### You must have statistical smarts to get the most from Analytica

Given a good palette and brush, a skilled artist paints a masterpiece. The same tools
can produce a mess in the hands of an amateur.

That pretty much sums up Analytica, a visual modeling tool whose mathematical gloss can
be dimmed by the user's ignorance of statistical theory.

The program relies on flowcharts to map every factor that goes into making a decision.
For example, an analysis on the value of enforcing a seat belt policy factors in usage
rate, policy cost and value of a life, all appearing as colored ovals and boxes linked by
arrows to the result: a purple trapezoid labeled Value of Policy.

Double-clicking on any of these icons retrieves underlying details--the variables and
formulas that churn out numerical values. When you click the Calculate button, the program
produces an x-y plot that represents each decision factor by a line.

Statisticians will be happy to know that Analytica lets you enter either a specific set
of numbers or one of a set of provided probability distributions.

You can display results in many ways, from single-number statistics to more informative
probability distributions.

This helps an analyst understand and explain variations in results and respond to the
contingencies that go along with any real-world decision.

Variables can be defined by arrays with up to 15 dimensions, making it easy to
incorporate complex interactions into simpler mathematical formulas. The results are
pleasing to look at, but the interface makes you read 20 pages of the manual just to draw
a box.

Analytica could be dangerous for a budding Rembrandt. Its friendly impression masks its
statistically precise machinations--a recipe for potential disaster.

After you gaze open-jawed at its stunning, equation-filled flowcharts, you start to
feel unduly confident that your assumptions make sense--that is, if you can comprehend the
graphed results.

Even most educated people don't know the difference between a parametric analysis and a
triangular distribution, particularly the managerial types who are responsible for
deciding whether, say, a new system of one-way streets would improve city traffic.

I believe the world will be safer if the average decision-maker avoids this program.
Now that I've excluded much of the target audience, does Analytica have the analytical
brains to match its beauty?

To find out, I enlisted the help of a defense analyst who holds a doctorate in
probability and statistics. After playing around with it, he called it a "powerful
graphical analysis package that lets the analyst specify each independent variable's
value."

He said he was happy to see that resulting data could easily be exchanged in text
format with other applications for further crunching. He found the interface fairly simple
to master.

Other pluses: a nice array of prebuilt, deterministic models such as the Markov chain;
an excellent tutorial; and probabilistic variables for modeling uncertainty.

But two omissions, both of them mathematical, turned him sour on Analytica. He deplored
the lack of exponential distribution and time-series analysis.

"If Analytica bills itself as a complete analysis package rather than another set
of math tools, such models should be included," he said.

Dan Pacheco is a Washington computer reviewer and journalist.