Better Research Planning Through Simulation
summary
Shows a concrete workflow: use fabricatr, randomizr and estimatr inside DeclareDesign to declare a population, sampling and assignment, and an estimator, then run simulations. Example: a clustered classroom experiment (10 teachers, ~20 students each) runs 10,000 sims in ~240 CPU seconds, finds bias ~0.36 and power ~0.50, so the planned design fails. Packages bootstrap by default, integrate with tidyverse and support parallelization via future for speed.
AI-written summary · about
chapters (13)
- 0:00Introduction and motivation
- 1:08Reproducibility crisis and pre-analysis plans
- 6:24Garden of forking paths explained
- 8:49Introducing declare design suite
- 10:15Fabricator for synthetic data
- 13:24Randomizer for sampling and assignment
- 15:34Estimator for common analyses
- 17:20Declare design integration
- 18:25Example study simulation
- 23:43Design diagnostics and power curves
- 26:09Real-world data simulation case
- 27:56Key takeaways and future work
- 29:18Q&A and acknowledgments
related talks
57mGood Practices for Applied Machine Learning
22mModelling, Tidyverse Way
05mTemplated Analyses within R Packages for Collaborative, Reproducible Research
17mModel Migration: from Excel to R
14mDesigning a Socially-Critical Data Science Course- 57mGood Practices for Applied Machine Learning
20mRpanda trading simulation - from an idea to a multi-user shiny app
15mBeyond Efficiency