1

I’m attempting to run a regression and test whether the slope is equal to 1. Most statistical packages only test of it's equal to zero. Is there a way to bypass this ? I’ve SPSS and minitab and can run the typical regression.

Will the regression coefficients and SE be the same irrespective of whether I’m testing (slope = 0 or 1)? Meaning are the t stats and p values the only things that depend on the hypothesis.

Sheila
  • 11
  • 2
  • The hits at https://stats.stackexchange.com/search?q=regression+offset+-Poisson+-Binomial+-Gamma+-logistic+answers%3A1 provide ideas. A general procedure is offered at https://stats.stackexchange.com/questions/180478. Does that answer your questions? But the practical matter is that if in a model $E[Y]=\beta_0+\beta_1 x$ you wish to test the hypothesis $\beta_1=c$ for a predetermined constant $c$ (like $c=1$), then simply fit the model $E[Y-cx]=\beta_0+(\beta_1-c)x$ by regressing $Y-cx$ against $x$ and read the p-value directly off the standard output. – whuber Mar 07 '19 at 16:09
  • Sorry I’m a bit confused here. Thanks for the links, though my lack of stats knowledge prevents me from understanding the posts. And thanks for your suggestion. Is there any other alternative ? Model is basic bivariate. Would it be possible to run a normal regression (test beta is zero), then use the results to manually recalculate t test to see if beta is one ? Slope minus 1 divided by standard error – Sheila Mar 07 '19 at 16:18
  • I have provided you with by far the simplest approach, because it requires no post-processing on your part. The next simplest alternative is the calculation presented in the link I gave you--the manual recalculation of the t-test. – whuber Mar 07 '19 at 16:26
  • 1
    Several alternatives are discussed in answers at the first indicated duplicate – Glen_b Mar 08 '19 at 00:56

0 Answers0