Munging the Ghosts in the Machine

Abstract

Data wrangling is typically treated as an obligatory, codified, and ideally automated step in machine learning that transforms qualities into quantities and quantities into algorithmic inputs. In contrast, this chapter proposes a different perspective, suggesting that archival data wrangling is a theory-driven process best understood as a practical craft. The chapter draws on empirical examples from contemporary computational social science to identify nine core modes of data wrangling: noticing, accessing, filtering, assembling, weaving, casting, visualizing, retracing, and prospecting. These modes, while presented as sequential, are actually iterative and nonlinear in practice. Treating data wrangling as a craft allows researchers to practice reflexivity in a process that is inherently political. Archival evidence and analytical methods can propagate enduring institutional biases if left unchecked. To this end, the chapter also discusses how data wrangling can help address issues of coded bias in scholarly research and beyond. While contemporary machine learning focuses on architectural engineering, a thorough engagement with it requires a deep of understanding of data wrangling.