Spent behind the Wheel: Drivers' Labor in the Uber Economy

the ‘‘doing’’ of the research. Of course, this iterative and sequential process is possible because of the careful attention to training and test data. Sequential and iterative research is perhaps clearest in the discovery section of the text. Here the authors devote a substantial portion of the text to explaining the excitement of allowing one to discover an unexpected concept while in the process of analyzing data. Beginning from the assumption that text data does not have one ‘‘truth’’ to tell but rather that there are myriad methods to represent what the text can tell us (some more useful than others), the authors demonstrate that by using different methodologies, researchers can discover distinct aspects of bodies of text. They explain in detail several methods (e.g., clustering, mixed-membership topic models, and embeddings) that allow a researcher to uncover a pattern in text data that might not have otherwise emerged. That is, using a subset of textual data, researchers uncover a theme that they may not have begun their project with. This exciting new finding can then spur additional inquiries without ‘‘starting over’’ or polluting the scientific process. The authors navigate a fine line here and emphasize that this process of discovery (as well as other analytical procedures such as measurement and causal inference) maintains integrity by splitting the textual data into groups—some that the researcher discovers with and some that the researcher validates with. Here, we encounter a key aspect of this text that links computer science and the social sciences as well as inductive and deductive scholarship: the process of validation. Much of this text is dedicated to validation—its definition, its implementation, and especially its importance in the analysis of textual data. Hesitant readers should rest assured that the authors are not circumventing methodological rigor. This ambitious project is particularly admirable for its pursuit of multiple audiences. At different points in the text, the content is well suited for an advanced undergraduate methods class. At others, the methodological detail is such that even a experienced practitioner may not find it entirely comprehensible. As with most guidebooks, Text as Data cannot be all things to all interested parties; but it provides guidance for social scientists at multiple points in their journey. Helpfully, the authors are also careful to credit the many innovators and innovations in text analysis, pointing eager readers to other sources to further their study. Very occasionally, introducing the research process from the perspective of textual data does not balance well with the methodological specificity that follows in each section. This text is a much-needed addition to methodological work in the social sciences— not just because of its niche application to textual data, but because it contributes an important argument amid our occasional obsession with methodological purity at the cost of substantive contributions to knowledge.