Qualitative Comparative Analysis (QCA): Calibration, Truth Tables, and Solution Terms

Qualitative Comparative Analysis (QCA) treats a research question as a search for configurations, not a search for average effects. Instead of asking how much a single condition changes an outcome holding others constant, QCA asks which combinations of conditions are consistently sufficient (or necessary) for an outcome to occur across a set of cases — usually somewhere between roughly 10 and 60, too many for close case-by-case comparison and too few for conventional regression. Developed by Charles Ragin and set out in The Comparative Method (1987), QCA formalizes a logic case-oriented researchers had used informally for decades: systematic, Boolean cross-case comparison grounded in set theory rather than correlation.

This guide walks the QCA workflow in the order you actually run it: calibrating raw conditions into set-membership scores, assembling the truth table, checking consistency and coverage, and running Boolean minimization to produce the three solution terms — complex, parsimonious, and intermediate. It assumes you already have a research question and a small-to-medium set of cases in hand; for choosing whether QCA fits your question in the first place, see CASRAI’s guides to research paradigms and case study research method.

Why configurational, not correlational

QCA exists because three assumptions built into most variable-oriented statistics don’t hold for a lot of real causal questions:

• Conjunctural causation. A condition rarely produces an outcome on its own — it does so in combination with other conditions. QCA calls a condition that is part of a sufficient combination but insufficient alone an INUS condition (an Insufficient but Necessary part of a combination that is itself Unnecessary but Sufficient for the outcome).
• Equifinality. More than one combination of conditions can produce the same outcome. A regression coefficient assumes one causal path; QCA is built to surface several.
• Causal asymmetry. The combination of conditions that explains the presence of an outcome is not simply the mirror image of the combination that explains its absence — QCA analyzes both directions separately rather than assuming symmetry.

These assumptions make QCA a natural fit for organizational, policy, and comparative-institutional research questions — “which combinations of funding structure, leadership support, and staff capacity are sufficient for a program to be sustained?” is a configurational question; “does funding level predict sustainment, controlling for the rest?” is a correlational one. If your question is genuinely about net average effects across a large sample, QCA is the wrong tool; if it’s about which paths reliably lead to an outcome across a modest number of cases, it’s built for exactly that.

Step 1: Calibrating conditions into set-membership scores

Calibration is the step that makes QCA a set-theoretic method rather than just a tally of raw scores, and it is the step most new users underestimate. Every condition and the outcome must be converted from a raw measure (a survey score, a count, a continuous indicator) into a degree of membership in a theoretically defined set — “high funding,” “strong leadership support,” “sustained program.” That conversion is a substantive, theory-driven decision, not a statistical default like z-scoring or splitting at the median.

Crisp-set vs. fuzzy-set calibration

The original variant, crisp-set QCA (csQCA), calibrates every condition to a strict binary: a case is either fully in the set (1) or fully out of it (0). It is easy to compute and easy to explain, but it forces a hard cut on conditions that are genuinely graded, and cases sitting near the cutoff get treated identically to cases far from it.

Fuzzy-set QCA (fsQCA), the variant used in most published applications today, allows degrees of membership between 0 and 1 while still preserving the qualitative distinction between being “more in than out” and “more out than in” of a set. Calibrating a fuzzy set means fixing three qualitative anchor points, each of which should be justified from theory or external benchmarks rather than the sample’s own distribution:

• Full membership — the value above which a case is unambiguously in the set (conventionally scored 0.95, sometimes rounded to 1).
• The crossover point — the value of maximum ambiguity, where a case is neither more in nor more out of the set (fixed at 0.5). This is the anchor that matters most: it is the point that later determines how a case’s raw value translates into set membership, not the sample mean or median.
• Full non-membership — the value below which a case is unambiguously out of the set (conventionally scored 0.05, sometimes rounded to 0).

Ragin’s direct method of calibration fits a logistic function through these three anchors to convert every raw value into a continuous fuzzy score. A simpler indirect method asks the researcher (or expert coders) to sort cases into a small number of ordered qualitative groups first and calibrates from those group assignments. Whichever method is used, the anchors — not the calibration formula — are where the real theoretical judgment sits, and they belong in the published methods section, not buried in a footnote or a software default.

Step 2: Building the truth table

Once every condition and the outcome are calibrated, QCA constructs a truth table: a table with one row for every logically possible combination of the conditions. With k conditions, there are 2^k possible rows — three conditions produce 8 rows, five conditions produce 32. This is the practical reason QCA studies rarely use more than five or six conditions: adding conditions grows the table exponentially, and most of those rows will end up with no empirical cases in them at all (a state called limited diversity, which becomes directly relevant again at the solution-derivation step).

