Shoulder to shoulder
My occasional colleague Hafeez Jeraj, better known as 'Hash', is doing work on the use of a bioluminescent pseudomonas aeruginosa isolate to quantify bacterial adhesion to contact lenses. All scientists are, by definition, nosy creatures given to inquisitiveness, and freelances like myself are the worst of all - so I've been unable to resist the temptation to paddle around in Hash's data. I've learnt a lot, some of which makes me glad that I don't use contact lenses, but that's another story.
Hash is working in a university where the standard desktop statistical software is the academic stalwart, Minitab. By a strange mischance, I happened to have a new version of the program (release 14) about my person when we started discussing his findings, so it immediately became our lingua franca. Part way through the conversation, a major version release of Systat (11.0) arrived, made itself at home on my desktop machine, and joined the exploration. Since both of these packages have a long track record, it was interesting to see their latest incarnations at work, side-by-side on the same task. Both have reputations as packages uniquely useful for particular audiences, but with idiosyncratic interfaces in the eyes of the mainstream market, evoking the clich? 'no gain without pain'. The differences are still visible, but the pain has now disappeared and both are well worth revisiting.
- Minitab (top left) shows its multiple window structure, with minimised windows stacked at the bottom. The Minitab Project Organiser tree is visible from behind Systat (bottom right) with its contrasting tabbed multiple pane approach.
Most statistical packages these days are diversifying, attempting to spread their bases beyond the territory from which they originally sprang. Minitab and Systat are no exception to this. Minitab had already made a significant shift before this release, and the momentum is still under way; Systat prepared the ground in the previous two releases and has been reborn here. When I wrote about Minitab 13 (a long time ago now, it seems) I particularly noted the new emphasis on industrial and commercial application in a program originally designed for student use. Systat has similarly looked outward from its social science roots to the larger science world. Particularly interesting was how their very different approaches had converged in this evolutionary expansion - not just with each other, but with other products and with trends in the larger software environment.
As a visual thinker, my first approach to Hash's contact lenses was through construction of multiple pictorial models. I'm not alone in this tendency, and the publishers of analytical software know it; both these products have seriously examined and overhauled their graphics systems within the context of larger design considerations. Minitab has focused on streamlining of graphics creation and tuning, expanded its graphics command language, and the combination of the results into compound reporting documents. Systat's graphics have moved radically away from the imperative mode towards interactivity (though the imperative environment is still there, if you prefer it that way) as part of an overall revamp of the user interface. In Minitab's case, there is further evidence of the emphasis on industrial applicability, which I mentioned above, in additions such as Hotelling and Multivariate EWMA, empirical cumulative distribution graphing, graphical reporting of residuals, and an expanded fishbone cause-and-effect plot. In Systat, the big changes in this department are in real-time manipulation of screen plots though joystick-style visual mouse control of pitch, roll, and yaw, right-click editing of characteristics from colours and coordinates to smoothing behaviour, and changing of variables from within a completed graphic. Both packages have made various improvements in detail as well.
- In Systat (left and upper centre) an initial visualisation of revealed bacterial adhesion levels within their three dimensional data space has been plotted, for convenient manipulation and exploration before refinement. Minitab is rotating a similar dataset in scattergram form for examination with no envelope (lower centre) and displaying a smoothed interactivity surface projection for cleaning solutions (top right).
Nobody with past experience of Systat can start up this new release without commenting on the new interface. In a past review of a previous version, I described Systat as being like a Lunakhod - form following the execution rather than the aesthetics of function. That's changed now, with a move to the familiar multiple-pane organiser model. The program opens in the Output Organiser, where the usual operational elements appear: the data sheet and the results of operations, graphic or analytic, conducted from it. Items selected from the outline tree, in the left hand pane, appear in detail on the right, and elements can be copied, cut, and pasted from place to place. The other tab, the Dynamic Explorer, is used for editing and manipulating artefacts in detail.
