Does Split-platform work correctly?

[#1]

Hello,

I have been mystified by the Split-platform (Table -> Split). I think that with some conditions, splitting does not work correctly. I provide a simplified example here. Lets assume that I have a following table (also file attached):

If I open now the Split-platform (Table -> Split) and give inputs as follow:

I would expect to get following outcome:

But I am getting the following output: 

If one looks the realized output table more carefully, one can see that values originally mapped for Product A  gets mapped to Product B. This is wrong. This happens if there are non-equal amount of rows per feature to be split. With larger datasets, this is very difficult to notice and can lead to very wrong conclusions. 

Can somebody confirm this so that I not alone with these thoughts?

I am running JMP 16.2.

Janne 

[#2]

Hi Jannel,

In your example, you did not specify the Grouping variable, resulting in the odd behavior you observed. Use Product as your Group variable for accurate split.

Best,

TS

[#3]

 Hi Thierry_S,

Thanks for your suggestion. However, no matter what I use as Grouping variable, I don't get the result what I would like to have (below):

If you think that grouping is the key, you need to give more detailed step by step instructions.

thanks;

Janne

[#4]

Could be that JMP has no idea how you want to split the data, as you have "duplicates".

If the data had "unique" products feature combinations like:

With these settings

You would end up with

[#5]

I was able to get your desired result but it required 2 steps.  First, split your original table by Feature and by Product (with price as the column to be split).  Then do a stack - multiple series (2, contiguous) of the 4 columns you obtained, and you get the desired results.  I'm not sure if there is a way to do this with a single command.

[#6]

Thanks Jarmo and Dale for your suggestions. Dale's solution works but at least for me, it is not very intuitive.

The first proposal by Thierry_S to use grouping variable would be intuitive. I am sad to see what grouping does; just last data rows per group get in and all other data rows get vanished (see captures below (Unless each row used for grouping is unique as Jarmo suggested)). Generally, when I do split, most certainly my intent is not to vanish the data and thus not very happy for need to have special tricks to avoid this. Maybe there is some logic which is just beyond of my understanding

[#7]

Split is working the way it is designed to work - unfortunately that is not how you might intuitively expect it to work.

My view is that you can only reliably use it to generate the split data and label columns - the option to keep columns doesn't always work as expected, which can make it dangerous to use.

Intuitively you would expect the data in the "keep" columns to remain aligned with the split data - but JMP doesn't do that.  Split makes no attempt to maintain the alignment between the split and keep data (this is not my opinion - this is the answer from JMP Support when I raised the same issue recently).

[#8]

The work around is to (1) create subset tables for each level of 'feature' (2) for each subset table rename the data column to be the feature name (3) perform an outer join in the subset tables.

[#9]

Just want to say I fully agree with Jannel; the current functionality is not intuitive and extremely likely to mess up your data. At the very least the software should notify the user when the data does not comply to the functionality as is.

[#10]

See also discussion here: Losing Unique Values When Doing Table Split