logistic Regression Variables Basic question,
If I have a bunch of independents variables and a dependant one, would I run a separate model for each independent variable against the dependant one to find their effect(OR, prob etc) before creating a multiple best fit model?
 A: No, that is sometimes called "univariable pre-filtering" and is generally a bad idea because the effects can be confounded.
Here is an example.  In the simulation, $z$ is a classic confounder for the effect of $x$ on $y$.  Model 1 is simply y~x, model 2 is y~z and model 3 is y~x+z (the correct model for adjusting for this kind of confounding).
The effects for model 3 are accurately estimated (the effect of x is 0.2 on the log odds scale, the effect of 3 is 0.3, the intercept is -0.8).  Note that in the univariable case, the precision is too narrow and the effect estimates are too extreme.  It could also be the case the effect might be attenuated from the confounding, meaning you may ignore an important covariate because of the pre-filtering.
Not only that, but the p values obtained from such a procedure would be uninterpretable since the Wald test has no way of incorporating the uncertainty from the selection procedure into the p value.
Specify a model a priori, then fit that model.  This sort of pre-filtering is not a good way to do inference.

A: It's not so much about running a separate model for each variable, but more about identifying the important covariates that you'll use to build your model.
If you have categorical independent variables, I believe you can use a contingency table instead. However, for other types of data, it's probably best to fit them into a logistic regression one at a time and analyse the fit each time.
This resource online is quite helpful is you need a little refresher: Building and Applying Logistic Regression Models
