07. Design probability samples and understand selection weights
Lesson
LEARNING OBJECTIVES
Distinguish target population, frame and sample, calculate a simple selection weight, and explain how a multistage design affects analysis.
START WITH COVERAGE
Describe who belongs to the target population and which operational frame will be used. A register may omit people without service access, contain outdated addresses or include duplicates. Document and assess these problems before selection. A probability mechanism applied to an inadequate frame does not solve its coverage gaps.
Specify each stage: areas, households and people, for example. Record how units are selected, their selection probabilities, eligibility and response. Stratification can ensure planned representation of important groups; clustering may reduce travel but creates dependence that the analysis must acknowledge.
WORKED FICTIONAL WEIGHT EXAMPLE
A complete frame contains 1,000 households and a simple random sample of 100 is selected. Each household has selection probability 0.10. If one eligible child is selected at random from a household with two eligible children, that child's overall selection probability is 0.10×0.50 = 0.05; the basic inverse-probability weight is 20.
In a selected household with one eligible child, the corresponding probability is 0.10 and the basic weight is 10. Selecting one child per household does not give every child the same chance when household eligibility counts differ.
These are base selection weights under the stated assumptions. Non-response adjustments, calibration, trimming and uncertainty estimation involve further choices and documentation. Never apply a convenient weight merely because it makes the result look plausible.
PLAN INFERENCE ALONGSIDE SELECTION
A sample-size total does not specify the number of clusters, observations per cluster, strata or domain precision. More observations within a cluster can provide less additional information when outcomes are correlated. Preserve cluster and stratum identifiers and selection information for authorized analysis.
A basic unweighted browser mean is not automatically a valid population estimate for an unequal-probability survey. Appropriate software should incorporate the design, weights and relevant variance assumptions. Explain whether any finite-population correction is justified by the actual sampling design.
YOUR TASK
Sketch a two- or three-stage selection plan for your question. Define each unit and frame, the selection rule, how eligible individuals are enumerated and how non-response is recorded. Calculate a selection probability and base weight for two hypothetical participants.
Write a paragraph explaining which design variables must be retained for analysis and which frame limitations remain. Add a rule for handling an absent household that does not substitute whichever household is easiest to contact.
SELF-CHECK
1. Is a base weight the same as a frequency count of observations? No; it reflects selection probability under the design.
2. Can probability sampling eliminate outcome misclassification? No; sampling and measurement address different problems.
3. Does a larger total ensure precise subgroup estimates? No; subgroup size, design and variability matter.
PORTFOLIO OUTPUT
A selection diagram on paper, worked probability/weight examples and an analysis-information checklist.
FURTHER READING
WHO STEPS Manual, Part 2: Planning and set-up.
https://www.who.int/docs/default-source/ncds/ncd-surveillance/steps/part2.pdf
Distinguish target population, frame and sample, calculate a simple selection weight, and explain how a multistage design affects analysis.
START WITH COVERAGE
Describe who belongs to the target population and which operational frame will be used. A register may omit people without service access, contain outdated addresses or include duplicates. Document and assess these problems before selection. A probability mechanism applied to an inadequate frame does not solve its coverage gaps.
Specify each stage: areas, households and people, for example. Record how units are selected, their selection probabilities, eligibility and response. Stratification can ensure planned representation of important groups; clustering may reduce travel but creates dependence that the analysis must acknowledge.
WORKED FICTIONAL WEIGHT EXAMPLE
A complete frame contains 1,000 households and a simple random sample of 100 is selected. Each household has selection probability 0.10. If one eligible child is selected at random from a household with two eligible children, that child's overall selection probability is 0.10×0.50 = 0.05; the basic inverse-probability weight is 20.
In a selected household with one eligible child, the corresponding probability is 0.10 and the basic weight is 10. Selecting one child per household does not give every child the same chance when household eligibility counts differ.
These are base selection weights under the stated assumptions. Non-response adjustments, calibration, trimming and uncertainty estimation involve further choices and documentation. Never apply a convenient weight merely because it makes the result look plausible.
PLAN INFERENCE ALONGSIDE SELECTION
A sample-size total does not specify the number of clusters, observations per cluster, strata or domain precision. More observations within a cluster can provide less additional information when outcomes are correlated. Preserve cluster and stratum identifiers and selection information for authorized analysis.
A basic unweighted browser mean is not automatically a valid population estimate for an unequal-probability survey. Appropriate software should incorporate the design, weights and relevant variance assumptions. Explain whether any finite-population correction is justified by the actual sampling design.
YOUR TASK
Sketch a two- or three-stage selection plan for your question. Define each unit and frame, the selection rule, how eligible individuals are enumerated and how non-response is recorded. Calculate a selection probability and base weight for two hypothetical participants.
Write a paragraph explaining which design variables must be retained for analysis and which frame limitations remain. Add a rule for handling an absent household that does not substitute whichever household is easiest to contact.
SELF-CHECK
1. Is a base weight the same as a frequency count of observations? No; it reflects selection probability under the design.
2. Can probability sampling eliminate outcome misclassification? No; sampling and measurement address different problems.
3. Does a larger total ensure precise subgroup estimates? No; subgroup size, design and variability matter.
PORTFOLIO OUTPUT
A selection diagram on paper, worked probability/weight examples and an analysis-information checklist.
FURTHER READING
WHO STEPS Manual, Part 2: Planning and set-up.
https://www.who.int/docs/default-source/ncds/ncd-surveillance/steps/part2.pdf
My study notes
Stored only in this browser. Do not enter patient information.Saved in this browser. No account needed.