Statistical Training Games
For comments, questions and suggestions about this page, contact Jesper Frickmann.
This page contains resources for the consultant who is training other professionals in the noble arts of statistics. Even though it is hard for us to understand, some people find the fascinating world of statistics boring and hard to learn ;-) Setting up a game to illustrate, e.g., the principles of Design of Experiments is a good way to motivate and engage the participants. If you want to suggest some good games for this page, please contact Jesper Frickmann.
The Helicopter Experiment
This seems to be a classic for Six Sigma Black Belt training, where the participants play with paper helicopters.
From Matt Rowson.
Illustrates some basic experimental design principles, correlation, statistical process.
We normally run a chocolate cake panel, since almost nobody has a problem with tasting chocolate cake but in theory you can use any foodstuff for which you can buy a variety of quite similar-looking products of different brand / price / quality. Ideally you need six or seven varieties, hence lots of walking around supermarkets beforehand...
The participants / students each taste a sample of each cake, and score it "out of 10" for preference. It is obviously important that the participants don't know what brand they are eating, so each sample should be labelled "A64", "B38" etc rather than the product name (letter-number codes are better than single letters, which can bias panel along the lines of "A must be better than B") - and of course normally a mouthful of each is enough. As we do quite a lot of food tasting panels here, we can add quite a lot of relevant detail... for example:
- each participant tastes the products in a different order. We then explain about "Order effects" (if products are always tasted in same order you are subjecting results to bias... eg first product tasted tends to get high score, maybe people differentiate less towards end as they get bored etc)
- water and / or cucumber provided to clean palate between samples (to help avert "carry over" effects)
- details of similar studies we run on a larger scale (eg 300 Spanish housewives tasting 20 Spinach samples etc.
You have to be careful with allergies etc (like nuts) but if anyone can't or doesn't want to participate they can assess the cake for "appearance" only.
When all panelists have assessed all cakes you enter their preference scores into a spreadsheet or statistical package (we use SAS/JMP) and attempt to correlate their preference scores individually with the price (per 10g) of the cakes they were tasting, thus establishing which panelists "have expensive tastes". Scatterplots of results are also useful here. If you have time, you can obviously combine the entire panel's results and perform (e.g.) an ANOVA to compare cake-to-cake, but then you have to either gloss over or expand to checking assumptions etc so we normally don't go there.
Washing Machine Experimental Design
From Matt Rowson.
Illustrates aliasing, blocking, design considerations.
Describe a scenario whereby you have a limited amount of time in which to compare the efficacy of four detergents. You have four washing machines available and have enough time to run four loads in each. You wish to compare the detergents over a range of temperatures (say 30, 50, 70, 90 degrees) and using four different fabrics.
You can introduce the idea of designing to avoid aliasing by suggesting that each machine runs one load at each of the four temperatures (explaining the diagnostic problems that might result if all 30 degree loads were in the same machine, for example).
Over the years we have built up two versions of the "kit" needed to run this exercise. In both cases we use disposable dishcloths (four each of four different colours) as "fabrics" and empty detergent bottles (four each of four different designs) as "detergents". The "high-tech" version then involves sixteen waste-paper bins in a four-by-four grid to contain each "trial" with cardboard boxes dressed up as washing machines at the end of each row and temperature labels at the end of each column (the low-tech equivalent is run on a table top with sheets of paper instead of waste paper bins and simple labels instead of cardboard-box washing machines).
The objective of the exercise for the participants is to arrange both detergents and cloths so that each detergent is tested once in each washing machine, once at each temperature and once with each cloth. Surprisingly, children can do this almost instantly but adults often take much longer. This exercise obviously gives you a platform to talk about other considerations (e.g. interactions - not catered for by this design, and statistical power) if appropriate.
The solution is a Greco-Latin Square, of course, but prior knowledge of this isn't really a lot of help. Our solution sheet (one of a number of possible solutions) is attached.