The Statistics Tutorial Day

Transcribed from the printed issue; not yet reviewed. Read it in the PDF of the issue, page 65

The Statistics Tutorial was of a very mixed standard in more ways than one.

The morning sessions were intended to be an elementary guide to statistical techniques but were marred by the hi-tech video projection equipment not locking on to the PC-generated graphics displays. These were prime examples of the need to have a low-tech back-up of slides or foils for when hi-tech things go wrong. Also the early sessions lost momentum (and many people's interest) with the use of too many complicated models (plastic fruit joined by dowel rods to illustrate the concept of stratification in populations - I think!!). Too much time was spent in constructing these during the presentations.

As is usual at conferences the published papers are more informative than were the actual presentations. The first was entitled "Statistics: An Inductive, Conceptual Approach", and the second was "An APL Sampler: A Tool for Illustrating the Central Limit Theorem". This latter paper could be of interest to many end-users of our APL systems.

The afternoon sessions were much better presented and aimed at a higher level of understanding. The papers presented were entitled "Graphical Techniques in Statistical Data Analysis and Inference", "Statistical Analysis of Two-way (and Higher Dimensional) Tables", "Statistical Linear Models Using APL", and "A Statistical Symmetry".

The first was not specific to APL but did illustrate a number of ways of graphing statistical data. The second could be of interest to end-users and developers of our APL systems, especially those in marketing and reservations. It is mainly about the analysis of multi-dimensional contingency tables of frequency data, for which APL is admirably suited. APL expressions for Chi-squared and Fisher's tests are given as well as for more sophisticated tests.

The third paper was about estimating the parameters of a linear model, and hence the 'signal', given a number of observations affected by 'noise' (or random errors). APL may be used very effectively for this since the least-squares regression operator (domino) is powerful and easy to use.

The "Statistical Symmetry" paper was about generating samples from various distributions using random number techniques, which can then be used for simulations and modelling.