLS analysis has selected as best predictors. 'Components" are identified with number 1,2 etc and I do not know which predictors they are equivalent to. I do know that predictor 1 is NOT equivalent to component 1... So basically my question is how to relate components and predictors? What statistics (p-value? coefficient of regression? etc) does Minitab uses to select components?
Thanks for reading my post and for helping me. Please let me know if you need more details.

Regards,
 

Miner

Forum Moderator
Leader
Admin
I am not an expert on Partial Least Squares (PLS), but I read Minitab's Help info on it. It described Components as being similar to the Factors in Factor Analysis of which I am familiar. It also described the purpose of PLS as condensing a large number of variables, some of which are highly correlated with each other to a reduced number of Components. Again, this is similar to Factor Analysis.

In Factor Analysis, Factors combine multiple variables that are correlated with each other into a single construct or factor that you must name. An example that I have seen took a study to "define" feebleness where many factors , such as gripping strength, arm movement, range of motion, etc. were measured. The study found that many of these measures were correlated and could be reduced to several macro-variables called factors. For example, gripping strength, arm strength, arm range of motion were grouped into a single factor named upper body.

Components appear to be the same concept. Highly correlated predictors are consolidated into a macro-predictor called a component. You must identify what they have in common and name the component.

Minitab uses Loadings to select components. Again, this is similar to Factor analysis. Look at the loading graph. Long lines have high loads. Lines that are very near each other are correlated.

That is as much as I can help you with.
 
Top Bottom