The Ultimate Guide to Coding Qualitative Data

Coding qualitative data for valuable insights

Qualitative researchers, at one point or another, will inevitably find themselves involved with the process of coding their data. The coding process can be arduous and time-consuming, so it's important to get a sense of how coding contributes to the understanding of knowledge in qualitative research.

Let's look at some of the basics of the qualitative coding process and how ATLAS.ti can help you with coding qualitative data.

• Qualitative data
• Coding qualitative data
• Coding methods
• Using ATLAS.ti for qualitative data coding
• Automated coding tools in ATLAS.ti

Qualitative data

Qualitative research tends to work with unstructured data that is either unorganized or organized in a way that is not useful to your research inquiry. Suppose you need to determine the most important aspects for deciding what hotel to stay in when you go on vacation. The information that go into choosing the "best" hotel can be located in various and separate places (e.g., travel websites, blogs, personal conversations) and scattered among information that may not be relevant to you. In qualitative research, one of the goals prior to data analysis is to identify what information is important, find that information, and sort that information in a way that makes it easy for you to come to a decision.

Until this process, raw data in qualitative research is essentially meaningless data, at least from the viewpoint of empirically generating knowledge. Unlike quantitative data, unstructured data for qualitative research requires not only reorganization but a reflection on what the data means in the first place.

Coding qualitative data

Qualitative coding is a necessary process before researchers can engage in the qualitative data analysis process. Coding provides a way to make the meaning of the data clear to you and to your research audience.

What is a code?

A code in the context of qualitative data analysis is a summary of a larger segment of text. Imagine applying a couple of sticky notes to a collection of recipes, marking each section with short labels like "ingredients," "directions," and "advice." Afterward, someone can page through those recipes and easily locate the section they are looking for thanks to those sticky notes.

Now, suppose you have different colors of sticky notes, where each color denotes a particular cuisine (e.g., Italian, Chinese, vegetarian). Now, with two ways to organize the data in front of you, you can look at all of the ingredients sections of all the recipes belonging to a cuisine to get a sense of the items that are commonly used for such recipes.

As illustrated in this example, one reason someone might apply sticky notes to a recipe is to help the reader save time in getting the desired information from that text, which is essentially the goal of qualitative coding. Coding allows a reader to get to the information they are looking for to facilitate the analysis process.

The use of codes has a purpose beyond simply establishing a convenient means to draw meaning from the data. When presenting qualitative research to an audience, a narrative summary of the data alone lacks the analytical process that is inherent to establishing research rigor. Moreover, such narratives might be too lengthy to grasp when the objective is to reach a consensus on valuable insights.

As a result, researchers in all fields tend to rely on data visualizations to illustrate their data analysis. Naturally, if such visualizations rely on tables and figures like bar charts and diagrams to convey meaning, researchers need to find ways to "count" the data along established data points, which is a role that coding can fulfill. While a strictly numerical understanding of qualitative research may overlook the finer aspects of social phenomena, researchers ultimately benefit from an analysis of the frequency of codes, combinations of codes, and patterns of codes that can contribute to theory generation.

Applying codes

Understanding what codes are supposed to do, a researcher then looks at the data line-by-line and develops a codebook by identifying data segments that can be represented by words or short phrases.

Taking a look at the example above, a set of three paragraphs is represented by one code, which is displayed in green in the right margin. Without codes, another reader would need to read all of the text to determine a particular meaning that the researcher has already interpreted. On the other hand, a reader can examine a fully coded project and get a sense of the main points of the data by looking at the margin of the document.

Analyzing codes

Think of a simple example to illustrate the importance of analyzing codes. If you are analyzing survey responses for people's preferences for shopping in brick-and-mortar stores and shopping online, you might think about marking each survey response as either "prefers shopping in-person" or "prefers shopping online." Once you have applied the relevant codes to each survey response, you can compare the frequencies of both codes to determine where the population as a whole stands on the subject.

Among other things, codes can be analyzed by their frequency or their connection to other codes (which ATLAS.ti calls co-occurrence). In the example above, you may also decide to code the data for the reasons that inform people's shopping habits, applying labels such as "convenience," "value," and "service." Then, the analysis process is simply a matter of determining how often each reason co-occurs with preferences for in-person shopping and online shopping by analyzing the codes applied to the data.

As a result, qualitative analysis relies on coding in research inquiries that might otherwise be difficult or impossible to accomplish. Qualitative coding transforms raw data into a form that facilitates the generation of more accurate insights through empirical analysis.

That said, coding is a time-consuming, if necessary, task in qualitative research, and one that researchers have elevated into a series of established methods that are worth briefly looking at.

Coding methods

Discussions of qualitative research methods have yielded multiple methods for assigning codes to data. While all qualitative coding approaches essentially seek to summarize large amounts of information succinctly, there are various approaches you can apply to your coding process.

Inductive coding

Probably the most basic form of coding is to look at the data and reduce it to its salient points of information through coding. Any inductive approach to research involves generating knowledge from the ground up. Inductive coding, as a result, looks to generate insights from the qualitative data itself.

Inductive coding benefits researchers who need to look at the data primarily for its inherent meaning, rather than for how external frameworks of knowledge might look at it. Inductive coding can also provide a new perspective that established theory has yet to consider, which would make a deductive analysis infeasible.

Deductive coding

That said, a deductive approach to coding is also useful in qualitative research. Contrast with inductive coding, a deductive coding approach applies an existing research framework or previous research study to new data. This means that the researcher applies a set of predefined codes based on established research to the new data.

