Method guide / Configurations

Temporal QCA

Represent timing or sequences explicitly before examining configurational relationships.

17 min read

What the method does

Temporal QCA introduces time-related information into the conditions being compared, such as a prior-period state or event order. The temporal construction is part of the research design. Rows from the same unit must be aligned correctly; a lag across unrelated units has no meaningful interpretation.

Within-unit lag: L(Aᵢ,t) = Aᵢ,t−1.

The lag refers to the previous eligible period for unit i under the stated ordering rule. Missing periods, first observations and unequal spacing need explicit treatment. A temporal transformation does not by itself establish causal timing.

Data and analysis in QuickPLS

Provide a unit identifier, an ordered time variable, condition values and an outcome. Define the time window and transformation before running the set analysis.

QuickPLS provides the temporal preparation and records the transformations. Inspect the prepared cases and time alignment before reading the resulting truth table.

Follow a worked example

The following example shows the data, settings and decisions for one analysis. Use the linked practice files to follow it. For your own study, keep the procedure and choose the variables, model and settings that fit your research question.

Download panel-time.csv to begin. It contains 9 rows and 8 columns. The example below identifies the variables to use, their roles and any rows to exclude; not every column is an analysis variable.

Choose which earlier observation the question needs

A one-period lag asks for the value at a specified earlier time. A one-observation lag asks for the previous recorded observation of the same unit. They agree when every period is observed, but can differ when a period is missing. Choose the rule that answers your research question rather than using spreadsheet row order.

The teaching file is deliberately unsorted. The predecessor of unit u1 at period 2 is u1 at period 1, not the previous CSV row. Inspect the original rows QuickPLS uses to create a value. A lag must remain within its unit and use the intended earlier observation, never a future value.

Recorded temporal transformations in the Temporal and panel diagnostics sample. The expressions describe the preparation of time-based variables.
Recorded temporal transformations in the Temporal and panel diagnostics sample. The expressions describe the preparation of time-based variables.

Identify the units, periods and incomplete histories

Prepare an explicit unit identifier and time field. State what one time unit means and how ties, absent events and missing values should be handled. Keep the full original histories available while creating temporal columns; select final analysis observations afterwards.

Create a separate temporal dataset before analysing it

  1. Choose the source and identifiers

    Open Data, More, Temporal workflows. Under Transformation source choose the original dataset, then Define temporal outputs. Select the unit and period fields and state the time unit.

  2. Define and preview a transformation

    Choose an exact-period lag, observation-order lag, binary event order or start/end change. Enter its rules, then choose Inspect generated values before saving.

  3. Check the original observations

    Inspect the source rows behind each generated value, including missing predecessors, event ties and absent endpoints. Check the units and periods covered.

  4. Save the new dataset separately

    Give the output a clear name, confirm that you inspected its values and source evidence, then Save separate temporal dataset. Select the saved dataset in Data and save a study draft for it.

  5. Save and run conservative temporal QCA

    Return to Temporal workflows and Load temporal setup for the derived study. Use at least one generated temporal column, set calibration and case choices, preview the truth table, save the temporal revision and calculate.

Check how a temporal value was created

For event order, specify the event columns, what counts as each first event and how ties or unobserved events are treated. For a start/end change, name the exact endpoints; a nearby observation answers another question. Resolve duplicate unit/time pairs rather than selecting whichever row happens to appear first.

Keep the derived dataset separate and connected to its source observations. Temporal QCA requires an actual generated temporal column and returns conservative solutions, meaning it does not use unobserved combinations to simplify the explanation. The preparation workflow can handle longer histories than the later analysis, so check both size limits.

For a lag, verify the unit and earlier-time rule. For event order, inspect the original first-event observations. For a change, inspect the exact starting and ending rows. Those checks make it possible to explain a surprising generated value before interpreting a solution.

Inspect the nine-row panel-time example and the original rows used for each generated value. For the capacity lag, first-period values are missing because no predecessor exists. For a start/end change, a value outside the chosen window is unavailable, not zero. Keep those different reasons visible.

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

OutputHow to read itWhat to check
Unit and period identifiersCheck that each unit-period pair identifies a single observation and that the chosen time order is meaningful.CSV row order cannot establish the time sequence for each unit.
Generated columns and original observationsCheck the original rows used to calculate each lag, change or other temporal value.A lag uses the specified earlier period. Show when that earlier observation is missing.
Temporal truth tableRead the table for the generated dataset and the saved rules for including cases.Calling a column temporal does not create repeated observations when the original data contain none.
Conservative solutionRead the saved conservative pathway and the memberships of its cases.This procedure does not offer parsimonious or intermediate temporal solutions.

Follow an earlier value through the unsorted file

Use panel-time.csv or a copy of Temporal and panel diagnostics. Its saved examples let you follow a generated value back to the original observations. Keep the original files when creating another transformation.

