Paper tutorial / Research in practice

Regulatory design and trust: an fsQCA comparison

Run two QuickPLS fsQCA analyses to compare trust in a regulatory system with trust in its lead organisation.

19 min read

We use this paper to show how a similar analysis can be carried out in QuickPLS. Follow the steps with the supplied practice data, then adapt the setup and calculations to your own study. The practice data are a separate teaching example.

Can people trust the system without trusting each organisation?

Trust in a regulatory arrangement and trust in a particular regulator are related but distinct. A person may approve of a system's safeguards while doubting one organisation's conduct. A study that combines these outcomes into a single score can hide that distinction.

Verhoest and colleagues used QCA to study 18 regulatory regimes across six countries and three sectors. They considered design choices including centralisation, independence, mandated interaction and rule appropriateness in relation to system and actor trust. This exercise uses original fictional service-regulation cases to teach two-outcome comparison. A publisher correction concerned the special-issue title, not a methodological correction identified in the inspected record.

Keep institutional design and two forms of trust separate

Use 36 fictional service-regulation arrangements. One row represents an arrangement with a defined jurisdiction and service remit. The four design conditions are concentrated decision authority, organisational independence, required stakeholder interaction and rules suited to the service. The two outcomes are trust in the arrangement and trust in its lead organisation.

The continuous 1–7 ratings are invented teaching assessments. They are not citizen survey responses, legal classifications or measures taken from the cited paper. If a real study collects several people's views about one regime, aggregation needs a reason and evidence; treating every respondent as an independent regime would change the unit of analysis.

Ask the two trust questions under the same four design conditions. This is a comparison of separate set relationships. It does not estimate a pathway from institutional design through system trust to actor trust.

Conditions and outcomes in this practice study. The named conditions enter configuration analysis together. Each outcome is a separate question; the arrow is not a regression effect.
The named conditions enter configuration analysis together. Each outcome is a separate question; the arrow is not a regression effect.

On a small screen, swipe the table sideways to see every column.

ColumnSet meaningRole
CENTRALConcentrated decision authorityCondition
INDEPStrong organisational independenceCondition
INTERACTRequired stakeholder interactionCondition
FITRules well suited to the serviceCondition
SYSTEMHigh trust in the arrangementOutcome
ACTORHigh trust in the lead organisationOutcome
case_idFictional arrangementLabel

Step 1. Define the cases and the assessment rubric

In a writable project choose File > Import Data… and Choose data file…. Select the CSV. In Review & import confirm raw data, Header row 1, comma separator and Decimal convention Dot (1.25). Set the analytical numeric columns to Continuous. Choose Import dataset and confirm Dataset imported and verified.

Check all 36 case identifiers and the six score ranges. The supplied data have no missing values or reverse-coded scores. A high CENTRAL value means concentrated authority, not good governance. Whether that design contributes to trust is the research question, so avoid labelling its high end desirable in advance.

Use exclusion 2, crossover 4 and inclusion 6 for the invented rubric. Interpret these separately for each concept: high independence and high rule suitability are different sets even when their numeric anchors match. In a real comparative study, institutional evidence must justify how the same rubric applies across jurisdictions.

Four conditions permit 16 configurations. Inspect the distribution before minimisation because 36 cases do not guarantee broad coverage of institutional combinations. Save a note about the intended population and comparison period alongside the study.

Screenshot to be addedImport the 36-case teaching file

Step 2. Calculate the SYSTEM baseline in QuickPLS

  1. Open the fsQCA settings

    With the imported dataset selected, choose Calculate…. In Find a method select Fuzzy-set Qualitative Comparative Analysis. Open Method settings. The form starts with Conditions and Outcome and case identity.

  2. Assign the conditions and outcome

    Under Conditions choose Add condition for each of CENTRAL, INDEP, INTERACT and FIT, in that order. Under Outcome and case identity select SYSTEM as Outcome variable, Presence of the outcome as Analyse and case_id as Case identifiers. Check that only the intended conditions appear; the case label is not an analytical condition.

  3. Calibrate every selected column

    For each condition and the outcome choose Direct fuzzy-set calibration. Enter exclusion 2, crossover 4 and inclusion 6. Explain the column’s set definition and the invented assessment rubric in its anchor rationale. These increasing anchors make a higher raw score correspond to greater membership. There is no separate inclusion-membership entry in this form.

