Weeding out or picking winners in open innovation? Factors driving multi-stage crowd selection on LEGO ideas

Abstract

Using unique data from the LEGO Ideas platform and a novel approach of algorithm-based abduction, we combine multiple methods to provide new insights into crowd selection. Through qualitative content-coding, interviews, and prior literature, we derive an initial set of variables. We then use machine learning for feature selection and to identify the most important factors for crowd selection. The findings are used to build theory on crowd selection, which we test on a hold-out sample. Our key finding is that ideator and idea characteristics suggested by prior research can predict crowd selection only at early stages. More specifically, crowds rely on ideator status, prior success, and a carefully crafted idea presentation with many images to weed out bad ideas in early stages. However, these characteristics have little bearing in predicting winners at later stages. We explain this with signaling theory, where these ideator and idea characteristics represent strong, costly signals for idea quality and value only in early stages but fade as new signals (such as social signals of popularity and gaining fact traction) emerge. Our study provides two main contributions to research on crowd selection and idea evaluation. First, our approach enables us to prune explanations for crowd selection and guide attention to the factors which matter most. Second, we extend prior work by considering multi-stage crowd selection and highlighting its dynamic nature.