Slight difference in p-values under “Combine” rule (before vs after entry)
Hello,
I noticed small but consistent differences in the p-values when using stepwise regression and the combined rule.
In my example (see screenshot below), the materials (B–C) appear as follows:
Before entering the model: Prob > F = 0.75127
After entering the model: Prob > F = 0.75103
The F ratio and degrees of freedom are exactly the same (F ≈ 0.104, df1 = 1, df2 = 16), so I wonder why the p-values are different.
My guess is that JMP is performing a "temporary full fit" or incremental projection when calculating candidate effects before they enter the model, which may produce small numerical differences from the final full model fit.
Can someone confirm whether this is indeed the way JMP calculates candidate p-values in the combination rule, and whether the slight (1e-4 order) difference is due to internal precision or rounding?
Thanks!
JMP Professional 17.1.0
Basic Data Analysis and Modeling
Re: Slight difference in p-values under “Combine” rule (before vs after entry)You could try changing the column formatting for the F-ration by right-clicking on it to verify if there is a difference
Re: Slight difference in p-values under “Combine” rule (before vs after entry)Thank you for your explanation!
After checking carefully as you suggested, I found that the small difference in the p-values actually comes from a slight difference in the F ratio between the two situations.
Could you please help me understand why the F ratio itself is not exactly the same in these two cases?
