Accepted Solutions
Congratulations, Bordini, you are specifying your model correctly. You matched the fixed effects.
Since you stated you are new to statistics and JMP, I'll try to keep it simple. Most classical statistics assumed independence (no correlation) within blocks (plots) and repeated measures. In the 1980's as compute power was radidly increasing so did the statistical methods that accomodate different assumptions, different covariance structure. For example, for an actual split plot design, the layout of the subplot could have spatial effects, etc. Independence no longer has to be assumed. Note, if you run your Mixed Model in JMP 12 you will get a zero effect for Shift not the -11.7776 result in JMP 13 and 14. JMP 13 and 14 uses a new method (Kackar-Harville), a correction that is applied to the variance matrix of the fixed effects.
I have not read through all the details, but it appears JMP 13 and 14 is using a random intercept covariance structure, where JMP 12 set the values to 0. See the left screenshot, this is where the -11.7776 is coming from. Minitab is likey using a different approximation method (since it's intercept is not zero). When using different software the parameterization can be different as well. JMP uses the kth effect to be the negative sum of effects 1-(k-1). From your screeshot of the output, it looks like Minitab uses the 1st block to be the sum of the others [I am guessing].
I know this "statistics" stuff might seem too complex, but don't give up yet. One of the reasons, I am a fan of JMP because it so easy to visualize variation and effects. See the variability plot of your data table. A good plot of the doesn't rely on best best methods. Note that, the size of the Temperature effect is large. Variation in the horizontal purple lines represent the Time effect. Look at the effect of Temperature at Time=30. It is different. That is the Temperature*Time interaction. Not shown here is the standard deviation plot. The variability of shift is smaller for high Temperature and high Time. The fixed effects model is an numeric estimation, this is the visual estimation.
I hope that helps.
Thank you. I already looked at the resources in JMP, but I did not not come up with the correct results. I have a MINITAB output with the correct results, but mine from JMP are completely different. I honestly don't know what I'm doing wrong in JMP.
See MINITAB output attached, please. Thank you.
The other attachments are here.
Congratulations, Bordini, you are specifying your model correctly. You matched the fixed effects.
Since you stated you are new to statistics and JMP, I'll try to keep it simple. Most classical statistics assumed independence (no correlation) within blocks (plots) and repeated measures. In the 1980's as compute power was radidly increasing so did the statistical methods that accomodate different assumptions, different covariance structure. For example, for an actual split plot design, the layout of the subplot could have spatial effects, etc. Independence no longer has to be assumed. Note, if you run your Mixed Model in JMP 12 you will get a zero effect for Shift not the -11.7776 result in JMP 13 and 14. JMP 13 and 14 uses a new method (Kackar-Harville), a correction that is applied to the variance matrix of the fixed effects.
I have not read through all the details, but it appears JMP 13 and 14 is using a random intercept covariance structure, where JMP 12 set the values to 0. See the left screenshot, this is where the -11.7776 is coming from. Minitab is likey using a different approximation method (since it's intercept is not zero). When using different software the parameterization can be different as well. JMP uses the kth effect to be the negative sum of effects 1-(k-1). From your screeshot of the output, it looks like Minitab uses the 1st block to be the sum of the others [I am guessing].
I know this "statistics" stuff might seem too complex, but don't give up yet. One of the reasons, I am a fan of JMP because it so easy to visualize variation and effects. See the variability plot of your data table. A good plot of the doesn't rely on best best methods. Note that, the size of the Temperature effect is large. Variation in the horizontal purple lines represent the Time effect. Look at the effect of Temperature at Time=30. It is different. That is the Temperature*Time interaction. Not shown here is the standard deviation plot. The variability of shift is smaller for high Temperature and high Time. The fixed effects model is an numeric estimation, this is the visual estimation.
I hope that helps.
