In fitglm(), logistic regression, when response is two columns matrix, the response in the table can not be recognized

age = categorical([1,1,1,1,1,2,2,2,2,2,3,3,3,3,3]');
stage = categorical([1,2,3,4,5,1,2,3,4,5,1,2,3,4,5]');
y = [11,15,2,5,18,3,3,5,0,1,4,4,3,3,2]';
m = [12,16,4,18,19,4,7,8,25,3,5,12,15,93,5]';
resp = [y,m];
data = table(resp, age, stage);
% data = table(age, stage, resp); % This works.
mdl = fitglm(data, 'resp ~ age + stage', 'Distribution', 'binomial', 'link', 'logit');

When 'resp' is at the last column in the table, this works well. But when resp is not in the last column, fitglm() doesn't work;

The error is "Response variable must be a numeric vector".

[the cyclist]

This is definitely a bug.

As best as I can tell, fitglm is failing to recognize that you have entered the formula that defines the response variable, and is instead doing the default behavior when there is no formula, which is to assume that the last variable of the table is the response variable.

I recommend submitting a bug report. And could you please post a link to the bug report here, after you have done that?

[Hang Du]

I submitted a bug report and the following is the reply:

%%%%%%%%%%%%

I understand that you are facing an issue while using "fitglm()" to perform linear regression on your data that has multi-variate response (i.e. the response variable has more than one column). Linear regression with "fitglm()" is performed between a single response variable (i.e. an n-by-1 vector) and one or more explanatory (or, independent variables).

Hence, in your use case, you could use the 'mvregress' function which is used to fit a model using multi-variate regression. The following documentation links might be useful to you:

https://www.mathworks.com/help/stats/mvregress.html

https://www.mathworks.com/help/stats/specify-the-response-and-design-matrices.html

However, I have also filed a bug report as to why the function "fitglm()" works differently when you place the response variable as the last versus when you place it elsewhere in your table.

%%%%%%%%%%%%%

[the cyclist]

Thanks for submitting the bug report, and reporting back!

Searching the bug reports for "fitglm" doesn't currently find anything, but hopefully it gets fixed in the next release.