A More User-Friendly Design Expert
by Stat-Ease inc.
200 MHz Pentium or better; Windows 95 with 16 MB RAM, 98 or NT 4.0 with 32 MB, or 2000 with 64 MB RAM; and 10 MB hard disk space.
Contact: Stat-Ease Inc.
2021 E. Hennepin Ave., Suite 191
Minneapolis, MN 55413
Telephone: (612) 378-9449
Fax: (612) 378-2152
Design-Expert Release 6.0.1
by Felix Grant
hese are exciting days for designed experiments in main-stream quality practice. Evaluating a standard statistical-
analysis package today means--as a matter of
routine--looking at the level of specific support. There is a growing population of small-scale tools out there for particular aspects of design.
Stat-Ease produces two levels of product in this area: Design-Expert (known as DX), with release 6.0.1 reviewed here, and
Design-Ease (or DE), a simpler version with a similar interface currently in release 5. The older release 5 of DX is also still actively supported and widely used; because
several of my clients and colleagues rely on it yet in the firing line, I put DX on a laptop and took it around with me for a couple of weeks to make live comparisons. I
also set it to work installing some real product-output experiments for a live consulting task.
The first (and, for many, most important) difference that struck everyone who tried it but which also took a while to pin down fully is usability improvements. This is
crucially important in an area of work where most users are still uncomfortable. It partly results from the move to full 32-bit architecture and a fuller compliance with
current Windows visual conventions (such as the standard toolbar), but there's far more to it than that.
The built-in help function is much more detailed and pervasive in version 5. Input and model decisions are well-supported, results are well-explained, and contextual
assistance has moved onto a new plane. The formal help system is structured to serve as a tutorial if required. A lot of thought has gone into this, evoking a relaxed "just folks" style.
All of release 5's rough edges have been removed. Presentation of the program's facilities, through a browser-style model, is now much more intuitive. The structure
of the whole project under study runs in tree form down a narrow left-hand frame. The detail of a selected facet is expanded into the larger right-hand panel resembling a
dashboard with appropriate buttons and displays; this expansion occasionally spills over, when required, into child windows for extra detail without clutter. The whole
process is delightfully well-managed.
The package is also now more amenable to tailored requirements across a range of
aspects. This is a difficult area for any program designer: What are the limits of balance in trading off user choice against validity constraints? DX has steered a
sensible path; choices are available within constrained limits which reflect sound practice. The significance threshold, for example, which defaults to 0.05, recognizes
only one discrete step either way (to 0.1 or 0.01) while the numeric optimization simplex fraction is allowed an effectively continuous variation between 0.01 and 0.2
around the 0.1 default. In-line usage is supple, too, allowing (for instance) the temporary elimination of data or model terms without requiring their deletion.
Exploratory changes can be made to model, factors and terms, then either confirmed or reversed at need.
The actual working parts of the package also offer numerous developments. Orthogonal Taguchi arrays go up through 19 choices to L64, and two-level blocked
designs allow up to 256 runs. Response surface designs accommodate nine and 10 factors, most mixture designs 24 factors. Numerical optimization--now far more
transparent, courtesy of the new environment layouts--includes categorical factors and the facility for setting constant levels. A more holistic and all-pervasive approach
to graphic methods supports the computational core and yields dividends in productivity. There is a good power calculation facility, an F-test on items and coefficients have confidence limits.
Out in the real world, all of my long-suffering guinea pigs were enthusiastic about DX; the development since their own release 5 installations were welcomed
unreservedly. Test runs made using existing, tried-and-tested problems revealed no weaknesses and consistently produced results as good as or better than those in use
and with less effort. On two new, large, complex problems, the package produced good outcomes from scratch quickly and easily. The outcomes proved unbreakable
under cross-testing. The propagation-of-error tools, understandably advertised as a Six Sigma plus, proved their worth in a different role--taming the impact of
unavoidable and uncontrollable environment variations.
If you've already invested in DX version 5, with a system and body of expertise built
around it, don't hesitate: At $395 the upgrade is all gain and no pain. If you're looking for a change or setting up a new regime, you know what you're after and will be
weighing several options in your own circumstances, but this should definitely be one of them. Download an evaluation copy, explore all of the aspects I haven't space to
mention here, and put it through its paces.
Felix Grant is a lecturer and consultant in the United Kingdom. E-mail him at email@example.com .
Global On-Demand SPC
by Statware Inc.
Server--Windows NT or 2000, 256 MB RAM. Browser--Netscape 4.08 or greater, Internet Explorer 4.03 or greater.
Base price--$35,000. Includes five dynamic users, unlimited viewers, three days of onsite training and set-up and one year of support.
257 S.W. Madison Ave., Corvallis, OR 97333
Telephone: (800) 478-2892
Fax: (541) 758-4666
by Dirk Dusharme
ou can never have too much information. The problem is that sharing up-to-date process information across multiple sites and various computing platforms can be a major
undertaking. Fortunately, Web browsers have broken this barrier, allowing users to access information from almost any type of computer.
Statware has taken advantage of the Web's capabilities with its Statit e-QC, an interactive, Web-based SPC tool that allows users to access real-time quality reports using a browser. Statit
e-QC allows users to specify what data they want to look at and in what form. Data analysis is performed on the server and appears in the browser's window. The
resultant process analysis is based on the most current data available.
Statit e-QC is a server-based application that pulls data from any local or remote
ODBC-compliant source and utilizes Statware's extensive library of statistical analysis tools to generate reports on demand.
Authoring a particular application requires either a copy of Statit, Statware's statistical analysis package, or at least knowledge of Statit's scripting commands. The author
uses Statit to develop the required analyses that will be made available to users. This involves setting up the analysis and determining which variables users can analyze.
The author cuts and pastes the resultant scripting code from Statit into Statit e-QC using a browser. In some cases, generating the macro scripts will involve adding some extra SQL commands.
Given that Statit and Statit e-QC are both Statware products, our only complaint is that there could be a little tighter integration between the two, eliminating the need to
cut and paste or hand script SQL.
Within the user's browser, macros, pregenerated reports and static Web pages appear
in a Windows Explorer-like navigation frame. Users click on the desired macro, then use a drop-down menu to select the variables to be analyzed. With one click, Statit
e-QC pulls current data from the appropriate source, does the analysis on the server, generates the charts and tables, and sends them to the user's Web browser.
Data drill-down allows users to click on graph items (e.g., points on an X-bar chart or bars on a Pareto diagram) to see actual process data or drill down through layers of
data. The user determines the amount of drill down and what data can be displayed. And here's a surprising feature: The user can e-mail a report complete with the same drill-down capabilities.
Users may also annotate displayed data: tag points as assignable cause or process change, or add descriptive text. If necessary, e-QC recalculates the limits.
Statit e-QC can be used to tie process diagrams, CAD drawings, documentation and so forth to statistical reports. For instance, Statit e-QC could display a CAD drawing
containing hot spots that, when clicked, cause it to collect and display the data associated with those points.
Statit e-QC is aimed at major manufacturers, divisions and smaller companies that deal with same, and the price reflects it. At $35,000, this is probably not the tool for
the mom-and-pop shop. However, companies with multiple sites and/or divisions that need to share up-to-date process data easily and effectively regardless of computing
platform should take a look at Statit e-QC.