Multiple chi-square or logistic regression?

#1
Hi,

I'm new to statistics and am working on a project for work. I have read extensively and believe I am on the correct path, but would like to check and make sure:

I have one dependant variable of outcome: had reaction or did not (nominal)

I have multiple independent variables, some nominal (have disease Y/N), some ordinal (number of medications) and some measured (age).

I was planning on converting the measured to ordinal based on groups and then running a logistic regression on the analysis to determine which factors influenced the final outcome.


an alternative approach would be to run multiple chi square 2x2 (or I guess that would be fisher, depending on sample size) on each variable. I thought tthis would take a long, long time.


Am I starting on the right path? Is there a better test to get started with?

Thank you for any starting tips and advice. Also, if there are any good reading references, I would appreciate it as well. I'm not just looking for an answer, but to learn.

Thank you.

Chris
 

CB

Super Moderator
#2
I have one dependant variable of outcome: had reaction or did not (nominal)

I have multiple independent variables, some nominal (have disease Y/N), some ordinal (number of medications) and some measured (age).

I was planning on converting the measured to ordinal based on groups and then running a logistic regression on the analysis to determine which factors influenced the final outcome.
Binary logistic regression seems sensible, given that (I'm assuming) you're wanting to look at multivariate effects. A couple of comments:

-Are you sure you want to treat number of medications as an ordinal variable? This strikes me as a discrete ratio variable (3 medications is as much more than 2 medications as 2 is than 1, and there is a meaningful zero point). Remember that considering it as ordinal means treating it as nominal in the binlog reg, and cluttering up the equation with a whole bunch of possibly unnecessary coefficients. I'd treat it as a metric variable, personally.

-Why convert the metric/measured variables to groups? Leaving them as metric makes the regression equation simpler, and also means you don't lose information by reducing variability down to groups.