Multiple Regression and Confounding Factors

I am working on my thesis project which involves analyzing whether a dichotomous independent variable has an effect on a continuous dependent variable, however there are a number of potentially confounding factors that may affect this relationship. How do I go about doing a multiple regression in SPSS to determine which variables have a significant confounding effect? And how do I control for those variables? Any help would be greatly appreciated!
My <.001 cents (alert: not a statistician).

When you remove a confounding variable and rerun the model, the co-efficent of the variable of the variable-outcome relationship it confounds will change by >10%.

So for example if removing exercise changed the co-efficient (parameter estimate) of diet by >10% and the outcome was blood pressure, you could say that excerise confounds the diet-blood pressure relationship, i.e. is related to both diet and blood-pressure and so should be kept in the model so that it is controlled for and you can know the effect of diet on blood-pressure keeping in a sense, excercise constant.

edit: no idea about SPSS, sorry.
Last edited:


Phineas Packard
Include your treatment variable and then all your potential confounding variables into an ANCOVA. The effect of your treatment variable on the outcome will then be conditioned on (i.e. net of) all confounders. If you have sufficient power then there is really no need to look at whether whether they have a significant effect or not as you have included them due to theoretical concerns about there impact and you are controlling for them in order to booster causal inference.


Fortran must die
IMHO confounds are best identified by theory not by statistics :) And removed by your design not by a statistical technique. But I like all the answers above....