|
5.
Process Improvement
5.5. Advanced topics 5.5.9. An EDA approach to experimental design 5.5.9.7. |Effects| plot
|
|||
| Engineering cutoff | Draw the horizontal line on the chart at that value which you as an engineer have declared beforehand as the engineering cutoff. Any effect larger than this cutoff will be considered as significant from an engineering point of view. | ||
| Specifying a cutoff value requires non-statistical thinking, but is frequently useful |
This approach requires preliminary, data-free thinking on the part
of the analyst as to how big (= what number?) an effect (any effect)
must be before the analyst would "care" as an engineer/scientist? In
other words, in the units of the response variable, how much would the
response variable have to change consistently before the analyst would
say "that's a big enough change for me from an engineering
point of view"? An engineering number, a cutoff value, is needed here.
This value is non-statistical; thie value must emanate
from the engineer's head.
If upon reflection the analyst does not have such a value in mind, this "engineering significance" approach would be set aside. From experience, it has been found that the engineering soul-searching that goes into evoking such a cutoff value is frequently useful and should be part of the decision process, independent of statistical considerations, of separating the effects into important/unimportant categories. |
||
| A rough engineering cutoff |
In the absence of a known engineering cutoff, a rough cutoff value is
commonly 5 % or 10 % of the average (or current)
production response for the system. Thus, if a chemical reaction
production process is yielding a reaction rate of about 70, then 5 % of 70 = 3. The engineer may declare any future effect that causes an average change of 3 or more units in the response (that is, any estimated effect whose magnitude exceeds 3) to be "engineering significant". In the context of the |effects| plot, the engineer would draw the line at a height of 3 on the plot, and all effects that are above the line are delared as significant and all below the line are declared not significant. |
||
