Using Negative Cases in Qualitative Analysis
Examine accounts that challenge an emerging explanation, then revise the finding or state the conditions under which it remains defensible.
A negative case is an instance that contradicts, complicates, or fails to fit a working explanation, and analysing it deliberately is how a qualitative study tests its claims rather than merely illustrating them. The discipline lies less in finding such a case than in recording what it challenged, what else could account for it, and what the analysis now says. This article sets out how to distinguish a contrary case from a differently situated one, how to preserve the reasoning behind a revision, and how to keep a challenge log that makes the process visible to readers and commissioners.
What a negative case is, and what it is not
A negative case is defined against a specific claim. If the draft finding is that a service goes unused because people do not know about it, the negative case is a respondent who knew about it and still did not use it. If the claim concerns how a decision was reached, the negative case is an account of that decision that runs the other way. Without a stated claim there is nothing to contradict, so the first task is to write the explanation in a form that could be wrong.
Two errors follow from treating every surprise as a contradiction.
The first is to convert a difference in circumstance into a refutation. A case drawn from a different local authority, a different eligibility group, or a period after a policy change may behave differently because the conditions differ, not because the explanation is false. That is a boundary condition to state, not a defeat.
The second error is the reverse: to treat a genuine contradiction as mere context to protect the original claim. The test is whether the contextual difference was specified in advance as relevant. If the explanation never claimed to cover people with existing professional advice, a case involving such a person limits the scope but does not refute the claim. If the explanation was meant to cover all non-users, the same case does refute it. Deciding which is true requires evidence about the scope of the claim, not preference for the tidier story.
Retaining the reasoning, not just the conclusion
Published findings typically present the revised explanation and drop the route taken to it. That is a loss. A reader cannot judge whether a revision was warranted, and a later analyst cannot tell whether an apparent contradiction was examined or overlooked.
Retaining the reasoning means keeping a record of four things: the explanation as it stood before the challenge, the account that challenged it, the alternatives considered, and the check that distinguished between them. Superseded wording should be kept rather than overwritten, because the sequence is the evidence of analytic work. Where a check was inconclusive, say so; an unresolved negative case reported honestly is more useful than a premature resolution.
This is also where qualitative appraisal standards bear on the question. The UK Government Social Research framework Quality in Qualitative Evaluation, published within HM Treasury's Magenta Book, asks appraisers to consider whether design and sampling are justified, whether coverage and missing perspectives are explained, and whether collection, analysis and reporting are transparent (Quality in Qualitative Evaluation). A negative case log responds directly to the second and third of those concerns, since it documents which perspectives were sought, which were absent, and what difference their absence makes to the conclusion. The framework sets an appraisal standard rather than a prescribed technique; the working format below is a practical suggestion for meeting it, not an instrument it validates.
A challenge log as a working format
The log below is offered as a suggested structure for recording each serious contradiction as it arises. It is not a validated instrument and no funder or ethics committee requires it in this form. Its value is that it forces the analyst to separate what was observed from what it might mean, and to specify in advance what would settle the question.
| Field | What to record | Common failure |
|---|---|---|
| Explanation under test | The claim as written, with its intended scope: which population, setting and period it covers | A vague claim that nothing could contradict |
| Contrary account | What the participant said or did, in their terms, with enough detail to be recognisable | Paraphrase that smooths the disagreement |
| Context | Who the person is, how they were recruited, what differs from the main sample, and any access constraint | Context used as an excuse rather than a condition |
| Alternative interpretation | At least two readings: the explanation is false; the explanation holds but its scope is narrower; the data are unreliable | Settling on the first reading that preserves the original claim |
| Further check | The specific step that would distinguish the readings: a new prompt, a re-read of existing transcripts, a search for comparable accounts, a targeted additional interview | A check that could not fail |
| Revised finding | The claim as it now stands, with confidence and limits stated | Revision recorded without the prior wording |
Two conventions make the log more useful. Date each entry and note who reviewed it, so that the record shows when a challenge entered the analysis. And keep a status column for entries that remain open, since unresolved cases are the ones a reader most needs to know about.
The further-check column deserves particular care. A check is only informative if the possible outcomes would lead to different conclusions. "Discuss with the team" is not a check. "Search all transcripts for respondents who mentioned awareness and then gave a reason for non-use" is one, because finding several such accounts changes the revised finding and finding none leaves the original claim standing with a noted exception.
A fictional worked example
The following example is invented to show the format in use. It does not describe a real study, client, or dataset.
A local-service study examines why an advice service has low take-up among residents on lower incomes. The draft explanation, based on interviews with non-users, is that non-use is chiefly a matter of low awareness.
One interview is conducted with a resident who is not a user but is unusually well informed: she receives the council newsletter, had read about the service, and had used a similar service in another area some years earlier. She describes deciding not to use it, citing a previous experience she found unhelpful and a preference for dealing with her housing issue through her landlord's solicitor.
Entered in the log, this becomes a challenge to the awareness explanation. The context field records that she differs from most participants in having prior experience of a comparable service, which may put her outside the population the original claim described. Two alternative interpretations are written down: that awareness is necessary but not sufficient, and that non-use among aware residents has its own reasons which the study has not yet characterised. A third possibility is also noted: that her account reflects a single strong prior experience rather than a settled view, and should not be over-read.
The further check is to re-read all transcripts for any mention of prior advice-seeking or of reasons for non-use given after awareness was established, and to add a prompt on prior experience to the remaining interviews. The revised finding states that low awareness is one barrier among several, that a distinct group of aware residents describes deliberate non-use, and that the size of that group cannot be estimated from this design.
Two boundaries apply to the example. The single contrary case does not support any statement about how common deliberate non-use is; estimating prevalence requires a design built for that purpose. And the disagreement should not be dissolved into a consensus formulation that loses what the participant actually said. "Residents have mixed views" would discard the specific reasoning the case was valuable for.
Implications for how a study is commissioned and read
The log works best when its requirements are agreed before fieldwork. Those commissioning analysis can ask for the claim to be stated in falsifiable form, for the sampling plan to name which dimensions of variation might produce contrary accounts, and for the negative case log to be delivered alongside the findings. Those conducting the analysis can keep the log from the first analytic memo, since reconstructing challenges after the report is drafted tends to produce justifications rather than records.
Readers of a completed report can apply a similar test. Look for the scope of each claim, for evidence that the sample included people positioned to contradict it, and for an account of what was done with the contradictions found. A single coherent explanation presented with no record of challenge is not necessarily wrong, but it gives the reader no way to tell.
The practical next step is modest: take the current draft finding, write it as a claim with an explicit scope, and list the accounts that would count against it. If the list is empty, the sampling or the claim needs revision. If it is not, the first entry in the challenge log can be written today. Organisations commissioning such work may find the guidance on working with the network useful when specifying what the analysis should document.