Understanding Firms’ Innovation Behavior Through Web-Scraped Data: A Two-Step Calibration Approach with Evidence from Florence

While interest in firms’ innovative behavior has steadily increased, comprehensive probability-based survey data remain scarce. At the same time, web-scraped data are becoming increasingly available, although they are inherently non-probabilistic and potentially affected by selection bias. This paper considers a setting in which web-scraped information is available for a subset of firms in the target population, enabling the construction of a proxy for their innovation behavior. For a smaller subset of these firms, additional data from their social media channels can also be collected. The combined use of web and social media information allows for the development of a more refined and informative proxy of firms’innovative behavior. To address the non-probabilistic nature of the data and enhance population-level inference, we propose an estimation strategy based on a two-step calibration procedure.