Author |
Message |
Klaas Faber (faber)
New member Username: faber
Post Number: 5 Registered: 9-2003
| Posted on Monday, September 03, 2007 - 6:55 am: | |
Hi! When developing a permutation test for a complex problem like component selection, one has to make a choice for test statistic. Fit of y would have been another choice (which I investigated as well; it gives similar results). The idea behind the choice we finally ended up with, is that it leads to a t-test for PCR (normalized scores) under standard regression assumptions (no noise in x). This permutation test therefore generalizes that particular t-test to (1) more general circumstances and (2) other component-based calibration methods. Klaas |
Christian Mora (cmora)
Junior Member Username: cmora
Post Number: 10 Registered: 2-2007
| Posted on Tuesday, July 10, 2007 - 1:37 pm: | |
OK, I found it: it was the inner product of the scores and the Y vector. Thanks anyways |
Ian Michael (admin)
Board Administrator Username: admin
Post Number: 11 Registered: 1-2006
| Posted on Tuesday, July 10, 2007 - 10:28 am: | |
In case anyone wants to read the article, it can be downloaded free-of-charge from http://www.spectroscopyeurope.com/td_col.html#18_5. |
Christian Mora (cmora)
Junior Member Username: cmora
Post Number: 9 Registered: 2-2007
| Posted on Tuesday, July 10, 2007 - 8:37 am: | |
I'm sorry: Spectroscopy Europe October/November 2006. Thanks |
Howard Mark (hlmark)
Senior Member Username: hlmark
Post Number: 113 Registered: 9-2001
| Posted on Tuesday, July 10, 2007 - 8:17 am: | |
Christian - in which periodical? JNIRS is only up to Vol. 15, and NIR News is only up to issue #4 of Vol. 18. Unless it's some other journal, it hasn't been published yet!! \o/ /_\ |
Christian Mora (cmora)
Junior Member Username: cmora
Post Number: 8 Registered: 2-2007
| Posted on Monday, July 09, 2007 - 9:29 pm: | |
Dear List Members; In Tony Davies's article Vol 18 No 5 2006, Faber and Rajko discussed the permutation test as an approach to evaluate the significance of the components of a PLS model. They say:"For each of these random models, a test statistic is calculated" Does anyone know what is the actual "test statistic" they mention? Thanks CMora |