Dynamic reporting

Statistics as a scientific discipline has a dynamic nature, which can be observed in many statistical algorithms and theories as well as in data analysis. For example, asymptotic theories in statistics are inherently dynamic: they describe how a statistic or an estimator behaves as the sample size increases. Data analysis is almost never a static process. Instead, it is an iterative process involving cleaning, describing, modeling, and re-cleaning the data. Reports may end up being re-written due to changes in the data and analysis.

A dynamic document is a mixture of normal text and program code; when we compile the document, the code is executed and turned into results. In the end, the results from computing and the original text are combined to form a report.

Literate Programming

Statistical reporting is often the last step of data analysis, and traditionally it is separated from the computing, i.e., we finish all the computations and collect the results before we write the report. This is not only more demanding in terms of manual work, but also error-prone. The paradigm of literate programming (Knuth, 1984) was brought forward in the 1980s to write software and documentation, and later it was incorporated into the R language as the well-known tool named Sweave (Leisch, 2002) for writing statistical reports. However, Sweave is not a complete implementation of literate programming, as pointed out by Schulte et al. (2012), and it lacks many useful features in statistical reporting such as caching and full integration with the capabilities of R graphics. Literate programming is often related to reproducible research, because the results are generated directly from computer code and it is easy to re-run the code to get the same results.

Animation

Animations can add insight and interest to traditional static approaches to teaching statistics and reporting, making statistics a more interesting and appealing subject.

Traditional static printed statistical reports have severe limitations, for conveying results with some degree of complexity to an audience that is not trained to understand the statistical ideas. Here, an active animated way to present results can help greatly. The human visual cortex is arguably the most powerful computing system to which we have access. Visualization puts information in a form that uses the power of this computing system. By virtue of our visual system, we may be able to quickly understand a complicated method or result, or at least a simplified case. Wender and Muehlboeck (2003) showed the advantages of animations in teaching statistical topics like least squares and type I & II errors. Velleman and Moore (1996) discussed a set of animation ideas that can be effective for a student to learn concepts actively. Animation can be helpful only if it is used appropriately; Fisher (2010) mentioned several successful applications of animations such as GapMinder (Rosling and Johansson, 2009), but also argued animation could be very bad when used poorly – it is usually good for presentation but not for exploration.

library(animation)
conf.int(0.9, main = "Demonstration of Confidence Intervals")

for (i in 1:50) {
    plot(x=i/10, y=i/10, xlim=c(0,5), ylim=c(0,5), pch=20, col="red", cex=5)
}