Yes, it's entirely possible that the divorce flowed backward through time and wiped the smiles off the future divorcees faces while they were posing for yearbook photos.
Thank you for the knee-jerk causation/correlation comment.
I think the person who complained about the correlation/causation comment was right, whether it was knee-jerk or not. You don't need time travel to see what the problem could be.
Say a typical person may or may not be the smiley type, and say that means nothing at all about their chance of marital bliss. But say oh, I don't know, all migraine sufferers frown, because they are in pain. And say that being a migraine sufferer puts a terrible strain on your marriage (look, it's just an example) and so you are more likely to divorce. Now if you look for the divorce rate of people who frown, it will be much higher than that of people who smile. But that has no predictive value for an individual who frowns, because (in this hypothetical example) it's not the frown that is the issue - it is whether they get migrains. If there is no predictive value, there is no real science.
The other fundamental problem with presenting research in this way is that it is very easy to pick two percentile ranges and find a difference even in random data. The question that one should ask is - was there a dose response. In other words, if you plotted divorce rate along this so called "smile strength" axis, would there be a linear correlation (the weaker the smile the higher the divorce rate)? Or is your conclusion based on the outliers?
Thank you for the knee-jerk causation/correlation comment.