<meta http-equiv="Content-Type" content="text/html; charset=UTF-8" />Hi all,
A customer discovered that JMP 10.0.2 yields different estimates than SAS 9.2 when applying linear mixed models for both programs.
Here is the example:
JMP 10.0.2: Least sq mean std error Lower CL Upper CL
A 99,073901 0,63026917 97,530305 100,61750
B 99,153272 0,63426425 97,608388 100,69816
C 99,766164 0,63105798 98,222362 101,30997
D 98,341544 0,54416752 97,067229 99,61586
E 97,177437 0,65625537 95,620335 98,73454
F 96,691972 0,63959974 95,147114 98,23683
SAS 9.2:
| Effect | F2 | Estimate | Standard Error | DF | t Value | Pr > |t| | Alpha | Lower | Upper |
| F2 | A | - 99.0739
- 0.6302
| | 83 | - 157.20
| <.0001 | - 0.05
- 97.8204
- 100.33
| | |
| F2 | B | - 99.1533
- 0.6341
| | 83 | - 156.38
| <.0001 | - 0.05
- 97.8921
- 100.41
| | |
| F2 | C | - 99.7662
- 0.6310
| | 83 | - 158.11
| <.0001 | - 0.05
- 98.5111
- 101.02
| | |
| F2 | D | - 98.3415
- 0.5433
| | 83 | - 181.02
| <.0001 | - 0.05
- 97.2610
- 99.4221
| | |
| F2 | E | - 97.1774
- 0.6549
| | 83 | - 148.39
| <.0001 | - 0.05
- 95.8749
- 98.4800
| | |
| F2 | F | - 96.6920
- 0.6396
| | 83 | - 151.18
| <.0001 | - 0.05
- 95.4199
- 97.9640
| | |
As you can see, they get different confidence intervals. Can you help me with an answer for the reason for this?
Data:
| V1 | F1 | F2 | F3 |
| 99,38463 | A | A | A |
| 98,99927 | A | A | A |
| 99,27793 | A | A | A |
| 99,29341 | A | A | A |
| 99,16183 | A | A | A |
| 98,86683 | A | A | A |
| 98,95378 | A | A | A |
| 98,80463 | A | A | A |
| 98,63573 | A | A | A |
| 98,87085 | A | A | A |
| 98,82634 | A | A | A |
| 98,99573 | A | A | A |
| 98,94268 | A | A | B |
| 99,22524 | A | A | B |
| 98,91561 | A | A | B |
| 99,28476 | A | A | B |
| 99,54232 | A | A | B |
| 98,61012 | A | A | B |
| 99,25841 | A | A | B |
| 99,3661 | A | A | B |
| 99,02293 | A | B | C |
| 98,76476 | A | B | C |
| 99,06098 | A | B | C |
| 99,19512 | A | B | C |
| 99,13061 | A | B | C |
| 99,36744 | A | B | C |
| 99,35598 | A | B | C |
| 99,03805 | A | B | C |
| 98,88524 | A | B | C |
| 99,72793 | A | B | C |
| 99,42183 | A | B | C |
| 99,69329 | A | B | C |
| 100,1484 | A | B | C |
| 99,64268 | A | B | C |
| 98,57 | A | B | C |
| 98,68439 | A | B | D |
| 99,02549 | A | B | D |
| 98,76195 | A | B | D |
| 99,06159 | A | B | D |
| 99,62293 | A | B | D |
| 100,9184 | A | C | E |
| 100,532 | A | C | E |
| 100,4365 | A | C | E |
| 99,64683 | A | C | E |
| 99,175 | A | C | E |
| 98,75463 | A | C | E |
| 97,95756 | A | C | E |
| 99,21073 | A | C | E |
| 99,74841 | A | C | E |
| 99,29878 | A | C | E |
| 98,73598 | A | C | E |
| 97,89524 | A | C | E |
| 98,81134 | A | C | E |
| 100,5904 | A | C | F |
| 99,94049 | A | C | F |
| 99,98195 | A | C | F |
| 99,90427 | A | C | F |
| 100,6167 | A | C | F |
| 100,4715 | A | C | F |
| 100,095 | A | C | F |
| 97,87988 | B | D | G |
| 97,91 | B | D | G |
| 99,33707 | B | D | H |
| 100,9166 | B | D | H |
| 99,03921 | B | D | H |
| 97,95787 | B | D | H |
| 98,01957 | B | D | H |
| 98,37085 | B | D | H |
| 97,32073 | B | D | H |
| 97,63201 | B | D | H |
| 98,3075 | B | D | I |
| 98,72665 | B | D | I |
| 98,86909 | C | E | J |
| 97,68666 | C | E | J |
| 97,97723 | C | E | J |
| 98,31459 | C | E | J |
| 98,86204 | C | E | J |
| 98,35888 | C | E | J |
| 99,50009 | C | E | J |
| 99,05164 | C | E | J |
| 99,25596 | C | E | J |
| 98,83933 | C | E | J |
| 95,81784 | C | E | K |
| 95,01733 | C | E | K |
| 96,34623 | C | F | L |
| 96,74948 | C | F | L |
| 97,20784 | C | F | L |
| 97,36222 | C | F | L |
| 97,19511 | C | F | L |
| 96,46179 | C | F | M |
| 96,149 | C | F | M |
| 96,94456 | C | F | M |
| 96,50015 | C | F | M |
| 96,35419 | C | F | M |
| 96,05544 | C | F | M |
| 96,46495 | C | F | M |
My best guess is that you customer doesn't set DDFM=KENWARDROGER in the Model statement in SAS Proc Mixed. The Kenward and Roger correction is the default used by JMP. I found the LS Means results from JMP match Proc Mixed.
