It's all too common for people to prefer a simpler-seeming model to one that actually works. Another way to put this is that people will typically choose a model with one less variable over one that predicts outcomes more accurately.
These kinds of simplistic models are seductive because they feel good. They are easy to understand and easy to rationalize when some of the data don't fit.
That good feeling comes with a heavy price-tag, though.
For the same reasons they are so attractive, simplistic models run the risk of being canonized as right (about which I have previously written). As a result, people will fight to hold a broken model in place when it should be replaced and, ironically, discard a partially-broken model entirely rather than just amending it.
This means we waste a lot of time dropping bad ideas, only to find out they were good again later, then dropping the bad ideas that supplanted them, only to find out they were good, too.
We need a better way of managing this stuff.
