Am I allowed to pool groups for GLM?

[gamadeus]

Here is my problem:

I am doing a stability study, for which I have several lots on different storage temperatures. In order to asses the stability I want to do a pooled regression analysis, but first I have to check in my GLM, whether there are differences in the intercepts and the slopes, otherwise I am not allowed to pool them.

So far so good, but here is the tricky part: I also want to see if there are differences between the storage temperatures. For this, I do no longer use individual lots, but the data for all lots combined. There is no concern doing this, if the slopes and intercepts are not significantly different for each temperature, but is it also valid to do this, if there actually is a significant different?

Thanks for any help in advance!

[jeja]

I`m having a similar problem, just many steps behind. I`ve always been terrible with stats and now it seems I can`t get what goes into which column . Problem, pretty much the same: want to check if I can combine the lots from the same temperature, using the same criteria: are there significant differences in their slopes and intercepts. It seems you are way passed this part. Can you, please, help me do it? I have 3 lots, 36 months data for all of them. The file is attached. Thank you!

[gamadeus]

Well, since nobody is answering my question I might as well answer yours.

So there is no difference in the slope between the 3 lots.


General Linear Model: Measure versus Lot



Factor Type Levels Values

Lot fixed 3 1; 2; 3





Analysis of Variance for Measure, using Adjusted SS for Tests



Source DF Seq SS Adj SS Adj MS F P

Time 1 69,215 69,215 69,215 16,88 0,001

Lot 2 17,616 11,900 5,950 1,45 0,260

Lot*Time 2 1,058 1,058 0,529 0,13 0,880

Error 18 73,804 73,804 4,100

Total 23 161,693





S = 2,02491 R-Sq = 54,36% R-Sq(adj) = 41,68%





Term Coef SE Coef T P

Constant 98,1591 0,6465 151,83 0,000

Time -0,15129 0,03682 -4,11 0,001

Time*Lot

1 0,00203 0,05207 0,04 0,969

2 0,02183 0,05207 0,42 0,680

However, when you remove the interaction term you can see that the intercepts are different. So you have to take the worst case intercept with the common slope.

General Linear Model: Measure versus Lot



Factor Type Levels Values

Lot fixed 3 1; 2; 3





Analysis of Variance for Measure, using Adjusted SS for Tests



Source DF Seq SS Adj SS Adj MS F P

Time 1 69,215 69,215 69,215 18,49 0,000

Lot 2 17,616 17,616 8,808 2,35 0,121

Error 20 74,862 74,862 3,743

Total 23 161,693





S = 1,93472 R-Sq = 53,70% R-Sq(adj) = 46,76%





Term Coef SE Coef T P

Constant 98,1591 0,6177 158,91 0,000

Time -0,15129 0,03518 -4,30 0,000

[jeja]

Thank you sooo much.