This post is based on a talk I gave at ICQI 2018, which itself leads on from a talk from last year on the Entomologies of qualitative coding.

Good qualitative data is rich, and detailed - a fertile medium for understanding and interpreting the world. But the detail of the data comes at a price, usually qualitative data sources are lengthy, and are about a lot of different things. You don't just ask a single question that can be answered with a one word answer, you inquire and explore a range of issues around the topic to draw out detail and explore the 'why' behind the answers.

This means that the analysis of qualitative data starts with reading the data, to get a sense of the landscape of it, but an intermediary stage is coding - and this is the part of qualitative analysis that qualitative software like Quirkos can help with. We create codes which are like themes, and read through the text and put sections of text which are relevant to them into each code. Tagging the data in this way lets us bring quotes together that fit a theme, to eventually support (or disprove) a hypothesis. But what should these codes be? What features do we highlight that help us see the similarities and differences in the data?

Broadly speaking these themes can of two types: very low order, basic descriptive codes, or 'higher level' conceptual codes. It's difficult to describe the process and difference between these types of coding, but you can conceptualise it as moving from the lowest common denominator across the data to the highest (although it's also possible to do it the other way around).

This concept is found often in high-school level math when trying to add fractions. If they have different denominators (the bottom bit which shows how many sections they are) you have to multiply them out to get a common denominator - in other words a number that can be used to divide both fractions. It’s also the same game that entomologists are playing when trying to create a taxonomy of insects or other animals. Think about how you would describe what features are common to the butterflies in the top image? It can be a specific spot of a certain colour (a basic low level feature) or a feature that may look very different, but has a similar purpose - like an antenna (high level).

This is a bit like what qualitative coding often tries to do - find common themes that occur across all the sources. At a very basic level these will probably be very simple. Everyone in our sources is talking about 'Politics' in a general sense, 'The Media' and 'Opinions'. Creating these descriptive codes, and putting text into them is a useful way to start the coding process. It creates a 'map' or list of everything everyone is saying about 'The Media' and we can then read all these quotes together and look for patterns.

But there is a risk of creating codes that are so 'common' and so basic that they are pretty meaningless on their own. The more vague the theme, the more data will fit into it, but the less useful filtering of the data you are getting. Remember, the end goal is to find data that will answer your research question, and it is unlikely that these are as vague as 'What do people say about Politics?'. Usually you are looking for a much more specific insight, such as 'How do libertarian leaning people distance themselves from the policies of the Republican party?'. To do this, and to make a meaningful conclusion, we need to move to something more akin to the highest common denominator. In other words, what are the highest level, most significant and specific insights that are common themes in the data?

These are the 'highest common divisors' - in math the largest number you can give that allows you to divide and compare numbers or fractions. Every number is dividable by itself, and one. In the same way, every thing an individual says is true about themselves, and each word or statement they make is true in itself. However, neither of these is particularly interesting or insightful in itself, without some point of comparison. It's not important that an opinion or experience must be common to everyone to be relevant in qualitative research, in fact that's the strength of this methodology. However, you could argue that dissenting or different views are only interesting in comparison.

However, the highest common denominator codes should be at a high enough conceptual level to cover a variety of opinions, but bring them together under a common theory. It's not just saying people think this, or some people think the other way about their political leanings, but how they create a political identity. High level codes should be a close match to a theoretical interpretation of the world, such as “Gender is a performative act” (Butler 1988). These may be an existing theory, or a new theory you are discovering by applying a grounded theory approach.

But it's usually pretty hard to jump straight to this level of understanding of your data. Maybe you can read though all the sources once and just see a new conceptual understanding of the world emerge. However, this is rare, and you would probably still want to have quotes to illustrate and support your understanding. That's why creating your coding structure of the lowest common denominator first can help you to get to the next levels. And there may be multiple levels of coding, involving moving up, grouping and refining codes to support a deeper hypothesis. It's one reason why qualitative analysis is often described as a cyclical, iterative process.

Quirkos is designed to help you create and manage these different stages of coding with the ‘levels’ feature. This lets you create groups of codes from different coding iterations, and even have some that belong to all or just some of the levels. There’s a lot more information about how they work (and can be used to do other things) in this blog post on levels and groups. However, you can also always download the full version of Quirkos for free and try it for a month. It’s the easiest qualitative analysis software package to learn, as well as being one of the cheapest and most visual.

But remember, that even once you have created and populated a high level coding framework, this is not the same as analysis. You still need to make the leap from coding to qualitative analysis and actually read through the coded data, keep re-conceptualising it, and eventually match it to your research questions so that they can be answered. However, if you can keep coming back to the butterfly categorisation or fraction addition metaphors above, it might help you keep your eye looking out for both the low and high level themes in your research, and developing a rich coding framework that will help your insights and conclusions bubble up from your data.

If you are doing inductive coding or grounded theory, your coding framework can get complex, quickly. If you have hundreds of codes, they can become difficult to mangage which can slow down your coding - the part of your analysis you want as efficient and effective as possible so you can focus on identifying bigger trends.

Fortunately, this is what qualitative analysis software is best at - and whether you are using Nvivo, Atlas.ti or Quirkos there are ways to organise and sort your coding themes. In Quirkos the whole interface is designed to give you great flexibility to group and work with your codes, and this week we are going to look at some of the different ways to do this.

