Lortz, Sebastian A.J. (2025) Advancing Replication Study Decision-Making with DISCOURSE: Data-simulation via Iterative Stochastic Combinatorial Optimization Using Reported Summary Estimates. Research Master thesis, Research Master.
|
Text
S4381319LortzMasterThesisSubmission.pdf Download (3MB) | Preview |
Abstract
In response to the replication crisis and the frequent unavailability of raw data sets in psychology and the social sciences, I developed DISCOURSE - Data‐simulation via Iterative Stochastic Combinatorial Optimization Using Reported Summary Estimates. This algorithmic framework simulates plausible data sets from published summary statistics when access to raw data is restricted. DISCOURSE comprises four modules tailored to various analysis contexts: Descriptives, ANOVA, Multiple Linear Regression, and Linear Mixed‐Effects Regression. Each module’s workflow initializes a candidate data set and through iterative cycles, the algorithm applies stochastic perturbations, heuristic adjustments, and permutation‐based moves to alter values. An objective function evaluates the alignment with reported summary estimates and a simulated‐annealing acceptance criterion with temperature schedules ensures robust exploration of the search space with convergence towards global optima. I validate the method on multiple benchmark data sets with known raw data, demonstrating that each module reproduces its target summary measures with very small discrepancies. I showcase the application of the algorithmic framework using published research articles and discuss the method’s limitations. DISCOURSE is available as R package and comprehensive ShinyApp, offering researchers a tool for generating synthetic data sets solely from summary estimates.
| Item Type: | Thesis (Research Master) |
|---|---|
| Supervisor name: | Ravenzwaaij, D. van |
| Degree programme: | Research Master |
| Differentiation route: | Understanding Societal Change [Research Master] |
| Date Deposited: | 17 Jul 2025 09:20 |
| Last Modified: | 17 Jul 2025 09:20 |
| URI: | http://gmwpublic.studenttheses.ub.rug.nl/id/eprint/5595 |
Actions (login required)
 |
View Item |