Interview Design in Project Research
Three examined works ran interview strands, and no two of them made the same decisions about who to approach, what to ask or what to keep. One of the three also mislabelled its own method in print, which is easier to do than you would like.
Four studies, three interview protocols
The supplied teaching source lists interviews as one of thirteen data collection methods and defines none of them. Its only applied guidance is a single worked example, where a question about how managers and workers apply ethical standards is answered by "interviews or questionnaires follow up of responses to specific applications" and labelled qualitative. Everything else about interview design has to come from elsewhere.
The examined works supply four points of comparison — three with described protocols and one where no protocol is described at all.
INTERVIEW STRANDS ACROSS THE EXAMINED WORKS
| Decision | Exemplar B (thesis) | Exemplar C, paper 3 (portfolio) | Exemplar D (journal paper) |
|---|---|---|---|
| Number conducted | 10 | 8 | 9 |
| Design named | One-on-one semi-structured | In-depth, developed into case studies | Ten open-ended questions; called 'unstructured' in the abstract |
| Mode and length | Telephone, about one hour each | Not stated in the material available | Not stated in the material available |
| How participants were found | 28 experts approached via a professional networking site | Volunteers, described as 'an ideal sampling profile' | Experienced R&D project managers from three organisations in three different industries, two of them from the case-study organisation |
| Participant profile reported | All 20+ years' experience; 80% held doctorates; each coded with two letters and profiled in a table | Professionals whose recent projects became the cases | Experienced project managers in the field studied |
| Size justified by | A stated target of 'at least ten' | An assertion that the volunteers were an ideal profile | Selection criteria rather than size |
| Recording | Recorded where permitted, otherwise captured by notes | Not stated in the material available | Not stated in the material available |
Every figure is a finding of an individual study. None is a minimum, a target or a norm for interview-based research. Exemplar A, the insider action research thesis, names seventeen data sources gathered over a three and a half year field period but no interview protocol is described in the material available, so it cannot be compared here.
Recruitment: who was approached, and who arrived
Exemplar B approached 28 experts and completed 10 interviews. That ratio is the interview equivalent of a response funnel, and reporting both numbers does the same work: it shows the reader how much selection happened between the sampling intention and the people who actually spoke.
The participant profile that follows is unusually specific — every interviewee had at least twenty years of experience and 80 per cent held doctorates. Note what that buys and what it costs. It buys authority for statements about a technical domain. It costs the perspective of anyone less senior, which is a limitation whether or not the thesis says so.
Exemplar D took the opposite approach to breadth. Its nine participants were drawn from three organisations in three different industries, with two from the case-study organisation itself. That spread is a deliberate hedge against a single-organisation account, in a paper whose other strand is a two-project case study inside one organisation.
Asking around the question instead of asking it
The most transferable single tactic in the examined set comes from Exemplar C, and it is stated explicitly in the work itself.
The failure mode this avoids is familiar to anyone who has run a lessons-learned workshop. Ask a project manager directly whether a project succeeded and you get a position, usually a defensible one, and usually the one they have already told their sponsor. Ask them to rate fifteen specific things and the position has to be assembled from the ratings — by you, afterwards, on the record.
The second half of the tactic matters as much. Because the item list came from nine named prior sources, the instrument is traceable to the literature rather than to the researcher's intuitions, and a reader can go and check where each item came from. That is the review-to-instrument link that most methodology chapters assert and few demonstrate.
When the label on your own method is wrong
Exemplar D's abstract describes unstructured interviews. Its methods section describes a fixed set of ten open-ended questions put to nine participants. Those two descriptions cannot both be right, and the second one is the accurate account of what was done.
A fixed question set asked of every participant is a structured protocol with open responses — a semi-structured design. Calling it unstructured claims a different kind of study: one where the agenda emerges in conversation and each interview can go somewhere the last did not. A reader deciding whether to trust the cross-participant comparisons needs to know which of those happened.
Which label your protocol has actually earned
Recording, coding and what was left undocumented
Exemplar B recorded interviews where permitted and captured the rest by notes. That is a mixed-fidelity dataset, and the honest move is exactly what the thesis did: state it, rather than implying a uniform recording standard that did not exist.
Analysis was done manually. The thesis states plainly that with only ten interviews, coding software was not needed. That is a defensible proportionality judgement, stated as one — considerably better than naming a software package to signal rigour that ten transcripts do not require.
What the interviews were actually for
The three strands play three different roles, and the role determines nearly everything about the design. This is the question to answer before writing a single question.
Three jobs an interview strand can do
Explaining a quantitative result
In Exemplar B the interviews follow the survey in a sequential explanatory design, and the qualitative chapter's sections deliberately mirror the quantitative chapter's. The interviews interpret specific survey findings rather than floating free. The design is unpacked in Mixed Methods Design: A Worked Example.
Producing the cases themselves
In Exemplar C each interviewee's recent project was developed into a case study. The eight cases split five successful and three unsuccessful, and the analysis contrasts the two groups. Here the interview is not evidence about a case — it is the means of constructing one.
Establishing that the problem is real
In Exemplar D the nine interviews establish that practitioners want a more structured risk methodology. None of the nine had used the technique the paper applies, so the interviews warrant the relevance of the work rather than evaluating it. That is a legitimate role, provided the paper does not later cite them as evidence the technique works.
Designing an interview strand you can defend
From role to protocol
Decide the job first
Interpret a quantitative finding, generate cases, or warrant relevance. The three produce different question sets, different sampling and different claims. Write the job down in one sentence.
