Coding the exit interview
Introduction to Social Research Methodology
Aim
This exercise puts today’s session into your hands. You will open-code a real-length passage of qualitative data, discover how differently two honest coders can read the same page, build a merged codebook, group codes into candidate themes, and practise the discipline that separates description from interpretation. Everything you do here is a miniature of the full analytical workflow — familiarisation, coding, categorising, theming, interpreting — applied to one Meridian exit interview.
The transcript
Meridian’s HR team records a short exit interview with every voluntary leaver. Below is the transcript of one of them: a warehouse operative from the e-commerce arm, leaving after fourteen months. I is the interviewer; R is the respondent. The transcript is printed with line numbers so that your codes can refer to exact locations. Read it once, in full, before you code anything — familiarisation comes first.
| Line | Transcript |
|---|---|
| 1 | I: Thanks for making time. You’re leaving us after fourteen months — |
| 2 | can you talk me through what led to the decision? |
| 3 | R: Honestly? No one thing. It builds up. The job itself |
| 4 | was fine, mostly. I liked the work — when I could plan it. |
| 5 | I: When you could plan it? |
| 6 | R: The rotas. My schedule changed three times last month alone, |
| 7 | and twice I found out from the team group chat, not from |
| 8 | anyone actually telling me. You can’t build a life around that. |
| 9 | My flatmate works at Domex and knows her shifts a month out. |
| 10 | I: Okay. What about pay — did that play a part? |
| 11 | R: I mean — yes and no. The money’s not terrible. But a |
| 12 | friend of mine does basically my exact job at Arkona and |
| 13 | she’s on, what, fifteen per cent more? Plus proper overtime rates. |
| 14 | When I found that out I just thought, why am I here? |
| 15 | I: Was there anything that made you want to stay? |
| 16 | R: The team, one hundred per cent. Best people I’ve ever |
| 17 | worked with. We covered for each other, we had a laugh even |
| 18 | on the awful days. Leaving them is genuinely the hard part. |
| 19 | I: And your manager? |
| 20 | R: Magda’s brilliant — I want that on record. She fought to |
| 21 | get us extra hands for the Christmas peak and got told no. |
| 22 | She’d fix the rota chaos herself if they let her. It’s |
| 23 | not her. It’s above her. |
| 24 | I: Above her meaning…? |
| 25 | R: Head office. Look — e-commerce is meant to be the growth |
| 26 | part of the company, right? It’s in every board presentation. But |
| 27 | nobody from HQ has ever once asked us what’s slowing the |
| 28 | picking line down. We’re a number on a dashboard to them. |
| 29 | I filled in that engagement survey last year — total silence after. |
| 30 | I: If you could change one thing, what would it be? |
| 31 | R: Give people their shifts two weeks ahead, minimum. That’s it. |
| 32 | That costs nothing. And maybe someone senior visits the warehouse once |
| 33 | a quarter and actually listens. Small things. But after fourteen months |
| 34 | of waiting for the small things, Arkona rang, and — well. Here we are. |
Task 1: open-code the transcript (15 minutes, solo)
Work alone — that is essential, because Task 2 depends on you and your partner coding independently. Do not confer.
Open-code the transcript: work through it line by line and attach codes freely, without any predetermined list. For each code, record:
- a short label (a word or short phrase — e.g.
scheduling chaos, not a full sentence); - the line number(s) it covers.
Requirements and guidance:
- Produce at least 8 codes. More is fine; early open coding is supposed to be generous.
- A code may cover one line or a span of lines (e.g. lines 6–8), and the same lines may carry more than one code.
- Mix altitudes if it feels natural: some codes may be descriptive (staying close to what was said) and some interpretive (naming what you take a passage to mean). Note for yourself which are which — you will need this in Task 3.
- Code what is in the data, not what you expect an exit interview to contain. If something surprises you, code it.
Set your list out like this:
| Code label | Line(s) |
|---|---|
| your label | e.g. 6–8 |
Task 2: compare codebooks and build themes (15 minutes, pairs)
Now pair up and put your two code lists side by side.
Step 1 — compare (about 7 minutes). Go through the transcript passage by passage and answer:
- Where did you both code the same passage — and did you give it the same label, or different ones?
- Where did one of you code a passage the other ignored entirely?
- Where did you apply codes at different altitudes — one descriptive, one interpretive — to the same lines? Decide together which altitude serves the research question (why are people leaving Meridian?) better, and why.
Treat every disagreement as information, not as an error: each one is telling you something about a vague label, an ambiguous passage, or a difference in interpretive stance. Discuss it before you resolve it.
Step 2 — merge and theme (about 8 minutes). Agree a single merged codebook: keep, rename, merge, or drop codes until you have one list you both accept. Then group your merged codes into 2–3 candidate themes and record them in this table:
| Theme (a claim, not a topic) | Codes it contains | Best supporting quote (with line numbers) |
|---|---|---|
Remember the standard from today’s session: a theme is a claim, not a topic. “Scheduling” is a topic; “unpredictable hours make staff feel interchangeable” is a theme. Your theme names should assert something a sceptic could dispute.
Task 3: description versus interpretation (5–10 minutes, pairs)
Choose one of your themes. Write two sentences about it:
- one sentence of description — something the transcript itself supports directly, which the respondent could read and endorse;
- one sentence of interpretation — an analytical claim that goes beyond the words, which only the analyst could assert.
Label which is which. Then apply today’s test to each sentence: could a participant have said this? If your “description” fails the test, sharpen it; if your “interpretation” passes the test, it is not interpreting enough.
Plenary discussion
Each pair briefly presents its themes. As a group, we compare:
- Did different pairs find substantially different things in the same 34 lines? Where do the theme sets overlap, and where do they diverge?
- Where pairs diverged: is that a flaw in the method, or exactly the problem that intercoder comparison, codebooks, and the audit trail exist to manage? What, concretely, would a research team do next about the divergence?
- One transcript cannot support a finding. What would you need to see across the other eleven transcripts before any of today’s themes could go into a report to Meridian’s board — and how would negative-case analysis figure in that?
Keep your merged codebook and theme table: they are worked examples of the concepts — codes, codebooks, themes, intercoder comparison, description versus interpretation — that the assessment may ask you to define and apply.