  4. Set the baseline thresholds

    Under Truth table and solution assumptions set Frequency threshold to 1, Consistency threshold to 0.80 and PRI threshold to 0.50. Leave Request intermediate solutions with explicit directional expectations unchecked for this baseline. The reference answer uses the conservative family from the completed output; it is not a family selected in this settings form.

  5. Calculate and check the result

    Choose Start calculation and wait for completion. Check the outcome, calibration, thresholds and included-case count in the completed result before interpreting it. The reference includes all 36 cases. If your result differs, inspect Exclusions and the selected settings rather than changing the answer to fit the reference.

  6. Read one complete solution

    In Results, select the completed calculation in Result. Under Estimates open Truth table and read frequency, consistency and PRI together. Then open Overall solution fit — every alternative. Use the conservative rows to compare with this lesson’s reference answers; keep the Solution identifier with the family. In Pathway fit and coverage — every alternative, read all pathways belonging to that same family and solution. Do not combine pathways from different complete alternatives. An alternative number identifies an expression; it does not rank its quality.

  7. Check the case memberships

    Under Model and data open Case memberships and source rows. Locate C0002 by its case identifier and compare FIT with SYSTEM. Use the lesson’s checked case table to follow the membership calculation. One case illustrates the pathway; its membership is not the pathway’s overall consistency.

Screenshot to be addedSelect the SYSTEM model in Method settings
Screenshot to be addedSet the truth-table thresholds

Step 3. Calculate ACTOR and compare it with SYSTEM

Check that the two completed results identify SYSTEM and ACTOR on the same 36-case file. Compare their settings, case inclusion and complete expressions before comparing fit. These ordinary calculations do not require a multiple-outcome planning workspace.

  1. Open another fsQCA calculation

    After SYSTEM finishes, reopen Calculate… and select Fuzzy-set Qualitative Comparative Analysis on the same dataset. Check CENTRAL, INDEP, INTERACT and FIT and their calibration anchors; do not assume the previous settings have carried over.

  2. Select actor trust

    Set Outcome variable to ACTOR and Analyse to Presence of the outcome. Choose Direct fuzzy-set calibration with exclusion 2, crossover 4 and inclusion 6 and explain the actor-trust set in its rationale. Keep case_id, frequency 1, consistency 0.80 and PRI 0.50.

  3. Calculate ACTOR

    Choose Start calculation and wait for completion. Confirm ACTOR and the 36-case dataset in the completed result. A finished SYSTEM result does not establish that ACTOR finished.

  4. Read the two results separately

    Use Result to open each completed calculation. Read its Truth table, then the conservative rows in Overall solution fit — every alternative and Pathway fit and coverage — every alternative. Keep one complete solution together for each outcome; do not combine favourite pathways from different alternatives.

  5. Follow the same arrangement

    In Case memberships and source rows, find C0002 in each result. Compare FIT and the other conditions with SYSTEM, then with ACTOR. Interpret the difference as two judgements about trust, not as two measurements of an interchangeable outcome.

Screenshot to be addedOpen the separate ACTOR result

Step 4. Check the answers for the 36-case file

Use direct logistic calibration for every selected condition and outcome: exclusion 2, crossover 4, inclusion 6; the calibration used by the reference assigns membership 0.95 at the inclusion anchor. The baseline uses frequency 1, row consistency 0.80 and PRI 0.50. Read the conservative rows in the completed solution tables. 36 of 36 cases are included; excluded case identities, if any, are in the download.

For SYSTEM, 9 truth-table rows are positive, 7 negative and 0 remainders. There are 1 complete conservative alternative(s). Alternative 1 has consistency 0.780, coverage 0.863 and PRI 0.618.

For ACTOR, 8 truth-table rows are positive, 8 negative and 0 remainders. There are 1 complete conservative alternative(s). Alternative 1 has consistency 0.831, coverage 0.794 and PRI 0.667.

About these answers: Reference answers calculated with R QCA and independently checked. These tables are not outputs from a QuickPLS app run.

On a small screen, swipe the table sideways to see every column.

OutcomeSelected alternativeSolution consistencySolution coverage
SYSTEM1 of 10.7800.863
ACTOR1 of 10.8310.794
Screenshot to be addedRead the completed SYSTEM result
What the separate checks cover

579 numerical comparisons passed tolerance 0.000000000001. Python independently recalculated memberships, case assignments, supported-row consistency and PRI, pathway and solution fit, and checked that each conservative alternative covers exactly the positive Boolean rows. This does not prove that the minimiser found every globally minimum expression and does not execute QuickPLS.

