‘To boldly go’: Exploring opportunity spaces

By Dr. Ian Cox, Marketing Manager, JMP

Imagine you are in a rowing boat, and have been given the task to survey a lake by taking depth-soundings using a rope marked every meter and with a heavy weight on the end.

“Rather than using a design generated by hand, the use of computer-generated designs will allow you to ‘fit the design to the problem’ rather than run the risk of ‘fitting the problem to the design’ and falling into the traps mentioned”
-Dr. Ian Cox, Marketing Manager, JMP

You row to a spot, throw the weight over the side and then count the number of marks as you pull the weight back up to find out how deep the lake is at this point. But this takes time and energy, and so, given the task at hand, two questions soon arise: How many depth-soundings can you take? Where should you take them? You have some prior information (the lake has a depth of zero all round its edge), but answering the two questions needs a bit more thought because it will depend on what you want to use the final survey for. For example, if you are interested in fishing, you might want to locate deep parts of the lake which you can cast to, so knowing the depth at the centre, which you can't reach from the shore, might not be of much use.

In this example we have an 'opportunity space' that we need to probe to get useful information: The places we decide to row to constitute a 'design', and the opportunity space is two-dimensional because we need a longitude and latitude to specify each point in this design. When we measure the lake depth at a design point we get a 'response'. For those that know the topic already, this is, of course, a gentle introduction to what is called experimental design, often shortened to DOE or DOX: DOE can provide an efficient way to probe an opportunity space to gather useful new knowledge, and this article is about some relatively recent advances in design that have great practical and business value.

Although the level of actual usage varies from sector to sector, or even within the same company, there can be no doubt of the practical value of DOE if applied correctly: Whenever any product is in development, testing or manufacturing, DOE can find valuable application to assure that customers get what they want, and don't get what they don't want (and at the lowest cost to you). Journals such as the Journal of Food Science and the International Journal of Food Science and Technology are full of DOE examples, and initiatives such as Six Sigma generally promote the use of DOE. Most tellingly, market leading companies such as Proctor and Gamble, Unilever and Kraft Foods use DOE at all stages of the product lifecycle and for a variety of purposes. Indeed, the idea of exploring opportunity spaces systematically through DOE can even be used at the product design stage, before physical prototypes exist to experiment with (through, for example, so called 'computer experiments' and 'discrete choice models').

But there are some problems. For example, you may have noticed that not all lakes are square. Like many areas of statistics, DOE originally was developed with the requirement that designs be made 'by hand'. Almost without exception, this heritage is still obvious in the software that DOE practitioners often use today: These so-called 'classical designs' have been compiled into libraries that the user then selects from to set up how their own experiment will be run. Although well tested and proven, these designs assume that the opportunity space we need to probe is always square (or some generalisation of this in many dimensions). A more serious problem is that classical designs can be inflexible in relation to the new knowledge you are after, simply because they are finite in number so can't fit every case exactly. Generally these problems lead to compromises such as probing a smaller region of the opportunity space than you want to, and either having too many or too few deign points, sometimes in the wrong places. Having too many points than you need has a clear business impact in terms of increased costs, cycle times and resources. In situations other than computer experiments there will be variation in the response when measured repeatedly at the same design point, and the impact of having a design with too few points will be to reduce the chance of detecting something that has practical importance. Finally, and as the lake example makes clear, the positions of the design points really should be tied to your objectives and the usefulness of the new knowledge you hope to gain.

Fortunately, these problems can be avoided, and in such a way that you or your technologists, scientists and technicians can easily get to grips with. Rather than using a design generated by hand, the use of computer-generated designs will allow you to 'fit the design to the problem' rather than run the risk of 'fitting the problem to the design' and falling into the traps mentioned above. The 'Custom Designer' in JMP produces a design that is always customised to your specific situation, and allows you to tackle many more design challenges from within a single environment. As one example, the Custom Designer can easily produce a design involving both 'process' variables and 'mixture' variables, a situation that is commonplace in the food industry but is not possible with a classical design. Furthermore, by providing one design environment, and an interactive, graphical means to understand the results of the analysis that follows, JMP can break down barriers to DOE usage within your company, allowing you to more quickly, easily and uniformly realise all the benefits that DOE is rightly famous for.

So, if you routinely have complex opportunity spaces you need to explore to make your customers happy, please take a serious look at JMP. It is the only commercially available software that tackles the design problem in a cohesive and powerful way, allowing you to 'boldly go' and explore, actively seeking out new, useful knowledge to drive innovation in your products and the way you make them.

Figure 1

Central Park, Row Boat Lake. Albert Kerr: www.oceansidegallery.com/HerrAuction.htm