Occasionally I will see coding frameworks that have not been sorted at all, and they can look something like this:

If you make no attempt to sort and group your codes, they can become very difficult to work with. The example above has around 100 codes/quirks, but finding the right one is hard, and there is no structure to help you find them thematically. If you are working with other people, it is also nearly impossible to understand what is being investigated - and the same is true for long lists of codes. However, in Quirkos you can move the bubbles around just by dragging them, so you can quickly create rudimentary clusters:

Already this is looking a lot better, and you can probably now see what the research topic is about... I've just created simple clusters by theme, with composers in the bottom left, playwrights to the right, and genres in the middle. This is quite usable now, and is well worth the 5 minute investment this took to sort. However, we can make even better. Quirkos supports millions of different colours, so you can also colour-code your quirks by theme:

Much prettier, but importantly it's also quicker to find Quirks. If I know that all my composers are shades of red, my eye is drawn to the bubbles immediately, and the colour of the highlight stripes in the coded text immediately tells me what has been coded. This is preserved in the exports as well, so reviewing and sharing data with others gets an extra dimension of organisation and information.

There are still no groups in the above example, and Quirkos makes these very easy to create - just drag bubbles onto each other to create hierarchies:

Now it's getting beautiful! And so much quicker to find the right codes. The only disadvantage to this view is that the sub-categories aren't directly visible. This isn't usually a problem because just hovering the mouse over the parent will allow the subcategories to pop out. But if you are doing a lot of direct coding to the sub and sub-sub categories, it can help to have them always expanded. You can do this with the 'Quirks as Tree view':

This is most similar to the list views in Nvivo and other qualitative analysis software, so can be a good way to get used with Quirkos, or use if you don't like the bubble paradigm. There are also options in the View button to arrange the bubbles by size, or alphabetically. Levels are another way to group themes in Quirkos, which can create non-hierarchical categories.

Finally, don't forget to merge! If you have too many codes, some of them are probably superfluous, and can either be merged into other more general topics or deleted all together. Quirkos gives you a lot of flexibility to work with your codes, and manage them so that you can keep focusing on your data. Download a free trial to see for yourself, and get in touch with us if you have any questions.

So you spend weeks or months coding all your qualitative data. Maybe you even did it multiple times, using different frameworks and research paradigms. You've followed our introduction guides and everything is neatly (or fairly neatly) organised and inter-related, and you can generate huge reports of all your coding work. Good job! But what happens now?

It's a question asked by lot of qualitative researchers: after all this bruising manual and intellectual labour, you hit a brick wall. After doing the coding, what is the next step? How to move the analysis forward?

The important thing to remember is that coding is not really analysis. Coding is often a precursor to analysis, in the same way that a good filing system is a good start for doing your accounts: if everything is in the right place, the final product will come together much easier. But coding is usually a reductive and low-level action, and it doesn't always bring you to the big picture. That's what the analysis has to do: match up your data to the research questions and allow you to bring everything together. In the words of Zhang and Wildemuth you need to look for “meanings, themes and patterns”

Match up your coding to your research questions

Now is a good time to revisit the research question(s) you originally had when you started your analysis. It's easy during the coding process to get excited by unexpected but fascinating insights coming from the data. However, you usually need to reel yourself in at this stage, and explore how the coded data is illuminating the quandaries you set out to explore at the start of the project.

Look at the coded framework, and see which nodes or topics are going to help you answer each research question. Then you can either group these together, or start reading through the coded text by theme, probably more than once with an eye for one research question each time. Don't forget, you can still tag and code at this stage, so you can have a category for 'Answers research question 1' and tag useful quotes there.

One way to do this in Quirkos is the 'Levels' function, which allows you to assign codes/themes to more than one grouping. You might have some coded categories which would be helpful in answering more than one research question: you can have a level for each research question, and  Quirks/categories can belong to multiple appropriate levels. That way, you can quickly bring up all responses relevant to each research question, without your grouping being non-exclusive. 

Analyse your coding structure!

It seems strange to effectively be analysing your analysis, but looking at the coding framework itself gets you to a higher meta-level of analysis. You can grouping themes together to identify larger themes and coding. It might also be useful to match your themes with theory, or recode them again into higher level insights. How you have coded (especially when using grounded theory or emergent coding) can reveal a lot about the data, and your clusterings and groupings, even if chosen for practical purposes, might illuminate important patterns in the data.

In Quirkos, you can also use the overlap view to show relationships between themes. This illustrates in a graphical chart how many times sections of text 'overlap' - in that a piece of text has been coded with both themes. So if you have simple codes like 'happy' or 'disappointed' you can what themes have been most coded with disappointment. This can sometimes quickly show surprises in the correlations, and lets you quickly explore possible significant relationships between all of your codes. However, remember that all these metrics are quantitative, so are dependent on the number of times a particular theme has been coded. You need to keep reading the qualitative text to get the right context and weight, which is why Quirkos shows you all the correlating text on the right of the screen in this view.

Compare and contrast

Another good way to make your explorations more analytical is to try and identify and explain differences: in how people describe key words or experiences, what language they use, or how their opinions are converging or diverging from other respondents. Look back at each of the themes, and see how different people are responding, and most importantly, if you can explain the difference through demographics or difference life experiences.

In Quirkos this process can be assisted with the query view, which allows you to see responses from particular groups of sources. So you might want to look at differences between the responses of men and women, as shown below. Quirkos provides a side-by-side view to let you read through the quotes, comparing the different responses. This is possible in other software too, but requires a little more time to get different windows set up for comparison.

Match and re-explore the literature

It's also a good time to revisit the literature. Look back at the key articles you are drawing from, and see how well your data is supporting or contradicting their theory or assumptions. It's a really good idea to do this (not just at the end) because situating your finding in the literature is the hallmark of a well written article or thesis, and will make clear the contribution your study has made to the field. But always be looking for an extra level of analysis, try and grow a hypothesis of why your research differs or comes to the same conclusions – is there something in the focus or methodology that would explain the patterns?

Keep asking 'Why'

Just like an inquisitive 6 year old, keep asking 'Why?'! You should have multiple levels of Why, with explanations in qualitative focus usually explaining individual, then group, and all the way up to societal levels of causation. Think of the maxim 'Who said What, and Why?'. The coding shows the 'What', exploring the detail and experiences of the respondents is the 'Who', the Why needs to explore not just their own reasoning, but how this connects to other actors in the system. Sometimes this causation is obvious to the respondent, especially if articulated because they were always asked 'why' in the interview! However analysis sometimes requires a deeper detective type reading, getting to the motivations as well as actions of the participants.

