nonnormal measurement error in linear regression
Hi,
Complicated stats question, but maybe someone out there knows how to proceed. I am trying to perform regression on two variables, the samples of which have significant, but known error components. Ordinary least squares regression cannot be used as it is assumed that measurements are made without error. As I understand it, the normal way to proceed would be to assume a maximum likelihood functional relationship (MLFR) and use some of the widely available iterative algorithms. However, in order to ensure even sample distrubution (i.e. not skewed) and homoscedasticity I performed logarithmic transforms on both variables. As a consequence the sample errors are lognormal. standard MLFR techniques assume normal error distributions. Is there any way of dealing with this problem. Specifically, is anyone aware freeware computer programs that would allow one to estimate parameter values and confidence intervals. 
hi,
i am also trying to perform non normal error dist in linear regression.. may i know what is your general equation for the error term? 
As far as I am aware, one way to correct for measurement error is to look for instruments that are correlated with the original independent variable(s) but do not have the measurement problem. See, e.g. Greene, 2nd Ed. Sec. 9.5.3.

All times are GMT 5. The time now is 01:24 PM. 
Powered by vBulletin Copyright ©2000  2014, Jelsoft Enterprises Ltd.
© 2014 Physics Forums