Nominal logistic regression

[#1]

For my master's thesis on sleep apnoea, I would like to create a model. In a first step, I want to use nominal logistic regressions to see which parameters are significant. However, when running these regressions, I find interpreting the results difficult. Is there anyone who can help me with this?

The images below contain the outcome of
1) Nominal Logistic Fit Model of a categorical variable (knowlegde CPAP: yes/no) and a continuous variable (BMI)
2) Nominal Logistic Fir model of a categorical variable (knowledge MMA: yes/no) and a 2nd categorical variable (ESS with 4 levels)

Finally, parameter estimates of certain outcomes are marked "unstable". What does this mean?

Thanks in advance! 

[#2]

@PV1 

The JMP documentation is a great place to go to get a basic understanding of these reports. Here is the link to the Logistic Fit Report 

A quick way to access the documentation is to use the "?" in the Tool Menu. This is one of my favorite tools when I am learning something new.

When your cursor changes to the ?, hover over any report window (it will turn blue). Click on the report and a new browser window will launch, taking you to the documentation that describes that report.

[#3]

How familiar are you with logistic regression? What is difficult for you to interpret in the reports?

Regarding the indication of unstable estimates, can you share the output for such a case?

[#4]

To be honest, not very familiar. I have already tried to figure out a number of things, but just don't understand how to interpret the results.

For example, in the example with the 2 categorical variables, the categories lower than normal and normal are significant, but what exactly does this mean? That knowledge around CPAP is different between these groups? Or within the groups themselves? I'm not really sure how to interpret these results.

For the other example I have the same Issue. The result is significant. The target level is set to "No", this means that those people do'nt have any knowledge about CPAP. What does this say about the relation between those 2 parameters? 

Regarding the "instable results", I added an output.

Thanks in advance!

[#5]

Did you know with nominal logistic regression, you are modeling the log ( odds ratio ) = the linear model? The odds ratio is the ratio of the odds of Knowledge CPAP = yesversus the odds of Knowledge CPAP = no. The odds are, in turn, the ratio of the probabilities for the two outcomes. So the coefficients might not be directly interpretable. You can save the prediction formulas and see a set of new data columns with formulas. The first formulas are the linear predictor. The second formulas are the inverse log ( odds ratio ), or probabilities. The Prediction Profiler on the probabilities might be the easiest way to interpret the model.

Is there a reason you did not use all the predictors simultaneously in your model?

Can you show a plot of the distribution of Knowledge CPAP and PSS10 CAT? It might be helpful to start Fit Y by X with Knowledge CPAP in the Y role and PSS10 CAT in the X role. You should launch the Contingency platform this way. Let's see what the cell counts are in the contingency table.