Multi-Factor AnalysisByRaymond D. Matkowsky
For those of you that have been reading my columns, you know that I am a big proponent of Analyses of Variance (AOV). Indeed, my employer at the time (47 years ago!) sent me to a special class on the subject once a week for six months.
An AOV Is A Powerful Tool Used Properly
An AOV is a very powerful tool that unfortunately is seldom used and often abused. It is not the holy grail of research. Like any other analysis, it is subject to interpetation. This is something that sellers of pre-packaged AOV programs do not tell their buyers. Interpetation experience is a big component any analysis. There are also many factors involved that never get measured. Some defy measurements. Many of these factors are human based. The effect of machineary is also one such factor. This is why I insist that clients run their own tests with the people ultimately involved with plant operations.
Increasing Factors = Increasing Error
The number of samples initially required depends on the number of factors analyzed. For example, a study utillizing three factors (one being a control) requires 24 samples. Forty eight to replicate the analysis. Add one more factor and this jumps to 39. Seventy eight with replication.
Theorectically, there is no limit to the factors analyzed. Clients are tempted to include everthing but the "kitchen sink." I must admit that many times I am tempted to add just one more factor to the mix. Wrong!
Look at your experiment as a pie. You can slice it an infinite number of times. But, sooner or later each slice becomes the same thickness as the knife making the cuts. Look at this knife as error. The more cuts you make, the closer you get to having each slice within the range of error.
What are you to do?
My recommendation is to settle on no more than three factors. Choose your factors carefully. Sometimes a factor that is insignificant in the formulations plays a big influence on the final outcome. After your pilot plant studies, pick your best combination and run with it in your plant for about six months. Caution is the watch word. Sometimes something that will run like a charm in the lab and pilot plant will bomb completely in the plant. A six month time frame should allow you enough time to visit all common scenarios. Of course, any necessary tweaks are to be made at this point. If after running with what you have you are satisfied with the results and you would like to study more factors, use your new operating procedure as the control and begin new tests. This way you should always be operating under improved conditions.
In summary, use no more than three factors and have your plant personnel run the final stages of the analysis so that the human factor is hopefully neutralized. You may have to spend a great amount of time upfront but overall you will spend less time while still operating safely.
Do you have any other suggestions, please share them with your fellow readers. Email me at rdm@datastats.com.
Copyright © 2015 Raymond D. Matkowsky
Raymond D. Matkowsky is the Chief Executive Officer of Data Stats, a consulting firm specializing in system or product improvement through mathematical and scientific modeling. He can be reached at rdm@datastats.com or through Data Stats’ web site at www.datastats.com |
