hypothesis test that regression estimate non zero constant
Aug 7, 2014 12:13 PM(1275 views)
If I want to test that a regression parameter b is equal to a nonzero constant, is there a script for that?
For example, in the Pendulum sample data available in JMP, let's say I fit Period = b + m*sqrt(Length). I want to test the hypothesis that m=2. The output tests whether m=0, which may not be that interesting.
Is there a script to test whether a parameter estimate is equal to some non-zero constant?
Not a script, but it would be fairly easy to create one.
Easiest, quickest solution is to right-click on the Parameter Estimates table and choose the Lower and Upper 95% confidence limits. This will show you the 95% confidence interval for parameter estimate. If that interval contains the nonzero constant, then you have a non-significant result. If the nonzero constant is outside of the interval, then you have a statistically significant difference from that constant.
A simple formula script would look something like this:
t_Ratio == (Estimate - Constant) / Std_Error;
p_value==If(t_Ratio >= 0, (1 - t Distribution(t_ Ratio, df)) * 2, t Distribution(t_Ratio, df) * 2);
You can certainly make this much fancier with dialogs and such or even reading directly from a report, but this is the essence of the formulas that would work.