I read the paper of these political scientists, and it is a rather grotesque example of what is called in the statistical trade as "data mining." What is data mining? It is creating a model based on past data, until one gets the best "fit," and then projecting it into the future, without some explanatory theory that suggests the past will be the future because the variables have some theoretical continuing validity. This model is so Rube Goldberg, that it is laughable. Sure, the more lopsided the generic polls, the more seats one party will get as an initial matter, but doing regression analyses, with a complex formula, with only 15 data points, is well, when it comes to voting behavior, just not very persuasive. Yes, there are too few data points for this puppy, particularly since the dynamics of each election has so many rather unique factors, regional, and otherwise.
Ya, I know, I am just a dirt bag lawyer. But statistics is one of my hobbies. I branch out!
And there you have it.