Process Coding 101 (With Examples)
Coding for action: how to capture what people actually do
The 30-second summary
Process coding labels what people do. The codes are verbs in the "-ing" form, so reading them in order gives you the sequence rather than a pile of topics.
- Use it when your research question is about processes, routines or how something changes over time.
- The craft is in the verb: "researching options", not "research" (that is a topic) and not "talking" (too vague to group with anything).
- Your codes become stages, not themes, which is what lets you show a process without quoting a single transcript.
- Not the right tool when your aims are about meaning, when the data has no sequence in it, or when you are applying a framework rather than letting the actions emerge.
If you want the topic of a passage rather than the action in it, you want descriptive coding instead.
Navigate
Process coding is the technique that codes for action: what people do, rather than what they think or feel. This post covers what it is, how to pick the right gerund, when it fits your aims, when it does not, and how it differs from the technique it is most often confused with.
First, the big picture
Qualitative coding is the process of categorizing and labeling textual data. It lays the foundation for identifying themes and patterns, and ultimately for extracting insights, which makes it the first step in the broader qualitative analysis process.
Coding comes in three overarching approaches: inductive (codes emerge from the data), deductive (codes come from theory) and hybrid. Process coding is inductive, since the actions you code can only come from what participants describe.
Which approach fits your study depends on your research aims. Our guide to inductive vs deductive coding compares all three, and works through the same extract coded inductively and then deductively. On a large set of semi-structured transcripts it is also common to run structural coding first, purely to index the data by research question before any of this starts.
What is process coding?
Process coding focuses on actions, events and processes, which in practice means coding verbs and gerunds: words ending in “-ing”, such as running, jumping or singing.
In a study of new education policy you might code ‘implementing reforms’ or ‘training teachers’; in one following patients through long-term treatment, ‘complying’ or ‘recovering’. The codes track what is being done, which is what lets you follow a sequence through a dataset. That makes the technique a good fit whenever your aims involve processes, routines or change over time.
What it looks like on the page
Here is an extract from an interview with a small business owner about adopting new software:
| Interview extract | Process code |
|---|---|
| “We started by looking at what was actually slowing us down.” | Evaluating needs |
| “I spent about a month comparing three systems” | Researching options |
| “and talking to other owners who’d used them.” | Seeking peer advice |
| “We mapped out how it would slot into what we already had” | Planning integration |
| “and I ran the team through it over two afternoons.” | Training staff |
Read the code column on its own and you have the sequence: evaluate, research, consult, plan, train. That is the thing process coding gives you and the other techniques do not.
Choosing the right gerund
The technique is easy to describe and fiddly to do well, because most of the work is in picking the verb. Three rules that will save you a re-code:
- Use the -ing form, not the noun. ‘Researching options’ is a process; ‘research’ is a topic. The moment your codes become nouns you have drifted into descriptive coding without noticing.
- Name what the action achieves, not the surface behavior. ‘Talking to other owners’ is what happened; ‘seeking peer advice’ is what it was for, and it will group with other codes meaningfully.
- Stay at one level of granularity. If ‘researching options’ sits beside ‘opening a browser tab’, your categories will not hold together. Pick the altitude that matches your research question and stay there.
How to do process coding
Process coding runs in four steps.
- Collect and compile your dataset. Interview transcripts, field notes, or secondary data such as policy documents.
- Read through once, without coding. Understand the dataset as a whole rather than getting tunnel vision on one or two portions of it, and note the patterns you see.
- Read again, applying codes. New codes surface as you go, so build the list and cycle back to apply them, until everything is coded consistently.
- Group your codes into categories. With process coding these categories usually turn out to be stages, because that is the shape actions take when you line them up.
Once you’ve applied your codes and categorized them into logical groups, your dataset should, in principle, be ready for analysis. This might take the form of something like thematic analysis or content analysis, depending on your specific research aims. Our guide to choosing a qualitative analysis method covers the main options.
Process coding vs descriptive coding
These two get mixed up more than any other pair, and the difference is what part of a passage you are labeling.
| Process coding | Descriptive coding | |
|---|---|---|
| What it labels | The action in the passage | The topic of the passage |
| A typical code | ‘Researching options’ | ‘Technology purchase’ |
| Question it answers | What is happening here? | What is this passage about? |
| Best for | Sequences, routines, change over time | Indexing a large dataset so you can navigate it |
A quick test: if your code would still make sense as a chapter heading, it is descriptive. If it only makes sense as something someone is doing, it is a process code.
When not to use process coding
Process coding is not a default, and the cases where it works against you are worth knowing before you commit a dataset to it.
- When your aims are about meaning rather than activity. If you want to know what an experience meant to participants, coding their actions will miss it. In vivo coding or values coding fit better.
- When the data has no sequence in it. Open-ended survey responses often describe opinions, not processes. Forcing gerunds onto them produces codes nobody can categorize.
- When you are testing a framework. Applying an existing framework is deductive work, and your codes need to be its constructs. There is a subtlety here worth getting right: if the framework is itself about processes, such as a stage model or a care pathway, your codes will look exactly like process codes, all gerunds and actions. They are not. What makes process coding process coding is that the actions emerge from the data, not that the codes end in “-ing”. Code them deductively, and say so in your methodology.
- When one action dominates everything. If most of your data is people doing the same thing, process coding flattens rather than distinguishes. Code the variations instead.
Still have questions?
Do process codes have to end in “-ing”?
Almost always, and it is a useful discipline rather than a rule for its own sake. The gerund forces you to name an action in progress, which is the whole point of the technique. If you find yourself wanting a noun, that is usually a sign the passage is better served by descriptive coding.
What if a participant describes something they did not do themselves?
Code it anyway, and note who the actor is. ‘Waiting for approval’ is a process even though the participant is not the one approving. Studies of bureaucracy and healthcare are full of actions done to people rather than by them, and those are often the most revealing codes in the set.
How is process coding different from a timeline?
A timeline records what happened in order for one case. Process codes describe kinds of action that recur across cases, which is what lets you compare. If your codes only make sense in the order one participant said them, you have written a timeline and will struggle to categorize it.
Can I use process coding on my whole dataset?
You can, but most studies use it on the parts where activity is the question and something else elsewhere. Coding a whole dataset for action when half of it is people describing how they felt produces a lot of codes that never earn a category.
Can’t find your answer here? Ask a Grad Coach directly – the initial chat is free.
Speak with a friendly coach →
