Search All of the Math Forum:
Views expressed in these public forums are not endorsed by
NCTM or The Math Forum.



Re: Prediction interval in regresion through origin
Posted:
May 16, 2012 11:17 AM


Just three aditions.
Concerning the main question (PI in regression through origin):
1. Please, skip the comments on (in)appropriateness of regression through origin. (I know all that and I could write a lengthy essay on that myself  please, please, just the maths).
2. I may have written wrongly that CI in regression through origin is triangle shaped. Now I think it may be also "slightly curved", i.e., the two lines (which, of course, start from the origin like the regression line) going ever further apart being only seemingly straight. Please enlighten me.
Concerning the additional question:
3. I know that imposing inequality constrainst is "easier" (in terms of software implementation) in nonlinear regression, but that's not an option ("overkill", "impractical", irrelevant). And I actually know my "naive" suggestion (of keeping OLS if you get a>0 and doing regression through origin if you get a<0) is not bad  Prof. Ripley (in my somewhat qualified view one of the smartest people on the planet) say so (see first URL below)  what I'm asking/hoping for is radical simplification of the wayabovemylevel literature (see second URL below) in the case of having just one predictor and constraining the intercept rather than a coefficient.
http://r.789695.n4.nabble.com/fitlinearregressionwithmultiplepredictorandconstrainedintercepttd841165.html http://www.gurulib.com/_user_manual_file/pic_1247578519497.pdf
Thanks in advance again, Regards, Gaj



