topic Re: Slight difference in p-values under “Combine” rule (before vs after entry) in Discussions

Slight difference in p-values under “Combine” rule (before vs after entry)

happy — Tue, 21 Oct 2025 08:35:30 GMT

您好，
我注意到在使用逐步回归和组合规则时，p 值存在细微但一致的差异。
在我的示例中（见下方截图），材料 (B–C) 显示的效果为：
输入模型前：Prob > F = 0.75127
输入模型后：Prob > F = 0.75103
F 比和自由度完全相同 (F ≈ 0.104, df1 = 1, df2 = 16)，所以我想知道为什么 p 值不相同。
我的猜测是，JMP 在候选效应进入模型之前计算候选效应时会执行“临时完全拟合”或增量投影，这可能会与最终的完整模型拟合产生微小的数值差异。
有人可以确认 JMP 在组合规则中计算候选 p 值的方式是否确实如此吗？以及这个细微的（1e-4 级）差异是否是由于内部精度或四舍五入造成的？
谢谢！

JMP 专业版 17.1.0

Re: Slight difference in p-values under “Combine” rule (before vs after entry)

jthi — Tue, 21 Oct 2025 08:42:19 GMT

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)

happy — Tue, 21 Oct 2025 08:52:27 GMT

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?