Use NVivo framework matrices to summarize small and large data sets
I’ve been working with the NVivo framework matrices feature for a couple of years now and wanted to share two of the approaches I’ve used. I’d also be really interested in hearing how other users have worked with the framework matrices feature.
"A particular strength of the matrix output is the ability to quickly compare across cases, while also retaining a sense of the whole case."
Coding and 'auto summarizing'
I recently worked on a project which involved thirty half-hour telephone interviews with staff delivering a service.
These interviews formed part of a scoping stage for the project and the primary aim was to map the range and diversity in provision so that twelve in-depth case studies could be selected for the main stage of the project. The study had a reasonably tight deadline and the interviews were fairly short and focused, so we decided to audio record the interviews and write up short (usually 3-4 page) summaries of each interview.
Because this data was already in a summarized format, we decided to code the data in NVivo (using a simple coding frame e.g. context of provider, approach to delivery, views on delivery etc).
Once the coding was completed, I used the ‘auto summarize’ feature to display the coded data in a matrix with each column representing a different theme of interest, and each row representing a respondent.
- ..............................My data is confidential so this demonstrates the same idea using sample data
This worked really well for this study because the coded data was already concise and didn't need further summarization. By having the data displayed in the matrix, I was able to see quickly the variation in delivery across the 30 service providers (by reading down the ‘approach to delivery’ column), while also retaining a sense of each individual case by reading across each row.
Tips for using the auto summarize feature
I don’t tend to use the ‘auto summarize’ feature with verbatim interview transcripts because the quantity of verbatim data is too much to fit neatly into a matrix output (the cells are so large it defeats the point of having a matrix output!).
Another tip I’d recommend if you do want to auto-summarize coded data, is to make sure the original coded source is in a format suitable for conversion (choose a small font, and remove double line spacing for example) as the auto-summarize feature puts the coded data into the matrix in the format it was coded in.
Using framework matrices without coding
The Framework approach does not rely on coding, but instead encourages the summarization of your data into a matrix display that facilitates both ‘across case’ and ‘within case’ analysis.
Coding the data as a pre-cursor to summarization can be useful because:
a) It can be a way of piloting the thematic framework you have developed for your data management – helping you refine the structure;
b) For some types of data it facilitates the use of the ‘auto summarize’ feature (as detailed above);
c) Once data has been coded you can select the option to only see ‘cell coding’ in the associated view on a framework matrix and then summarize the coded data.
However, I will not always code my data prior to using the Framework matrices features.
Summarizing nine case studies and 100 in-depth interviews
For example, a recent study I worked on that was evaluating a health service involved nine case studies. In each of the case studies a range of staff (both clinical and non-clinical) and patients were interviewed with a final dataset of over 100 in-depth interview transcripts.
To facilitate the management of this amount of complex data, we used the Framework Matrices feature in NVivo.
Each individual interview was assigned a case node and was displayed in as a row in the Framework matrix. Each column in the matrix represented a different theme that was relevant to the study (some themes were relevant across all respondent types e.g. views of service impact, while other themes were specific only to staff e.g. implementation of service, or specific to patients e.g. motivation for seeking treatment).
The data management phase of the project involved summarising data from each case into the matrix cells and creating summaries links for quick reference back to the original data. The first few cases were used to pilot the thematic framework we had developed which was then amended (see Kandy Woodfield's forum post for more detail on developing a thematic framework).
Once happy with the structure, all the cases were summarised into the matrix. No coding was done prior to this, as the process of deciding where data belonged in the thematic framework was part of the summarisation process.
Quick comparison of cases
A particular strength of the matrix output is the ability to quickly compare across cases, while also retaining a sense of the whole case. By displaying the patient data alongside the data from the staff, it was straightforward to compare views and experiences from these different perspectives.
We also used a node classification to facilitate data organisation. Attributes were created to differentiate between each case study area, to differentiate between staff and patients, and to identify which treatments patients received. This allowed us greater flexibility when viewing data in the matrices as it was quick and easy to filter or order the data according to these attributes.
While summarization does involve an investment of time, by the end of this process, a complex large data set had become a managed set of framework matrices which can then be used to facilitate analysis.
If you want to know more, you might like this video Framework Analysis in NVivo or this help topic About framework matrices.
I’d be really interested in hearing how others have used the framework matrices features, and any hints/tips other users may have to make the most of these features.
This article started life as a post on the QSR Forum.