Don't panic!

Your work was not in vain. Even if you end up for some reason scrapping your coding framework and starting again, you will have become so much more engaged with your data by reading it through so closely, and this will be a great help knowing how to take the data forward. Some people even discover that coding data was not the right approach for their project, and use it very little in the final analysis process. Instead they may just be able to pull together important findings in their head, the time taken to code the data having made key findings pop out from the page.

And if things still seem stuck, take a break, step back and print out your data and try and read it from a fresh angle. Wherever possible, discuss with others, as a different perspective can come not just from other people's ideas, but just the process of having to verbally articulate what you are seeing in the data.

Also remember to check out Quirkos, a software tool that helps constantly visualise your qualitative analysis, and thus keep your eye on what is emerging from the data. It's simple to learn, affordably priced, and there is a free trial to download for Windows, Mac and Linux so you can see for yourself if it is the right fit for your qualitative analysis journey. Good luck!

In the analysis of qualitative data, it can be easy to fall in the habit of creating either very descriptive, or very general theoretical codes. It’s often a good idea to take a step back, and examine your coding framework, challenging yourself to look at the data in a fresh way. There are some more suggestions for how to do this in a blog post article about turning coding strategies on their head. But while in Delhi recently to deliver some training on using Quirkos, I was struck by a couple of exhibits at the National Museum which in a roundabout way made me think about coding qualitative data, and getting the balance right between analytical and emotional coding frameworks.

There were several depictions of Hindu deities trampling a dwarf called Apasmāra, who represented ignorance. I loved this focus of minimising ignorance, but it’s important to note that in Hindu mythology, ignorance should not be killed or completely vanquished, lest knowledge become too easy to obtain without effort.

Another sculpture depicted Yogini Vrishanna, a female deity that had taken the bull-head form. It was apparently common for deities to periodically take on an animal head to prevent over-intellectualism, and allow more instinctive, animalistic behaviour!

I was fascinated between this balance being depicted between venerating study and thought, but at the same time warning against over thinking. I think this is a message that we should really take to heart when coding qualitative data. It’s very easy to create coding themes that are often far too simple and descriptive to give much insight in the data: to treat the analysis as purely a categorization exercise. When this happens, students often create codes that are basically an index of low-level themes in a text. While this is often a useful first step, it’s important to go beyond this, and create codes (or better yet, a whole phase of coding) which are more interpretive, and require a little more thinking.

However, it’s also possible to go too far in the opposite direction and over-think your codes. Either this comes from looking at the data too tightly, focusing on very narrow and niche themes, or from the over-intellectualising that Yogini Vrishanna was trying to avoid above. When the researcher has their head deeply in the theory (and lets be honest this is an occupational hazard for those in the humanities and social sciences), there is a tendency to create very complicated high-level themes. Are respondents really talking about ‘social capital’, ‘non-capitalocentric ethics’ or ‘epistemic activism’? Or are these labels which the researcher has imposed on the data?

These might be the times we have to put on our imaginary animal head, and try to be more inductive and spontaneous with our coding. But it also requires coding less from the head, and more from the heart. In most qualitative research we are attempting to understand the world through the perspective of our respondents, and most people are emotional beings, acting not just for rational reasons.

If our interpretations are too close to the academic, and not the lived experiences of our study communities, we risk missing the true picture. Sometimes we need a this theoretical overview to see more complex trends, but they should never be too far from the data in a single qualitative study. Be true to both your head and your heart in equal measure, and don’t be afraid to go back and look at your data again with a different animal head on!

If you need help to organise and visualise all the different stages of coding, try using qualitative analysis software like Quirkos! Download a free trial, and see for yourself...

“It’s where you make up as you go along!”

For a lot of students, Grounded Theory is used to describe a qualitative analytical method, where you create a coding framework on the fly, from interesting topics that emerge from the data. However, that's not really accurate. There is a lot more to it, and a myriad of different approaches.

Basically, grounded theory aims to create a new theory of interpreting the world, either when it’s an area where there isn’t any existing theory, or you want to challenge what is already out there. An approach that is often overused, it is a valuable way of approaching qualitative research when you aren’t sure what questions to ask. However, it is also a methodological box of worms, with a number of different approaches and confusing literature.

One of my favourite quotes on the subject is from Dey (1999) who says that there are “probably as many versions of grounded theory as there are grounded theorists”. And it can be a problem: a quick search of Google Scholar will show literally hundreds of qualitative research articles with the phrase “grounded theory was used” and no more explanation than this. If you are lucky, you’ll get a reference, probably to Strauss and Corbin (1990). And you can find many examples in peer-reviewed literature describing grounded theory as if there is only one approach.

Realistically there are several main types of grounded theory:

Classical (CGT)
Classical grounded theory is based on the Glaser and Strauss (1967) book “The Discovery of Grounded Theory”, in which it is envisaged more as a theory generation methodology, rather than just an analytical approach. The idea is that you examine data and discover in it new theory – new ways of explaining the world. Here everything is data, and you should include fieldwork notes as well as other literature in your process. However, a gap is recommended so that literature is not examined first (like when doing a literature review) creating bias too early, but rather engaging with existing theory as something to be challenged.

Here the common coding types are substantive and theoretical – creating an iterative one-two punch which gets you from data to theory. Coding is considered to be very inductive, having less initial focus from the literature.

