Minitab is, in many ways, the archetypical statistics package. Like many seminal products, it was designed for ease of academic use, but the move to applied industrial relevance was made very early on. A decisive shift in this direction came with release 11, and the new release 12 places the package firmly in the quality and process control market.
Robustness, traditionally associated with Minitab, is a foremost characteristic of this release and a powerful reason for purchase; the cost will rapidly be made up in support-cost savings. An early beta build, seven months before production, was virtually unbreakable, and this remained true right through to the final release. A Minitab executive commented that their policy is to "go right down to the wire with bug checking"; this obviously pays dividends. The review team was impressed enough to use the beta for live data, with no regrets. Some programs in this market fall over with depressing regularity; Minitab clearly shows that this needn't be so.
Continuity also characterizes this release, with further productivity payoffs. There is a solid body of new development "under the hood" but no radical rearrangement of the interface. New tools are simply there, when wanted, within the familiar framework. The nature and location of enhancements confirm development directions already discernible in release 11. Regression analysis, multivariate analysis, tables, time series analysis and nonparametrics have been left alone.
General management issues have received considerable attention, tightening up on coherence and providing efficiency gains. Basic statistical tools have been tuned for flexibility and depth, while industrial and pre-study concerns have been considerably beefed up. The emphasis throughout is on solid, reliable expansion of capacity within the existing structure rather than scene-stealing new toys.
Although for this software a math coprocessor is not essential on an SX machine, graphics performance is noticeably sluggish without it. Computation on an SX is fine, and on a full 486DX or above, graphics follow suit.
In terms of general handling, a great deal of work has gone into making Minitab intuitive and productive, bringing specialist statistics handling more closely in line with general numeric practice and other software expectations. With this release, Minitab makes that indefinable move into the community of programs that are an active pleasure to use. Multiple worksheets, dynamically sized to deal with pasted data, add a new level of usability. Easier generation of those worksheets from subset data provides an additional gloss.
Complementing and unifying these changes is the biggest and most obvious of all: the introduction of "projects" as the primary Minitab file save unit, storing all worksheets and their associated session output and graphics in a single entity. The final touch, font control in the data window, may sound like a side issue, but it is one of those usability details that make a real difference to operator effectiveness and efficiency.
Quality control tools are the rational extension of good design into operational practice. For industrial R&D controllers and their production line colleagues, issues of design efficiency and reliability are (or should be!) priority concerns these days; use of a statistics package for post hoc analysis of chance sampling is a luxury. Objective fed study and application loops are the order of the day, and the tools for such control, in place with version 11, have been significantly strengthened in the current release.
The most widely used analyses relating to design sensitivity and power (z, t and proportion tests, one-way ANOVA, two-level full or fractional factorial and Plackett-Burman) can now be handled within Minitab. This places the means for maximum efficiency where there is no excuse for avoiding them -- and, of course, provides the means to draw post mortem lessons from the wreckage if they are ignored.
Designed experiment tools, logically contiguous with sensitivity studies, see a well-planned expansion in the areas of design production and analysis, botched runs and multiple response optimization. An overall concern with well-founded intuitive exploration, noticeable throughout the package, manifests itself here in the combination of numerical, interactive graphic and overlaid contour plot views onto the optimization process. Modification and updating of designs is better facilitated, offering further incentive for well-designed and evolutionary study practice.
Traditional quality control tools are comparably strengthened in release 12. Control charts deal with normal and non-normal data, and process capability embraces attribute data. Centrality estimations are separately illustrated for different groups on the same historical chart, with effective graphical treatment of symmetry and ANOVA.
Survival and reliability prediction is equally vital, and the software provides a solid range of core methods for both parametric and nonparametric distribution analyses. Full provision for left, right and interval censoring accompany exact failure; eight parametric distributions are backed up by appropriate nonparametric estimates. Life data regression and accelerated testing are well-engineered, as is probit analysis. The modified Newton-Raphson calculation consistently provides the best estimated parameters, although manual substitution is allowed if required.
Moving to the general statistical backup for these specialist tools, Minitab now has a range more than adequate for quality assurance and improvement use. Paired tests and intervals for t, one or two proportion estimation and tests, and p-values for correlation have been added to the basic statistics kit, and descriptive statistics can now be simply stored in worksheet columns. A new high resolution dot plot has also appeared. ANOVA capability now encompasses unbalanced and fully nested designs for expected mean squares, multiple comparisons and a good implementation of approximate F-tests.
In the real world of organizational cooperation, full OLE editability for graphic Minitab objects within other applications is a major asset. Software can no longer be an island; its productivity must be assessed in the overall context of the organizational environment within which it will be used. Increasingly, provision of static graphic content to other components of that environment will no longer do. The ability to update such material in situ is becoming a vital part of the live communications culture through which statistics now serves process. Less dramatically, graphic file save options have been expanded to include TIFF, PNG and JPEG formats, reducing the need for peripheral utilities.
Minitab's documentation remains structured and lucid. The education origins are visible here: First principles are provided but easily skipped by those familiar with the techniques involved. There is no distinction between graphical user interface and command line usage, allowing users to employ the mix that best suits their needs, experience or circumstances. The program comes with a good tutorial booklet for first-time users, providing rapid access whether at entry level or on transfer from other packages.
