How to Triangulate Expert Interview Evidence
Test interview claims against independent records and other perspectives, recording disagreement without treating repeated views as proof.
Triangulating expert interview evidence means testing each proposition that may enter your findings against independent lines of evidence, then recording where those lines converge, diverge or fail to appear. The practical unit is the claim, not the interview. A claim ledger makes that discipline visible and prevents a repeated account from being mistaken for corroboration.
This article is about the evidential work that follows interviewing. It is not a consent guide or a guide to interview length, and it does not assume a single correct number of interviews. The scope is policy, regional and social research in which expert interviews inform a written finding, recommendation or briefing.
Start with propositions, not transcripts
An interview contains observations, interpretations, recollections and predictions. A finding contains propositions. The triangulation task is to move from one to the other without losing the distinction.
A proposition should be written so that evidence could contradict it. "Implementation is difficult" is too vague to test. "Applicants cannot complete the verification step without an adviser" can be tested against interview accounts, process documents, case records and the experience of applicants who did not use an adviser.
The same interview can be a primary source for one proposition and a secondary source for another. An expert describing their own decision is a witness. The same expert describing how another organisation behaves is reporting, even when the report is well informed. The ledger should record that difference, because it affects how much independent corroboration the proposition needs.
The UK government's Magenta Book chapter on quality in qualitative evaluation sets out appraisal questions for qualitative evidence within UK evaluation guidance. Its focus is the appraisal of a qualitative study; it does not supply a claim-level record of which propositions rest on one account. That record has to be built for the study at hand.
Which claims need corroboration?
Corroboration is not a uniform requirement. It should rise with the load a proposition carries and with the vulnerability of its source. A claim that will appear in a recommendation, a claim about a third party's conduct, a precise claim about dates or numbers, and a claim that would surprise a well-informed reader all need more than a single interview.
The following comparison is a suggested starting point, not a rulebook. It distinguishes claim types by the kind of independent line that would materially strengthen or weaken them.
| Claim type | Example form | What interview evidence can establish | What usually provides independent corroboration |
|---|---|---|---|
| Legal, statutory or procedural fact | "A verification step is required before payment" | How practitioners understand or apply the rule | Primary legal or procedural text; official guidance; the document itself |
| Descriptive implementation account | "Files wait at verification for several weeks" | What a witness observed in their own area of work | Case records, process data, direct observation, an account from outside the same reporting chain |
| Prevalence or frequency | "Most applicants abandon at verification" | That the speaker believes this, or has seen cases | Administrative data or a defensible sample; interviews are weak for frequency |
| Causal claim | "The portal caused the delay" | A perception of cause and a proposed mechanism | Evidence that distinguishes the mechanism from alternatives; disconfirming cases; documentary sequence |
| Actor motive or intent | "The team delayed deliberately" | An interpretation offered by the speaker | Documents, correspondence, independent accounts; treat motive claims as interpretations unless directly evidenced |
| Prediction | "The change will reduce waiting times" | The reasoning behind the prediction | Track record, pilot data, model assumptions; interviews support reasoning, not validation |
| Normative recommendation | "The step should be abolished" | Values, priorities and practical objections | Corroboration of the factual premises; values disagreement is not resolved by counting sources |
| Contested political claim | "The process is deliberately obstructive" | How the claim is framed and defended | Documents from more than one side; attention to what each source has access to and an interest in |
Two questions help set the corroboration demand. First, what happens to the recommendation if this proposition is wrong? Second, would the proposition still be credible if the most interested source had not said it? A claim that survives neither question should not be written as settled.
Are the sources dependent on one account?
Multiple interviews are not multiple sources when they draw on the same account. Dependency is a spectrum. At one end is a single person speaking twice; at the other are witnesses with separate access, separate interests and no shared documentary basis.
A useful test is counterfactual: would the content of source B change if source A had never spoken? If the answer is no, B may still be useful, but it is not independent corroboration of A.
Look for these forms of dependence:
- Referral-chain dependence. Interviewees were recruited through one organisation or one gatekeeper, so the sample reflects that network's view of who is relevant.
- Documentary dependence. Several interviewees cite the same report, briefing or internal summary. The document, not the interviews, may be the real source.
- Interest dependence. Sources share an institutional interest in one account of the problem, even if they describe different details.
- Reporting dependence. One interviewee is reporting what another interviewee told them, or what a meeting concluded.
- Period and scope dependence. Accounts refer to the same period, office or caseload, so they cannot illuminate variation elsewhere.
- Vocabulary dependence. Interviewees use the same distinctive phrases, which may indicate a shared briefing rather than independent observation.
The practical response is not to discard dependent accounts. They may be the best evidence of how a network understands a problem. The response is to label them accurately. In the ledger, record the basis of the interview account and the dependency note separately. Then state what independent line, if any, supports the proposition.
Documentary evidence is not automatically independent. A document produced by the same unit, or a summary of the same interview, belongs to the same account. Conversely, a document can be independent of an interview while still being aspirational, outdated or drafted for an external audience. Independence and reliability are different questions.
A claim ledger as a working format
The ledger below is a suggested working format, not a validated instrument or an institutional requirement. It is designed to be used before drafting findings, when the research team can still commission a follow-up interview, request a document or narrow a sentence.
