Time Elapsed and Collaborative Expressions in Crowdsourcing Contests

Abstract

Much work in the crowdsourcing contest domain has typically focused on competition—in so doing assuming that all contestants(individual or teams) independently possess the knowledge to solve a given task. However, studies elsewhere have demonstrated that even in the presence of competitive pressures, collaborative behavior and the ensuing knowledge exchange matter for performance outcomes. This paper attempts to fill this gap in the crowdsourcing literature by examining the antecedents of collaborative expressions, a subset of and a more subtle form of collaborative behavior apt for online contests and having the potential to influence contest-level collaborations. Collaborative expressions refer to the voluntary self-expressions of solvers (individuals or teams) related to discussions, knowledge sharing, and feedback-seeking publicly during an ongoing contest (e.g., public code-sharing in a live contest). We propose that time elapsed in a contest has a negative influence on the number of collaborative expressions because a solver is less likely to receive suitable feedback from its competing peers as the contest nears its end. Furthermore, we examine the contingent effects of the number of contestants over time within a contest, operating via either ‘incentive effects’ or ‘parallel path effects.’ We test hypotheses using a panel dataset of 83 data science contests conducted on Kaggle.com from 2015 to 2018. Consistent with our hypotheses, we find that as more time elapses in a crowdsourcing contest, collaborative expressions by contestants decline. Furthermore, we find that the number of competitors weakens the negative baseline association between time elapsed in the contest and collaborative expressions, indicating the presence of a parallel path effect. Taken together, our findings offer significant implications for research and the practice of managing crowdsourcing contests.