There is no universal sample size rule for a political science or international relations PhD: King, Keohane and Verba’s Designing Social Inquiry (1994), the field’s most cited methodology text, treats the number of cases as a design decision tied to the specific causal inference being claimed, not a fixed target — a single-country case study and a thirty-country comparative panel can both be methodologically sound, provided the case-selection logic and the scope of the claim actually match.
| Design type | Typical N | Sampling logic |
|---|---|---|
| Single case study | 1 | Deviant, typical, or theory-testing case, selected for its relationship to existing theory |
| Small-N comparative | 2–6 | Most-similar or most-different systems design, matched or contrasted on key variables |
| Elite interview study | 15–40 (field-typical, not a rule) | Purposive sampling to reach relevant position-holders, often followed by snowball referral |
| Cross-national quantitative panel | 50+ countries, multiple years | Population-level or near-population sampling from an existing country-year dataset |
Case selection is the sampling decision in small-N political science research
In a single-case or small-N comparative design, the sampling decision is which case or cases to select, not how many individual respondents to survey within a case — a distinction that trips up candidates arriving from a survey-based social-science background. King, Keohane and Verba’s central methodological argument is that a case selected only for convenient access, rather than for its specific relationship to the theory under test, undermines the causal inference the thesis wants to draw regardless of how thoroughly that one case is subsequently studied. A deviant case — one that existing theory would not predict the observed outcome for — is selected precisely because it stresses the theory in a way a typical case cannot; a most-similar-systems design selects cases that are alike on most background variables so the one variable that differs can plausibly be isolated as doing the explanatory work; a most-different-systems design does the reverse, selecting cases that differ on almost everything except the outcome and the hypothesised cause.
A worked illustration, invented for this example and not a real submitted thesis: a candidate researching why some post-communist states consolidated democratic institutions while others reverted to authoritarian rule might select a most-similar-systems pair — two states sharing a common institutional legacy, similar GDP per capita at the point of transition, and comparable EU accession trajectories, but differing sharply on the outcome — precisely because the shared background variables let the analysis isolate a smaller set of candidate explanatory factors than a comparison across two states differing on everything at once. The case-selection paragraph in the methodology chapter should name these matched and differing variables explicitly, not simply assert that the cases are comparable.
Elite interview sampling: purposive selection and the access problem
Elite interview studies in political science and IR — interviewing legislators, diplomats, civil servants, party officials or NGO leaders — use purposive sampling almost by definition, because the population of relevant position-holders on a given question is small, identifiable, and not amenable to random sampling. Build the sampling frame from the specific institutional positions your research question requires, not from whoever responds first to a general request, and expect a substantial non-response rate: elite respondents are time-poor and gatekept by staff, and a thesis’s ethics application and fieldwork timeline should budget for this explicitly rather than assuming the sampling frame and the achieved sample will be close to identical. Snowball referral from an initial contact is a legitimate and common way to extend access once the first interviews are secured, but state in the methodology chapter how this could bias the sample toward one network or faction, and address that limitation directly rather than leaving it for an examiner to raise unprompted.
Consent and anonymity in elite interview research run in the opposite direction from most other fields’ research ethics conventions: many elite respondents — politicians, senior officials, published commentators — actively prefer to be named and attributed, since the interview is often an extension of their public role, not a disclosure of private information a vulnerable-population study would need to protect. A blanket anonymisation policy imported uncritically from a different field’s ethics template can therefore be inappropriate here; the consent form should offer the respondent an explicit choice between attribution, partial identification (role and institution but not name), and full anonymity, and the ethics application should explain why elite-interview conventions differ from the standard anonymisation default.
Access negotiation: the step that has to happen before ethics approval, not after
Access to elite respondents, government archives, or an international organisation’s internal documents is frequently the binding constraint on a political science or IR doctorate’s design, more so than the statistical logic of the sample itself. Negotiate access — a letter of support from a gatekeeping institution, a research visa if fieldwork requires travel to another jurisdiction, formal clearance for a national archive’s restricted holdings — before finalising the sampling frame in the ethics application, because an ethics committee cannot meaningfully approve a fieldwork plan whose access is still hypothetical, and a plan built around access that never materialises is a rebuildable but genuinely costly setback mid-candidature.
Documentary and archival sampling for comparative historical work
Where the research design is comparative-historical rather than interview-based, the sampling unit shifts to documents, archival files or historical episodes rather than people, and the same case-selection logic still applies: which archive, which time period, which set of government files is selected, and on what explicit basis relative to the research question. This documentary sampling logic overlaps closely with the source-selection problem a history doctorate faces — see how a history PhD chooses between archives and catalogues for the parallel discipline’s own version of the same access and selection problem — though a political science thesis typically frames the selection explicitly around a causal or comparative claim, where a history thesis’s selection criteria are more often organised around period, place or theme.
