Working with Imputed Data in JMP - Elastic Net

Question

Hi Everyone,

I am working with a dataset that went through multiple imputation in stata. I am trying to do elastic net regression on the data set in JMP because Stata does not seem to support elastic net in it's multiple imputation estimate command.

I was wondering, does JMP have any way of identifying the data as MI data, or does it even matter? The mi estimate command in stata conducts a combined analysis and pooling process. Is it appropriate to just run the raw imputed data set through elastic net without the analysis and pooling phase? Appreciate any help and any insight folks can provide.

PoissonCivet938 · Answer

I have discovered that JMP can consider the name of the imputation variable as such. On the webpage https://www.jmp.com/support/help/en/17.2/?os=win&source=application#page/jmp/launch-the-explore-missing-values-platform.shtml# for handling multiple imputation data, there is a validation window which accepts the SPSS variable 'imputation_' as a valid input and produces a missing data report.

Could this validation window for the imputation_ variable please be added to other types of analyses, for instance, the multivariate analyses types such as MANOVA with repeated measures? I work at an institute that has mainly longitudinal data so there is always attrition that needs to be compensated with multiple imputations. Thus, it would be just great if JMP could solve this problem that otherwise only R appears to be able to muster.

PoissonCivet938 · Answer

Hi I have the same problem. My repeated measures MANOVA is being calculated by JMP but only as if the sample has multiplied via the 100 imputed dataset in one spreadsheet, rather than for the pooled data set. It would be excellent if this could be implemented - in fact I think it is a must - and a recorded webinar could be made available.

LauraCS · Answer

I don't have a solution for the Elastic Net but @BTD, you might consider using full information maximum likelihood (FIML) instead of multiple imputation (MI) for your multivariate analysis. FIML is the default estimator in the Structural Equation Models platform in JMP Pro when there are missing data, and given that you have repeated measures data, you might want to look at this example of a latent growth curve model (makes less stringent assumptions about the data than RM ANOVA).

As a side note, this article compares FIML and MI and affirms that,

"Our study confirmed well-known knowledge that the two estimators tend to yield essentially equivalent results to each other and to those from analysis of complete data when the postulated model is correctly specified."

HTH,

~Laura