Hello,
I've a problem and desperately need help. Once I already posted a question here, but now I need to repeat what I am doing. I have 74 fractured specimens with fracture area of 3 to 5 mm^2 each. There are 7 fracture modes (1 to 5 per specimen). I measured area of each fracture mode on each specimen, and now I need to test my hypotheses that force, needed to break specimen, could be described as F = C + K1*S1 + K2*S2 + . + K7*S7 Where S1-S7 - areas of different fracture types (predictors). Results I got from SPSS and like them very much. But our statistical guru told me that I can throw away my results because predictors do not have normal distribution. That was the first time I posted question here and - thanks a lot - got an answer that for multiple regression normality of predictors is not required. Now our statistical guru is telling me that I can throw my results away because variables are dependant: "If one of the fracture types goes all the way up, others go all the way down". I told him that I checked on multicollinearity by computing R^2 for each of variables described by all other variables as described in http://www.graphpad.com/instatman/Ismulticollinearityaproblem_.htm He told me that dependency and multicollinearity are different things. I asked how to check variables on dependency, he told me there are no mathematical methods to do this. My main question: should I throw my results away? If I should not, then I have an additional question: SPSS has an option for computing multiple regression without coefficient C. It makes a lot of sense for my model: I need force equal to zero to "break" specimen with zero thickness. When I turn this option ON R increases. Why it happens? Thank you very much, Vladimir . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
