| Newsletter 10, October 1995 |
A computer, properly applied, will certainly help in data analysis. But a computer is not able to understand the story in your data for you. (Only a human can understand the statistical processes to which the data have been subjected and any assumptions on which they rest.)
“Most people’s knowledge of statistics is either rudimentary or non-existent. For many, ‘statistics’ is a dirty word: one hears such allegations as ‘Anything can be proved by statistics’, but this is only true if they are misused. Indeed, decrying statistics is usually merely a device used by the ignorant to protect their self-esteem. Many [readers] may feel that people cannot be expected to know about statistics or elementary probability theory, but all of us make intuitive statistical judgements much of the time and our false intuitions lead to grievous errors.”
— Stuart Sutherland, Irrationality: The Enemy Within, page 210; pub. Penguin 1994.
Minitab 10 Xtra is the latest version of Minitab for Windows. LMU receives this in accordance with our licence agreement.
LMU now has 2 extra modules for SPSS for Windows: CHAID and Exact tests.
(Chi-square automatic interaction detector)
The purpose of CHAID is to perform Segmentation Modelling, which is a statistical procedure to establish subgroups of a sample that differ with respect to a designated criterion. For example, suppose you are performing market research by means of a survey in order to discover which subgroups of the population are most likely to purchase your commodity. The chosen dependent variable is “expressed interest in the product”.CHAID will divide the sample of respondents into a number of distinct groups based on the best combinations of the recorded categorical predictor variables — say, age-group, sex, area of residence, etc. It produces a “tree”-diagram of the segmentation showing which category combinations indicate the highest levels of “expressed interest”.
This module gives exact p-values. Traditional tests, such as Chi-square, do not as they are based on the large sample approximation. Exact tests allow calculation of statistical significance to be made exactly, even for small samples. Approximately 30 different exact tests are available
Members of staff are welcome to discuss projects of a statistical nature with me. It is old wisdom that consultation is best done at the very beginning!
Also, any member of staff thinking of setting a student project involving collection and analysis of data is invited to discuss it with me, so the aims might be more readily fulfilled within the limitations. It is important that specific, appropriate, statistical procedures are considered before any data is collected. (Rather than collect some data and then think what to do with it!)
Some data sets are held locally. These cover a wide range of subject areas and could be useful in teaching. Some are described in the document Example Data Sets Held on the PC Networks, STATS-UG-3.
A new document is in preparation, Example Survey Analysis Using SPSS for Windows, a fairly self-explanatory title. My feeling is that SPSS, particularly the Windows version, is well suited to analysing survey data as, for example, it has a good labelling facility. This is not to say, however, that I would not use a different package, such as Minitab, when other objectives are in mind, or even transfer data to Minitab where a particular feature is needed. The Guide Sheet Moving Data Between SPSS and Minitab for Windows, SPSS-GS-3, shows how such transfer can be effected.
I have some programs which give instruction in various aspects of statistics (e.g. Quercus which is aimed at Bioscientists). Anyone interested in this area is invited to contact me.
— Dr Paul Marchant