Hi Steve
The 5000 men were randomly drawn using the Danish CPR register
which is a
database containing all Danish citizens. Everybody is assigned a
unique
10-digit number making it possible to select people 100% at random.
The number has the form DDMMYY-vxyz with DDMMYY being date of
birth. Z an
equal number for women and unequal for men.
The 5000 men were selected in a way that they reflected the back
ground
population regarding age and zip code.
They were aged 60 to 74 years at the date of data acquisition.
So the information we have on the 5000 men is 100% accurate and
they are a
100% match for the back ground population but the only information
was age
and zip code.
The questionnaire we mailed to the 5000 men contained a lot of
information.
For the 3750 who filled out the questionnaire we know all chronic
diseases
on ICD10 codes, all medication on ATC codes, partner status, level of
education, job situation, housing, smoking habits, physical activity,
height, weight, number of children, sexual problems.
The study is a cross sectional study examining androgens and
relations to
body composition, health status, life style, quality of life, sexual
dysfunction, physical performance, genetics etc.
You can see a PowerPoint file with the inclusion procedure here:
www.euphonium.dk/Inclusion.ppt
I sought equal numbers in each group in order not to end up with
too few
people amongst the eldest. I wanted a good representation in all
age strata
since we wanted to be able to make reference intervals for the
different
androgens for healthy 60-74 year olds and because we mainly
investigate
associations and we are not primarily interested in the
distribution in the
back ground population. But I'd like to be able to make some
estimates. As
an example I'd like to know the prevalence of erectile dysfunction
in the
back ground population or estimate the prevalence of hypogonadism or
diabetes using the data from the 600 very thoroughly examined men.
If you can help me I'll be very grateful
Best regards
Kristian Wraae
-----Oprindelig meddelelse-----
Fra: [email protected]
[mailto:[email protected]] På vegne af Steven
Samuels
Sendt: Saturday, December 06, 2008 5:04 PM
Til: [email protected]
Emne: Re: st: Survey - raking - calibration - post stratification -
calculating weights
--
Kristian, before I can answer your questions, I need details of how
you selected the original 5,000 men. Describe the target
population; the sampling "frame" (or list) from which you drew the
sample of 5,000; what information you have about the population (age
distribution, for example); what information about each man on the
list was available; exactly how you selected the 5,000. Also, what
is the purpose of the study? Why did you seek equal numbers of men
in the 15 age groups at later stages?
-Steve
questionaire.
On Dec 6, 2008, at 5:32 AM, Kristian Wraae wrote:
Hello all
I have a question regarding how to weight a data set.
The data is from a population based cross sectional study.
5000 randomly selected men reflecting the backgound population were
mailed a
questionaire.
75% responsrate. 3750 questionnaires filled out. We know the age
and the zip
code for non-responders.
The questionnaire contained several sociodemographic parameters s1,
s2,
..... Sn
Then 1845 men from the group that had filled out a questionnaire were
invited to take part in a scientific project. The men were randomly
selected
with an equal number in each age group (15 age groups of one year
intervals). So 123 men in each group.
946 men accepted a telephone call. 768 men never responded and 131
refused
to be interwied on telephone.
864 men of the 946 were then randomly contacted with equal numbers
in each
age group and 697 men agreed to take part in the project. 97 men
later
cancelled or never showed up.
So 600 men were included for further studies.
Now I would like to weight these 600 men so they reflect the
background
population in order to estimate the distribution of different
measures in
the background population (X1, X2, .... Xi) based on measures
amongst the
600 men (Y1, Y2, ..... Yi).
How do I do this?
As far as I can tell I need to compensate for the differences
between the
5000 and the 3750 and between the 3750 and the 600. Since the 1845
were
randomly selected and an equal number in each age group were
contacted I
assume that all men had an equal probability of being included so
design
weights are not really needed. Right?
But how do I compute a pweight that takes the two steps into account?