Modified (Straussian)
The way most people think about grounded theory probably links closest to the Strauss and Corbin (1990) interpretation of grounded theory, which is probably more systematic and concerned with coding and structuring qualitative data. It traditionally proposes a three (or sometimes two) stage iterative coding approach, first creating open codes (inductive), then grouping and relating them with axial coding, and finally a process of selective coding. In this approach, you may consider a literature review to be a restrictive process, binding you to prejudices from existing theory. But depending on the different interpretations, modified grounded theory might be more action oriented, and allow more theory to come from the researcher as well as the data. Speaking of which…

Constructivist
The seminal work on constructivism here is from Charmaz (2000 or 2006), and it’s about the way researchers create their own interpretations of theory from the data. It aims to challenge the idea that theory can be ‘discovered’ from the data – as if it was just lying there, neutral and waiting to be unearthed. Instead it tries to recognise that theory will always be biased by the way researchers and participants create their own understanding of society and reality. This engagement between participants and researchers is often cited as a key part of the constructivist approach.
Coding stages would typically be open, focused and then theoretical. Whether you see this as being substantively different from the ‘open – axial – selective’ modified grounded theory strategy is up to you. You’ll see many different interpretations and implementations of all these coding approaches, so focus more on choosing the philosophy that lies behind them.

Feminist
A lot of the literature here comes from the nursing field, including Kushner and Morrow (2003), Wuest (1995), and Keddy (2006). There are clear connections here with constructivist and post-modern approaches: especially the rejection of positivist interpretations (even in grounded theory!), recognition of multiple possible interpretations of reality, and the examination of diversity, privilege and power relations.

Post-modern
Again, a really difficult segmentation to try and label, but for starters think Foucault, power and discourse. Mapping of the social world can be important here, and some writers argue that the practice of trying to generate theory at all is difficult to include in a postmodern interpretation. This is a reaction against the positivist approach some see as inherent in classical grounded theory. For where this leaves the poor researcher practically, there are at least one main suggested approach here from Clarke (2005) who focuses on mapping the social world, including actors and noting what has been left unsaid.

There are also what seem to me to be a variety of approaches plus a particular methodology, such as discursive grounded theory where the focus is more on the language used in the data (McCreaddie and Payne 2010). It basically seeks to integrate discourse analysis to look at how participants use language to describe themselves and their worlds. However, I would argue that many different ways of analysing data like discourse analysis can be combined with grounded theory approaches, so I am not sure they are a category of their own right.

To do grounded theory justice, you really need to do more than read this crude blog post! I’d recommend the chapter on Grounded Theory in Savin-Baden and Howell Major’s (2013) textbook on Qualitative Research. There’s also the wonderfully titled "A Novice Researcher’s First Walk Through the Maze of Grounded Theory" by Evans (2013). Grounding Grounded Theory (Dey 1999) is also a good read – much more critical and humorous than most. However, grounded theory is such a pervasive trope in qualitative research, indeed is seen by some to define qualitative research, that it does require some understanding and engagement.

But it’s also worth noting that for practical purposes, it’s not necessary to get involved in all the infighting and debate in the grounded theory literature. For most researchers the best advice is to read a little of each, and decide which approach is going to work best for you based on your research questions and personal preferences. Even better is if you can find another piece of research that describes a grounded theory approach you like, then you can just follow their lead: either citing them or their preferred references. Or, as Dey (1999) notes, you can just create your own approach to grounded theory! Many argue that grounded theory encourages such interpretation and pluralism, just be clear to yourself and your readers what you have chosen to do and why!

Following on from the last blog post on creating weird and wonderful categories to code your qualitative data, I want to talk about an often overlooked way of creating coding topics – using direct quotes from participants to name codes or topics. This is sometimes called “in vivo” coding, from the Latin ‘in life’ and not to be confused with the ubiquitous qualitative analysis software ‘Nvivo’ which can be used for any type of coding, not just in vivo!

In an older article I did talk about having a category for ‘key quotes’ - those beautiful times when a respondent articulates something perfectly, and you know that quote is going to appear in an article, or even be the article title. However, with in vivo coding, a researcher will create a coding category based on a key phrase or word used by a participant. For example someone might say ‘It felt like I was hit by a bus’ to describe their shock at the event, and rather than creating a topic/node/category/Quirk for ‘shock’, the researcher will name it ‘hit by a bus’. This is especially useful when metaphors like this are commonly used, or someone uses an especially vivid turn of phrase.

In vivo coding doesn’t just apply to metaphor or emotions, and can keep researchers close to the language that respondents themselves are using. For example when talking about how their bedroom looks, someone might talk about ‘mess’, ‘chaos’, or ‘disorganised’ and their specific choice of word may be revealing about their personality and embarrassment. It can also mitigate the tendency for a researcher to impose their own discourse and meaning onto the text.

This method is discussed in more depth in Johnny Saldaña’s book, The Coding Manual for Qualitative Researchers, which also points out how a read-through of the text to create in vivo codes can be a useful process to create a summary of each source.

Ryan and Bernard (2003) use a different terminology, indigenous categories or typographies after Patton (1990). However, here the meaning is a little different – they are looking for unusual or unfamiliar terms which respondents use in their own subculture. A good example of these are slang terms unique to a particular group, such as drug users, surfers, or the shifting vernacular of teenagers. Again, conceptualising the lives of participants in their own words can create a more accurate interpretation, especially later down the line when you are working more exclusively with the codes.

Obviously, this method is really a type of grounded theory, letting codes and theory emerge from the data. In a way, you could consider that if in vivo coding is ‘from life’ or grows from the data, then framework coding to an existing structure is more akin to ‘in vitro’ (from glass) where codes are based on a more rigid interpretation of theory. This is just like the controlled laboratory conditions of in vitro research with more consistent, but less creative, creations.

However, there are problems in trying to interpret the data in this way, most obviously, how ubiquitous will an in-vivo code from one source be across everyone’s transcripts? If someone talks about a shocking event in one source as feeling like being ‘hit by a bus’ and another ‘world dropped out from under me’, would we code the same text together? Both are clearly about ‘shock’ and would probably belong in the same theme, but does the different language require a slightly different interpretation? Wouldn’t you loose some of the nuance of the in vivo coding process if similar themes like these were lumped together?

