Thank you.
Could you be more specific on which misconceptions exactly I harbour on
basic statistical concepts and distributional assumptions? The answers are
not helpful if they are not specific, and the problem I described is very
specific.
---------- Forwarded message ----------
From: Brian G Miller <[log in to unmask]>
Date: 2009/5/19
Subject: RE: continuous vs. categorical
To: - <[log in to unmask]>
Dear contributor
From your description of the problem, it appears that you harbour a
number of misconceptions about basic statistical concepts concerning
regression and distributional assumptions. I think you'll continue to
struggle unless you get professional advice from someone familiar with
the topic. If you're still at college, that would be your supervisor or
statistical adviser.
Brian Miller
-----Original Message-----
From: - [mailto:[log in to unmask]]
Sent: 15 May 2009 13:52
Subject: continuous vs. categorical
hello all,
I wonder if someone could help me with the following:
- I am doing some analysis on weight and BMI examining association with
gastrointestinal cancer. Weight and BMI are both continuous variables
and I
am also using them as grouped variables, grouping based on the control
distribution. I do logistic regression. To get the p for trend we know
that
we fit the grouped variable as continuous. In this case I also have the
original values of the variables, i.e. I have the continuous variable A
and
the generated grouped variable B. I tried not only fiting B as
continuous,
but also using the original variable A. I know that when the
distribution is
skewed you are better off using the grouped variable. In this case, when
I
examine weight either using A or B, I get significance for both cases.
When
I examine BMI using A I get significance, but using B I get no
significance.
Looking at the distribution of the cases and controls together, it is
indeed
skewed. So, I'm thinking that the valid results are the ones showing no
significance i.e. those I get from fiting the grouped variable as
continuous. What do you think?
- I have a very large sample so it is expected that the distributions of
weight and BMI will be normal, under a null hypothesis. But, since we
know
that nowadays people tend to be more fat than was the case decades ago,
irrespective of medical conditions, is it correct to expect normality
under
the null? My point being that if the distribution is skewed already
under
the null then maybe we don't have to remedy the skewness in the actual
data,
so maybe it's not best to use the grouped variable here instead of the
continuous.
I would appreciate your comments on this, and will compile the answers I
get.
Thank you all
Dr Brian Miller
Principal Epidemiologist
Institute of Occupational Medicine
Research Avenue North
Riccarton
Edinburgh
EH14 4AP
Tel: 0131 449 8044
Fax: 0870 850 5132
Email: [log in to unmask]
Web: http://www.iom-world.org
--------------------------------------------------------------------------
The Institute of Occupational Medicine (IOM) is a company limited by
guarantee, registered in Scotland (No.SC123972) and a Registered Scottish
Charity (No.SC000365). IOM Consulting Ltd is a wholly owned subsidiary of
IOM and a private limited company registered in Scotland (No. SC205670).
Registered Office: Research Avenue North, Riccarton, Edinburgh, EH14 4AP,
Tel +44 (0)870 850 5131.
This email and any files transmitted with it are confidential and intended
solely for the use of the individual or entity to whom they are addressed.
If you have received this email in error please notify the sender -
[log in to unmask] - immediately and delete the message from your
system. All IOM email is virus scanned but the recipient should also check
this email and any attachments for the presence of viruses.
|