Oh, and the paper's title is "The Robust Beauty of Improper Linear Models in Decision Making" and, in my brief skimming, had little to do with marital happiness although there is mention of that at one point.
Which is why it can be important to include interactions to your linear models, which would most probably have resolved this. This also allows better quantification of individual and joint effects than handcrafting composite variables (as far as your sample size etc allows it).