I realize I'm a little late to this discussion, but I haven't heard anyone
mention the "Extra Sums of Squares" or "Additional Sums of Squares"
principal which can be used to compare slopes and/or intercepts of
different regression models.  I don't have a good reference for the
procedure used, and it can require some care in the way the data is set up
to test different hypothesis about how models differ, but I know it is
another possible approach to this problem.

Jane F.



> Your approach is valid ONLY IF you are willing to ignore the fact that the
> slope to which you are comparing your slope is itself an estimate.  That
> is
> - you can use your CI to compare to a particular hypothesized value -
> basically testing the hypothesis Ho: beta = beta_0, where beta_0 is some
> hypothesized value, possibly from the literature.  However, if you really
> want to see if two slopes are equal, say Ho: beta_1 = beta_2, you are
> better
> off using the test on p. 360 of Zar.  This essentially looks at the CI of
> the difference in slopes (b_1 - b_2) to see if it includes 0.
>
> On 8/16/06, David Whitacre <[EMAIL PROTECTED]> wrote:
>>
>> While we're on regression--I know this is a really dumb question and I
>> should know the answer. But here goes, my ignorance on display:
>>
>> In comparing some regressions to published ones, how do I test for
>> significant difference in slope? I have calculated the 95% C.I. of my
>> slope by using the t distribution applied to the SE of the slope, as
>> described on p. 331 of Zar (1996, 3rd edition).
>>
>> If somebody else's slope is outside of this C.I., are the two slopes
>> significantly different at p = 0.05? That is, I don't have to consider
>> the
>> C.I. on their slope?
>>
>> Thanks much for any enlightenment on this very basic issue.
>>
>> Dave W.
>>
>
>
>

Reply via email to