22 Sep
2003
22 Sep
'03
11:42 p.m.
Based on today's discussion of distributions, do we think Plan B is better than Plan
C in problem 2? Based on the initial distribution, these distributions appear to be the
random distributions that are most similar to the original distribution, even though the
numbers in the two groups are different. i think the distribution is more important than
the total number of cases...other thoughts?
23 Sep
23 Sep
12:57 a.m.
New subject: histogram
Our group reached the same conclusion, that distribution across space was
important--reasoning was that if there is not a balanced distribution
across age groups, then we don't control for the relative effects that age
of participants might have upon the outcome, whereas balanced distributions
(however imperfectly) tend to minimize the interaction between age and
effect and thus also make the original inference less spurious by making it
more linear, that is, controlling for the relative effects of prior
conditions. Total number doesn't matter much--and this is where I am
weakly conjecturing--because we have no way of knowing given testing
conditions, if in both plans people are randomly picked, the plan
population size necessary to minimize the effects of other variables (in
this case age).
That's it so far. But just like the "R" stuff there is a really good
chance that everything above is completely wrong... There will be a group
of us in Littauer tomorrow at 8PM to hammer it out.
-Sean
At 11:42 PM 9/22/2003 -0400, Jason Lakin wrote:
Based on today's discussion of distributions, do
we think Plan B is better
than Plan C in problem 2? Based on the initial distribution, these
distributions appear to be the random distributions that are most similar
to the original distribution, even though the numbers in the two groups
are different. i think the distribution is more important than the total
number of cases...other thoughts?
*****
Cons Gradu Prov et Cantab et Nov Ebor:
Non Licet Omnibus Adire Corinthum.
8:55 a.m.
New subject: histogram
Well it might depend on whose interests you're trying to promote, i.e. who
would find the outcome preferable.
Plan B matches for age but puts twice as many people in the treatment group.
This means, ceteris paribus, a smaller confidence interval for the smoking
group and a larger one for the control group. So I guess this makes it easier
to get a result showing a control group cancer rate which approaches that of
the treatment group. However, the opposite is also true and you could get a
result which shows control group rates well below the mean. So I'm not sure
that we have the answer here.
It seems to me that Plan C has no age matching. Is age an important baseline
variable? I would assume it is, given a) common sense, b) no specific
information on other (possibly confounding) variables, and c) taking the hint
from the question about Plan A. Perhaps as an additional signal, the smoking
group in plan C is more skewed to older participants than the control group.
So I would hypothesize that while neither design is great, Plan B is preferable
because in Plan C you're choosing to ignore a variable which is known to have
an effect on outcome. And Philip Morris (incidentally now called Altria-- how's
that for a feel-good name) would probably reach the same conclusion since the
treatment group in Plan C is older and therefore will be prone to showing
higher cancer rates anyway.
But I could be completely wrong.
Paul
Quoting Jason Lakin <jlakin(a)fas.harvard.edu>du>:
Based on today's discussion of distributions, do
we think Plan B is better
than Plan C in problem 2? Based on the initial distribution, these
distributions appear to be the random distributions that are most similar to
the original distribution, even though the numbers in the two groups are
different. i think the distribution is more important than the total number
of cases...other thoughts?
7538
days inactive
7538
days old
gov1000-list@lists.fas.harvard.edu
2 comments
3 participants
participants (3)
-
Jason Lakin -
Paul Bodnar -
Sean L. Yom