Triangulating documents and interviews without double-counting your evidence
Many political science and IR doctorates combine elite interviews with documentary sources — meeting minutes, parliamentary records, diplomatic cables where declassified — as a deliberate triangulation strategy, and the sampling justification then needs to cover both sources coherently rather than treating them as two unrelated methods bolted together. State explicitly what each source type is being used to establish: interviews often surface motivation, internal disagreement and process detail that formal documents do not record, while documents provide a dated, less memory-dependent record of what was formally decided or stated. A thesis that uses an interview quotation to corroborate a documented decision, and is explicit about which claim rests on which evidence type, is markedly stronger at viva than one that blends the two without distinguishing what each is actually capable of establishing.
Cross-national quantitative panels: population sampling, not case sampling
Where the design uses an existing cross-national dataset — a democracy index, a conflict event dataset, an economic panel spanning dozens of countries and years — the sampling logic changes entirely: the thesis is typically working with the full available population of countries and years the dataset covers, or a defined subset excluded on stated, defensible criteria (data availability, a specific regional or temporal scope tied to the research question), rather than drawing a smaller sample from a larger population in the survey-sampling sense. State explicitly which countries or years were excluded and why, because an unexplained gap in cross-national panel coverage — a specific region or period missing with no stated justification — is one of the first things a quantitatively literate examiner checks. Missing data within an included country-year panel raises a separate question from missing countries altogether: state whether missingness is likely random or systematically related to the outcome being studied — a conflict dataset with worse coverage in exactly the states experiencing the most instability is a substantively different problem from data missing for unrelated administrative reasons, and the two require different handling in the analysis chapter.
Justifying your sample at the viva
Whichever design is used, the viva question is consistent across all of them: why this case, this set of interviewees, or this dataset, and not a different one that might have supported a different or stronger claim. A methodology chapter that states the case-selection logic explicitly — deviant case, most-similar systems, purposive elite sampling with a defined institutional frame, or population-level cross-national coverage with stated exclusions — answers that question before it is asked. The same underlying justification discipline applies across doctoral sample-size reasoning generally — see how sample size is reasoned and justified for postgraduate health research for how a different field makes the equivalent argument with a formal power calculation rather than a case-selection logic — and, for a comparative-law angle on the same selection problem, how a comparative law PhD chooses between jurisdiction and function as its organising logic.
FAQ
Is a single-case-study PhD thesis in political science methodologically weaker than a comparative one?
No, provided the single case is selected for a specific, stated relationship to existing theory — as a deviant, typical, or theory-testing case — rather than for convenience, and the claim made from it is scoped to match.
What is the difference between most-similar and most-different systems design?
Most-similar systems design selects cases alike on most background variables so a single differing variable can be isolated; most-different systems design selects cases that differ on almost everything except the outcome and hypothesised cause.
How many elite interviews does a political science PhD typically need?
There is no fixed rule; the number is driven by reaching theoretical saturation or covering the relevant institutional positions identified in the sampling frame, commonly in the range of 15 to 40 in published UK doctoral work, but this varies by research question.
Does snowball sampling bias an elite interview study?
It can, toward the initial contact’s own network or faction — state this limitation explicitly in the methodology chapter rather than presenting the achieved sample as fully representative.
What should I do if a country or period is missing from my cross-national dataset?
State explicitly which cases or years were excluded and the defensible reason — data availability or a stated scope tied to the research question — rather than leaving an unexplained gap for an examiner to identify.
When should I negotiate access to elite respondents or archives?
Before finalising the sampling frame in your ethics application — an ethics committee cannot meaningfully approve a fieldwork plan whose access is still hypothetical.
What is the single question examiners ask about sample or case selection at viva?
Why this case, interviewee set or dataset was selected and not an alternative that might have supported a different or stronger claim — answer this explicitly in the methodology chapter before it is asked.
Should elite interview respondents always be anonymised?
Not necessarily. Many elite respondents prefer attribution since the interview extends their public role; offer an explicit choice between attribution, partial identification and full anonymity rather than defaulting to a blanket anonymisation policy built for a different kind of research.
How should I combine interview and documentary evidence without weakening either?
State explicitly what each source type is being used to establish — interviews for motivation and process detail, documents for a dated formal record — rather than blending the two without distinguishing what each is capable of showing.
Build the case-selection argument into your methodology chapter
The justification a political science or IR thesis needs for its case, interview or dataset selection is, in miniature, an argument the methodology chapter has to make explicitly rather than leave implicit. Tesify’s thesis writing workspace helps you draft that justification against your own research question and available access, before fieldwork begins rather than as a retrospective defence at viva.