Forced experimentation: An analysis of multi-armed bandits with an inconsistent option

This paper investigates behavior in environments with forced experimentation. We analyze behavior in novel bandit problems where individuals repeatedly choose between an inconsistent option that may randomly be blocked in any given decision and a consistent option that is always available to be implemented. We analyze two environments with forced experimentation: an “Ex-Ante” environment and an “Ex-Post” environment. In the Ex-Ante environment, an individual knows at the start of a decision whether the inconsistent option can be implemented. In the Ex-Post environment, an individual only observes if the inconsistent option can be implemented by trying to implement it. We conduct an experiment where individuals in these two environments are predicted to choose the inconsistent option more often than the consistent option. Contrary to this prediction, we find evidence that subjects choose the consistent option more often. Structural analysis suggests that this result is driven by subjects who exhibit inertia after being forced into the consistent option and subjects who are averse to choosing an inconsistent option. Our results suggest that subjects may avoid inconsistently available options and that the presence of blocking in bandit problems leads to new behavioral factors to consider.