| Just as social media has permeated communication in our public and private lives, it has also become a widely used source of data and object of study in academic and commercial research. Despite widespread use, relatively little is known about how social media datasets change when observed at different points over time or how collection methods may impact the data at the core of our research projects. For example: Will results differ if social media data are collected in real-time, a few minutes after production, hours, days, or weeks later? What happens to the metadata, links to web pages, photos, and videos embedded in this content over time? If data collection methods do not preserve and archive social media posts, metadata, and linked content; are researchers venturing into a new dataset each time they engage with it? In this dissertation, a combination of quantitative and qualitative approaches are used to examine how social media datasets change over time and how change impacts the reliability and authenticity of this data. Three Twitter-based case studies, each exhibiting prototypical elements social scientists encounter in their research are used to demonstrate the impact of research design and data collection choices. This work advances the field of information science by empirically investigating how the ephemeral nature of social media data, metadata, and linked content have significant and lasting effects on the reliability and authenticity of datasets used in research. By situating research design decisions of how and when to observe data within the frameworks of process theory, infrastructure studies, and archival theory, this work brings the importance of methodological considerations to the forefront of studies of digital and social media. Empirical observations inform a set of implications for social media research, offering researchers practical considerations to inform their research designs. |