The answer to all of these issues is probably ‘yes’. However, they are not insurmountable. In fact, Johnny Saldaña suggests that an in vivo coding process works best as a first reading of the data, creating not just a summary if read in order,  but a framework from each source which should be later combined with a ‘higher’ level of second coding across all the data. So after completing in vivo coding, the researcher can go back and create grouped coding categories based around common elements (like shock) or/and conceptual theory level codes (like long term psychological effects) which resonate across all the sources.

This sounds like it would be a very time consuming process, but in fact multi-level coding (which I often advocate) can be very efficient, especially with an in vivo coding as the first process. It may be that you just highlight some of these key words, on paper or Word, or create a series of columns in Excel adjacent to each sentence or paragraph of source material. Since the researcher doesn’t have to ponder the best word or phrase to describe the category at this stage, creating the coding framework is quick. It’s also a great process for participatory analysis, since respondents can quickly engage with selecting juicy morsels of text.

Don’t forget, you don’t have to use an exclusivly in vivo coding framework: just remember that it’s an option, and use for key illuminating quotes along side your other codes. Again, there is no one-size-fits-all approach for qualitative analysis, but knowing the range of methods allows you to choose the best way forward for each research question or project.

CAQDAS/QDA software makes it easy to keep all the different stages of your coding process together, and also create new topics by splitting and emerging existing codes. While the procedure will vary a little across the different qualitative analysis packages, the basics are very similar, so I’ll give a quick example of how you might do this in Quirkos.

Not a lot of people know this, but you can create a new Quirk/topic in Quirkos by dropping a section of text directly onto the create new bubble button, so this is a good way to create a lot of themes on the fly (as with in vivo coding). Just name these according to the in vivo phrase, and make sure that you highlight the whole section of relevant text for coding, so that you can easily see the context and what they are talking about.

Once you have done a full (or partial) reading and coding of your qualitative data, you can work with these codes in several ways. Perhaps the easiest is to create a umbrella (or parent) code (like shock) to which you can make relevant in vivo codes subcategories, just by dragging and dropping them onto the top node. Now, when you double click on the main node, you will see quotes from all the in vivo subcategories in one place.

It’s also possible to use the Levels feature in Quirkos to group your codes: this is especially useful when you might want to put an in vivo code into more than one higher level group. For example, the ‘hit by a bus’ code might belong in ‘shock’ but also a separate category called ‘metaphors’. You can create levels from the Quirk Properties dialogue of any Quirk, assign codes to one or more of these levels, and explore them using the query view. See this blog post for more on how to use levels in Quirkos.

It’s also possible to save a snapshot of your project at any point, and then actually merge codes together to keep them all under the same Quirk. You will loose most of the original in vivo codes this way (which is why the other options are usually better) but if you find yourself just dealing with too many codes, or want to create a neat report based on a few key concepts this can be a good way to go. Just right click on the Quirks you want to keep, and select ‘Merge Quirk with...’ to choose another topic to be absorbed into it. Don’t forget all actions in Quirkos have Undo and Redo options!

We don’t have an example dataset coded using in vivo quotes, but if you look at some of the sources from our Scottish Independence research project, you will see some great comments about politics and politicians that leap out of the page and would work great for in vivo coding. So why not try it out, and give in vivo coding a whirl with the free trial of Quirkos: affordable, flexible qualitative software that makes coding all these different approaches a breeze!

For the first time in ages I attended a workshop on qualitative methods, run by the wonderful Johnny Saldaña. Developing software has become a full time (and then some) occupation for me, which means I have little scope for my own professional development as a qualitative researcher. This session was not only a welcome change, but also an eye-opening critique to the way that many in the room (myself included) approach coding qualitative data.

Professor Saldaña has written an excellent Coding Manual for Qualitative Researchers, and the workshop really brought to life some of the lessons and techniques in the book. Fundamental to all the approaches was a direct challenge to researchers doing qualitative coding: code different.

Like many researchers, I am guilty of taking coding as a reductive, mechanical exercise. My codes tend to be very basic and descriptive – what is often called index coding. My codes are often a summary word of what the sentence or section of text is literally about. From this, I will later take a more ‘grand-stand’ view of the text and codes themselves, looking at connections between themes to create categories that are closer to theory and insight.

However, Professor Saldaña gave (to my count) at least 12 different coding frameworks and strategies that were completely unique to me. While I am not going to go into them all here (that’s what the textbook, courses and the companion website is for!) it was not one particular strategy that stuck with me, but the diversity of approaches.

It’s easy when you start out with qualitative data analysis to try a simple strategy – after all it can be a time consuming and daunting conceptual process. And when you have worked with a particular approach for many years (and are surrounded by colleagues who have a similar outlook) it is difficult to challenge yourself. But as I have said before, to prevent coding being merely a reductive and descriptive act, it needs to be continuous and iterative. To truly be analysis and interrogate not just the data, but the researcher’s conceptualisation of the data, it must challenge and encourage different readings of the data.

For example, Professor Saldaña actually has a background in performance and theatre, and brings some common approaches from that sphere to the coding process: exactly the kind of cross-disciplinary inspiration I love! When an actor or actress is approaching a scene or character, they may engage with the script (which is much like a qualitative transcript) looking at the character's objectives, conflicts, tactics, attitudes, emotions and subtexts. The question is: what is the character trying to do or communicate, and how? This sort of actor-centred approach works really well in qualitative analysis, in which people, narratives and stories are often central to the research question.

So if you have an interview with someone, for example on their experience with the adoption process, imagine you are a writer dissecting the motivations of a character in a novel. What are they trying to do? Justify how they would be a good parent (objectives)? Ok, so how are they doing this (tactics)? And what does this reveal about their attitudes and emotions? Is there a subtext here – are they hurt because of a previous rejection?

