Paper 3 question 3 is worth 6 marks and sets one of three tasks. How the researcher could improve credibility. How they could reduce bias. Or how far the findings transfer to other people or settings.
All three are asking the same underlying thing: can you judge qualitative research by qualitative standards, rather than by standards borrowed from an experiment?
That is the whole question, and most lost marks come from bringing the wrong toolkit.
Two vocabularies for related questions
Quantitative and qualitative research ask overlapping quality questions and answer them with different words. Question 3 needs the right-hand column.
| The question | Quantitative word | Qualitative word |
|---|---|---|
| Do the results reflect what really happened? | Internal validity | Credibility or trustworthiness |
| Does this apply to other settings? | Ecological validity | Transferability |
| Is the process documented well enough to follow? | Reliability | Dependability |
| Are findings grounded in the data rather than assumptions? | Objectivity | Confirmability |
Treat this as a useful parallel rather than an exact conversion. The point is having the right vocabulary available, not memorising a table.
Three sentences that cost marks
"This reduces the validity of the study."
For question 3 the expected vocabulary is credibility, trustworthiness and transferability. Generic validity language signals that you are evaluating a qualitative study as though it were an experiment. Name the actual concept instead. If the issue is whether the account reflects what really happened, that is credibility. If it is whether findings might apply to a comparable setting, that is transferability.
"The sample was only nine people, so it cannot be generalised."
Small N on its own is not a sufficient criticism. Ask what kind of claim the study is making, then judge the sampling against that. If the study claims something about a wider population, discuss whether its sampling logic supports that leap. If the question is about transferability, discuss whether the researcher gave enough detail about who these nine people were for a reader to judge whether findings might carry across.
Generalisability is not a banned word. Qualitative work does discuss generalisation, in different forms. What fails is treating statistical representativeness as the only kind that counts.
"The researcher's presence may have introduced bias."
True, and it stops at the obvious. In an interview the researcher is effectively the measuring instrument, and that influence is part of the method. Ask what the researcher did about it.
The rule is worth memorising: in an experiment, researcher influence is a problem to design out. In qualitative research it is part of the method, to be reduced where possible and made transparent where it cannot be.
Know what kind of qualitative source you are looking at
Thematic analysis is the most common in Paper 3. The researcher reads transcripts repeatedly, codes sections, and groups codes into themes. You spot it because the source presents two to four named themes with participant quotes underneath. The key point: themes are the researcher's interpretation, not something sitting in the data waiting to be found. Two researchers could reasonably arrive at different themes. That is not automatically a flaw, but it is why transparency about the process matters.
Content analysis defines categories first, then applies them systematically. You spot it by defined categories reported as counts or percentages. The weak point is that the categories themselves are a judgement.
Grounded theory builds a theory from the data, cycling between collecting and analysing. You spot it through several rounds of data collection or mention of saturation. The weak point is that the theory comes from one sample in one context, so applying it elsewhere needs argument rather than assumption.
Credibility, and how researchers build it
Credibility asks whether the account is a believable, well-supported reflection of what participants actually experienced.
It does not happen by accident. The tools are specific.
Triangulation. More than one source, method or researcher, checked for whether they point the same way. Member checking. Taking findings back to participants to ask whether the interpretation fairly represents their accounts. It strengthens credibility. It does not prove one reading correct, and where participants disagree, that disagreement is itself information. Prolonged engagement. Enough time with participants to get past surface answers. Iterative questioning. Returning to unclear or contradictory answers later, rephrased, to see whether the account holds. Rapport. Enough trust for full answers. Thin answers make a thin account.
Transferability and thick description
Transferability asks whether findings might reasonably apply to a similar context. Not statistically, but plausibly.
The crucial point is that this is the reader's judgement to make, and the researcher's job is to make that judgement possible.
That is why thick description matters: a rich account of who the participants were, where the study happened, and what the circumstances were. Without it, transferability cannot be assessed at all, because there is nothing to compare against.
Reflexivity
Reflexivity is the researcher openly reflecting on their own background, assumptions and position, and how these shaped what they noticed and how participants responded to them.
It is the direct answer to the fact that the researcher cannot be removed. They are not expected to pretend they had no influence. They are expected to be transparent about what that influence probably was.
A study that is transparent about the researcher's role is doing what is expected. A study that presents its findings as though the researcher were invisible is the one with a credibility problem.
Say which side the bias comes from
Do not write that there could be bias. Say which side, and how.
Researcher bias comes from the person asking. Leading questions. Selective attention to what fits expectations. Confirmation bias when deciding what counts as a theme.
Participant bias comes from the person answering. Social desirability. Demand characteristics. Acquiescence, meaning agreeing with whatever is said.
Naming which one and explaining how it operates is often the difference between two marks and five.
Match the fix to the flaw
This is the move that separates a strong answer from a list.
Weak answers name tools at random. "The researcher could use triangulation, member checking and reflexivity." That is a shopping list. It shows recognition, not understanding.
Strong answers name the problem first, then the tool that solves that specific problem, then the mechanism.
The researcher coded all nine interviews alone, which means no one checked her interpretation of the data. She could involve a second researcher in the coding and compare the themes each identified. Where their interpretations differed, that disagreement would itself be informative rather than being settled by one person's judgement.
Problem. Tool. Mechanism. That is the shape.
How to write the answer
Six marks. Aim for two or three developed, source-specific points rather than six brief ones. The mark band is holistic, so there is no one-point-per-mark formula.
Orient briefly with one line on what kind of source this is. Then for each point: name the specific feature of this source, name the concept properly using qualitative vocabulary, explain the mechanism, and if the task asks how to improve it, give the tool that fits that particular problem.
Then close honestly. Acknowledge that researcher influence is part of qualitative inquiry, and that the aim is to reduce and disclose it rather than pretend it is absent. That closing move is worth real marks because it shows you understand what qualitative research is trying to do. For a transferability question it is usually unnecessary.
Do not skip the improvement. Two of the three tasks ask how the researcher could improve credibility or avoid bias. An answer that lists problems without saying what could be done differently has answered half the question, however accurate the problems are.
On timing, questions 2 and 3 share around 35 minutes, so roughly 17 each including reading.
The test to run before you finish
If your answer would apply to literally any interview study ever conducted, you have not used the source.
If you want the full worked source with graded answers at each band, that is what the PsychExcel Paper 3 Complete Guide is built for.