Predictors in Logistic regression, SPSS

#1
Hello. I have the problem I can’t deal with. Maybe you can help me? You see, I have 2 samples in retrospective cohort study – cases (include 29 patients with an outcome) and control (328 without an outcome). As I’ve read due to the small amount of cases in logistic regression I can include maximum 3 predictors. In order to define them, I’ve compared 2 independent groups and want to check 6 possible predictors. But how can I do this? Should I first include in LR only categorical variables and in the second – only continuous. Or should I include all 6 predictors in one logistic regression analyse and ignore the rule? I’m waiting for your answer. Please, help me to solve this problem…

Best, Julia
 

Karabiner

TS Contributor
#2
If you ignore the rule not to include too many predictors, then your results
probably won't be generalizable.
The usual recommendation is to base your decision which predictors to use
on theoretical considerations.

Kind regards

K.
 
#3
So, the only way is to choose 3 predictors out of 6 and do the logistic regression? But what if I don't know what predictors are more important? As I understand I should choose 3 predictors which have the strongest correlation according to the results of correlation analyse? Am I right?

Thank you for your reply!

Best, Julia
 

Karabiner

TS Contributor
#4
Unfortunately, I do not know what the objective of your study is, or what the
context of the study is, or what will be done with the results, or how strong the
theoretical context is. So I am unable to say what criteria "should" be used.
But performing 6 univariate correlation analyses and then using the 3 predictors
with the strongest associations would not solve the underlying problem
that 6 predictors are a bit too many for just 29 cases.

Kind regards


K.