Home > Research > Activities > Experimental design for multi-level data: Impro...
View graph of relations

Experimental design for multi-level data: Improving our approach to power analysis using Monte Carlo simulation-based parameter recovery estimation

Activity: Talk or presentation typesOral presentation


Abstract: Ensuring that our experimental studies are adequately powered to detect an effect of interest is a central concern across scientific disciplines. Multi-level experimental designs present a particular challenge for power analysis in its traditional sense, with most formulaic power analysis calculations silent to these internal data structures. In recent years, simulation-based approaches to power analysis have become more accessible, through R packages such as SIMR (Green & MacLeod, 2016). While this represents a welcome step-change, in their current form, these new approaches are limited in a number of ways. Power analysis is generally defined as the ability to recover an effect different from 0, involving a significance test in which a parameter estimate from a simulation-model is contrasted with null. In effect, this approach can tell us if our effect of interest is different from 0, but not by how much. Additionally, more user-friendly simulation-based power analysis methods typically offer limited flexibility in the range of model classes they can accommodate. I will demonstrate a general framework that can be used to overcome these issues: allowing for effective and informative calculation of parameter recovery across a range a model classes, including Bayesian approaches.

Event (Conference)

Title12th International Multilevel Conference
LocationUtrecht University