Alternative to multiple regression with non-independent observations?

Can you pool the non-independent values? Is there an interest in looking at father vs mother, or can you call them both a repeated measure of "parent"? How is your data categorized? Can you take a mean of the non-independent values and use the mean as a pooled response in the overall model?

What is the hypothesis of your model? If you go back to the questions you are trying to answer then the modeling dilemmas sometimes answer themselves.