For single users of version 11 or earlier, running (or planning to run) Windows 95 or NT, release 12 must be a recommended upgrade. For any setting where statistics is a primary concern but Windows 3.1 is still in use, the additional power and productivity potential in release 12 are justification enough for upgrading Windows itself. For multi-user sites, upgrades on 486DX machines or above will pay for themselves, and retention of release 11 on SXs and Windows 3.1 PCs will cause no confusion. Users can move back and forth between the two releases without noticing the difference much until the new facilities are required.
For those with no existing commitment to statistical software, an additional selling point is Minitab's availability on other platforms. It is very unlikely that Minitab cannot be applied across the mix of your hardware and software environment, and its portable worksheet format ensures transparency of data transfer.
Minitab Release 12 by Minitab
System requirements: 486 computer or better; a Pentium is recommended. A math coprocessor also is recommended. Windows 95 or NT. Minimum of 16 MB RAM for Windows 95, 16 MB for Windows NT. 25 MB free disk space.
Price: $975. Network and academic pricing available.
Contact: Minitab Inc.
3081 Enterprise Drive
State College, PA 16801-3008
Ph: (800) 448-3555
Fax: (814) 238-4383
About the author
Felix Grant is a freelance journalist, lecturer and research consultant in the United Kingdom.
Streamlined Calibration Tracking
GAGEpack for Windows
Review by Craig S. Howell
ith current ISO and ANSI standards requiring increased accountability and record keeping, calibration software packages are finding their way into more companies. Requirements vary with each company but typically include a records database showing unique controls for each gage, location or owner, size or range, nomenclature, date of last calibration and date due for recalibration.
GAGEpack offers a multilevel system that allows users to set up systems as basic or in-depth as they require. It can be configured to store an unlimited number of gage histories, procedures, schedules, repairs, vendor information, R&R gage studies, and usage and location details. The data required to get started can be as simple as gage numbers and due date intervals, but the program perhaps impresses most when handling more detailed information.
The program maintains three different databases: gage, user and choice list. Users can create secondary databases to track gages used by different departments.
The gage database stores gage information such as descriptions, procedures, histories, calibration and R&R studies. Interval and due date calculation offer choices based on time, gage usage or both. Automatic or manual adjustments to increase or decrease the interval can be made depending on pass/fail factors.
Calibration instructions and calibration steps can be manually input for each gage, or you can call up existing documents from your computer or a network. When entering the specifics, a choice of variables or attributes, unilateral or bilateral tolerancing, and pass/fail conditions are available. Correction factors can be set up for the standards used. Gage events include calibration, check-in and check-out, R&R studies, repair information and user-defined history details.
The user database contains user names, password and security options, and access rights information. Security levels and user access rights can be created for each user or left unrestricted.
The list database includes choice list options for different gage types, status, locations, job and part numbers, manufacturers as well as any user-defined lists. Separate user-defined fields are available and can be tied into the choice lists. A vendors area stores information about the company the gage was purchased from as well as details on the company that repairs the gage. The gage list summarizes information for all gages in the current database. Listed gages can be sorted -- a commonly used feature to generate calibration due lists -- filtered, found, deleted, copied and added to. The gage list is accessible to the gage form.
The amount of information available in this program is exceptional. For instance, the report list offers 63 different reports, ranging from calibration certificates to R&R study results. Changing the format of the reports requires a fair amount of programming knowledge, however, making them rather rigid to use.
In our evaluation copy, the calibration certificate shows six lines for measurement standard information and NIST number -- information required by most ISO and ANSI standards. However, the information can't be entered from within GAGEpack but must be entered manually after printing the certificate. The company has fixed this in the latest version. Existing databases can be imported into GAGEpack using its Access-based functions.
A superb feature on the gage form is an option to use existing calibration procedures and data sheets from any system drive, network disk drive or CD-ROM by entering the appropriate path information in the calibration instructions. This feature allows users to take advantage of the most current procedure offerings available from NAVAIR, GIDEP, etc. on CD-ROM without rekeying them. A complete section on R&R study features a number of statistical analysis functions and a comprehensive section on measurement system analysis and variables.
The software's biggest drawback is that users can specify only one master gage per gage in the calibration instructions. Most gages require more than one measurement standard/master gage to perform a calibration. This prevents reverse tracing to determine which gages were calibrated with a specific master gage should it be found to be out of tolerance. A process to provide that information is required by most ISO and ANSI standards. The calibration steps page has an area to assign a separate master gage to each calibration step, but the program's current version doesn't allow sorting gages by this information.
Online help is comprehensive and easy to use, although the printed user guide does not go into much depth. However, the technical support was very helpful -- and free for registered users. Overall, GAGEpack for Windows is well-thought-out and provides a choice of detailed information. Users will find that operating the program is intuitive, and the extras included make this package a useful tool and a reliable investment.
GAGEpack by PQ Systems
System requirements: 486 computer or better. Minimum of 4 MB RAM, 16 MB recommended. Windows 3.11, 95 or NT. 12 MB free disk space.
Price: First license -- $695. Additional licenses -- $595
Contact: PQ Systems Inc.
10468 Miamisburg-Springboro Road
Miamisburg, OH 45342
Phone: (800) 777-3020
Fax: (937) 885-2252
About the author
Craig S. Howell is owner of CPM Labs, a full-service calibration and ultra-precision measuring lab located in Rancho Cordova, California. Howell can be reached by fax at (916) 635-4434.