Step 5. Read the complete SYSTEM alternative

The table gives every pathway in reference alternative 1. Read * as AND, ~ as absence, and the rows together as OR. An omitted condition can be present or absent within that term. Keep the complete expression together. The downloadable calculation files include any other returned alternatives.

The SYSTEM outcome is a calibrated membership, not a pass/fail prediction. Raw coverages can overlap; adding them does not give solution coverage.

About these answers: Reference answer from R QCA 3.25, with independent checks of calibration, pathways and fit.

On a small screen, swipe the table sideways to see every column.

PathwayConsistencyRaw coverageUnique coverage
FIT0.8140.8030.396
~CENTRAL*INDEP*INTERACT0.8490.4670.060

Step 6. Read the complete ACTOR alternative

The table gives every pathway in reference alternative 1. Read * as AND, ~ as absence, and the rows together as OR. An omitted condition can be present or absent within that term. Keep the complete expression together. The downloadable calculation files include any other returned alternatives.

The ACTOR outcome is a calibrated membership, not a pass/fail prediction. Raw coverages can overlap; adding them does not give solution coverage.

About these answers: Reference answer from R QCA 3.25, with independent checks of calibration, pathways and fit.

On a small screen, swipe the table sideways to see every column.

PathwayConsistencyRaw coverageUnique coverage
CENTRAL*INTERACT0.8380.5730.028
~INDEP*INTERACT0.8650.5730.025
INTERACT*FIT0.9030.5870.034
~CENTRAL*~INDEP*FIT0.8990.4400.072

Step 7. Follow one actual practice case: C0002

For case C0002, the membership in FIT is 0.759. Take the smallest membership across the conditions in that term, complementing a membership first when it has ~. The outcome membership is 0.684. This is one case inside a pathway, not the pathway’s overall consistency.

About these answers: Values calculated from this CSV using the stated logistic calibration.

On a small screen, swipe the table sideways to see every column.

Condition or outcomeRaw scoreCalibrated membership
CENTRAL2.81840.149
INDEP4.31690.615
INTERACT6.4530.974
FIT4.77770.759
SYSTEM4.52440.684
ACTOR5.01130.816
Screenshot to be addedFind case C0002 in the result

Step 8. Compare the planned threshold changes

The frequency / consistency column lists the rules applied to truth-table rows. The final consistency column measures the minimised solution. These are different quantities; a row threshold of 0.85 does not require solution consistency to exceed 0.85. A frequency increase can remove observed support and create more remainders.

These runs change frequency or row consistency while holding calibration fixed. They do not test other calibration anchors, outcome negation or another solution family. If the lesson asks you to explore those choices, save and report them as additional runs.

About these answers: External reference runs, with every returned conservative alternative retained.

On a small screen, swipe the table sideways to see every column.

Outcome / changeFrequency / consistencyPositive rowsAlternativesSolution consistencySolution coverage
SYSTEM: baseline1 / 0.80910.7800.863
SYSTEM: frequency2 / 0.80610.8290.747
SYSTEM: consistency1 / 0.85910.7800.863
ACTOR: baseline1 / 0.80810.8310.794
ACTOR: frequency2 / 0.80510.8280.685
ACTOR: consistency1 / 0.85810.8310.794

Step 9. Compare configurations without ranking the outcomes

In the checked SYSTEM alternative, FIT appears on its own. In the checked ACTOR alternative, FIT appears with other conditions. Open both complete alternatives and read those terms alongside their cases. This difference concerns the defined system-trust and actor-trust sets; it does not establish that suitable rules alone will cause trust.

Look at conditions omitted from a pathway as unrestricted within that simplified term. Look at conditions explicitly negated as required absences. This matters for CENTRAL, whose absence means relatively dispersed authority under the stated calibration, not the absence of government.

Read solution coverage separately for SYSTEM and ACTOR. The memberships being covered differ. A larger coverage for system trust does not mean system trust is more important, more desirable or easier to produce. Inspect the same cases across both outcomes to understand the distinction.

Where multiple alternatives exist, report the alternatives or justify which one you discuss. Do not select different terms from different alternatives merely to tell a coherent institutional story.

