Thesis Issue - Confounding in fully adjusted model

Hi, my thesis is due soon and I am having a problem with confounding.

Background: I am testing the relationship between parity (independent) and BMI (dependent) with linear regression. The unadjusted relationship is significant (p<0.001). When I do a multiple regression analysis with parity and monthly household income (which also has a significant unadjusted association with BMI) as the independent variables and BMI as the dependent, parity is no longer significant but income remains significant. Therefore, is income a confounder in the relationship between parity and BMI? Or if no relationship exists between parity and BMI after adjustment, is it not confounding? The unstandardised B coefficient of parity changed by more than 10% from the unadjusted to adjusted models.

Any help is greatly appreciated as I am under time pressure. Thank you!
Last edited:


Less is more. Stay pure. Stay poor.
You are getting trapped in the old guard of confounder detection. The best approach is to map out the structural relationship between the three variables given temporality and known theory.

The results can be a multitude of possible relations, but you need to use your best knowledge to presume the relationships. The following are possible scenarios, however it could be more tricky than this if income effects parity which effects income. Right? Please provide what you think the relationship looks like and we can continue the discussion on how to identify the effects. P.S., what is the listed objective in your thesis for this question? What are you hoping to answer?



Less is more. Stay pure. Stay poor.
P.S., Should age be in your model. In that it impacts parity, income, and BMI (directly and indirectly)?

Also, what is your sample size?

How is parity formatted and defined? How can it be standardized effectively?