Transcript Sampling - Terveyden ja hyvinvoinnin laitos

Sampling
Presentation on workshop in
Luxembourg
10.April 2008
Johan Heldal
Target Population
The TP in a survey is the finite set of
units we want to study and draw
inferences on.
In HES: Individual persons at
specified ages within a specified
geographical area: A country
It will sometimes have to be
restricted by practical considerations
defined by the sampling frame.
Recommendations
for TP
Geographical coverage: The entire
territory of a country.
Time of reference (t): A fixed date
(e.g. January 1.) of the year of the
survey.
Core pop: Permanent residents
filling ≥ 25 and ≤ 65 in (t, t+1).
Recommended extension: Down
to age ≥ 18. No upper bound.
Critical topics
Type of sampling frame
Coverage of sampling frame
Timeliness of the sampling frame:
Not too old.
Sample size
Goal: Nationally representative data.
A good frame
Should have good
coverage of target units.
Should contain few nontarget units that cannot
be identified.
Should contain
information relevant for
stratification
Should contain good
contact information.
Should be up-to-date
Sampling frames I
Population Registers
Have individual persons as units
Contain information on age and gender
Contain information on addresses.
Of 32 countries listed, 23 say they have a PR
16/23 want to use it as a HES frame
14/16 update continuously, 2/16 yearly
11/16 cover entire TP
9/16 do both
Census Lists
7 countries want to use CLs as frame
5/7 cover the entire TP
3/7 update continuously, 2 annually
Sampling frames II
Electoral Rolls
3/26 countries want to use ER as frame
1/3 update continuously
Centralized GP list of patients.
May only cover persons having a doctor.
May be subject to professional secrecy(?).
Telephone directory
Not complete enough
Fixed phones are related to households, not
persons.
Cell phones not always registered on the
carrier
Better used as supplementary frame
Sampling frames III
Postcode Address Files (England,
Scotland)
May have good coverage of households
But do not contain information on
eligible residents
Have no information on age and gender
Maps combined with small area
population statistics (NHANES, USA)
Other frames may be supplementary
Frame
recommendations
Preferred frame:
Updated register or census lists with
complete coverage, whenever
available.
Second choice: Updated postcode
address files (England).
Third choice: Maps in combination
with census statistics (NHANES)
Sampling design
recommendations I
Primary Sampling Units (PSUs):
Clinic areas
Secondary Sampling Units (SSUs):
Invitees to sampled clinics.
PSU-strata: Subgroups of PSUs by
geography and health related factors.
Demographic strata: Target
population by gender and 10 year age
groups.
Sampling design
recommendations II
PPS-sampling: Sampling of PSUs in
PSU-strata with Probabilities
Proportional to population Size.
Equal sample size for each gender
by age stratum
Equal probability within each
gender by age stratum
About equal sample size to each
sampled PSU (clinic).
Sample size
Two sample sizes:
1. Number of PSUs sampled (clinics): m
2. Number of persons to be invited: n
Recommendations:
– Minimum size for n: 3000 invitees.
– Larger sample recommended, required
if detail below national level desired.
– m and n = n/m set by cost efficiency
criteria
That’s all.
Thank you for your
attention.