Other techniques talked about the importance of creating codes which were based around emotions, participant’s values, or even actions: for example, can you make all your codes gerunds (words that end in –ing)? While there was a distinct message that researchers can mix and match these different coding categories, it felt to me a really good challenge to try and view the whole data set from one particular view point (for example conflicts) and then step to one side and look again with a different lens.

It’s a little like trying to understand a piece of contemporary sculpture: you need to see it up close, far away, and then from different angles to appreciate the different forms and meaning. Looking at qualitative data can be similar – sometimes the whole picture looks so abstract or baffling, that you have to dissect it in different ways. But often the simplest methods of analysis are not going to provide real insight. Analysing a Henry Moore sculpture by the simplest categories (what is the material, size, setting) may not give much more understanding. Cutting up a work into sections like head, torso or leg does little to explore the overall intention or meaning. And certain data or research questions suit particular analytical approaches. If a sculpture is purely abstract, it is not useful to try and look for aspects of human form - even if the eye is constantly looking for such associations.

Here, context is everything. Can you get a sense of what the artist wanted to say? Was it an emotion, a political statement, a subtle treatise on conventional beauty? And much like impressionist painting, sometimes a very close reading stops the viewer from being able to see the brush strokes from the trees.

Another talk I went to on how researchers use qualitative analysis software, noted that some users assumed that the software and the coding process was either a replacement or better activity than a close reading of the text. While I don’t think that coding qualitative data can ever replace a detailed reading or familiarity with the source text, coding exercises can help read in different ways, and hence allow new interpretations to come to light. Use them to read your data sideways, backwards, and though someone else’s eyes.

But software can help manage and make sense of these different readings. If you have different coding categories from different interpretations, you can store these together, and use different bits from each interpretation. But it can also make it easier to experiment, and look at different stages of the process at any time. In Quirkos you can use the Levels feature to group different categories of coding together, and look at any one (or several) of those lenses at a time.

Whatever approach you take to coding, try to really challenge yourself, so that you are forced to categorise and thus interpret the data in different ways. And don't be suprsied if the first approach isn't the one that reveals the most insight!

There is a lot more on our blog about coding, for example populating a coding framework and coding your codes. There will also be more articles on coding qualitative data to come, so make sure to follow us on Twitter, and if you are looking for simple, unobtrusive software for qualitative analysis check out Quirkos!

In previous blog articles I’ve looked at some of the methodological considerations in developing a coding framework. This article looks at top-down or bottom-up approaches, whether you start with large overarching themes (a-priori) and break them down, or begin with smaller more simple themes, and gradually impose meanings and connections in an inductive approach. There’s a need in this series of articles to talk about the various different approaches which are grouped together as grounded theory, but this will come in a future article.

For now, I want to leave the methodological and theoretical debates aside, and look purely at the mechanics of creating the coding framework in qualitative analysis software. While I’m going to be describing the process using Quirkos as the example software, the fundamentals will apply even if you are using Nvivo, MaxQDA, AtlasTi, Dedoose, or most of the other CAQDAS packages out there. It might help to follow this guide with the software of your choice, you can download a free trial of Quirkos right here and get going in minutes.

First of all, a slightly guilty confession: I personally always plan out my themes on paper first. This might sound a bit hypocritical coming from someone who designs software for a living, but I find myself being a lot more creative on paper, and there’s something about the physicality of scribbling all over a big sheet of paper that helps me think better. I do this a lot less now that Quirkos lets me physically move themes around the screen, group them by colour and topic, but for a big complicated project it’s normally where I start.

But the computer obviously allows you to create and manage hundreds of topics, rearrange and rename them (which is difficult to do on paper, even with pencil and eraser!). It will also make it easy to assign parts of your data to one of the topics, and see all of the data associated with it. While paper notes may help conceptually think through some of the likely topics in the study and connect them to your research questions, I would recommend users to move to a QDA software package fairly early on in their project.

Obviously, whether you are taking an a-priori or grounded approach will change whether you will creating most of your themes before you start coding, or adding to them as you go along. Either way, you will need to create your topics/categories/nodes/themes/bubbles or whatever you want to call them. In Quirkos the themes are called ‘Quirks’ informally, and are represented by default as coloured bubbles. You can drag and move these anywhere around the screen, change their colours, and their size increases every time you add some text to them. It’s a neat way to get confirmation and feedback on your coding. In other software packages there will just be a number next to the list of themes that shows how many coding events belong to each topic.

In Quirkos, there are actually three different ways to create a bubble theme. The most common is the large (+) button at the top left of a canvas area. This creates a new topic bubble in a random place with a random colour, and automatically opens the Properties dialogue for you to edit it. Here you can change the name, for example to ‘Fish’ and put in a longer description: ‘Things that live in water and lay eggs’ so that the definition is clear to yourself and others. You can also choose the colour, from some 16 million options available. There is also the option to set a ‘level’ for this Quirk bubble, which is a way to group intersecting themes so that one topic can belong to multiple groups. For example, you could create a level called ‘Things in the sea’ that includes Fish, Dolphins and Ships, and another category called ‘Living things’ that has Fish, Dolphins and Lions. In Quirkos, you can change any of these properties at any time by right clicking on the appropriate bubble.

Secondly, you can right click anywhere on the ‘canvas’ area that stores your topics to create a new theme bubble at that location. This is useful if you have a little cluster of topics on a similar theme, and you want to create a new related bubble near the other ones. Of course, you can move the bubbles around later, but this makes things a bit easier.

If you are creating topics on the fly, you can also create a new category by dragging and dropping text directly onto the same add Quirk button. This creates a new bubble that already contains the text you dragged onto the button. This time, the property dialogue doesn’t immediately pop-up, so that you can keep adding more sections of data to the theme. Don’t forget to name it eventually though!

