Steve Sailer's definition of race -- an extended family, inbred to some extent -- is as good as any, and better than many. But it's important to keep in mind that size makes a difference. Just as we see things going on in big lakes (noticeable tides, for example) that we don't see in small lakes, we also see important things going on in the large extended families known as races that we could get away with ignoring when looking at regular family-sized families.
One of the things that goes on at the scale of races is that selection affects gene frequencies over and above what you'd expect on the basis of genealogy. For example, if Al is Betty's uncle, then I can figure out right away the probability (1/4) that Al and Betty share a given allele (over and above the probability that it's shared by the population as a whole). On the other hand, if I know that Al and Betty belong to the same race, and know the average coefficient of inbreeding for that race (relative to some larger population, maybe humanity as a whole), then I can convert this coefficient of inbreeding into a coefficient of relatedness. But I have to be a lot more cautious in my guesses about how likely they are to share an allele by common descent. Because of selection, they may be much more likely to be similar at a locus for skin color, or lactose malabsorption than at, say, the ABO blood group locus.
One implication is that trying to apply Hamilton's theory of kin selection to make predictions about racial altruism is extremely dodgy. One of the assumptions that Hamilton and his successors make in deriving his famous equation B/C > 1/r is that there is no selection between the time a gene is present in a common ancestor and the time it expresses itself among two descendants. This assumption may be plausible enough in extended families -- probably not enough of Alan's siblings have died young while committing altruistic deeds to affect the probability of Betty getting an altruism allele. It's less plausible -- not really plausible at all -- that if Alan and Betty merely belong to the same race, and are genetically similar by virtue of a great many very distant genealogical connections, that selection hasn't had a big impact on any altruism genes.The implication is not that genes are irrelevant to large-scale altruism. Rather, the problem of altruism on large scales is like one of those physics problems where you relax some simplifying assumptions, and suddenly everything gets a lot messier, and you have to start running simulations, and so on.