Minitab implemented a similar reorganisation of approach a while back, although its implementation is based on multiple windows rather than multiple panes. Interfaces, of course, are a means to an end; a way in which to access most fluently the statistical tools on offer behind them. In this respect, both products have moved significantly forward. In the past, each was more easily used as the only window onto any particular project, but there are now no problems with use in tandem, with other analytic tools, or with the software environment in general. That has to be an advantage; interoperability is key to development of cooperative working and, while plurality of tools is essential, isolationism is not. Minitab and Systat offer different palettes of strengths and viewpoints; long term retention of those vital differences in the 'gene pool' of the market depends on ability to integrate without requiring a retooling of mental processes developed elsewhere. That becomes particularly important as analysis develops; if different tools are to offer complementary synergy rather than mutual incomprehension, it's important that similar concepts are accessible in similar intuitive relation to similar interface functions, albeit from different perspectives. Both Systat and Minitab now fall fully within the larger sense of community without having lost their distinct identities.
Still heading for the ultimate destination
The analyses behind those interfaces are grouped in ways that are different, offering no barrier to migration back and forth, but do reflect the different development backgrounds. Peering through Hash's contact lens data, I started from conventional hypothesis testing options and worked outward. In Minitab, the main menu heading is 'Statistics', and Student's t is under 'basic statistics'; for Systat it's 'Analysis' then 'Hypothesis Testing'. Both have Wilcoxon under a 'Non Parametric' subheading at the first level. So it goes on, as you work up the scale of complexity; different but equally clear access structures that take only a moment's thought to align. These two products both have a history of first-rate documentation, and that tradition continues; but the necessity has waned to the point where I doubt that any statistician needs it before getting productive work done. Minitab reached me as a CD only, with the documentation in electronic form. Systat had the traditional small library of paperbacks but, apart from briefly admiring their thoroughness and completeness on arrival, I've left them untouched in the box. This is not a suggestion that extensive high-quality documentation is no longer needed; it is the cornerstone guarantee of the software that it accompanies, but its shift from instruction text to reference source is an evolutionary landmark that both Minitab and Systat have safely passed.
With all this enthusiasm for the journeys on which both products have embarked, and the distances which they have already travelled, it's also true that neither has yet reached their ultimate destination.
Where a set of tools are placed is an important signal about intended audience. Minitab's core Stat menu is less compact than Systat's Analysis equivalent, mainly because it places at top level a set of groupings (control charts, quality tools, reliability/survival) that might come under the heading of 'new industrial audience'. I've talked to some industrial users, and the decision seems to be a good one in marketing terms: there is a fairly large population of managers out there, for whom statistical analysis is not the favourite way to spend a day and who welcome signposting. The graph menu is similarly expanded; where Systat collapses the whole graph type selection process into a single Gallery option and five classes at the top level, Minitab enumerates 18 types. The top-level menu also has more entries. Where Systat still assumes that its users are experienced in such software, Minitab aims to welcome those who are doing this work from necessity rather than inclination.
One of Systat's key strengths has always been the psychoperceptual range of its visualisation methods - uniquely valuable, in my opinion, and the reason for bringing it to bear on this study. But much of this range is, so far as I can discover, still accessible only from the command line. In looking at the behaviour of Hash's contact lens adhesions over time in different environments, I have been using many of my favourite spatial plots to visualise what's going on; but they are not always easy to get at. I look forward to the day when I can click a graphics menu option and see a relevant subset of the graphics gallery; select (say) a Voronoi tessellation from that gallery; and then interactively fine tune the result, as I already can with other displays.
Suggestions for the future
Minitab's next move, in my opinion, should be to tame its Window structure. This needn't be a major undertaking. The p. aeruginosa dataset on which I am working is, in the general scheme of things, tiny; but it has numerous ramifications. As I work on it, exploring avenues that open up as its secrets are unpacked, the Minitab windows proliferate. There are 32 of them so far, peeking out at me from behind this wordprocessor as I write: data sheets, plots, output, command history, several runs of the different experiments. As I minimise these windows, they stack themselves very neatly as small identifier blocks from the bottom left of the screen; a delight to see. But as I open them, move them, resize them, close them again ... after a while, as happens with physical resources in my study, the blocks are all mixed up and I can't remember where I put any given item. They are, of course, instantly accessible from the Project Manager window... but I can't remember where I put that one, either! I don't want the very flexible multiple window model abandoned, but I would welcome an option (for example) to dock permanently the Project Manager along a chosen edge of the screen or workspace. To minimise Windows from their Project Manager entries would be nice, too.
That I have closed with suggestions for the future should not obscure my enthusiasm. On the contrary; the suggestions are worth making because they follow good work already done. This is a thoroughly welcome pair of upgrades, making a clear view of data more painless than it was before.