Social Science Computer Review (SSRC) has just published a paper by yours truly, Dr Pat Harpur and Dr Corrie Uys to https://doi.org/10.1177/08944393231204163. As the article's title suggests, we focus on the contrasting the Qualitative Data Analysis Software (QDAS) packages that currently support live Twitter data imports. 

QDAS tools that support live data extraction are a relatively recent innovation. At the time of our fieldwork, four prominent QDAS provided this: only ATLAS.ti™, NVivo™, MAXQDA™ and QDA Miner™ had Twitter data import functionalities. Little has been written concerning the research implications of differences between their functionalities, and how such disparities might contribute to contrasting analytical opportunities. Consequently, early-stage researchers may experience difficulties in choosing an apt QDAS to extract live data for Twitter academic research.

In response to both methodological gaps, we spent almost a year working on a software comparison to address the research question (RQ) 'How do QDAS packages differ in what they offer for live Twitter data research during the organisational stage of qualitative analysis?'. Comparing their possible disparities seems worthwhile since what QDAS cannot, or poorly, support may strongly impact researchers’ microblogging data, its organisation, and scholars’ potential findings. In the preliminary phase of research, we developed a features checklist for each package, based on their online manuals, product descriptions and forum feedback related to live Twitter imports. This checklist confirmed wide-ranging disparities between QDAS, which were not unexpected since they are priced very differently- ranging from $600 for an ATLAS.ti subscription, to $3,650 for a QDAMiner (as part of the Provalis Research’s ProSuite package, which also includes WordStat 10 & Simstat).

To ensure that each week's Twitter data extractions could produce much data for potential evaluation, we focused on extracting and organising communiqués from the national electrical company, the Electricity Supply Commission (Eskom). ‘Load-shedding’ is the Pan South African Language Board’s word of the year for 2022 (PanSALB, 2022), due to it most frequent use in credible print, broadcast and online media. Invented as a euphemism by Eskom’s public-relations team, load-shedding describes electricity blackouts. Since 2007, planned rolling blackouts have been used in a rotating schedule for periods ‘where short supply threatens the integrity of the grid’ (McGregor & Nuttall, 2013). In the weeks up to, and during, the researchers’ fieldwork, Eskom, and the different stages of loadshedding strongly trended on Twitter. These tweets reflected the depth of public disapproval, discontent, anger, frustration, and general concern.

QDAS packages commonly serve as tools that researchers can use for four broad activities in the qualitative analysis process (Gilbert, Jackson, & di Gregorio, 2014). These are (a) organising- coding sets, families and hyperlinking; (b) exploring - models, maps, networks, coding and text searches; (c) reflecting - through memoing, annotating and mapping; and (d) integrating qualitative data through memoing with hyperlinks and merging projects (Davidson & di Gregorio, 2011; Di Gregorio, 2010; Lewins & Silver, 2014).

Notwithstanding the contrasts in the costs for different QDAS packages, it was still surprising how much the QDAS tools varied for the first activity, (a) ‘organising data’ in our qualitative research project: Notably, the quantum of data extracted for the same query differed, largely due to contrasts in the types and amount of data that the four QDAS could extract. Variations in how each supported visual organisation and thematic analysis also shaped researchers’ opportunities for becoming familiar with Twitter users and their tweet content. 

Such disparities suggest that choosing a suitable QDAS for organising live Twitter data must dovetail with a researcher’s focus: ATLAS.ti accommodates scholars focused on wrangling unstructured data for personal meaning-making, while MAXQDA suits the mixed-methods researcher. QDA Miner’s easy-to-learn user interface suits a highly efficient implementation of methods, whilst NVivo supports relatively rapid analysis of tweet content.

We hope that these findings might help guide Twitter social science researchers and others in QDAS tool selection. Our research has suggested recommendations for these tools developers to follow for potentially improving the user experience for Twitter researchers. Future research might explore disparities in other qualitative research phases, or contrast data extraction routes for a variety of microblogging services.  More broadly,  an opportunity for a methodological contribution exists regarding research that can define a strong rationale for the software comparison method.

The authors greatly appreciate the SSRC's editor, Professor Stephen Lyon, advice on improving our final manuscript. We also thank The Noakes Foundation for its grant AFSDV02- our interdisciplinary software comparison would not have been possible without funding to cover subscriptions to the most extensive versions of MAXQDA Analytics Pro and QDA Miner. All authors are affiliated with the Cape Peninsula University of Technology (CPUT) and appreciate CPUT's provision of licensed versions of ATLAS.ti.

Please comment below if you have any questions or comments regarding our paper?