Each row should contain:
- Proposition. A testable statement, preferably one sentence.
- Claim type. Using the categories above or your own.
- Interview basis. Who said it, in what capacity, and whether they were a witness or reporting.
- Dependence note. Shared network, document, interest, period or vocabulary.
- Independent corroboration. A document, a separate interview outside the chain, administrative data or observation.
- Disconfirming evidence sought. What was checked that could have contradicted the proposition.
- Status. Corroborated, converging but dependent, single-source, contested or unsupported.
- Use in report. Assert, attribute, caveat, use as background or drop.
A compact version is easier to maintain than a long form. The point is to make the evidential basis of a sentence inspectable by someone who did not conduct the interviews.
Fictional worked example. The study, organisations and figures below are invented. They describe a fictional licensing process; no real law, country or programme is intended.
A fictional research team conducted 12 expert interviews on implementation barriers in a licensing process. Three propositions were extracted for the ledger.
| Proposition | Interview basis | Dependence note | Independent corroboration | Disconfirming evidence sought | Status | Use in report |
|---|---|---|---|---|---|---|
| The online portal is the principal cause of delay. | Mentioned in 9 of 12 interviews. | Seven interviewees came through one advisory network; two repeated an adviser's account. | An applicant-support charity's anonymised case log showed the largest elapsed-time segment at document verification, not portal use. | Three applicants outside the advisory network said fee sequencing, not the portal, caused their delay. | Contested | Report as two mechanisms, not one principal cause. |
| A verification step must be completed before payment. | Described by five interviewees. | Two described the same published guidance. | The published process guidance states the sequence. | Checked whether an exception was documented; none found. | Corroborated | Assert with a document citation. |
| Staff can waive verification at their discretion. | Raised by two interviewees. | Both from the same office; one was reporting a conversation. | None found. | Asked the other office; no corroboration and no denial. | Single-source | Do not use as a finding; retain as a question for follow-up. |
The first row changes the writing. The original draft might have said: "The portal is the main barrier." The ledger supports a more bounded formulation: "The portal was the most frequently mentioned barrier among a connected group of adviser interviews. Case records point to document verification as the largest elapsed-time segment, while applicants outside that group described fee sequencing. The evidence does not support a single principal cause."
That formulation is longer, but it is also more useful. It tells the reader what was said, by whom, with what dependency, and where the contradiction lies.
How contradictions should appear in findings
Contradictions should appear as findings, not as noise to be smoothed away. The common failure is to write "some stakeholders felt X, while others felt Y" and leave the reader unable to judge which account has the stronger evidential basis. That sentence records disagreement without analysing it.
A better convention is to classify the contradiction before deciding how to present it.
- Definitional contradiction. The sources use the same word differently. "Delay" may mean time from application to decision for one source and time from a complete file to decision for another. Report the definitions.
- Scope or period contradiction. The accounts refer to different offices, caseloads or years. The disagreement may be variation, not conflict.
- Positional contradiction. A front-line account and an oversight account describe different parts of the same process. They may be complementary rather than incompatible.
- Interest-based contradiction. One source benefits from one account of the problem. Report the interest without treating it as proof of falsehood.
- Factual contradiction. Two accounts cannot both be true. State what would resolve the difference and whether it was resolved.
- Values contradiction. The sources agree on facts but disagree on priorities. Do not present this as an evidential failure.
When a contradiction remains unresolved, the finding should carry four elements: the proposition, the convergent evidence and its independence, the divergent evidence and its basis, and the implication for the recommendation. A short finding note can follow this form:
Proposition: Verification is the main elapsed-time barrier.
Convergent evidence: Case records and two independent applicant accounts.
Divergence: Adviser interviews emphasise the portal; most come from one network.
Assessment: The portal is a reported burden, but the evidence does not establish it as the principal cause of elapsed time.
Implication: The recommendation should address verification capacity and portal usability separately.
Silence also belongs in the record. If a document that should exist cannot be found, or if no interviewee outside one network can describe a step, that absence may be relevant. It should be reported as an absence of corroboration, not as proof that the step does not exist.
Avoid vote counting. Five interviews against three do not settle a factual question when the five share a briefing and the three have direct access to the records. The ledger should show the basis of each account, not a tally.
Limitations and judgement
Triangulation cannot repair a sampling frame that excludes the people who would disagree. It cannot validate a shared assumption held across an entire policy network, because every source may have learned the same account. It cannot make a retrospective interview into a contemporaneous record.
Ethical constraints also limit corroboration. Confidentiality may prevent naming a source or approaching another office without risking identification. In that case, record the category of corroboration and the reason for the limitation, but do not put identifying detail into the report. A corroborating source can be described by role and basis without being named.
The ledger is a record of evidential support, not a truth machine. Its value is that it forces the writer to distinguish assertion from attribution, convergence from repetition, and a genuine contradiction from a difference in scope. That distinction is the difference between an interview-informed finding and an interview-supported impression.
Before the next round of drafting, take the five propositions most likely to appear in recommendations and give each a ledger row. Mark any proposition with no independent line. If none exists, the finding is a single-account finding, and the sentence should say so. For a related comparison of how written research and expert calls contribute different kinds of evidence, see written research versus expert calls.