Thanks Don! 1. Using natural language, it should have been Z is equal to the difference in the parameters divided by the square root of the sum of the squared standard errors.
2. This formula is useful for comparing parameters across regressions for different groups in OLS. I missed the fact that the original poster was referring to a nonlinear model, in which case this formula may not be appropriate. Brett --- Donald Burrill <[EMAIL PROTECTED]> wrote: > The formula below appeared in my mailer as > Z = ( b1 ^ b2 ) / ( SEb1^2 + SEb2^2 ) ^ 1/2 > (with four instances of the symbol ^); although > the editor in which > this reply is being composed shows ^V for the > first of these. > Presumably that first symbol is a minus sign ? > > And does not this formula presuppose that b1 and b2 > are uncorrelated? > That may be a reasonable assumption for the original > poster's > (Msherif's) situation, but there really wasn't > enough information > provided to know whether that's reasonable or not -- > depends on what > that "nonlinear model" actually is, and on what > "sector" means in the > original post. > > On Wed, 8 Oct 2003, Brett Magill wrote: > > > Here is a formula to compute Z, which can then be > > referred to a normal distribution. > > > > Z = ( b1 � b2 ) / ( SEb1^2 + SEb2^2 ) ^ 1/2 . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
