Author |
Message |
Ole Rasmussen (oler)
New member Username: oler
Post Number: 1 Registered: 2-2009
| Posted on Wednesday, February 18, 2009 - 10:57 am: | |
Mary Lou I can show you how to use indicator variables - if you still have questions after the various postings here then email me at [email protected] Thanks Ole |
Tony Davies (td)
Moderator Username: td
Post Number: 184 Registered: 1-2001
| Posted on Monday, February 09, 2009 - 2:41 pm: | |
Message for David Russell, Did you recieve my e-mail? If not please e-mail your address to: [email protected] Thanks, Tony |
Peter Tillmann (tillmann)
Member Username: tillmann
Post Number: 15 Registered: 11-2001
| Posted on Thursday, February 05, 2009 - 4:21 am: | |
Hello, as Pierre told >The difference between intercepts could be >interpreted as differences between the ref >methods (some other factors can interact). This is a pitfall using the indicator variables. If the bias between the two groups are caused by other reasons (i.e. "real difference") it is attributed to the indicator variables, i.e. methods. So to use the indicator variables properly the datasets have to be "nice". Nice with the meaning: equal distribution (at least the mean). Yours Peter |
David Russell (russell)
Senior Member Username: russell
Post Number: 42 Registered: 2-2001
| Posted on Wednesday, February 04, 2009 - 1:17 pm: | |
Dr Davies - Could you please post a link to the Unscrambler Octane Dataset? It apparently is no longer included in Camo's demo distribution. |
Paolo Berzaghi (pberzaghi)
New member Username: pberzaghi
Post Number: 1 Registered: 11-2008
| Posted on Wednesday, February 04, 2009 - 11:48 am: | |
Marylou, in your case you just need one indicator variable (the one you called ID1), then under OPTIONS and then COSTITUENTS you link your digestibility variable to the ID1. Then Select "All groups" as Pierre already indicated. you are ready to go!! Paolo |
Pierre Dardenne (dardenne)
Senior Member Username: dardenne
Post Number: 40 Registered: 3-2002
| Posted on Wednesday, February 04, 2009 - 11:46 am: | |
Marylou, Correct : 2 groups, 1 ID with 0 and 1. Bear in mind that you can correct for an "unknown" bias bit not for a slope. So if the invitro method has a just bias regarding invivo, Indicator Variables can help. If the bias is known, it easier to correct one of method by this bias. If there is a slope between the methods, Indicator variables will not work. the only way is to correct one set the best you can before calibrating using a subset measured by both methods. Pierre |
Mary Lou Swift (pacificagri)
New member Username: pacificagri
Post Number: 2 Registered: 2-2009
| Posted on Wednesday, February 04, 2009 - 9:57 am: | |
Thank you all very much for your replies. Just as an aside for Tony, it was is article that spurred my interest in using the indicator variables. To be a bit more specific, I am trying to calibrate for energy digestibility in barley using 501 samples. For 300 of these samples, energy digestibility was determined using animal studies. For the remaining 201, energy digestibility was determined using a enzymatic in-vitro test. Without going into detail, these values are "corrected" to remove the invitro bias and simulate animal digestibility. I wanted to use indicator variables to designate the two different methods. I have set up two additional variables called ID1 and ID2 but I have a feeling there should only be ID1 and that the animal study would be coded 0 and the in-vitro coded 1. Is this correct? Thanks again. Marylou |
Tony Davies (td)
Moderator Username: td
Post Number: 183 Registered: 1-2001
| Posted on Wednesday, February 04, 2009 - 9:11 am: | |
Hello Mary, You may well have all the information you need but if you would like to see a worked example from my column in Spectroscopy Europe you can download it from www.spectroscopyeurope.com/TD_11_2.pdf This is run in Unscrambler not in WinISI but it demonstrates the priciple. Hope you have fun! Best wishes, Tony |
Pierre Dardenne (dardenne)
Senior Member Username: dardenne
Post Number: 39 Registered: 3-2002
| Posted on Wednesday, February 04, 2009 - 2:34 am: | |
DPLS is fine if there are differences in the Xmatrix. Indicator variable is especially dedicated when there are suspected biases in Ref values. If there are differences in the X, then DPLS is probably better providing you have enough information to calibrate in each group. The advantage of Indicator variable is the use of the full X matrix. All Bcoefficients are the same except the intercepts. Pierre |
Jose Miguel Hernadez Hierro (jmhhierro)
New member Username: jmhhierro
Post Number: 5 Registered: 4-2008
| Posted on Wednesday, February 04, 2009 - 2:15 am: | |
Hi This method works in the way related by Pierre I have just used this method and it has provided the same results in the two groups. These results are not too different to the results provided in a general model. On the other hand, I have used DPLS model and I have developed individual models. These results are better than the others Do you think that this approach is better than the use of indicator variables? I think this is another way to solve your problem Thank you |
Pierre Dardenne (dardenne)
Senior Member Username: dardenne
Post Number: 38 Registered: 3-2002
| Posted on Wednesday, February 04, 2009 - 12:02 am: | |
Hi, Howard told us about history, which is fine, but it does not help too much. To use Indicator variables, new variables must be added in the matrix Y as many as groups minus 1 and coded 0 and 1 for 2 groups, 0,0 ; 0,1 and 1,0 for 3 groups, 0,0,0; 0,0,1; 0,1,0 and 0,1,1 for 4 groups, etc�. . The windows the select the variables is split in two parts. It is obvious to set up the variables. Click the option �All groups� in the output window and the EQA file will have as many models as groups with each having a specific intercept. The difference between intercepts could be interpreted as differences between the ref methods (some other factors can interact). Pierre |
Howard Mark (hlmark)
Senior Member Username: hlmark
Post Number: 218 Registered: 9-2001
| Posted on Tuesday, February 03, 2009 - 5:51 pm: | |
Don - WinISI is the software package developed by John Shenk and Mark Westerhaus, and sold under the umbrella of ISI, the company they formed. Originally it was developed for forage analysis, and from what I understand they made a good living selling it as third-party software to the forage industry and secondarily to the agricultural industry in general. Mark was (and probably still is) a crackerjack statisticn and developed several algorithms for specialized needs of those industries, that are in the WinISI package, including indicator variables. The software was built around FOSS instrumentation, and some of their special algorithms were made to use the FOSS capabilities. Eventually they merged with (or taken over by, I never knew the details) FOSS. When John got forced out of ISI, Mark was made president of the company, and, as far as I know, still sells the software. I think that Mary Lou isn't asking so much what indicator variables are, as to how to get WinISI to apply them. As a function of the software, you have to know what buttons to push, and since I never used that package, I can't help her, but I think her best bet is to contact Mark and ask him. \o/ /_\ |
Don Burns (burns)
New member Username: burns
Post Number: 4 Registered: 1-2006
| Posted on Tuesday, February 03, 2009 - 4:20 pm: | |
I don't know what WinISI is, but indicator variables are described in all three editions of the Handbook of Near-Infrared Analysis. If you can't get a copy, send me an e-mail at [email protected]. |
Mary Lou Swift (pacificagri)
New member Username: pacificagri
Post Number: 1 Registered: 2-2009
| Posted on Tuesday, February 03, 2009 - 3:46 pm: | |
Could someone provide direction on how to use indicator variables in WinISI. I am working with animal digestibility that has been determined with two different methods and would like to see if using indicator variables would improve the calibration. Thankyou |