comparing confidence intervals for the odds ratio. If a
meaningful gain of precision is made by dropping a
nonconfounder, then such a nonconfounder may be
dropped. Otherwise (i.e., no gain in precision), the
nonconfounder should remain in the model with all
other variables needed for controlling confounding.- 1
- 1
- 2
- 1
- 2
- Unpooled; the analysis treats all five strata (matched
pairs) as unique. - 2.5
- Pooled.
- Pooled; treating the five strata as unique is artificial
since there are exchangeable strata that should be
pooled. - W¼1,X¼1,Y¼0, andZ¼2.
- mOR(unpooled)¼undefined; mOR(pooled)¼2.5.
- Only choice a is correct.
428 11. Analysis of Matched Data Using Logistic Regression