In quantitative research tradition, scholars are interested on operationalisation: connecting the theoretic concept into a measurable variable. Similarly, programming and computational data analysis tries to make sense of the phenomena, and (social) scientists often seek to discuss about them through conceptual lenses. This means that conceptual idea, computational thinking transforming the idea into code and the code itself should be aligned to address the conceptual research challenge. This ensures that scholars end up discussing about the phenomena they originally sought to discuss.
For some research challenges, the operationalisation is straightforward â but for others, this is not always the case. For example, many scholars have used unsupervised machine learning with textual data to understand topics (Jacobi et al., 2016), discoursive contexts (Pantti et al., 2019), issues (Levy and Franklin, 2014), frames (Ylä-Anttila, 2018), and conduct content, discourse or narrative analysis (for further discussion Isoaho et al., 2019). All of them have used the same unsupervised machine learning method. However, it is hard to justify that the same method used to analyse frames could also analyse discourses or narratives. Rather, it appears that extensive human interpretation is put to formulate abstract results from an algorithm into social science relevant concepts (e.g., Baumer et al., 2017; Jacobi et al., 2016). Similarly, in organizational settings social networks can model contagion, environmental shaping, structural capital or access to resources, depending on what nodes and ties represent (Borgatti and Foster, 2003). Therefore, any results from network analysis relate to any of these distinct research directions.
These examples demonstrate that computational approaches do not define correct or incorrect use in relation to specific research method or theoretical concepts-at least yet, as the research processes are still evolving (see Section 10.5). Therefore, the difficult role of connecting code and its work into concepts and theories will remain for humans - a computer will always execute what it is asked to do, even if the outcome is not conceptually clear. Put more bluntly: if someone plans to shoot oneself in the foot, computational methods allow using bazooka instead a smaller gun.
The challenges of solving the right question is not unique to academia or computational methods. For example, Passi and Barocas (2019) show how a high-level business formulation - supporting car sales - is transformed into different objectives which were easier to transform into algorithmic problems. Such problems included determining clients' credit rating to examine who would be likely to receive a positive credit decision. However, the problems did not link to the high-level business challenge, leading to disconnect and eventual failure of the work. What this shows is the danger of doing “something” with computational methods: it might be technically highly sophisticated but might not discuss about the relevant conceptual thinking and therefore, lacks a theoretic contribution. In academic settings, similar challenges may emerge if the connection between concepts and analysis work itself is not sufficiently good. Quantitatively oriented researchers are focused on operationalisation to ensure conceptual clarity before the research starts. Qualitatively oriented researchers work in different ways: often they do not articulate them in a similar manner before the research has started, but rather work iteratively with the data and their knowledge to consider suitable concepts. However, in both approaches include thinking about the research, asking questions such as what is done and why and what kinds of claims or results can emerge as the outcome of the research project. Computational methods are still seeking to understand how it should be done to ensure clarity of concepts in its work. Underlying are deeper differences between what is the role of theory in research.