You need to look at it in a Bayesian sense. Consider all the possibilities and then look at the outcomes they produce. BB BG GB GG These are all equally like

You need to look at it in a Bayesian sense. Consider all the possibilities and then look at the outcomes they produce. BB BG GB GG These are all equally likely, right? They each have P = 1/4. If we took 1000 fathers who had two children each, and lined them up, we would expect 1/4 of them to have children matching each of the above pairings. That means 250 of each. With me so far? Now, we know that we are dealing with a father who has at least one son. If we went along the line and asked each father 'Do you have at least one son: Yes/No?', then 750 would answer yes, and 250 (the GG fathers) would answer no. If a random father comes up to us and says 'I have at least one son', we know that he is from those 750 - we have selected a subset to deal with. Of those 750, only 250 have a second son, so the probably is 250/750 or 1/3.