Set a target and a stopping rule before recruiting
Exemplar B set 'at least ten' and reported approaching 28 to get there. A target stated in advance survives contact with a low acceptance rate; a number arrived at afterwards looks like whatever you managed.
Write the selection criteria, not the wish list
Exemplar D justified nine interviews by criteria — experienced managers in the relevant project type, spread across organisations and industries. Criteria you can state are criteria a reader can check.
Build the instrument from the literature
Derive rateable items from named prior sources, so each item has a provenance. Add one residual question so material outside the literature has somewhere to go.
Decide whether to ask the headline question at all
If asking directly would produce a rehearsed position, approach the construct indirectly and assemble it from the parts. Then say in the methodology that you did this, and why.
Fix the recording and consent position per interview
Recorded where permitted, notes otherwise, is workable — as long as the dataset's mixed fidelity is stated rather than smoothed over.
Reproduce the guide
One appendix page. It is the single cheapest thing you can do to make a qualitative strand checkable, and almost nobody does it.
Before the first interview
- The job the strand does is written in one sentence, and the questions serve that job
- A target number and the criteria for inclusion exist in writing, dated before recruitment
- Every item can be traced to a source, a prior finding or an explicit design decision
- There is at least one open residual question for material the literature has not named
- The label you will use — structured, semi-structured, unstructured, in-depth — matches what the protocol actually does
- Consent and recording arrangements are settled, including what happens when recording is refused
- You know how participants will be identified in the write-up, and how many voices a claim will need before you make it
For the strand that usually precedes or follows the interviews, see Survey Design and Response Rates in Practice; for what the teaching source does and does not supply on methods generally, see Data Collection Methods: Data as Evidence.
What to carry forward
- Decide whether your interviews interpret, generate or warrant before you draft a single question.
- Report both numbers — approached and completed — the way one examined work reported 28 approached and 10 completed.
- Defend the sample by a target set in advance or by selection criteria. Calling your volunteers an ideal profile is an assertion, not a defence.
- Approaching a construct indirectly, by rating items rather than asking the headline question, is a stated tactic in one examined work and it is worth copying.
- Derive items from named prior sources and add one residual question for what the literature missed.
- Check that your abstract's method label matches your protocol. One published paper in the examined set does not.
- Reproduce the interview guide. It costs a page and it is the only way a reader can see what was asked.
Frequently asked questions
How many interviews are enough?
The supplied teaching source gives no guidance and the examined works give three different answers with three different justifications: ten against a stated target of at least ten, eight volunteers asserted to be an ideal profile, and nine defended by selection criteria rather than by number. None of these is a standard. Set a target in advance, justify it by the job the strand does, and report how many people you approached to reach it.
What is the difference between unstructured and semi-structured interviews?
The supplied teaching source defines neither; interviews appear only as a method label. Operationally, a fixed question set asked of every participant with open-ended answers is semi-structured, while unstructured means no prepared question set at all. One examined paper calls its interviews unstructured in the abstract while describing ten fixed open-ended questions in the methods section, which is a semi-structured design.
Should I ask participants my research question directly?
Not always. One examined work deliberately never posed its headline question, approaching the construct indirectly by asking participants to rate items drawn from nine named prior sources, specifically to avoid influencing responses. Direct questions on a contested construct tend to return rehearsed positions; indirect ones let you assemble the position from the parts and show your working.
Do I have to record every interview?
One examined thesis recorded where permitted and captured the remainder in notes, and said so. That is a workable arrangement provided the mixed fidelity of the resulting dataset is stated rather than implied away. Settle the recording and consent position before the first interview, including what you will do when a participant declines to be recorded.
Do I need coding software for a qualitative strand?
Not necessarily. One examined thesis analysed its interviews manually and states plainly that with only ten interviews, coding software was not needed. Naming a package adds nothing to a small dataset; what matters is that the coding procedure is described and, ideally, that someone else has checked a sample of it.
Should the interview guide go in an appendix?
Yes, and it usually does not. Across the examined works, survey instruments are reproduced at length while interview guides, transcripts and consent forms are largely absent, so the quantitative strands are documented for replication and the qualitative ones are not. A one-page guide is the cheapest way to make your qualitative strand checkable.
References and source attribution
- Bryman, A. 2016, Social Research Methods, 5th ed., Oxford University Press, Oxford.
- O'Leary, Z. 2017, The Essential Guide to Doing Your Research Project, 3rd ed., Sage Publications, London.
- Naoum, S. G. 2013, Dissertation Research & Writing for Construction Students, 3rd ed., Routledge.
- Quinlan, C. 2011, Business Research Methods, 1st ed., Cengage Publishing, Chapter 1, p. 50 — the methodological pyramid, on which interviews appear as a data collection label.
- The examined exemplar set supplied with this library: five research works comprising two doctoral theses, a portfolio-format professional doctorate, a journal paper and a conference paper. Authors, institutions, employers and publishers are not named, per the terms under which this library was written.
- The supplied teaching source: consolidated week 2 to week 4 teaching notes and slide material on research methods, used here as the basis for all statements attributed to "the teaching source".
Suggested questions for Ask KEVOS
- Help me decide whether my interview strand should interpret a survey, generate cases or warrant relevance.
- Draft selection criteria and a target number for an interview study with project managers in two industries.
- Turn my research question into a set of rateable items so I never have to ask it directly.
- Check whether my protocol is structured, semi-structured or unstructured based on how I plan to run it.
- Write the recording and consent paragraph for my methodology chapter.