24294_en-US

liliconcom

Incorrect Imputation of Power Terms when “Standardize X” is chosen

In PLS:

Saved imputed value for X*X is not Mean(X)*Mean(X) nor Mean(X*X) but something else.
Inconsistency between imputed values for main effects and higher order terms.

mjoner, varbanov.ar

Re: Incorrect Imputation of Power Terms when “Standardize X” is chosen - Status changed to: Open for VotingRe: Incorrect Imputation of Power Terms when “Standardize X” is chosen

I agree with Michael! This issue needs further discussion and hopefully an action from JMP. It does not make sense the saved data after imputation to have values on different scales within the same column. This is simply wrong! This could have strong implications on business decisions. I am not sure why JMP takes a position not to make a change here to correct the issue. 

 

Jeff,

Please would you let us know if there is a way to make a progress on this issue? 

Thanks,

Alex

Re: Incorrect Imputation of Power Terms when “Standardize X” is chosen

OK. This inconsistent behavior (whether it comes from SIMCA or not) in how higher order terms are saved in the imputation table is causing quite a bit more trouble than I thought it would in my attempts to script around this. It would really be nice to address this differently. Can we get into a deeper discussion on this?

Re: Incorrect Imputation of Power Terms when “Standardize X” is chosen<meta http-equiv="Content-Type" content="text/html; charset=UTF-8" />

Correct.

Just to be absolutely clear here. If I fit the model (SAS notation here for brevity) Y = X X*X, and save the imputation, I get:

(when standardize X is checked): for X, imputed value = Mean(X); for X*X, imputed value = Mean(StdX*StdX)

(when standardize X is not checked): for X, imputed value = Mean(X); for X*X, imputed value = Mean(X*X)

So Standardize X doesn't affect the imputations of the original variables, but it does affect the imputations of any cross-products involving the variables.

As long as this remains unchanged, no problem, I know how to get what I need.

Thanks.

Re: Incorrect Imputation of Power Terms when “Standardize X” is chosen

To be clear, JMP standardizes the X before squaring it.  Therefore, the 0.9923 is the mean of StdX * StdX .  If you uncheck the Standardize X examine the values for the imputed X and X^2 you will see the values I think you were expecting such as 2.3437.

Re: Incorrect Imputation of Power Terms when “Standardize X” is chosen

There is no change planned here.

The current implementation matches Simca which is what many customers are coming from.

Re: Incorrect Imputation of Power Terms when “Standardize X” is chosen

Checking on this one too. No apparent change coming in JMP13. That should be fine, but if a change is coming we do have scripts that depend on this.