All software packages allow you to group your themes in some way, usually this is in a list or tree view, where sub-categories are indented below their ‘parent’ node. For example, you might have the parent category ‘Fish’ and the sub-categories ‘Pike’, ‘Salmon’ and ‘Trout’. Further, there might be sub-sub categories, so for example ‘Trout’ might have themes for ‘Brown Trout’, ‘Brook Trout’ and ‘Rainbow Trout’. This is a useful way to group and sort your themes, especially as many qualitative projects end up with dozens or even hundreds of themes.

In Quirkos, categories work a little differently. To make a theme a sub-category, just drag and drop that bubble onto the bubble that will be its parent, like stacking them. You will see that the sub-category goes behind the parent bubble, and when you move your mouse over the top category, the others will pop out, automatically arranging like petals from a flower. You can also remove categories just by dragging and pulling it out from the parent just like picking petals from a flower! You can also create sub-sub categories (ie up to three levels depth) but no more than this. When a Quirk has subcategories clustered below it, this is indicated by a white ring inside the bubble. This method of operation makes creating clusters (and changing your mind) very easy and visual.

Now, to add something to the topic, you just have to select some text, and drag and drop it onto the bubble or theme. This will work in most software packages, although in some you can also right click within the selected text where you will find a list of codes to assign that section to.

Quirkos, like other software, will show coloured highlighted stripes over the text or in the margin that show in the document which sections have been added to which codes. In Quirkos, you can always see what topic the stripe represents by hovering the mouse cursor over the coloured section, and the topic name will appear in the bottom left of the screen. You can also right-click on the stripe and remove that section of text from the code at any time. Once you have done some coding, in most software packages you can double click on the topic and see everything you’ve coded at this point.

Hopefully this should give you confidence to let the software do what it does best: keep track of lots of different topics and what goes in them. How you actually choose which topics and methodology to use in your project is still up to you, but using software helps you keep everything together and gives you a lot of tools for exploring the data later. Don’t forget to read more about the specific features of Quirkos here and download the free trial from here.

The terms ‘unstructured data’ and ‘qualitative data’ are often used interchangeably, but unstructured data is becoming more commonly associated with data mining and big data approaches to text analytics. Here the comparison is drawn with databases of data where we have a defined field and known value and the loosely structured (especially to a computer) world of language, discussion and comment. A qualitative researcher lives in a realm of unstructured data, the person they might be interviewing doesn’t have a happy/sad sign above their head, the researcher (or friend) must listen and interpret their interactions and speech to make a categorisation based on the available evidence.

At their core, all qualitative analysis software systems are based around defining and coding: selecting a piece of text, and assigning it to a category (or categories). However, it is easy to see this process as being ‘reductionist’: essentially removing a piece of data from it’s context, and defining it as a one-dimensional attribute. This text is about freedom. This text is about liberty. Regardless of the analytical insight of the researcher in deciding what relevant themes should be, and then filtering a sentence into that category, the final product appears to be a series of lists of sections of text.

This process leads to difficult questions such as, is this approach still qualitative? Without the nuanced connections between complicated topics and lived experiences, can we still call something that has been reduced to a binary yes/no association qualitative? Does this remove or abstract researchers from the data? Isn't this a way of quantifying qualitative data?

While such debates are similarly multifaceted, I would usually argue that this process of structuring qualitative data does begin to categorise and quantify it, and it does remove researchers from their data. But I also think that for most analytical tasks, this is OK, if not essential! Lee and Fielding (1996) say that “coding, like linking in hypertext, is a form of data reduction, and for many qualitative researchers is an important strategy which they would use irrespective of the availability of software”. When a researcher turns a life into 1 year ethnography, or a 1 hour interview, that is a form of data reduction. So is turning an audio transcript into text, and so is skim reading and highlighted printed versions of that text.

It’s important to keep an eye on the end game for most researchers: producing a well evidenced, accurate summary of a complex issue. Most research, as a formula to predict the world or a journal article describing it, is a communication exercise that (purely by the laws of entropy if not practicality) must be briefer than the sum of it’s parts. Yet we should also be much more aware that we are doing this, and together with our personal reflexivity think about the methodological reflexivity, and acknowledge what is being lost or given prominence in our chosen process.

Our brains are extremely good at comprehending the complex web of qualitative connections that make everyday life, and even for experienced researchers our intuitive insight into these processes often seem to go beyond any attempt to rationalise them. A structuralist approach to qualitative data can not only help as an aide-mémoir, but also to demonstrate our process to others, and challenge our own assumptions.

In general I would agree with Kelle (1997) that “the danger of methodological biases and distortion arising from the use of certain software packages is overemphasized in current discussions”. It’s not the tool, it’s how you use it!

When doing analysis and coding in a qualitative research project, it is easy to become completely focused on the thematic framework, and deciding what a section of text is about. However, qualitative analysis software is a useful tool for organising more than just the topics in the text, they can also be used for deeper contextual and meta-level analysis of the coding and data.

Because you can pretty much record and categorise anything you can think of, and assign multiple codes to one section of text, it often helps to have codes about the analysis that help with managing quotes later, and assisting in deeper conceptual issues. So some coders use some sort of ranking system so they can find the best quotes quickly. Or you can have a category for quotes that challenge your research questions, or seem to contradict other sources or findings. Here are 6 suggestions for these meta-level codes you could create in your qualitative project (be it Quirkos, Nvivo, Atlas-ti or anything!):

Rating
I always have a node I call ‘Key Quotes’ where I keep track of the best verbatim snippets from the text or interview. It’s for the excited feeling you get when someone you interviewed sums up a problem or your research question in exactly the right way, and you know that you are going to end up using that quote in an article. Or even for the title of the article!

