- View SourceHi there,Still fighting with that "save" problem. Wonder if any of you know the "trick" to this. If I open something then decide that I didn't want it and try to close it, JMP forces me to do a "save as". It doesn't give me an option just to close it without saving or allow me to "save" material in the "normal" manner as in updating what I already have done. In the cases where I am working on something and just want to save the work I can only do a "save as" instead of having JMP update what I've done. The only "work around I can figure out is to have a general "trash this" file to heard all the junk I don't want to save and another "figure out how to do this" file to save the work I would have rather updated to an already existing file. The problem with this approach is that I have a gazillion files that I have to figure out how to name in a manner that I can actually find the most updated version of my work.So what's the secret?I've also stopped using Journal because it saves things in a redundant manner. If I have something open that I want to save to the journal it saves that item and EVERYTHING associated with it, even though I already have all those other things saved on the Journal. So I end up with a great big journal with about 5 useful items and then the same "supporting" files repeated for each and every one of those 5 items. Gobbs of redundancy. So I no longer use Journal.
*CJ* - View SourceYes that is very helpful. I knew that the VIF showed collinearity, but the information about "seeing" it is very useful information that I'll add to my growing "
*CJ's Dummies guide to DOE*".I just couldn't figure out what I was doing that was creating the collinearity. Found out though,...and of course, with every discovery there arises another question.Seems I WAS trying to make the Augment feature do something it wasn't designed to do. I now know that Augment can be utilized to add runs*within*the factor ranges that I set for the original design. i.e., Factor A at a range of 2 to 4, and factor B at a range of 20 to 40. However, it wasn't designed to extend the range of factors in the manner I was trying to do. I was trying to change the range of factor A from 2 to*8*instead of using A at it's original range of 2 to 4. So, JMP was looking at ALL the information for factor B as a whole, (original 15 runs and added 8 runs thru augmentation) but it was only using the range designated for the added 8 runs for factor A, ... because that is what I was unknowingly telling it to do, look at 22 runs for factor B and only 8 for factor A.So, I went in and changed the coding for factor A to include the ranges for all the combined runs and that took care of the problem. I don't know if you're*suppose*to do this, but it worked.So here is the question that this spurs; If I run an experiment and the resulting data tell s me that I set the range of one factor too narrow, what is the correct way to*fix*it? Consider that you don't have the resources to start the whole thing over, just to add a few runs to*complete*the picture.Possible solutions that come to mind, but I have no clue if they'd work properly are;Sit down at the computer, make a new design which encompassed the correct ranges, then go into the data table and change the first 15 numbers that JMP generates for each factor. Input the factor and range data obtained by the completed DOE and complete the experiments for the remaining runs. The problems that come to mind, you would have to be careful which JMP designated runs you replaced with the actual runs already completed, and would changing the values of the factors in the data table screw up the math and give less accurate data?Or design a matching DOE with the requirements of the new factor range, join the data to the original DOE and proceed. Problem with this is that it seems as if I'd run into the same problem in analysis that changing the coding to extend the range might present.Input?.....Thanks :-)*CJ**(847)-808-3525*

**From:**GLJUG@yahoogroups.com [mailto:GLJUG@yahoogroups.com]**On Behalf Of**Mark A Anawis**Sent:**Thursday, February 19, 2009 8:44 AM**To:**GLJUG@yahoogroups.com**Cc:**GLJUG@yahoogroups.com**Subject:**Re: [GLJUG] VIFs

Hi CJ,

High VIF is a sign that you have collinearity of factors. That is, 2 of your factors are related to one another such as if you had length and surface area as x variables. You can often see collinearity in the leverage plots which will appear as a scrunching up of points in the x axis. Other methods are: examine correlations and associations between variables, regression coefficients change wildly when variables are included or excluded, standard errors of the regression coefficient are large, predictor variables with strong relationships to response don't show significance. Your remedy is to remove one of the variables.

As to blocking, you do want to check the blocking variable since if your augmented data set shows a difference seen as a statistical significance between block 1 (original DOE) and block 2 (augmented data set), then you are not controlling some variable which is not part of your design.

Hope this helps,

Mark**Mark A Anawis, MA, CSSBB**

Senior Scientist

ADD

On Market Quality EngineeringAbbott

100 Abbott Park Road

Bldg. AP8B-3/Dept. 04Z7

Abbott Park, IL 60064-3500

U.S.A.office (847)-937-4347

fax (847)-938-2219__Mark.Anawis@ abbott.com__The information contained in this communication is the property of Abbott Laboratories, is confidential, may constitute inside information, and is intended only for the use of the addressee. Unauthorized use, disclosure or copying of this communication (or any part thereof) is strictly prohibited and may be unlawful. If you have received this communication in error, please notify Abbott Laboratories immediately by replying to this e-mail or by contacting __postmaster@abbott. com__, and destroy this communication (or any copies thereof) including all attachments.**"Nash, CJ (C)" <cnash2@...>**

Sent by: GLJUG@yahoogroups. com02/18/2009 08:26 AM

Please respond to

GLJUG@yahoogroups. comTo<GLJUG@yahoogroups. com> ccSubject[GLJUG] VIFs

Question on the VIF's, I'm doing an Augmentation. I did the RSM (three factors, A, B, and C) and the resulting data indicated that we may be able to use Factor B at a lower volume than expected when paired with the Factor A.. Both products were factors in the original RSM. So, I'm doing an Augmentation. Two of the factors (Factors A and C) are remaining unchanged and the values for B need to be lowered. My goal is to run new test runs and then take another look at the RSM which will contain all three original factors with more runs of course, but with factor B at a wider spread than the original RSM.Problem, when I try to develop the Augmentation model I get good numbers in my G Efficiency, above 50; and below 1.0 in the Fraction of Design plot, but my VIF's are throwing me. No matter how many runs I add I get consistently bad VIF numbers for the effects of factor B (100, 440) and B*B. Those two are 6.9 and 6.8 respectively in the model I am looking at right now, but have been anywhere form 7.0 and higher (sometimes REALLY higher) on others I've run. All the other VIF's are consistently lower than 2. I have to be doing something wrong because I ran a model with 100 runs and it gave me a G efficiency of 45, the Fraction of Design Space Plot gave me a sigmoidal curve and the two BIT terms were 3.6 and 3.6. 100 is WAY more than I need I just wanted to see if there WAS a number of runs that would give me a good model. JMP is trying to tell me something's wrong, but I've been unable to figure it out thus far.

Another question, when I do the fit model for these Augmented designs the FIT Model box that comes up where you input your factors and responses comes up with "Block" in the model effects. Block isn't a factor or a response, it shows up after I hit "

*Group new runs into separate block*". I've run the models with this in it and with it removed because I'm not sure if it should be there. I don't think so because when I leave it in and look at the VIF numbers "Block" gets a VIF consistently above 12, B*B are above 6.00 and B(100, 450) are in line with the other effects, below 2.Thanks for any suggestions you may be able to offer. Remember though Mathematically, you guys have more functional brain cells tuned to this stuff than I ,so Stats for Dummies please. lol :-)

LOADS of thanks!

*CJ*(847)-808-3525