I received the following comment from Betsey Stevenson, the creator of the divorce calculator (see my post below). I haven't had time to digest this yet, but thought I'd put it up while I'm thinking about it.
I fear that the divorce calculator whirled through the blogosphere so quickly that what it actually is intended to measure got lost. The criteria that I use are not based on a theory of what predicts divorce and I am not arguing that these are the most important predictors of divorce.
What the calculator does is uses large scale data collected by the US Census to calculate what percent of people actually divorce. We can do this for all people in the US currently (i.e. what percent of all adults have experienced a divorce) or we can do it by criteria (what percent of college educated people who married in their 30s and during the last 20 years had divorced within 10 years). Either way, the numbers are just facts. While facts can be a whole lot simpler than the theories that social scientists come up with to understand divorce, I believe that facts have their place and it can be informative to know them.
Unfortunately large, representative datasets containing information on marital history rarely (ever?) contain information on the many important correlates of divorce. Indeed, as I argue in much of my research on divorce, understanding the causal implications of divorce is difficult precisely because the factors predicting divorce are typically not collected in datasets making it impossible for us to control for them (i.e. having an alcoholic spouse likely increases the chances of divorce and is probably bad for the kids education) and also difficult for us to even know the facts regarding their association with divorce rates.