However, another way you can manage quotes is to give them a ranking scheme. This was suggested to me by a PhD student at Edinburgh, who gives quotes a ranking from 1-5, with each ‘star-rating’ as a separate code. That way, it’s easy to cross reference, and find all the best quotes on a particular topic. If there aren’t any 5* quotes, you can work down to look at the 4 star, or 3 star quotes. It’s a quick way to find the ‘best’ content, or show who is saying the best stuff. Obviously, you can do this with as little or much detail as you like, ranking from 1-10 or just having ‘Good’ and ‘Bad’ quotes.

Now, this might sound like a laborious process, effectively adding another layer of coding. However, once you are in the habit, it really takes very little extra time and can make writing up a lot quicker (especially with large projects). By using the keyboard shortcuts in Quirkos, it will only take a second more. Just assign the keyboard numbers 1-5 to the appropriate ranking code, and because Quirkos keeps the highlighted section of text active after coding, you can quickly add to multiple categories. Drag and drop onto your themes, and hit a number on the keyboard to rank it. Done!

Contradictions
It is sometimes useful to record in one place the contradictions in the project – this might be within the source, where one person contradicts themselves, or if a statement contradicts something said by another respondent. You could even have a separate code for each type of contradiction. Keeping track of these can not only help you see difficult sections of data you might want to review again, but also show when people are being unsure or even deceptive in their answers on a difficult subject. The overlap view in Quirkos could quickly show you what topics people were contradicting themselves about – maybe a particular event, or difficult subject, and the query views can show you if particular people were contradicting themselves more than others.

Ambiguities
In qualitative interview data where people are talking in an informal way about their stories and lives, people often say things where the meaning isn’t clear – especially to an external party. By collating ambiguous statements, the researcher has the ability to go back at the end of the source and see if each meaning is any clearer, or just flag quotes that might be useful, but might be at risk of being misinterpreted by the coder.

Not-sures
Slightly different from ambiguities: these are occasions when the meaning is clear enough, but the coder is not 100% sure that it belongs in a particular category. This often happens during a grounded theory process where one category might be too vague and needs to be split into multiple codes, or when a code could be about two different things.

Having a not-sure category can really help the speed of the coding process. Rather than worrying about how to define a section of text, and then having sleepless nights about the accuracy of your coding, tag it as ‘Not sure’ and come back to it at the end. You might have a better idea where they all belong after you have coded some more sources, and you’ll have a record of which topics are unclear. If you are not sure about a large number of quotes assigned to the ‘feelings’ Quirk (again, shown by clustering in the overlap view in Quirkos), you might want to consider breaking them out into an ‘emotions’ and ‘opinions’ category later!

Challenges
I know how tempting it can be to go through qualitative analysis as if it were a tick-box exercise, trying to find quotes that back up the research hypothesis. We’ve talked about reflexivity before in this blog, but it is easy to go through large amounts of data and pick out the bits that fit what you believe or are looking for. I think that a good defence against this tendency is to specifically look for quotes that challenge you, your assumptions or the research questions. Having a Quirk or node that logs all of these challenges lets you make sure you are catching them (and not glossing over them) and secondly provides a way to do a validity assessment at the end of coding: Do these quotes suggest your hypothesis is wrong? Can you find a reason that these quotes or individuals don’t fit your theory? Usually these are the most revealing parts of qualitative research.

Absences
Actually, I don’t know a neat way to capture the essence of something that isn’t in the data, but I think it’s an important consideration in the analysis process. With sensitive topics, it is sometimes clear to the researcher that an important issue is being actively avoided, especially if an answer seems to evade the question. These can be at least coded as absences at the interviewer’s question. However, if people are not discussing something that was expected as part of the research question, or was an issue for some people but not others, it is important to record and acknowledge this. Absence of relevant themes is usually best recorded in memos for that source, rather than trying to code non-existent text!

These are just a few suggestions, if you have any other tips you’d like to share, do send them to daniel@quirkos.com or start a discussion in the forum. As always, good luck with your coding!

In framework analysis, sometimes described as a top-down or 'a-priori' approach, the researcher decides on the topics of interest they will look for before they start the analysis, usually based on a theory they are looking to test. In inductive coding the researcher takes a more bottom-up approach, starting with the data and a blank-sheet, noting themes as the read through the text.

Obviously, many researchers take a pragmatic approach, integrating elements of both. For example it is difficult for a emergent researcher to be completely naïve to the topic before they start, and they will have some idea of what they expect to find. This may create bias in any emergent themes (see previous posts about reflexivity!). Conversely, it is common for researchers to discover additional themes while reading the text, illustrating an unconsidered factor and necessitating the addition of extra topics to an a-proiri framework.

I intend to go over these inductive and deductive approaches in more detail in a later post. However, there is also another level in qualitative coding which is top-down or bottom-up: the level of coding. A low 'level' of coding might be to create a set of simple themes, such as happy or sad, or apple, banana and orange. These are sometimes called manifest level codes, and are purely descriptive. A higher level of coding might be something more like 'issues from childhood', fruit, or even 'things that can be juggled'. Here more meaning has been imposed, sometimes referred to as latent level analysis.

Usually, researchers use an iterative approach, going through the data and themes several times to refine them. But the procedure will be quite different if using a top-down or bottom-up approach to building levels of coding. In one model the researcher starts with broad statements or theories, and breaks them down into more basic observations that support or refute that statement. In the bottom-up approach, the researcher might create dozens of very simple codes, and eventually group them together, find patterns, and infer a higher level of meaning from successive readings.

So which approach is best? Obviously, it depends. Not just on how well the topic area is understood, but also the engagement level of the particular researcher. Yet complementary methods can be useful here: the PI of the project, having a solid conceptual understanding of the research issue, can use a top-down approach (in both approaches to the analysis) to test their assumptions. Meanwhile, a researcher who is new to the project or field could be in a good position to start from the bottom-up, and see if they can find answers to the research questions starting from basic observations as they emerge from the text. If the themes and conclusions then independently reach the same starting points, it is a good indication that the inferences are well supported by the text!