0

My goal is to compare published regression weights estimated from a sample in another country, with those of the same model established on our data.There is ample evidence on the web that a Chow test is the best method to test equality of regression parameters, by combining two groups. But a problem emerges when the data for the 1st study is not at hand.

Alternatively, z-test for comparison of regression slope has been suggested. but as addressed in a similar question, this test is only robust when the ratio of two SEs for the coefficients does not exceed 3. Unfortunately this ratio of SEs is about 10 in case of my study.

Again in the question attached, Welch-Satterthwaite t-test or Cochran test (Cochran and Cox 1950) is recommended for Behrens-Fisher problem. Since I don't have the data for the first group I only can perform the tests by hand, but I could not find the formula for Welch t test or CBF for comparison of slopes. The only Welch or Cochran t test I found is for comparison of the means, not the slopes.

So four questions arise here:

  1. Are these two tests the same?

  2. Are there any alternatives in such a situation?

  3. Or can I compare the R squared in the same model across the groups instead?

  4. If yes, what test can be applied?

Community
  • 1
Sadjjad Riyahi
  • 113
  • 2
  • 6
  • which Cochran test are we talking about? – Glen_b Mar 05 '14 at 01:40
  • @Glen_b I edited the question, but yet I'm not exactly sure. I found that SAS has an COCHRAN option in TTEST procedure to produces p values for the unequal variance situation by using the Cochran and Cox (1950) approximation. although I don't use SAS. these all seem confusing, because when I modify the denominator for the z-test to SQRT(SE1^2/n1 + SE2^2/n2) as in approximation t-testsI get very large z values and for the formula for t-test proposed on the attached similar question,not only I don't know how to modify the denominator,but also I even couldn't find it anywhere else. – Sadjjad Riyahi Mar 05 '14 at 07:51

0 Answers0