Researchers can benefit from using both approaches in tandem if their research questions call for a synthesized analysis. Returning to the example of a cookbook, a person may mark the different sections of each recipe because they have prior knowledge about what a typical recipe might look like. On the other hand, if they come across a non-typical recipe (e.g., a recipe that may not have an ingredients section), they might need to create new codes to identify parts of the recipe that seem unusual or novel.

Employing both inductive coding and deductive coding, as a result, can help you achieve a more holistic analysis of your data by accounting for the typical features of a phenomenon while generating new knowledge about the less familiar aspects.

Thematic analysis coding

Whether you decide to apply an inductive coding or deductive coding to qualitative data, the coding should also be relevant to your research inquiry in order to be useful and avoid a cumbersome amount of coding that might defeat the purpose of summarizing your data. Let's look at a series of more specific approaches to qualitative coding to get a wider sense of how coding has been applied to qualitative research.

The goal of a thematic analysis arising from coding, as the name suggests, is to identify themes revolving around a particular concept or phenomenon. While concepts in the hard sciences such as temperature and atomic weight can be measured with numerical data, concepts in the social sciences often escape easy numerical analysis. Rather than reduce the beauty of a work of art or proficiency in a foreign language down to a number, thematic analysis coding looks to describe these phenomena by various aspects that frequently occur in the data.

Looking at the recipe again, we can describe a typical recipe by the sections that appear the most often. The same is true for describing a sport (e.g., rules, strategies, equipment) or a car (e.g., type, price, fuel efficiency, safety rating). While later analysis might be able to numerically measure these themes if they are particular enough, the role of coding along lines of themes provides a good starting point for recognizing and analyzing qualitative concepts.

Process coding

Processes are phenomena that are characterized by action. Think about the act of driving a car, rather than describing the car itself. In this case, process coding can be thought of as an extension of thematic coding, except that the major aspects of a process can also be identified by sequences and patterns, on the assumption that some actions may follow other actions. After all, drivers typically turn the key in the ignition before releasing the parking brake or shifting to drive.

Structural coding

The "structure" of a recipe in a cookbook is different from that of an essay or a newspaper article. Also think about how an interview for research might be structured differently from an interview for a TV news program. Researchers can employ structural coding to achieve a greater understanding of how cultures shape a particular piece of writing or social practice.

Longitudinal coding

Studies that observe cultures or practices over time do so to capture and understand changes in dynamic environments. The role of longitudinal coding in studies such as these is to illustrate how the frequencies and patterns represented by codes change from one observation or interview to the next. This will help researchers illustrate differences over time in an empirical manner.

Using ATLAS.ti for qualitative data coding

Qualitative data analysis software should effectively facilitate qualitative coding. Researchers can choose between manual coding and automated coding, where tools can be employed to suggest and apply codes to save time. ATLAS.ti is ideal for both approaches to suit researchers of all needs and backgrounds.

Manual coding

At the core of any qualitative data analysis software is the interface that allows researchers the freedom of assigning codes to qualitative data. ATLAS.ti's interface for viewing data makes it easy to highlight data segments and apply new codes or existing codes quickly and efficiently.

In vivo coding

Interpreting qualitative data to create codes is often a part of the coding process. This often means that the names of codes may differ from the actual text from the data itself.

However, the best names for codes sometimes come from the textual data itself, as opposed to some interpretation of the text. As a result, there may be a particular word or short phrase that occurs frequently in your data set, compelling you to incorporate that word or phrase into your qualitative codes. Think about how social media has acronyms like "YOLO" or "YMMV" and the likelihood that there is no way to succinctly summarize the intended meaning of such acronyms beyond using the acronyms themselves.

In vivo coding is a handy feature in ATLAS.ti for when you come across a key term or phrase that you want to build your codebook on. Simply highlight the desired text and click on "Code in Vivo" to create a new code instantly.

Code Manager

One of the biggest challenges of coding qualitative data is keeping track of dozens or even hundreds of codes, where a lack of organization may hinder researchers in the main objective of succinctly summarizing qualitative data.

Once you have developed and applied a set of codes to your project data, Code Manager gives a bird's eye view of all of your codes so you can develop and reorganize them hierarchically. Your list of codes can then be exported for use in spreadsheet software for quantitative analysis.

Automated coding tools in ATLAS.ti

Traditionally, qualitative researchers would perform this manual coding on their data manually by hand, which usually involved reading data line by line, page by page and using highlighters or bookmark flags to mark the key points in their data for later reference.

Researchers use qualitative data analysis software, however, to save researchers' time and effort in the coding process. As a result, a number of automated coding tools in ATLAS.ti such as AI Coding, Sentiment Analysis, and Opinion Mining use machine learning and natural language processing to apply useful codes for later analysis. Moreover, other tools in ATLAS.ti rely on pattern recognition to facilitate the creation of descriptive codes throughout your project.

AI Coding

One of the most exciting implications for recent advances in artificial intelligence is its potential for facilitating the research process, especially in qualitative research. The use of machine learning to understand the salient points in data can be especially useful to researchers in all fields.

AI Coding, available in both the platform and Web version of ATLAS.ti, essentially performs inductive coding on your qualitative data. It can process data through OpenAI's language models to suggest and apply codes to your project much more quickly than a manual coding approach.

Sentiment Analysis

Textual data, especially data that comes from in-depth interviews and open-ended survey questions, often contain sentiments that are positive or negative in nature. To conduct automated coding for these sentiments, ATLAS.ti employs machine learning to process your data quickly and suggest codes to be used in later analysis.

Opinion Mining

This tool is similar to Sentiment Analysis, but synthesizes the understanding of sentiments with key phrases in your textual data. The codes generated from Opinion Mining can provide a useful illustration of how language in interviews, focus groups, and surveys is used when discussing certain topics or phenomena.