Profesional Development Workshop

Day 1: Tuesday, 13 September


Many Models: Conducting Sensitivity Analyses in R

Session convener: Liam Wright, University College London

Typically, research papers in the health and social sciences include the results of one, or a few, models. These are drawn from a wider universe of defensible models that could have been run, but results may not be robust to these other model specifications. Other times, answering a research question requires running a model over different subsets of the data or with different sets of variables (for instance, repeating a model across a set of ages or performing an outcome-wide analysis). However, writing the code to run multiple models can be daunting and error-prone. In this interactive workshop, we will demonstrate simple ways to run many models using the tidyverse package within the programming language, R. We will also show how the results of many models can be combined and presented in simple, comprehensible ways using the ggplot2 plotting package, and we will show how computationally expensive models can be run cost-effectively in the cloud with Amazon Web Services. Attendees will require access to R and RStudio, which are both freely available, and will need to install the tidyverse package before joining the session (install.packages(“tidyverse”)). Syntax and data will be provided.