Divorce Rate

I have heard this statistic so many times (over the last 5-8 years) that, "50% of all marriages end in divorce."

I want to know who is calculating this and how they are calculating it. I know it CAN be calculated, but it seems very difficult to do properly and if done properly, may not be as bad as it sounds. Does anyone have any information on the formula or method used to make this claim? I think many people interpret this statistic to mean "if I get married there is a 50% chance that it will end in divorce" and I don't think it means that (depending of course on how it is actually calculated - or if it is actually true!)

A few thoughts about why it is difficult to calculate:
1. Do annulments count as divorces?
2. Assuming they do, are there any other ways to end a marriage other than divorce or death?
3. Does that mean 50% of marriages end by one or both parties dying? (That doesn't sound all that bad that way)
4. Marriage is a contract survival model, so if you take a slice of the population at a given time you will not be counting all the marriages that have yet to end in divorce
5. It would seem that it may be more likely to say, 50% of marriages get a divorce in the first x years or something like that, otherwise how could you compare a marriage in 2012 to a marriage in 1987?

A few thoughts about why it may not be AS bad as it sounds:
1. Consider 10 people. They all marry to create 5 marriages. 3 remain married until death. 2 end in divorce leaving 4 people single. They swap partners and create 2 new marriages. 1 ends in divorce, the other ends in death. The yield is 7 marriages created, 3 end in divorce and 4 end in death. 4 people of the 10 (40%) got a divorce while 3 marriages of the 7 ended in divorce (42%). When you start modeling people like Newt Gingrich who have had 4 divorces (?) you can see why the statistic may be misleading. Some people are much more inclined to get divorces than others and if they have multiple divorces it affects the overall metric.

A more extreme example: Consider 4 people. 2 marry and remain until death. 2 get married and divorced 5 times. There are 6 total marriages, 5 of which end in divorce. 83% of the marriages ended in divorce. Only 50% of the population experienced one or more divorces.

Alternatives:
It would seem a much more meaningful metric would be the % of people to experience one or more divorces within the first 10 years of marriage, 20 years of marriage, 30 years, 40, 50, etc. It would be a growing percentage for each timeframe and you could compare different marriage entry years at each time threshold to compare the trend.

Another method would be considering % of first marriages as a separate metric from second marriages, third and so on.