A Nudge or a Crutch?Temptation, Learning, and Backward Induction in Sequential Decision Tasks

Research in a variety of settings has shown that providing interim incentives can help people achieve long-term favorable outcomes. For example, researchers have found that paying children to read books can lead to higher performance on reading tests. At the same time, businesses have offered interim incentives, such as teaser rates on cable TV packages, to entice people to make decisions that may not be in their best long term interest. Both of these patterns attest to the ability of short-term incentives to have short-term effects in the desired direction. However, they leave unanswered the question of whether the learning they prompt transfers to new situations in a changing environment. This study uses a laboratory experiment to determine how interim payments, both on and off the optimal decision path, impact learning. Subjects play a simple game against a computer in which winning a prize requires five or six correct moves. Interim payments, smaller than the prize for winning, are inserted in some games, and treatment effects concern whether these interim payments lie on or off the optimal path, and whether they occur early or late in the game. After 30 rounds, the game is changed so that a new path becomes optimal.

We find that interim payments on the optimal path help subjects learn game-specific patterns, but impede them from learning to backward induct. In contrast, interim payments off the optimal path hinder learning, but those who do learn are able to backward induct when the game changes. The most favorable results, however, arise from the untreated game: learning in the absence of teaser payments is highly transferable. These results may have implications for social programs that try to reward good behavior in dynamic environments. They are also consistent with the premise that harder learning is more transferable.