Michael E. Kummer

Spillovers in networks of user generated content: Evidence from 23 natural experiments on Wikipedia

Endogeneity in network formation hinders the identification of the role social networks play in generating spillovers, peer effects and other externalities. This paper tackles this problem and investigates how the link network between articles on the German Wikipedia influences the attention and content generation individual articles receive. Identification exploits local exogenous shocks on a small number of nodes in the network. It can thus avoid the usually required, but strong, assumptions of exogenous observed characteristics and link structure in networks. This approach also applies if, due to a lack of network information, identification through partial overlaps in the network structure fails (e.g. in classrooms). Exogenous variation is generated by natural and technical disasters or by articles being featured on the German Wikipedias start page. The effects on neighboring pages are substantial; I observe an increase of almost 100 percent in terms of both views and content generation. The aggregate effect over all neighbors is also large: I find that a view on a treated article converts one for one into a view on a neighboring article. However, the resulting content generation is small in absolute terms.

Identification of Causal Effects in the Context of Mass Collaboration

Several instances of successful online mass collaboration have recently generated large amounts of data. These datasets are very appealing for empirical research on patterns and drivers of mass collaboration in a wide range of social science disciplines. However, their complexity, the presence of network effects, and multidirectional nature of the causal mechanisms at play often raise substantial challenges to empirical researchers. In this chapter, we discuss the econometric approach to mass collaboration, focusing on the methodological challenges of causal identification and the interpretation of how some factors affect others. Our chapter provides methodological tools for causal identification of effects in observational data from mass collaboration platforms. Specifically, we present two quasi-experimental methods, natural experiments and instrumental variables, in detail and show applications using examples from our own research.

Market Entry in E-Commerce

We analyze the behavior of start-ups in e-commerce, namely on Austrias leading price-comparison-site, a multi-product environment with almost complete information. We use weekly panel data on price-quotes of digicams, Audio/HiFi-equipment and hardware. We furthermore use advanced estimation methods, which, having only recently been introduced to IO, aim at using a minimum of modeling assumptions. Thus, being able to trace the behavior of roughly 350 start-up companies and 600 incumbents, we investigate whether start-ups have a different composition of product-portfolios, charge lower prices and offer fewer goods.