2. Content
Sampling method & sample size
for survey
What is complex sampling
method
Sampling weight
3. About sampling
Not feasible to selectALL population
Best sampling should be able to represent population
Sampling error occurs when statistics ≠ parameters
Sampling error is not sampling bias
Sampling error is random, sampling
bias is predictable(systematic)
Sampling design affects sampling error
Standard error measures sampling error
4. The aim of any sampling
plan should be to reduce
sampling error,and to
avoid sampling bias
5. Describe the sample
Target population – inferred population
Study population – representative of the target
population
Sampling frame – list of sampling unit
Sampling unit – unit to be sampled
Observation unit – unit to be observed/measured
6. Sampling method
Random vs. non-random
Random ensures representativeness
Simple vs. complex
SRS = all samples have equal chance to be selected i.e.
equal probability of selection
Anything not SRS is complex sampling
8. Stratified versus cluster sampling
Stratified for heterogeneous groups
e.g. male-female, age groups
Cluster for homogenous groups – rarely homogenous,
only in ideal situation e.g. schools, districts
10. Design Effect (deff)
Design Effect = Variance estimate (complex)
Variance estimate (SRS)
How much the sample differ from population
Different value for different variable
Usually, deff for complex survey >> 1
If > 1.5, meaning effective loss 50% of sample if
designed using SRS
11. Sampling Weight
aka Probability Weight
N/n (inverse of sampling fraction)
Two stage = (N1/n1)*(N2/n2)
The sum of PW = population
Weighting can increase standard error
12. Sampling weight…
Why? There is always imperfection in sampling
Weighting will try to correct
1. Unequal probability of selection – base/design
weight
2. Non-response bias
3. Stratification in population – trying to
represent true characteristics of population
e.g. by sex, ethnic etc. – poststratification