On April 10, 2018, I gave a talk entitled Developing your first R package: A case study with esvis for the Eugene R Users Group. Although I discussed my esvis package, the focus of the talk was really on tools and tips for developing R pacakges. In this post, I’ll go over some of the content in that talk, and discuss why I think you should develop an R package.

This post has needed to be writtend for a little while, but I’ve been busy with the actual work of redesigning my website (in fact, I have a number of posts that are backlogged). This post will have a little bit of code (all of it CSS, rather than R), but mostly it’s just about my journey and things I’ve learned. Getting started with Blogdown In Yihui’s introduction to blogdown he advocates for simpler themes.

In this post I basically just wanted to share some recent developments that I’ve made to my personal R package {sundry}. All of the recent advancements have been made to work with the tidyverse, so things like group_by should work seamlessly. If you feel like giving the package a whirl, I’d love any feedback you have or bugs you may find. At this point the package is only on github. If there seems to be interest from others in using any of this functionality, I may submit it to CRAN.

A relatively routine process for me is to combine multiple files into a single data frame by row. For example, the data might be stored in separate CSV files by grade and content area, but I want to load them all and treat it as a single data frame with a grade and content indicator. A good default for this sort of process, is to keep all the variables that are present in any data file, and pad with missingness for the files that don’t have that specific variable.

In this post, I’d like to share one of the more unique plots from esvis - the binned effect size plot. The overall purpose of the binned effect size plot is to evaluate if the differences between two distributions are consistent across the scale. We’ll start with a quick example from the package, and then move to a simulated data example. Quick example The API for esvis is consistent across functions, so we use the same outcome ~ predictor forumula as the first argument, followed by the data for all functions.