When, why and how to use coding to interpret qualitative information for insights

Shapes sortedPublished by Ray Poynter, 6 June 2023

In this post, I am going to talk about what coding is, when it can be useful (and when it isn’t), and how to use it. I am going to be looking at coding in the context of qualitative information, such as text, photos and video. This blog focuses on using coding to find insights, particularly in the context of market research.

What is ‘qualitative information’?
Before I talk about coding, it makes sense to clarify what I mean by qualitative information. Qualitative information describes the qualities or characteristics of things. Examples of qualitative information include transcripts of conversations, open-ended survey responses, blogs and social media comments. Qualitative information also includes non-text information such as photos and videos.

Direct interpretation of qualitative information
In simple cases, qualitative information can be easy to interpret. If we are testing a new menu with a group of eight people, and all eight participants: express disgust on their faces, do not finish eating their serving, and tell us that it was horrible, then we can interpret that this meal is a failure for this sort of customer. In some cases, a few simple questions might also indicate why. Perhaps it was too spicey, too salty, or had an unpleasant consistency.

However, in most real-world insights and research situations, it is not possible to assess the implications of qualitative information directly without some form of analysis. Coding is can be a stepping stone into the analysis.

A simple example of coding to extract meaning
I will start with a simple type of coding in the context of survey open-ended responses.  In this example, imagine we have 500 interviews from a survey. One of the questions on the questionnaire was: “What drink did you have with your dinner yesterday?”. The answers might include things like the following: Coke, Coca-Cola, Orange juice, Tea, Beer, Coke, Coffee, White Wine, Cappuccino, Pepsi, Coffee, Red Wine, Diet Coke …..

We can convert this qualitative data into quantitative information by replacing each open-ended response with a code from a limited set of codes. For example, we might characterize each one as: Soft Drink, Hot Drink, or Alcohol. Or, we might use a longer list of codes, for example, Cola, Other Soft Drinks, Beer, Wine, Tea, & Coffee.

Once we have replaced the open-ended text with codes, we can count the frequencies or use the codes in cross-tabulations.

Coding Topics
In a slightly more complicated example of coding, imagine another question from the same study. This question is” “What do you think the key food trends are in 2023?”. Possible answers might include:

‘A shift towards gluten-free foods’
‘More vegan options’
‘Avoiding red meat to improve my health’
‘Using an air fryer’
‘Reducing salt and sugar’
‘Thinking about the carbon footprint of meals’
‘Eating less meat’
‘Following a pescatarian diet’

After reading some (perhaps 100) of the responses, the analyst might decide there were a fixed number of topics. This list is referred to as a code frame. From the list above, the code frame might include:

  • Vegan/Vegetarian/Pescatarian
  • Health concerns
  • Cooking related
  • The environment

Once a code frame has been established, all of the responses can then be assessed and allocated. Counts or tabulations of these codes can be valuable. However, when coding open-ended responses into codes, the analyst should still pay attention to the underlying text as well as the codes.

Coding Emotions
Another type of coding looks at the emotional message implied by responses. In this example, we are looking at a set of restaurant reviews, in order to assess sentiment. For example:

“The food tasted great”
“It was too expensive and the service was really slow”
“I had the tuna bowl and chips”
“They brought me the wrong meal”

Typical sentiment coding process seeks to code each response as one of: Positive, Negative, or Neutral.

For example, the coding of these reviews could be expanded to combine topic and sentiment, for example

“The food tasted great” – Food – Positive
“It was too expensive and the service was really slow” – Price, Service – Negative
“I had the tuna bowl and chips” – Food – Neutral
“They brought me the wrong meal” – Service – Negative.

Emotional Coding
Sentiment coding is simply one form of emotional coding, there are lots of alternative ways of coding emotion. For example, reviews or open-ended comments from a customer satisfaction study could be coded in terms of how people felt about the experience. One coding that is often used is based on Paul Ekman’s seven basic emotions: Anger, Contempt, Disgust, Enjoyment, Fear, Sadness, and Surprise.

The codes can be counted to provide a quantitative picture of the information, but counting is just one way of looking at what coding provides. For example, the analyst might filter the responses to look at all of the items that were coded as Sadness, to help understand what tended to cause Sadness and to review which words people were selecting to reflect sadness.

Coding but Not Counting
We can use coding to help understand qualitative information without even planning to try to turn it into numbers. This is particularly relevant when the information is deeper (which is often the case with qualitative research) and/or the sample size is small (which is also often the case with qualitative information).

For example, if we are looking at interviews with people about how they cook meals, we can use coding to crystalize meaning in our analysis. We might read the text and apply codes related to the question we are seeking to answer and our underlying knowledge of human behaviour. Looking at the text of people talking about cooking, we might highlight signals relating to ‘internal reward’, ‘status seeking’, ‘risk avoidance’, ‘confident’, ‘Insecure’, ‘collaborative’, and ‘self-focused’. Using this coding approach obliges the analyst to read the text with a specific lens. Whilst counting is not relevant, the broad categories of Frequent, Common, Rare and Absent can be useful.

Using Word Cloud Plus to kickstart your coding
I do a lot of coding, and as a consequence, I have helped create a product that can make it easier to kickstart the coding of open-ended information. Word Cloud Plus generates word clouds and in the pro version, you can access a simple coding tool. The big benefit is that this approach speeds up finding the key codes that you should be using.

If you would like to know more about this way to code open-ended information, check out this recorded presentation.

If you would like to try Word Cloud Plus, then click here.

When not to use Coding
At the top of the blog, I mentioned that I would talk about when not to use coding. Here are some examples:

  • When your data set is so small that you can grasp the meaning without having to decompose the information into its components.
  • In cases where you are using an alternative technique such as semiotics, discourse analysis, conversational analysis, and narrative analysis (these tend to use their own form of coding).
  • In situations where the frequency of concepts is not connected with the type of meaning we are looking for. If we were looking at a classic ‘whodunnit’ murder mystery novel, for example, one by Agatha Christie, and if we were trying to predict who had committed the murder, it is unlikely that codes and counts would be helpful (the author carefully lays false trails). However, if we wanted to conduct a review of which structures were more popular with readers, then codes might be very useful.