Each case is sorted into exactly one truth-table row based on which side of 0.5 it falls on for every condition (a case with membership above 0.5 in a condition is assigned to the “1” side of that condition’s column for truth-table purposes, even though its fuzzy score is retained for the consistency calculation in the next step). For each row that does have empirical cases, the analyst then checks:

• Frequency — how many cases fall into that row. Rows below a minimum case-count threshold (commonly 1 for small-N studies, sometimes 2–3 for larger ones) are typically treated as having too little empirical support to code confidently and folded into the remainder rows for the minimization step.
• Consistency — whether the cases in that row agree on the outcome (see Step 3). A row where cases disagree on the outcome is a contradictory row, and it has to be resolved — usually by adding a condition that distinguishes the disagreeing cases, revisiting calibration, or returning to the case knowledge to check for a coding error — before minimization can proceed.

Rows with no empirical cases at all are logical remainders. They are not dropped from the analysis; how they’re handled is exactly what distinguishes the parsimonious, intermediate, and complex solutions in Step 4.

Step 3: Consistency and coverage

Before minimizing anything, QCA checks whether a combination of conditions is actually a good empirical fit for the outcome — the set-theoretic equivalent of a goodness-of-fit statistic.

• Consistency measures how closely a subset relationship holds in the data — the extent to which cases sharing a condition combination also share the outcome. It is calculated as the sum of the minimum of each case’s membership in the combination and the outcome, divided by the sum of the case’s membership in the combination. A consistency of 1.0 means every case with that combination also fully has the outcome; published applications generally treat consistency at or above roughly 0.75–0.8 as an acceptable threshold for a sufficient combination, with the exact cutoff set and justified by the researcher rather than read off a single universal rule.
• Raw coverage measures empirical importance — what proportion of the outcome’s membership across all cases is accounted for by that particular combination. A combination can be highly consistent but explain only a small slice of the outcome (low coverage), which is normal under equifinality: several combinations can each cover a different subset of cases with the outcome.
• Unique coverage is the portion of outcome membership explained by a combination that no other combination in the solution also explains — it shows which paths are doing genuinely distinct explanatory work versus overlapping with another path.

Analysts also commonly report the Proportional Reduction in Inconsistency (PRI) alongside raw consistency, since raw consistency alone can look artificially high when a combination is simultaneously consistent with both the outcome and its absence (a problem with simultaneous subset relations). A combination should clear both the consistency and PRI thresholds before it’s treated as sufficient.

Step 4: Boolean minimization and the three solution terms

With a truth table in hand and a consistency threshold applied, QCA runs Boolean minimization (based on the Quine–McCluskey algorithm) to reduce the sufficient combinations to their simplest logically equivalent expression — combining any two rows that agree on the outcome and differ by only one condition, since that condition cannot be doing causal work. This step produces one of three solution terms depending on how the analysis treats logical remainders — the truth-table rows with no empirical cases:

• The complex solution makes no simplifying assumptions at all: every logical remainder is treated as “don’t know” and excluded from minimization. It stays closest to the raw data but is often the least parsimonious and hardest to interpret, since it can’t be reduced past what the observed cases directly support.
• The parsimonious solution allows the minimization algorithm to use any logical remainder, however implausible, if doing so produces a simpler expression. It is maximally reduced but risks including “easy” counterfactual assumptions the researcher would not actually defend if asked directly — a remainder assumed away purely because it made the formula shorter, not because there’s a substantive reason to think that combination would produce the outcome.
• The intermediate solution is the version most published QCA work reports as its primary result. It uses only the logical remainders consistent with the researcher’s directional expectations — substantive, stated-in-advance judgments about whether each condition is expected to contribute to the outcome when present or when absent. Remainders are incorporated only if assuming them lines up with theory; the rest are left out. This makes the intermediate solution a middle ground: more reduced than the complex solution, but constrained by theory rather than by whatever happens to minimize the formula.

Because the parsimonious and intermediate solutions can differ in which conditions appear, a common and informative reporting convention is to present them together and mark, for each condition, whether it is a core condition (present in both the parsimonious and intermediate solutions — strong evidence it matters) or a peripheral condition (present only in the intermediate solution — evidence it matters, but with a theoretical assumption doing some of the work). Reporting the complex solution alongside them, even in an appendix, lets a reader see exactly how much simplification the other two solutions rest on.

