[jadence]

This article relies heavily on findings by the Gottman Institute however the Gottman Institute's findings are suspect because they built their prediction models after the results are known and they never verified that their models hold up w/ additional data.

In machine learning parlance, it's equivalent to deriving a formula using training data and not checking if the formula is accurate w/ a validation data. I doubt many (anyone?) here would trust an autonomous car built w/ only training data and never tested w/ validation data. Similarly, I don't believe/trust the Gottman's prediction rate (94%) for a second.

More info: http://www.slate.com/articles/double_x/doublex/2010/03/can_y... (2010)