The most neglected phase of nearly all process studies is sampling design. Quality decisions are only as good as their initial data, yet sampling frequently gets left to convenience or worse. Many failures of quality assurance can be traced back to initial or production-line selection; however, nQuery Advisor, written by UCLA statistician Janet Elashoff for Statistical Solutions of Boston, is a software package designed to eliminate this weak link. It manages to do this effectively and efficiently, without unnecessary frills.

The program's initial impression on startup is deceptively Spartan: on the opening screen, only four menu options and two tool-bar icons greet the user. A box advises selecting "File, New" for structured help; doing so opens the door to an impressively competent system.

The detailed manual could moonlight as a college study text, and the help system, progressively layered, assists different users in appropriate ways. Notes panels below the working area encourage documentation of each stage, and these annotations are available for the final report.

The default new file screen can calculate minimum sample size tables for means or proportions. The user is asked for goal, number of groups and method of analysis; short option menus are offered for each, as is guidance (if required), by means of context-sensitive explanatory prompt cards. Alternatively, users can enter sample sizes to see what the implications will be for other factors. A range of values in  spreadsheet form answers "what if" queries about different sampling strategies.

The program also offers a distribution function menu. With this, the user is guided through a required design's formulation, structuring and outcome definition via a possibilities-tree structure, using results to refine subsequent choices and allowing only suitable ranges of response. A subsidiary options menu, accessed from the sample size table, allows variation of appropriate factors within sensible limits. Results of parameter variations are tabulated and, from the plot menu, graphically illustrated. The Assistants menu provides various aids, as needed.

The software's range of choices may seem restricted, but this is a strength. Its step-by-step approach, allowing only solidly based choices, eliminates both uncertainty and questionable research bases, and leads the design to well-chosen analysis types. All results, tables, plots and associated notes can be transferred to the clipboard, encouraging the production of reports that thoroughly document their genesis. The rationale for decisions made therefore will always be accessible to other professionals at any future time.

The current version, 2.0, includes a range of valuable additions to what was already a valuable tool in its first release. The tables for one-sample tests, and confidence intervals for sampling from finite populations, are of direct relevance to process studies. Estimation of standard deviation in cluster sampling allows client managers to exert closer cost control over high throughput monitoring. Tabulated power computations for survival, accrual and dropout shave uncertainty margins from process and production decisions.

For review, the program was used to design a study of its own effectiveness. Two groups of small-scale commercial investigations--sizes defined by nQuery   itself--were identified; one group was given access to nQuery Advisor, while the control group was left to choose its own methods. Subsequent analysis showed significantly better design, with identifiable knock-on benefits in both cost and quality outcomes, amongst those studies designed using the software.

The required tools are, of course, present in many high-end general statistics packages; but how many quality assurance clients have the necessary background to make good use of them? Weak initial design devalues subsequent quality programs, but software such as this, simple and easy to use, will more than pay for itself on the first run. If you are a consultant, buy a copy for yourself--then encourage your clients to buy one, too.

nQuery Advisor Release 2.0
by Statistical Solutions

System requirements: nQuery Advisor requires a 386 or higher processor, 3.5 inch floppy drive (for installation), 3 MB of free disk space, 8 MB of RAM, VGA graphics and a mouse. Software runs under DOS 3.1 or higher and Windows 3.1, 95 or NT. The Windows version must have small fonts installed.

Price: $675 for commercial single user
$475 for academic single user
Multi-user discounts available

Contact: Statistical Solutions
60 State St., Suite 700
Boston, MA  02109
Telephone: (800) 262-1171
Fax: (617) 854-7473
Web: www.statsolusa.com

About the author

Felix Grant is a freelance journalist, lecturer and research consultant in the United Kingdom.