29505_en-US

liliconcom

Statistical comparison for multiple response columns

I am building a categorical report for a survey question which has several missing rows because not everyone who completed the survey was asked the question.  When looking at only the rate per case responding only the statistical comparisons don't look correct.  Digging further it looks like the comparison is only for rate per case and not any of the other options.  I don't see a note in the report to indicate which rate the comparison is based on, so it is confusing.  Having rate per case responding is a great feature for JMP13 and one which I already use alot, but it is less helpful if there isn't a statistical comparsion.  There is a work around where I can filter anyone who answered the question so that the value for total cases and total cases responding matches, but if I didn't know I needed to do that I could be reporting data incorrectly.

 

Attached is a file as an example and is the data where I realized that the statistics weren't changing for rate per case responding.  I realize that it is complicated because there can be up to three rows of percentages reported and the statistical comparisons are below.  There is a need to survey analysts for that comparison.  Many of our tests will have censored questions - especially now that surveys are done on-line and it is very easy to build in skip patterns. 

Re: Statistical comparison for multiple response columns

Jeff-  Two follow up questions on this subject.

 

1.  To help reduce the chance that an analyst unknowingly reports that there is a significant difference in rate per case responding, would it be possible to include in the notes for a categorical report what values the statistical comparison is based on?

 

2.  This question is more for my understanding.  There is an option to "count missing data".  It will change the counts for total case responding (and total responses).  If I select "count missing data" the rate per case and rate per case responding will agree.  If I unselect "count missing data" and there is missing data in the column then rate per case responding will be different.  Would it be an option to have "count missing data" affect the total cases instead of total cases responding?  

 

Amy

Re: Statistical comparison for multiple response columns

Thanks for your post here. You are correct that the comparisons are not based on the Rate Per Case Responding. To get that you'll need to use the Local Data Filter to include only the non-missing responses. At that point the comparisons will be on the responding cases only.

 

I have entered an idea in your Wish List for this request and noted it internally also. I can't commit to a timeframe for addressing this at this point.

 

The attached data table demonstrates using the Local Data Filter to get only the non-missing responses.