I had to compute an indicator this week. It had confidence intervals that relied on taking 100,000 samples from the indicator’s approximate distribution. I had to repeat this over multiple GP practices and for twelve different demographic groups. I decided to use dplyr1 because I thought it would help me organise all subgroups involved. I used mutate_at() heavily and thought that dplyr was keeping everything organised. However, when I moved from the 10 samples I’d used for testing to the 100,000 samples required by the specification of the indicator, my code moved to a crawl.

I love making plots in R with ggplot. However, there are always a few niggles that I forget about between plots. I wrote this post so that I have somewhere to look the next time I need to tweak a few things in my plots. I intend to come back and add updates in the future as I learn more things. If I keep coming back, I might also remember a few of these too.

The shiny app embedded below allows you to explore the number of packets you need to complete the Panini sticker album for this summer’s Euro tournament. This builds on the results I presented in an [earlier post](link to earlier post) and allows you to explore how many fewer packets you need to buy when you have more friends to swap with. You can also vary the number of runs performed because the model runs considerably slower with more swappers involved.

As a follow-up to my post about the Euro 2016 Panini Stickers, I’ve now completed the collection with the help of an online swapping site and by buying the last 39 stickers directly from Panini. I also managed to write a new simulator, this time with additional collectors involved. I assume that each of N collectors will buy a packet of stickers and add any new stickers to their album. Then they attempt to swap any leftovers with the rest of their friends.

This post deals with how to generate random numbers in R. It is good to know how to generate random numbers with a particular language or software package for at least one of the following three reasons: You want to test something that depends on a particular distribution. You’re running a stochastic process of some kind (Branching process, random walk etc) and you need random numbers for deciding whether an event occurs.