Hello Analytica Users,
In this issue we share some details about our greatest training resource, discuss Analytica’s options for choosing sampling methods and share a case study for Analytica Decision Engine. As always, feel free to share your questions and requests.
Paul Sanford, Newsletter Editor
Profiling Robert D. Brown III, Decision Strategist and Analytica Trainer
Those attending the training session in New York this month will get to know our most experienced training expert Rob Brown. Rob Brown has worked with Analytica since before it was Analytica (the early ancestor was called Demos). He first discovered the influence diagram modeling environment in 1996 while working for Scientific Atlanta. Rob found the platform both powerful and easy to use, and has been an enthusiastic advocate ever since. The modern version of Analytica contains many of Rob's improvement ideas. Teaching is in Rob's blood as he covered secondary level math and physics early in his career. Now, as president of the strategic planning consulting group Incite Decision Technologies, Rob continues to use Analytica on a daily basis. He develops decision support tools and performs quantitative analysis for strategic planning and capital allocation. Recognized as a talented teacher and Analytica expert, Robert agreed to host his first Analytica training session in Los Gatos in 2000. Since then he has hosted 3-4 classes per year. His enthusiasm has been rubbing off on trainees ever since. See our website for more details on Analytica Training.
TIPS & TRICKS
How to Dance the Latin Hypercube
Perhaps you haven't given much thought to how Analytica generates random samples for simulating uncertain quantities. If you explore Uncertainty Options... in the Result Menu and click More Options, you'll see three sampling methods: Latin hypercube sampling (LHS) is the default rather than Monte Carlo. David Vose recently posted in LinkedIn to argue that Monte Carlo is superior to LHS to explain why it's the only option in his Modelrisk software. Lumina CTO Lonnie Chrisman refutes most of these arguments point-by-point in his recent blog post.
In standard Monte Carlo sampling, it's as if you were to draw a probability density curve on the floor and toss sand over it. Each grain of sand under the curve represents a sample. You get some random clustering, and a slightly different result for each sand toss. Latin Hypercube Sampling reduces the error from n samples by spreading them more evenly across each uncertain distribution. It divides the distribution into n intervals of equal probability. Median Median Latin hypercube sampling (MLHS) just uses the median of each interval. Random Latin Hypercube Sampling selects a random value from each interval. If the model has only one uncertain variable, MLHS generates distributions with minimal random noise. If just a few uncertainties dominate the results (which is often the case), LHS (of either type) converges faster than Monte Carlo. If there are many uncertain inputs contributing to the results, LHS does about as well as Monte Carlo. That's why Analytica offers MLHS as the default.
Vose also criticizes LHS as being inherently slower than Monte Carlo for the same sample size. That might be true for standard LHS methods for inverting CDFs used in other simulation tools, but Analytica uses a sophisticated, little-known, and lightning-fast algorithm (invented by Peter Acklam) for inverting normal and related CDFs. After the sand settles, Analytica lets you choose whichever method you prefer. We think that's more friendly than forcing you to choose Monte Carlo. (To learn more, see Choosing a sampling method in the Analytica User Guide.)
Case Study Highlight: Marketing Evolution
Marketing Evolution helps leading consumer product companies spread their advertising efforts among TV, magazines, online, social media, and other channels. Combining their own research with sophisticated decision analytics, their tools find the mix across the full range of media to maximize effectiveness for a given budget. Marketing Evolution's web-based tools enable their clients to explore interactively the effectiveness of alternative strategies. These tools are powered by Analytica Decision Engine with Optimizer.
"Analytica is the backbone of our entire decision analytics platform. Its role in our success is immeasurable." says Graeme Pillemer, Senior Modeling Analyst, at Marketing Evolution. Lumina has supported Marketing Evolution since 2010 with both software and consulting services. Lumina is proud of their success and enjoy our continuing partnership. For more.
Come visit us at INFORMS
Come meet us in person at the Lumina exhibit at the INFORMS Annual Meeting in San Francisco, November 9-12. Get a demo of Analytica or Cubeplan. Let us know if you'd like an Exhibit pass or want to set up a meeting. Our CEO, Max Henrion, is also giving talks on 'Deciding how to decommission California's offshore oil rigs' and 'How to Recalibrate and Combine Overconfident Experts'. For more