A compact illustrative walkthrough

The scenario below is a simplified, illustrative example built to show the mechanics end to end — it is not drawn from a real study, and the “cases” are labeled generically rather than naming any real organization. Suppose a researcher has six programs (A–F) and wants to know which combination of conditions is sufficient for a program being sustained past its initial funding period. Three conditions are calibrated as fuzzy sets: stable multi-year funding, active leadership sponsorship, and dedicated staff capacity.

1. Calibrate. Each program’s funding stability, sponsorship, staff capacity, and sustainment outcome are converted to 0–1 fuzzy scores using anchors set before looking at which programs actually succeeded — for example, “funding secured for 3+ years” as the full-membership anchor for the funding condition.
2. Build the truth table. With three conditions there are 8 possible rows. Programs A, B, and D all fall on the “high funding, high sponsorship, high staff capacity” row and are all sustained — a consistent row with no contradiction. Program C has high funding and sponsorship but low staff capacity, and was not sustained.
3. Check consistency and coverage. The “funding · sponsorship · staff capacity” combination shows a consistency of 1.0 across the three cases that share it and covers half of the sustained programs — high on both measures, so it clears the threshold as a sufficient combination.
4. Minimize. Boolean minimization across the confirmed rows shows that whenever both funding and sponsorship are present, staff capacity — while present in every fully sustained case — is not logically required to distinguish sustained from non-sustained rows once the remainder rows are incorporated under the researcher’s directional expectations, producing an intermediate solution centered on funding and sponsorship together, with staff capacity as a peripheral condition strengthening but not solely driving the result.

A real application would report the calibration anchors, the full truth table, consistency/coverage for every row and solution, and both the intermediate and parsimonious solutions with core/peripheral conditions marked — the compressed version above only shows where each number in that eventual write-up comes from.

Common pitfalls

• Too many conditions for too few cases. Each added condition doubles the truth table’s row count; with more than five or six conditions and a small case set, most rows end up empty and the “logical remainder” problem dominates the analysis rather than the empirical evidence. Keep the condition count disciplined and theory-driven, not exploratory.
• Calibrating from the sample instead of from theory. Setting the crossover point at the sample’s own mean or median (rather than a theoretically or externally justified value) collapses the distinction between calibration and simple standardization — and different researchers calibrating the same raw data from different anchors can get materially different truth tables.
• Reporting only the parsimonious solution. Because it is the most reduced, the parsimonious solution can look like the cleanest result, but it may rest on counterfactual assumptions the researcher hasn’t actually examined. The intermediate solution, justified with explicit directional expectations, is the conventional primary result for a reason.
• Skipping contradictory-row resolution. Minimizing straight through a truth table that still has rows where cases disagree on the outcome produces a solution that looks precise but is built on an unresolved coding or measurement problem underneath it.

Frequently asked questions

How many cases does QCA need?

There’s no fixed minimum, but QCA is built for the small-to-medium-N range — commonly cited as roughly 10 to 60 cases — where there are too many cases for close qualitative comparison of every pair but too few for the asymptotic assumptions behind most inferential statistics. csQCA has been applied to fewer than 10 cases; fsQCA applications more commonly sit in the dozens.

What’s the difference between fsQCA and csQCA?

Crisp-set QCA (csQCA) calibrates every condition to a strict binary — a case is either in a set or out of it. Fuzzy-set QCA (fsQCA) allows degrees of membership between 0 and 1, calibrated against theoretically defined anchor points, which preserves more information about cases near a set’s boundary. fsQCA is the more commonly used variant in current published work.

What is a logical remainder?

A logical remainder is a row in the truth table — a logically possible combination of conditions — for which no case in the dataset is actually observed. Because the number of possible combinations grows exponentially with the number of conditions, most truth tables have far more remainder rows than rows with empirical cases. How remainders are treated during Boolean minimization is what produces the complex, parsimonious, and intermediate solutions.

Is QCA qualitative or quantitative?

Both, deliberately. QCA formalizes case-oriented, configurational thinking using Boolean and set-theoretic logic and produces numeric consistency/coverage statistics, but it depends on close, qualitative case knowledge at every stage — calibrating conditions against substantive theory, resolving contradictory rows, and setting directional expectations for the intermediate solution all require the kind of case familiarity a purely variable-oriented method doesn’t ask for.

What software runs a QCA analysis?

The dedicated fsQCA software (developed by Ragin and colleagues) and the open-source QCA package for R are the two most commonly used tools; both build truth tables, compute consistency/coverage, and run Boolean minimization to produce complex, parsimonious, and intermediate solutions.

For related methodological ground on this site, see CASRAI’s guides to case study research method, qualitative research methods, and research paradigms, or return to the Research Methods & Statistics hub for the full cluster.