Despite being often discussed both in practice and academic circles, the sunk cost effect remains empirically elusive. Our model based on reference point dependence suggests that the traditional way of testing it—by assigning discounts—may not produce the desired effect. Motivated by this, we evaluate it across the gain-loss divide in two pre-registered experiments. In an online study with N=1,806, we randomize the price (low, medium, or high) of a ticket to enter a real-effort task, and observe its effect on play time. Our intervention, which varies the sunk cost by $2 for a 14-minute task, results in a moderate sunk cost effect (0.12 SD or 1.3 minutes). We further explore the economic applications of the effect in a field experiment on YouTube with N = 11,328 videos in which we randomize whether the time until a pre-video ad becomes skippable is shortened (0 s), default (5 s), or extended (10 s). The intervention has an overall insignificant effect on video engagement. This is driven by a sizable negative effect on the extensive margin, a channel which is not present in the online study. Specifically, more users leave before the video starts in the extended treatment (5.2 pp or 28% more relative to the shortened treatment). Taking the results of both studies together, we offer evidence of the sunk cost effect in a controlled environment, but its application in policy settings may prove challenging.
full text