Jackson
Better Research Planning Through Simulation (303795)
Graeme Blair, University of California at Los AngelesAlex Coppock, Yale
*Neal Fultz, Factual
Keywords: research, design, planning, simulation
Research plans need to specify the population of interest, a sampling method, an experimental assignment method, and a method to estimate and test parameters of interest. In all but the most simple designs, these steps can interact, unintentionally leading to biased or under-powered studies, and ex-ante diagnostics (such as power calculations) may not have easy, closed form solutions.
Instead, researchers turn to simulation methods. We have implemented DeclareDesign, a set of free and open-source R packages, to make due diligence much easier:
* fabricatr - create realistic synthetic data sets, with hierarchical, panel and correlated structure. * randomizr - conduct random sampling and assignment with clusters, blocks or strata. * estimatr - fast estimators for design-based inference (with clustered and robust standard errors) * DeclareDesign - an extensible framework for declaring, simulating, and diagnosing research designs.
The suite of packages allows researchers to mix and match different strategies, and run simulations to determine the best plan to answer their research questions.