Last week I had a chat with one of Kerrie Mengersen’s new MSc students about the UPTECH work I’ve been doing. Aleysha Thomas will be working on scientific visualisation of statistical modelling/inference/analysis and has been talking to some of the other students in our stats group to get an idea of what sort of work we do and what sort of stuff there is to visualise. For example, my doing semi-parametric regression with splines means that the marginal densities of the parameters aren’t as interesting as the fitted smooth which is comprised of those parameters multiplied by B-spline basis vectors. A tensor product of two splines is better off visualised with a contour plot than with a list of coefficients and their credible intervals.
My work involves spatio-temporal modelling but I don’t present it as a spatial map if the temporal variation is more interesting. At some point, though, I’ll be publishing graphs of continuous space spatial random effects and that will require some thought as to how the uncertainty in the estimates is represented. At the moment I’m just using levelplot() to plot the mean and standard deviation, but there are other options such as stacked surface plots and contour plots. Matt Moores does spatial statistics as well (cone beam and fan beam CT scans for radiotherapy treatment) but deals with completely different data, methods and visualisation techniques. In addition, our audiences’ only overlap is statisticians.
Pitching to your audience is a very important part of presenting your work, whether it’s at a conference, in a journal article or to your PhD seminar panel. It’s no good me coming up with a beautiful way of representing my inference if it’s impossible for others to understand. Hopefully with Aleysha on board our stats group will be exposed to new ways of visualising data and inference. My stats group is certainly better than my aerosols group when it comes to graphical representation of data and results, but we can still be doing better (myself included).