Abstract:
Actors who learn slowing are less sensitive to feedback from recent actions. Such slow learning is particularly beneficial in environments where it is not clear if an action will lead to a global maximum. We argue that fast learning can outperform slow learning when the role of audience is incorporated. Relative to slow learners, the performance samples offered by fast learners are more unreliable because fast learners tend to prematurely converge to alternatives that are likely inferior in the long run. Nevertheless, audiences may give favorable feedback to fast rather than slow learners if this sampling bias is not corrected for. Using simulation modeling, we demonstrate that the differential sampling process between fast and slow learners determines how audiences form expectations and react to performance deviations, and potentially reward fast learners for unreliability. We then empirically examine our argument using the Google Search data regarding the Canadian software firms from 2004 to 2013. The results support that fast learners tend to attract more (less) attention when outperforming (underperforming) the expectations than slow learners. More generally, we present a sampling account for why fast learning is not necessarily a suboptimal strategy because of the way bounded rational audiences respond to the performance differences among bounded rational actors.
