Scaling Up Qualitative Research Methods with Natural Language Processing Tools: The Case-Study of Colombia’s 2018 Presidential Election on Twitter

Traditional qualitative methods have the capacity to create rich and nuanced understandings of participants’
lived experiences, but this richness in understanding has often been at the expense of scale. For instance,
Mason (2010) completed an empirical analysis of the average number of interviews contained within
qualitative PhD theses, using data collected from theses.com (Mason, 2010). Through an analysis of 560
PhD theses, that Mason coded to be theses that utilized qualitative methods, they determined the mean
number of qualitative interviews was 31 (Mason, 2010: 13) but when looking at defined methods, theses
using a grounded theory approach used (mode: 25) and (median: 32) interviews (Mason, 2010: 9). Other
empirical work cites that the average number of qualitative interviews needed to reach saturation within
qualitative research to be 16 interviews, where meta-themes could begin to be established reliably at around
six interviews (Guest et al., 2006: 83). Are there tools that we can begin to deploy, in addition to qualitative
methods, that might allow researchers to expand sample sizes, while maintaining the richness and nuance of
understanding that are the hallmarks of qualitative analysis? Here we argue yes, and that Natural Language
Processing is a tool that qualitative researchers need to explore and deploy when working with large-scale
textual datasets, such as those produced through social media.