Qualitative analysis in NVivo

Get the transcripts ready before importing them

Type every transcript in Word and keep one file for one participant. I tell my scholars to make the interviewer’s lines bold. Then while reading you can see at once where the participant starts talking. Type it word for word, with pauses or laughter where they change the meaning.

Do the anonymising before you import anything. Change every name to a code like P01 to P18, and take out the names of the hospital, village or company if a person could be identified from them. Who is P01, who is P02, that key list you keep in another file only, locked with password. Your ethics approval most likely asked for this.

In South India many theses have interviews in Tamil, Telugu, Kannada or Malayalam. One way is to code straight in Tamil or Telugu, whatever the person spoke, and later translate just the quotes going into the chapter. Or translate all the transcripts first and code in English. The first keeps more meaning, and the second is easier when your guide or the second coder can’t read that language. Both are accepted in viva. Only thing is, the method chapter must say which way you went and who translated. Planning to code in the original script? Then first put one transcript in NVivo as a trial and check the letters come out correct and the search is finding words.

For most qualitative theses 12 to 30 interviews is a common range, but don’t fix a number and aim for it. The number has to come from saturation, the point where new interviews stop giving new codes. Our guide on sample size for a thesis touches on how to argue it.

Setting up the NVivo project

Start a new project and put the date in its name. After that, bring your transcripts in through Import → Files. Each transcript comes in as one file under Data → Files.

Then make every participant a case, so that one transcript is tied to one participant. Select the transcripts, right-click and create cases from them, or use Create → Case. Add a case classification, for example “Participant”, and give it attributes like gender, age group, years of service and district. It takes twenty minutes, and later you can ask NVivo something like “what did nurses with under two years’ service say about support, and how is it different from the others?”

Open an analysis memo also at this stage. Go to Create → Memo and name it “Analysis log” or something close. Every time you merge codes, rename a theme or change your mind, add a dated line here. This memo becomes your audit trail, and examiners’ questions on how the analysis developed you can answer from it.

Save the project many times while working. Keep a backup copy of the project file in pendrive or on a cloud drive, not only on your laptop. NVivo project files can become big, and a file getting corrupted one week before submission is a real risk.

Braun and Clarke’s six phases, done in NVivo

Thematic analysis is the method most Indian qualitative theses use, and most of them cite Braun and Clarke (2006). Their six phases go well with NVivo.

1. Get familiar with the data. First you sit and read the transcripts, each one full, at least twice, without coding anything yet. Note your first impressions as annotations against the exact lines (right-click → New Annotation).

2. Generate initial codes. Now go through each transcript again and code the parts that carry some meaning. Select a passage and drag it onto a code in the Codes list, or right-click and code it to a new code. In the first round, code freely. From 15 transcripts a first pass mostly gives 80 to 150 codes, and that is quite normal at this stage.

3. Search for themes. Look at which codes go together, then make parent codes and drag the related codes under them. Each parent code is a possible theme, and the codes under it is the building blocks of that theme.

4. Review themes. Open each possible theme and read all its coded text in one sitting. See whether it fits together as one theme, and whether it is overlapping too much with some other theme. Some themes you will merge, some you will split into two, some you will drop. Whatever you change, put one line about it in the analysis memo that same day.

5. Define and name themes. For every theme write two or three sentences on what it covers and also what it does not cover. After that you give the name. Suppose you name it just “Support”. The examiner may ask, support of what, from whom? A name like “Holding on through each other” already answers that.

6. Write up. We have explained this in the last section, on the findings chapter.

Showing how codes became themes

The table is only an illustration, from an imagined study of newly joined staff nurses in a private hospital. Examiners will ask how you got your themes, and a table like this (a real one is longer) answers them before they ask. Keep it in the method chapter or an appendix, and the viva discussion on method becomes easier.

See that the codes stay close to what the participants actually said. The themes are your own reading of the data, and the two levels should stay separate. “Fear of asking” tells the reader what the nurses felt, so it works as a finding. But if you write “Communication”, that is only the topic. Examiner will take it as a chapter heading, not a theme.

NVivo can export your code list along with the descriptions, which is useful for the appendix. For that, write a short description for each code when you create it, in the code’s properties.

Using NVivo queries, and what not to claim from them

NVivo has many queries under its Explore tab, and we use them for checking our coding and comparing groups of participants. Finding the themes is still your job. For a thesis, four queries are enough.

A Word Frequency query gives a list of the most common words. Use it for familiarisation or a word cloud. Word counts are not themes, so please don’t put a word cloud in the findings as if it is a result.

A Text Search query brings up every place where a word or phrase appears, so you can see if you missed coding some passage.

A Matrix Coding query crosses your codes with the case attributes. For example it can show how many participants from each gender or experience group talked about a theme. Good for comparing, but keep in mind that with 18 participants the counts gives a description, and you cannot treat them as statistics.

A Coding Comparison query compares the coding done by two coders on the same files. It reports percentage agreement and Cohen’s kappa. As per Landis and Koch (1977), a kappa from .61 to .80 is substantial and above .80 is almost perfect. If your design has a second coder, give the Coding Comparison query results in the thesis. Braun and Clarke’s later reflexive approach (2019) says coder agreement is not the right test of quality. If you are following that approach, explain your position in the method chapter. Don’t add a kappa just for show.

Showing the analysis can be trusted

Nobody will ask you for a reliability coefficient in a qualitative thesis. What the examiner checks is whether he can believe your findings, and whether somebody else reading your method chapter could follow your path from transcripts to themes. Many Indian theses put this under the four headings of Lincoln and Guba, which are credibility, transferability, dependability and confirmability.

The analysis memo is your audit trail. The code-to-theme table shows the path from the data to the themes. For credibility you can do member checking. Here you send a short summary of themes to a few participants and ask, does this match what you went through? Transferability comes from thick description. You write about your setting in good detail. Then a reader working in some other place can see for himself if your findings may hold there also. Add a short reflexive note on your own position too. If you are a nurse studying nurses, for example, write a few lines on how that could have affected the data.

Give all this a separate heading in the method chapter. Mostly two pages are enough for it.

Writing the findings chapter

Arrange the findings chapter theme by theme. Don’t arrange it participant-wise or by interview question. Open each theme with two or three sentences on what it is about. Then take the subthemes one by one, give your explanation and add one or two quotes from the transcripts. Before going to the next theme, connect this one back to your research question in a small paragraph.

Keep the quotes short, many times one to four lines. After the quote put who said it, with the code and some detail, for example “(P07, female, 3 years’ service)”. Take quotes from many participants. If half the quotes come from P03 because P03 spoke well, examiners will notice. Also don’t leave the quotes to do all the work, because examiners read your explanation between them as the analysis.

Quote translated from Tamil or Malayalam? Say that in a footnote. The original words you can give in the appendix. If your study has a survey part also, our questionnaire design guide covers that side. Our data analysis service gives NVivo coding support along with SPSS work.