This is an interesting question. Many people respond, and many are wise enough to remain silent. Those who remain silent know that it all comes down to circumstance, and they couldn't possibly know how they might react if any small variable of the situation were different, etc..
Richard Dawkins' book, The Selfish Gene, explains how evolution answers this question. Though the book is titled 'Selfish Gene', it actually talks a lot more about altruism. It explains how your relation to the person in danger directly effects your decision to save them, weighed against the likelihood of your death. For example:
If a man you don't know falls into the Niagara River and is rushing towards the Falls to his death, it's very unlikely that you will jump in and try to save him.
Replace the man with your child, parent or sibling (50% gene relation), and you're very likely to jump into the river (50% more likely), despite the high likelihood of your death.
This is an integral function of gene propagation, in that you're more likely to sacrifice your life for others that have your genes (you're statistically 25% less likely to jump in after a first cousin or a grandparent) than others who don't.
Regardless of all this, if the likelihood of your death is 100%, you're almost certainly not going save the family member. The only mechanism of evolution that is stronger than gene propagation is survival. That being said, in an irrational state (due to a sudden onset of violence or danger), the 'fight-or-flight' reflex might cause some to take an offensive position, in turn protecting their loved ones. It's also not unreasonable to imagine a mother doing just about anything to protect a child. But keep in mind, in neither of these situations have the 'protectors' rationally calculated a 100% chance of death-- they either think they can win or at least survive.