topic Re: Could Fit model test dataset with missing data ? But a lot Nonestimatable effect and incomplete plots in Discussions

Could Fit model test dataset with missing data ? But a lot Nonestimatable effect and incomplete plots

Diana — Fri, 09 Jun 2023 00:30:48 GMT

Hello all,

I'm new to JMP (JMP15) and only familiar with limited analysis.

	position	instant	24h	48h	72h	100h
test A	A, B	Y	Y	Y	Y	Y
test B	A, B	Y	Y	Y	Y	Y
test C	A	Y	Y	Y	Y	NA

To simply explain my dataset, with me have data at most time point. only test C didn't have 100h data. Test A and B had tested on both A&B position. Test C tested on A only.

The output from Fit model contains a lot Non-estimable. no results for all the items related to timepoint or position. Table and plot. How should I set in JMP to analyze dataset with missing data smoothly (should be achievable I believe)?

Thank you!

Re: Could Fit model test dataset with missing data ? But a lot Nonestimatable effect and incomplete plots

Phil_Kay — Fri, 19 Mar 2021 09:38:00 GMT

It would help if you can attach an anonymised version of your data. There are different reasons why some of the model effects might be NonEstimable. It is a complex model that you are fitting so it might be that you don't have enough observations. Or it could be because of aliasing. It is difficult to know from the information that you have provided.

Re: Could Fit model test dataset with missing data ? But a lot Nonestimatable effect and incomplete plots

Diana — Fri, 26 Mar 2021 09:46:11 GMT

Hi Kay,

There are over 2000 lines in the original dataset. I extract ~200 lines and the issue is still here.

Thank you for your attention and help.

Cheers,

Diana

Re: Could Fit model test dataset with missing data ? But a lot Nonestimatable effect and incomplete plots

Dan_Obermiller — Fri, 26 Mar 2021 12:22:27 GMT

Based on the small dataset you included, I believe the issue is Site2. Not all of the time points were observed AND not all of the positions were observed with Site2. Your model includes the Position*Site interaction and the Site*Time Point interaction. You cannot estimate all of the levels of those interactions because you do not have enough data (specifically, the missing combinations with Site2).

If you hide and exclude the Site2 data from the analysis, the non-estimable terms go away. Alternatively, you could remove the Position*Site interaction and the Site*Time Point interaction from the model.