Stepwise regression "match"

[#1]

Hello,

I have a stepwise regression and I have the following results when I save the model in the database :

What does that mean ? Indeed, I have several coefficients for the same modality in a variable and it makes me "match". For example the modality "39" has 2 coefficients, "1" and "-1"
What are the "match"?
Thank you for your help

[#2]

I'll take a crack at this, but it would help if you attach a snapshot of the data file.  It appears that your variable Sem. fab was grouped by JMP into two bins - some of the values were assigned +1 and others -1.  I believe the stepwise model will group a nominal variable, according to groups that work "best" in the model, so this variable appears to have been split into two groups.  That is fine, if Sem. fab is really a nominal variable (I'm guessing it refers to two designated groups of manufactured parts, for example).  Since it has numeric values, I'm not sure - and if it really is a continuous measure, then you should change the data type and model it that way.

Assuming it is a nominal variable, then you model is simply estimating 2 constant values - one for one of the groups and another for the other group.  So, your model is predicting the intercept +/- and constant value, depending on which group it is in.  This suggests to me that a stepwise regression is not the most useful way to look at this data.  I'd suggest doing some more exploratory graphical analysis, since there appear to be 2 groups of observations, each with a predicted value.  Are there no other explanatory variables for you to model?  With 2 predicted continuous outcomes, and 1 nominal variable (identifying the 2 groups of observations), a simple t test may suffice.

[#3]

Here is a sample file I want to make a logistic regression model to predict the "Quality" variable representing the quality of cheeses.

In the example, I only put 6 variables but actually there is much more.

Before making my logistic regression model, I wanted to make a stepwise regression in order to select the best variables to integrate them into my model

However, I don't understand why it separates me into several groups the Variable Year, Month and Week

Indeed, these can not be continuous digital variables, they are in the correct format: ordinal digital

Thanks for your help

[#4]

I'd suggest using Predictor Screening (under the Screening platform) to explore the various predictors.  I believe Stepwise will automatically group "similar" nominal variable values together (when they have similar impacts on the response variable), which is what I think you are seeing.  That can be useful when a nominal variable has a lot of levels, and there are groups of levels that have similar impacts.  But, from my experience, it makes it harder to see the importance of the different potential explanatory variables, so I'd start with predictor screening.