Simulation-Assisted Ex Post Facto Design
Simulation-Assisted Ex Post Facto Research Design · Also known as: simulation-enhanced causal-comparative design, ex post facto with simulation, retrospective simulation design, SAEPF design
Simulation-assisted ex post facto design is a non-experimental observational approach in which the researcher examines already-occurred events or conditions using existing records and then supplements the empirical analysis with computational simulation to approximate counterfactual scenarios that cannot be observed in reality. The design retains the retrospective, naturalistic character of classic ex post facto research while leveraging agent-based, Monte Carlo, or system-dynamics simulation to address the inherent confound limitations of purely archival work.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use this design when the research question is retrospective and causal, when experimental or prospective quasi-experimental data collection is impossible or unethical, and when the available archival data are rich enough to calibrate a meaningful simulation model. It is well-suited to policy evaluation after the fact, historical comparative analysis, epidemiology, and educational research where natural variation in past conditions can be exploited. Do not use it when the archival data are too sparse or inconsistent to calibrate a model reliably, when the causal mechanism is poorly understood (leaving the model unconstrained), or when stakeholders will interpret simulation outputs as confirmed facts rather than model-dependent estimates. If randomization is feasible, prefer a true experiment; if prospective data collection is feasible, prefer a prospective quasi-experimental design.
Strengths & limitations
- Enables retrospective causal inquiry when true experimentation is ethically or practically impossible.
- Simulation augmentation allows explicit counterfactual reasoning, reducing the 'no-manipulation, no-causation' critique of pure ex post facto work.
- Sensitivity analysis within the simulation framework quantifies how robust findings are to unmeasured confounders.
- Can integrate heterogeneous archival sources — administrative records, historical datasets, existing surveys — into a unified analytical model.
- Transparent parameterisation forces the researcher to make causal assumptions explicit rather than implicit.
- Causal conclusions remain conditional on model assumptions; if the simulation model misspecifies the underlying mechanism, inferences may be systematically biased.
- Requires both observational data expertise and computational modeling competence — a combination not always available in a single research team.
- Model calibration depends on data quality; sparse, inconsistent, or selectively recorded archival data undermine the entire approach.
- Simulation results can be misread as stronger evidence than they are — findings must be communicated with explicit uncertainty bounds.
- Publication and replication standards for the simulation component are less settled than for standard statistical designs.
Frequently asked
Does adding simulation make an ex post facto study into a quasi-experiment?
No. The ex post facto design remains non-experimental because the independent variable was not assigned by the researcher and occurred in the past. Simulation augments the analysis by modeling counterfactual scenarios, but it does not create a comparison group through prospective design. The causal inference is stronger than a purely descriptive archival study but weaker than a true experiment or a prospective quasi-experiment.
What kind of simulation model is most appropriate?
The choice depends on the research question and data. Agent-based models suit phenomena driven by heterogeneous individual interactions. System-dynamics models suit aggregate feedback mechanisms. Monte Carlo simulation suits sensitivity analysis of parameter uncertainty in statistical models. The model type should be grounded in substantive theory about the causal mechanism, not chosen for technical familiarity.
How do I validate the simulation model?
Validation involves at minimum face validity (does the model behave as theory predicts?), historical replication (does it reproduce key features of the observed data it was not directly fitted to?), and sensitivity analysis (do conclusions change substantially when uncertain parameters are varied?). A model that passes only the first test should not be trusted for causal inference.
How should I report uncertainty in simulation-assisted findings?
Report the full distribution of simulation outcomes across runs, not only the mean or most likely value. Present sensitivity analyses showing how findings shift across plausible parameter ranges. Clearly label all simulation-derived numbers as model estimates, not observations, and state the key assumptions on which the counterfactual scenarios rest.
Is this design suitable for small archival datasets?
Caution is warranted. Simulation does not create statistical power from thin data; a poorly calibrated model amplifies rather than corrects data limitations. If the archival dataset is too small to reliably estimate even a baseline statistical model, simulation adds uncontrolled uncertainty. Consider whether a descriptive ex post facto study — without the simulation layer — is a more honest representation of what the data can support.
Sources
- Kerlinger, F. N. (1964). Foundations of Behavioral Research. Holt, Rinehart and Winston. link ↗
- Shadish, W. R., Cook, T. D., & Campbell, D. T. (2002). Experimental and Quasi-Experimental Designs for Generalized Causal Inference. Houghton Mifflin. ISBN: 978-0395615560
How to cite this page
ScholarGate. (2026, June 3). Simulation-Assisted Ex Post Facto Research Design. ScholarGate. https://scholargate.app/en/research-design/simulation-assisted-ex-post-facto-design
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Agent-Based ModelingSimulation↔ compare
- Ex Post Facto DesignResearch Design↔ compare
- MONTE-CARLO-SIMULATIONDecision-making↔ compare
- Retrospective Cohort StudyEpidemiology↔ compare