cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
JMP is taking Discovery online, April 16 and 18. Register today and join us for interactive sessions featuring popular presentation topics, networking, and discussions with the experts.
Choose Language Hide Translation Bar
gustavjung
Level III

Difference between GLM and Generalized Regression (binomial, logit)

Hi!

I have analyzed the same data using 2 approaches in JMP. 

1) Generalized Linear Model, binomial, logit

2) Generalized Regression, binomial, logit

 

As you can see the p-values are different. Can you elaborate please why?

How can I interpret these results having different p-values?

 

gustavjung_0-1586979920711.png

 

 

The initial dataset is:

gustavjung_0-1586980356772.png

 

 

Learning DOE
1 ACCEPTED SOLUTION

Accepted Solutions

Re: Difference between GLM and Generalized Regression (binomial, logit)

The GLM model was fit using MLE. The Generalized Regression model was fit using LASSO. LASSO is a form of penalized regression. It uses L-1 norm for the penalty on the estimates for variable selection. Even if a variable is selected, the estimates are affected by the penalty. The differences that found are due to the penalty. It changes the estimate and the standard error of the estimate so the Wald Chi-square also changes.

 

But I think that there might be a mistake in the way you set up the model. You need to supply both the successes and the failures for a binomial response. It appears that you only supplied the successes. Also, the number of observations is wrong. You have many more than 4 observations as demonstrated by the N Trials.

 

I would correct this model as follows. Subtract Successes from N Trials to get Failures. Stack the Successes and Failures columns. Use the column with the counts in the Freq analysis role and the label column (Successes and Failures) as the response or Y role.

 

The two factors are categorical so make sure that they use the nominal modeling type. The data table looks like this now.

 

Screen Shot 2020-04-16 at 6.39.59 AM.png

 

Select Analyze > Fit Model and fill out the launch dialog as follows:

 

Screen Shot 2020-04-16 at 6.41.01 AM.png

 

I left the nominal logistic regression choice for the fitting personality but at this point you could select GenReg or GLM with the binomial distribution and the login link function.

 

 

View solution in original post

4 REPLIES 4

Re: Difference between GLM and Generalized Regression (binomial, logit)

The GLM model was fit using MLE. The Generalized Regression model was fit using LASSO. LASSO is a form of penalized regression. It uses L-1 norm for the penalty on the estimates for variable selection. Even if a variable is selected, the estimates are affected by the penalty. The differences that found are due to the penalty. It changes the estimate and the standard error of the estimate so the Wald Chi-square also changes.

 

But I think that there might be a mistake in the way you set up the model. You need to supply both the successes and the failures for a binomial response. It appears that you only supplied the successes. Also, the number of observations is wrong. You have many more than 4 observations as demonstrated by the N Trials.

 

I would correct this model as follows. Subtract Successes from N Trials to get Failures. Stack the Successes and Failures columns. Use the column with the counts in the Freq analysis role and the label column (Successes and Failures) as the response or Y role.

 

The two factors are categorical so make sure that they use the nominal modeling type. The data table looks like this now.

 

Screen Shot 2020-04-16 at 6.39.59 AM.png

 

Select Analyze > Fit Model and fill out the launch dialog as follows:

 

Screen Shot 2020-04-16 at 6.41.01 AM.png

 

I left the nominal logistic regression choice for the fitting personality but at this point you could select GenReg or GLM with the binomial distribution and the login link function.

 

 

rlw268
Level III

Re: Difference between GLM and Generalized Regression (binomial, logit)

@Mark_Bailey Hi Mark! Can you also explain the difference to Binomial GLZ with the logit vs Just doing Nominal Logistic Regression. I understand the test the options provided are different but can you tell me why that is?

Re: Difference between GLM and Generalized Regression (binomial, logit)

I do not understand your questions.

Re: Difference between GLM and Generalized Regression (binomial, logit)

Binomial generalized linear model with a logit link function is a logistic regression model for categorical response variables with 2 outcomes (e.g. sick/healthy, male/female, purchase/no purchase).  Nominal Logistic Regression usually refers to logistic regression with 3 or more levels of the categorical response variable (e.g favorite ice cream flavor: chocolate/vanilla/strawberry/other).  Is this what you were asking?