Swine Flu and Ebola: A Twitter analysis using NVivo
13 December 2016 - BY Wasim Ahmed
PhD candidate and Research Associate at the Information School, University of Sheffield, Wasim Ahmed writes about how he analyzed Tweets related to the Swine Flu and Ebola outbreaks.
NVivo supports the thematic analysis of social media data – the ability to dig deep and analyze the themes in the conversations that are taking place online. This added an important dimension to my exploration of Tweets about infectious disease outbreaks.
Combining automated methods with in-depth analysis
Before deciding to use a qualitative methodology, I focused on the more automated methods of analyzing social media data using a combination of sentiment analysis, machine learning and network analysis – all of which can be done using NVivo.
I found that the results being returned by these automated methods were useful but only touched the surface of what people were saying. I decided to use a combination of automated methods (which would filter the data for me) alongside a more in-depth qualitative methodology (thematic analysis).
Five phases of thematic analysis
Braun and Clarke (2006) have written that:
“Thematic analysis is a method for identifying, analyzing and reporting patterns (themes) within data. It minimally organizes and describes your data set in (rich) detail. However, it frequently goes further than this, and interprets various aspects of the research topic”.
The process of thematic analysis is performed over five phases:
1. Familiarizing yourself with your data
2. Generating initial codes
3. Searching for themes
4. Reviewing themes
5. Writing up the final analysis (Braun and Clarke, 2006).
Moving from Excel to NVivo
I used Microsoft Excel to categorize Tweets until NVivo was recommended to me by several colleagues as a very easy to use tool. I also found that using qualitative methodologies such as thematic analysis was made easier by the software. The various phases of thematic analysis can be performed very well by using NVivo.
NVivo supports the importing of Twitter data directly via NCapture but since I had already retrieved the Tweets using DiscoverText, I went ahead and imported them into NVivo.
Creating an initial set of codes
I coded the Tweets to generate an initial set of codes – this corresponds to the second phase of coding as described by Braun and Clarke (2006).
The coding process was straight forward – I could highlight text and enter new codes. For example, when a Tweet referred to face masks in a humorous manner, I added two initial codes: ‘face masks’ and ‘humor’.
The screen shot below displays the list of initial codes that were generated:
Organizing codes into themes
NVivo also has a number of neat features that help with the generation of themes - this forms the latter stages of thematic analysis.
You can easily merge codes (or nodes as they’re called in NVivo) into broader themes.
Features like the hierarchical chart helped me to see and explore dominant ideas:
NVivo also provides the ability to perform a cluster analysis – the grouping of sources based on similar characteristics. The results can be viewed as either a dendogram or a cluster map.
This was useful when I was attempting to see if there are any correlations between the different codes, for instance, whether Tweets that mention ‘pigs’ also mention ‘transmission’ and so forth.
Overall, I have been finding NVivo to be fantastic software for the analysis of large amounts of social media data.