<< part 1, Experimental Design: Work Smarter
Quick, Visual Analysis
The chart below contains both the results of each trial (bottom row) and the results of our evaluation in the two results columns, on the right. Look at the bottom row, Yield Percentage. You can see that trials five through eight brought better results than trials one through four. Why? Looking at the top row of the grid, Concentration, we can see that when Concentration is set to the High level, Yield Percentage improves.
Taking a deeper look, you will see that all other factors in experiment, under columns five through eight, are “balanced”. In other words, in columns one through four, as well as five through eight, all factors other than Concentration have two H settings and two L settings. It is this balance, which exists through all factors, that allow us to solve the grid.
Tying Things Up with a Simple Mathematical Analysis
All that’s required here is simple addition and division. The Avg L and Avg H columns on the right are simply averages of the trial results for each setting within the experiment. Look at the pH row, for example. Add the four columns together that have the pH factor is set to level L, and then divide by four, you get the Avg L results. (50 + 65 + 95 + 80)/4 = 72.5. Likewise, the average pH for the level H is the average yield percentage for columns 3, 4, 6 and 7.
Once the values for the two result columns have been calculated, its easy to spot the factors that contribute to a higher yield from the manufacturing process. In Conculsion: Concentration and Additive make a significant difference and the other factors do not.
Before You Get Started
I’ve done my best to keep this article simple and at the same time help to raise your level of thinking in regards to the possibilities of using one’s time and resources more effectively. Be forewarned that an experiment such, as efficient as it is, does consume a significant amount of time and resources. In addition, there are issues to consider that have not been raised in this article. For example, the number of times one must perform each experiment to get statistically significant results. Another issue that has not been discussed is the potential for interaction between various factors. Finally, its often decided, in the midst of an experiment, to redesign one or more elements of what is being tested and to start once again from scratch.
I strongly recommend that before undertaking a project such as this that you hire me directly to review and possibly consult on your project.