Turn the sequence of observations into an explicit research choice

The synthetic panel-time.csv file has units u1, u2 and u3 observed at periods 1, 2 and 3. Its rows are deliberately out of time order. The columns are case_id, unit, period, capacity, event_a, event_b, X and Y. Capacity is a raw score, event_a and event_b mark recorded events, and X and Y are binary memberships used in the separate panel-fit lesson.

You might ask whether event A occurred before event B, or use earlier capacity to define a later condition. Both require an explicit time rule. The previous spreadsheet row may belong to another unit. Build temporal columns from the full unit histories first, then choose the observations appropriate for the final outcome analysis.

Check one observation per unit and period

Three units observed over three periods give nine distinct unit-period pairs. Here period is the time index. For real data, state whether it represents years, months, survey waves or another interval. Duplicate unit-period pairs leave the sequence ambiguous and need a documented resolution before transformation.

Keep case_id as the original observation label. For u1, t02 is period 1, t07 is period 2 and t05 is period 3. The CSV does not put those observations next to one another in time order. Their labels let you trace which earlier row supplied each new value.

Build the histories before selecting final observations

  1. Start from the original panel file

    Open a working copy of fsqca_temporal_panel.qpls. Choose Data > More > Temporal workflows. Select the original panel-time dataset as Transformation source and choose Define temporal outputs. Set Unit identifier column to unit, Time column to period and Time unit to Named integer periods; describe one period as a teaching period.

    What you should see: The original panel file has the intended unit and time columns.

    How to read it: Inspect ordering and duplicate unit-period rows.

  2. Define the capacity lag

    Define capacity_previous_period from capacity. Choose Lagged value and the Exact elapsed time lag rule with one period. An absent exact predecessor remains missing. Do not use Previous observed records within the unit for this exercise; that is a different rule when periods are missing.

    What you should see: The lag is defined within units.

    How to read it: A value from the preceding unit must never become the current unit’s lag.

  3. Inspect the generated values

    Choose Inspect generated values before saving. For u1, check the period-2 lag against t02 and the period-3 lag against t07. Inspect the first-period rows, which have no earlier period.

    What you should see: Prepared values show the expected first-period and missing-period behaviour.

    How to read it: Resolve incorrect alignment before analysis.

  4. Define event order separately

    Define a_before_b from event_a and event_b using Event A before event B. Match the saved teaching policy: Event not observed and Events at the same time use Set absence (0). Inspect u1, u2 and u3. The tied u3 case is coded 0 by this policy; it does not show that B happened first.

    What you should see: Event order is defined separately from the lag.

    How to read it: These constructions answer different timing questions.

  5. Save a separate dataset and study

    Name and Save separate temporal dataset after reviewing the preview. Keep the original observations. Select the saved derived dataset in Data and save a study draft for it.

    What you should see: The derived dataset and its study are saved separately.

    How to read it: The original measurements remain available for checking.

  6. Choose the analysis observations

    After preparing the complete histories, choose Analyze an explicit original-row list and enter 3, 5, 8 in Original rows, increasing and comma separated. These are t03, t05 and t08: the period-3 observations of u3, u1 and u2. Use the saved rationale: Explicit period-3 endpoint for each unit: original rows 3, 5 and 8, irrespective of file order. Select these rows after transformation; earlier observations still supply the histories.

    What you should see: The analysis observations satisfy the declared time window and eligibility rule.

    How to read it: Record excluded periods.

  7. Choose the derived condition and outcome

    Load Event order at explicitly selected final observations. Select a_before_b as the only condition and Y as the outcome. Use existing membership for both, outcome presence and case_id as the case label. capacity_previous_period and capacity_change_1_to_3 are transformation examples; they are not conditions in this saved analysis.

    What you should see: The selected derived condition and outcome match the temporal question.

    How to read it: Confirm which time each column represents.

  8. Match the analysis settings

    Set frequency to 1, consistency to 0.80 and PRI to 0.50. Reject missing input. Leave crossover exclusions empty, the optional contradiction band unchecked and directional expectations unset. Select the conservative solution. Confirm original rows 3, 5, 8.

    What you should see: Calibration and thresholds match the planned analysis of those derived observations.

    How to read it: Do not silently reuse incompatible raw-variable anchors.

  9. Preview, save and calculate

    Use Preview temporal truth table, Save temporal revision and Calculate temporal QCA in that order. Read the completion status and conservative result. The hand-checked transformation values below are not a newly calculated solution.

    What you should see: The completed result retains the temporal preparation.

    How to read it: Interpret the configurations only after checking the generated case rows.

Screenshot to be addedthe exact setup

Trace each lag to its actual predecessor

For u1, capacity is 10 at period 1, 20 at period 2 and 30 at period 3. Its one-period lag is missing at period 1, then 10 and 20. The period-2 lag comes from t02, and the period-3 lag comes from t07. These are hand-derived CSV values, not a newly executed app transformation.

