Subscribe Bookmark RSS Feed

Strange output from Stepwise fit modelling

peterv

Community Member

Joined:

Apr 22, 2014

Hi everyone,

I’m pretty brand new to JMP and I’m trying to use it for fit modelling with stepwise regression. I was instructed to use P-value Threshold (prob to enter 0,15; prob to leave 0,2) from an expert in the area.

However my output seems kind of strange compared to the raw data I am using. To explain it a bit more the parameters in the attached pictures means following:

Andel_p7 / Andel_p8= Sales share of product 7 and 8 (Continuous, 0 to 1)
Pris_p# = Price of product # (Continuous, varies between -2 and 2 in the data)
Kampanj S07-S08 = Campaign on products 7 and 8. (Nominal, 1 for yes, 0 for no)

6075_poultry7.JPG6076_poultry8.JPG

The problem is that the Campaign parameter (Kampanj S07-S08) has a negative impact on the sales shares of product 7 and 8, which it shouldn’t since the sales share actually increase in the raw data. This is just an example among other analyses I’m doing for this project, and the problem occurs almost every time.

I have noticed that on the Campaign parameter, the {0-1} switches – and when it is {1-0} the answer seems correct. But why is this and how do I control it? When I get {1-0} instead of {0-1} the value consists of the same figures but positive instead of negative.

Can someone explain this to me? (I have read a lot of JMP documentation, but I still don’t get the hierarchical tree of linebackers etc – so I can’t interpret the output really).

Thanks a lot in advance!

/Peter