Follow a regime with contrasting trust memberships

Hypothetical example

Illustrative reading example

Imagine arrangement R with INDEP membership 0.80, FIT 0.70, INTERACT 0.30, SYSTEM 0.85 and ACTOR 0.40. Its membership in INDEP * FIT is 0.70, while membership in INTERACT * FIT is 0.30. Under the illustrative expressions, R is more strongly covered by the system-trust route.

The outcome memberships also differ: system trust is high, while actor trust is below the crossover. This is not a contradiction to erase by averaging. The two ratings describe distinct judgements. In a real case, investigate the reasons using independent institutional and perception evidence.

Averaging SYSTEM 0.85 and ACTOR 0.40 gives 0.625, but that number is not automatically membership of a defensible overall-trust set. A new set would require a conceptual definition and calibration. Arithmetic convenience cannot supply either.

The memberships and expressions above are invented for interpretation practice. Their minimum operations are hand calculations. They are not expected results from the CSV, findings from the cited regimes or a validated trust index.

Step 10. Test the comparisons without selecting a preferred institutional design

Run another calculation for each outcome with frequency 2. Compare each with its frequency-1 baseline to identify dependence on single-case configurations. Keep the same conditions, outcome definitions, calibration and consistency rule so the change has one clear explanation.

Next, if justified by the rubric, inspect a stricter definition of high trust in a separate calibration revision. Do not redefine independence or centralisation simply because their presence creates an awkward political interpretation. A conceptual change must be explained before the resulting solution is read.

Consider whether one jurisdiction or service type dominates a pathway. The practice file has no real jurisdictions and cannot answer that question substantively. In a real study, case composition is part of the explanation, especially when institutional settings differ greatly.

Neither set sufficiency nor a neat expression establishes that redesigning a regime will cause trust to improve. Timing, selection and omitted institutional conditions still matter. Frame design implications as propositions to investigate within the observed scope.

Investigate apparent institutional contradictions

On a small screen, swipe the table sideways to see every column.

SymptomCheckAction
Trust outcomes point in different directionsTheir definitions and the same cases' membershipsKeep the distinction and investigate why system and actor judgements differ.
Centralisation appears positively in one route and negatively in anotherThe surrounding conditionsExplain conjunctions rather than assigning one universal sign.
The comparison uses different case countsOutcome-specific exclusions and calibrationReport the difference beside the fit values.
A route disappears at frequency 2Its single-case supportDescribe sensitivity and the relevant case instead of suppressing the comparison.
A report averages trust membershipsWhether an overall set was definedRetain separate outcomes unless a new concept is justified.
A design recommendation sounds causalThe observational case designLimit the claim to the observed configurations and needed follow-up evidence.

Step 11. Write two results with one clear comparison

Write one Results paragraph per outcome, naming the solution family, complete alternative, pathways and fit. Follow with a comparison paragraph about shared and differing conditions, the cases behind those differences and sensitivity to support rules. Do not turn a shared condition into a universal design prescription.

OutcomeReport alongside its solution
System trustDefinition, included cases, selected alternative, fit and illustrating cases
Actor trustDefinition, included cases, selected alternative, fit and illustrating cases
Cross-outcome comparisonShared conditions, contrasting cases and differences in scope
SensitivityChanges under frequency and any justified calibration alternatives

Methods writing outline

We analysed 36 fictional regulatory arrangements using four design conditions and separate system-trust and actor-trust outcomes. We calibrated [definitions and anchors] and completed one calculation for each outcome on the same dataset. Baseline thresholds were [values]. We compared separate frequency-2 calculations and recorded [case and calibration decisions].

Step 12. Write the result for this teaching file

We calibrated 36 synthetic cases using direct logistic calibration with anchors 2, 4 and 6 and inclusion membership 0.95. We used a frequency requirement of 1, row consistency of 0.80 and PRI of 0.50. We retained the conservative solution family.

Report the complete selected alternative and the changes shown in the sensitivity table. Explain case examples alongside aggregate coverage. These results describe the original teaching dataset, not the source paper’s observations. If several alternatives exist, explain the selection and preserve the others; do not combine favourite terms.

Keep the institutional claim precise

  • Identify the case unit and comparison period.
  • Explain why high centralisation is not labelled good in advance.
  • Keep system and actor trust separate.
  • Trace one contrasting case across both outcomes.
  • Report single-case dependence.
  • Distinguish an observed configuration from a causal reform claim.

Further reading

Related guides