doe analysis
Six Sigma – iSixSigma › Forums › Old Forums › General › doe analysis
 This topic has 3 replies, 3 voices, and was last updated 13 years, 4 months ago by Robert Butler.

AuthorPosts

January 13, 2009 at 5:14 pm #51667
Hello,
I have recently completed a doe where I have measured 10 outputs per run. I am trying to maximise my output.
What is my best way to analyse the results. Should I put the average response for each run (repeats) or should I copy the doe 10 times and put them in as replicates. and put in the individual values. I did not run replicates during the doe. Which is the most accurate and correct method of analysing the DOE.
Thanks in advance
0January 13, 2009 at 6:00 pm #179685
Robert ButlerParticipant@rbutler Include @rbutler in your post and this person will
be notified via email.As described and noted your measures are repeats not replicates. If you don’t have repeat measures capability in your software you will have to go back to the “old days” of the 1980’s, take the averages and run the analysis on those. Since you are running an analysis on averages you really want to plot the data (you should do this in any case) to make sure the averages are being unduly influenced by one or two data points.
If you try to pretend the measures are replicates (this will happen if you can’t tell the machine otherwise) and treat the observations as individual independent measures your estimate of residual error will, most likely, be very small and incorrect. As a result you will find a host of statistically significant differences that have nothing to do with reality and everything to do with analytical error.0January 14, 2009 at 12:38 am #179702
ObserverParticipant@Observer Include @Observer in your post and this person will
be notified via email.While you are at it, you might also calculate the standard deviation of your repeats and do an analysis on those as well. You might find that certain factors will influence the average in your desired direction but negatively influence variation or vice verse. Or, you might find that none of the factors influence the average yet will impact variation.
0January 14, 2009 at 2:17 pm #179716
Robert ButlerParticipant@rbutler Include @rbutler in your post and this person will
be notified via email.The standard deviation of replicates will have something to do with changing factors in an experimental design but not the standard deviation of repeats so any correlation between repeated measure variation and design variables will be fortuitous.
The way to think about this is the following:
Let’s assume two design factors A and B
Experiment Level A Level B Repeated Measures
1 Low Low a11 a12 a13…..
For experiment #1 and for any other experiment the level of A and of B is fixed – no variation – and yet there is variation in the repeats – that variabilty is most likely due to something other than A or B. True, A and B might not be at exactly the same values during the time it takes to gather repeated measures but it is very likely that the amount of variation in the two is far less than the amout of variation that would be exhibited if the values of A and B were completely reset (a replicate) before taking the second measurement.
As with genuine replicates, what you can do with the averages of the repeats is use the method of BoxMeyers and assess the impact of factor changes on the variation of those average values where the average of the repeated measures is treated as a single measurement for the experimental condition.0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.