There are many reasons why we carry out this behavior, and a paper by Wolfgang Gaissmaier and Lael Schooler in the December, 2008 issue of Cognition suggests a new reason. They find that people who tend to probability match are better able to detect changes in the environment than people who find the option that is most highly rewarded and stick with it.
A while back, I wrote a post about unpredictability in human behavior. I discussed the concept of probability matching in choice. A simple case of probability matching is one in which you are given a choice between two cups, only one of which will have an M&M in it. If the M&M is in the left cup 80% of the time, and the right cup 20% of the time, you can maximize the number of M&Ms you are likely to receive by always selecting the left cup. That way, you will get M&Ms 80% of the time. People (and most other animals) tend not to make this optimal response. Instead, we tend to probability match. That is, if the left cup has the M&M 80% of the time, then we pick the left cup 80% of the time.