Interaction plots

[lilyclements]

I quite like using interaction plots when exploring the data and thought perhaps something like this could be appropriate for R-Instat especially since it is quite an easy couple of functions to implement.

Three Way Interaction There's a really easy function to do a three-way interaction plot (And it's ggplot!). It's with the dae package and is interaction.ABC.plot. I've used it several times but will show their example as then I have code which can be replicated;

interaction.ABC.plot(response = MOE, x.factor = A, groups.factor = B,
                     trace.factor = C, data=ABC.Interact.dat,
                     title = "Three-Way Interaction Plot")

Two-Way Interaction I haven't been able to find a function like the one above to do this. However, it can be done quite nicely with the qplot function outlined here - https://sebastiansauer.github.io/vis_interaction_effects/

So on that link there is this code to do a two-way interaction plot

library(ggplot2) 
library(dplyr) 
tips <- read.csv("https://sebastiansauer.github.io/data/tips.csv")
qplot(x = total_bill, y = tip, data = tips, color = sex) +
  geom_smooth(method = "lm") 

And this is the plot that gives (again this is given on the link above)

It's not a necessary feature, but one I know I use a lot in R and I think could be quite appropriate in R-Instat! Especially since the three-way-interaction plot is so simple.

In fact if the lecturer strike continues onto next week then I might be able to implement it. Although, hopefully I'll have lectures again by then.

[rdstern]

That's really interesting. We do need to get back to our modelling dialogues and that certainly includes having interaction plots. Recently I have become more impressed with the outputs from the mosaic project, see, for example here.

They have the main package called mosaic, also mosaiccore, and mosaicmodel. I'd be interested in your views on that. They have recently also added ggplot2 to their original use of the lattice plotting system.

[lilyclements]

Sorry it's taken a while to reply! I couldn't get it to work on my macbook for some reason and wasn't reunited with my laptop until Tuesday.

plotModel is a nice function. It's not one I've come across before, but is definitely one which looks really useful.

• It uses ggplot2
• The code for the two-way interaction plots are much simpler than what I was suggesting and additionally the function for the two-way interaction plot is the same as the three-way function so there's that consistency there
• I can't find how to get confidence intervals with it but would be very surprised if there wasn't a way to do so!
• With the three-way plot you can do it either faceted (like with the interaction.ABC.plot option) or not have it faceted. I personally prefer it faceted but it's nice it gives that option
• It works nicely with polynomial/log/transformed terms

If I can/Once I do work out confidence intervals on it then it's definitely a preference to the one I was suggesting.

[dannyparsons]

These all look interesting. Just one comment that the two way interaction plot we can do all through our scatter plot dialog as we have the geom_smooth options there now.

[Ogik99]

5164