HI all,
Please see attached for the STATA code slides that were discussed during small group last week. Slides 2 and 3 were shown by Monica in class. Slide 1 summarizes what we discussed in class about how/when one can use a model with a cross-product term (to test for statistical interaction, multiplicative), then moving on to modeling cross-classified exposure categories (with dummy variables) to present effect mod/interaction in your tables with a single reference group.
best,
JMC
thank you for the slides. I am still unclear on exactly what comparisons STATA is making when you use this command which we are taught in BIOSTAT208 (using two ##):
logistic case sex colgrad age snp##smoking
limited output is:
1.snp |
.9239497 .1433513 -0.51
0.610 .6816852 1.252313
1.smoking | 3.76362
.5128197 9.73 0.000
2.881533 4.91573
snp#smoking |
1 1 |
1.825605 .3284794 3.35
0.001 1.283072 2.597543
VERSUS this command which you used in HW6 and made a point about the # vs ## but did not provide additional discussion in the hw key as far as I could tell:
logistic case sex colgrad age snp#smoking
limited output is VERY different:
snp#smoking |
0 1
| 3.76362 .5128197
9.73 0.000 2.881533 4.91573
1 0
| .9239497 .1433513
-0.51 0.610 .6816852
1.252313
1 1 | 6.348352 .8402967 13.96 0.000 4.897699 8.228675
I cannot figure out why the differences between the interaction term output between these two commands and I also shared this with Dr. Shiboski in our lab session, but I still don't have an understanding as to why you are calling the second command output "correct" on the homework vs the output for the first command.
given that the OR of 1.8 for the first command (the way we are taught in 208) is SO different from the OR of 6.3, it seems very important to understand what STATA is doing in these two commands as the answers are very different.
