17 December 2009
Reviewed by Felix Grant
The 12th edition of GenStat for Windows has been out for a few months, now, but I’ve been examining it in the wild (so to speak) rather than on a review bench. One reason for this has been an upcoming SCW feature, which has had me schlepping around a variety of life science teams, for which many of 12's features were useful, and in which GenStat happens to feature prominently.
New in 12th edition is provision for linkage to the third-party experimental design package CycDesigN. GenStat itself has one of the best inboard design suites around, but the option to call on an external tool that generates GenStat code can never be a bad thing. Such linkage is part of a general VSNi policy of ecumenical openness and interworking. GenStat has a reputation for 'opening anything', often through prepackaged solutions or, failing that, flexible tailored input filtering. With this release, four more file formats join the ever growing ready-to-roll category: Cornell ecology (from the program series of that name), MapQTL, R/QTL (both of particular interest to geneticists or those working in related fields) and Stata 10. During the time I’ve been carrying GenStat 12 around with me, the first three of those (I haven’t yet had occasion to import any Stata data in anger) have, along with the CycDesign linkage, shown themselves to be solid, reliable, and useful.
The worksheet gains a number of refinements, including a very useful set of table manipulations. One of these allows display as a 3D cube, with data rearranged by two factors across row and column and the third in tabbed layers.
Genetics continues to get considerable attention, well beyond file format access. The dedicated QTL stats menu option gathers QTL tools together in one place, including an intelligent and user friendly workbook from which data structures can be managed fluently. Microarrays and metanalytics also get new menu space.
More generally, methods continue to be added to the menu structure across the board. As a conservative statistician of a certain age, I have my reservations about imputation of missing survey data, but I have to confess that both model-based and hot deck methods are soundly implemented with a well thought out set of options and controls. Multivariate analyses gain easy access to MANOVA permutation tests and biplots for principle components or canonical variates. A rich set of quantile methods joins the regression menu, as does an option to run numerous differently modelled analyses in parallel. ANOVA has trellis mean plots and standard difference errors only a click away.
Despite all this work on ease of use, of course, there is a great deal more to GenStat than its WIMP interface. Below the GUI, it is still a language driven environment and 12th edition sees 40-plus new procedures (of which R0INFLATED, which fits zero-inflated regression models, has become a great favourite of mine) a similar number of new functions (many of them image related), and two directives. There are numerous tweaks in detail, as well, which collectively smooth use of the whole. Running a review over this prolonged period, on live work, has brought home to me what a rational, ordered and planned upgrade path VSNi has for its flagship product. The very success of this approach, incremental over frequent releases rather than trumpeted at longer intervals, can sometimes leave one overlooking its quietly impressive quality.