For u2, capacities 20, 30 and 40 give later lag values 20 and 30. For u3, capacities 50, 70 and 90 give 50 and 70. The three first-period rows lack predecessors. Exact-period and previous-observation lags agree in this complete file; a gap in the periods can make them differ.

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

UnitCapacity at periods 1, 2, 3One-period lag at periods 1, 2, 3Period-3 minus period-1 change
u110, 20, 30Missing, 10, 2020
u220, 30, 40Missing, 20, 3020
u350, 70, 90Missing, 50, 7040
Screenshot to be addedCheck the matching saved example

Separate an observed tie from its chosen coding

For u1, event A first appears at period 1 and B at period 2, so A precedes B. For u2, B appears at period 1 and A at period 2, so B precedes A. For u3, both appear at period 1 and the first events are tied. You can check these facts directly in event_a and event_b.

A tie does not establish the order within that period. Choose an explicit supported tie policy and explain its meaning. Apply the same care when an event never appears or its information is missing. Unavailable order is not automatically evidence for the opposite order.

The saved columns are capacity_previous_period, a_before_b and capacity_change_1_to_3. For teaching, the saved a_before_b policy maps simultaneous or unobserved events to absence. It therefore codes tied u3 as 0. That code records the policy; it does not establish that B occurred first. Keep the observed tie and the coding decision separate in your interpretation.

Keep the earlier history when choosing later outcome rows

The final-observation exercise selects t03, t05 and t08, the period 3 observations for u3, u1 and u2. Earlier rows still supply the lags and event histories. Filtering the source to period 3 before preparation would remove the evidence needed to create those variables.

Selecting final observations is one possible design, not a guarantee of independence or causation. Explain why that outcome observation is appropriate, which history precedes it and whether the observation window captures the relevant events. Temporal order can support a causal argument, but order alone does not establish it.

Keep a completed transformation separate from a completed analysis

Temporal preparation supports 50,000 source rows, 64 generated outputs and one million generated values. The later set analysis supports 12 conditions and 10,000 original cases. Check the relevant limit at each stage rather than assuming that a successful preparation guarantees an admissible analysis.

If preparation is interrupted, do not assume that reopening will resume an unknown output automatically. Record the returned copy path and dataset name. Inspect the saved derived columns and their original-row evidence before calculating. An opened source-only project does not prove the transformation was saved. Clear dataset names help when comparing several time definitions.

Follow the unexpected value back to its source rows

Check the unit, time rule and original observations before changing a generated value. A missing predecessor, a tied event and a missing saved dataset are different problems and need different actions.

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

SymptomWhat to checkAction
A lag comes from another unitUnit identifier and source observationRecreate the transformation with the correct unit field.
An earlier value is missingBoundary periods and missing predecessor periodsApply and report the intended exact-period rule.
A tied event is counted as clearly orderedFirst event periods and tie policyUse the explicit supported treatment and state its meaning.
Final rows have no usable historyWhether filtering happened before preparationPrepare full unit histories, then select analysis observations.
A reopened project lacks temporal columnsWhich generated dataset was saved and where its project is storedOpen the saved dataset with the temporal columns and verify it before calculating.
The temporal analysis rejects the setupWhether a generated temporal column is actually usedSelect the intended derived column and its calibration.

Report the history rule and the analysis window

Describe the unit and time identifiers, transformation rule, treatment of missing periods or ties, and the observations selected after preparation. State which findings are checks of generated variables and which come from an actual completed temporal solution.

DecisionWhat to report
Time structureThree units; periods 1, 2, 3; one row per unit-period
LagCapacity, one exact period; boundary values unavailable
Event orderA before B, B before A, or tied; explicit policy
Analysis observationst03, t05, t08 for the saved final-observation exercise
Numerical origin hereHand-checked transformations from the CSV

Example Methods paragraph

We used nine synthetic observations from three units over three periods. Temporal columns were prepared from complete unit histories using explicit unit and period fields. We checked one-period capacity lags and first-event ordering against original observations before selecting final-period observations for the teaching analysis.

Explain the prepared variables before claiming a solution

You could report: The checked histories showed A before B for u1, B before A for u2 and a tie for u3. Capacity rose by 20, 20 and 40, respectively, from period 1 to period 3. One-period lags were unavailable at each unit's first observation. These statements describe the hand-checked teaching variables; they do not claim a newly calculated temporal solution.

  • Trace one lag to its original case label.
  • Explain what a missing period would do under the chosen lag rule.
  • State the treatment of tied and absent events.
  • Prepare the required history before selecting final outcome observations.
  • Verify the saved temporal dataset before reading a calculated solution.

Check identifiers and the two different size limits

Missing identifiers and duplicate unit/time pairs prevent a clear ordered history and must be resolved. Data derivation supports up to 50,000 rows. The later conservative temporal analysis has limits of 12 conditions and 10,000 cases. Preparing the dataset does not enlarge the analysis limits.

Calculation comparisons

Further reading

Related guides