8.2 Stratified and Cluster Sampling

Stratified and cluster sampling are key techniques for gathering representative data from complex populations. These methods divide the population into groups, either for targeted sampling or cost-effective data collection, improving precision and efficiency over simple random sampling.

Understanding these techniques is crucial for designing effective sampling strategies in real-world research. They allow researchers to balance statistical rigor with practical constraints, ensuring valid inferences about diverse populations while managing resources and logistics effectively.

Stratified Sampling

Stratified Sampling Methodology

• Stratified sampling divides population into distinct subgroups called strata before sampling
• Strata consist of homogeneous groups based on specific characteristics (age, income, education level)
• Each stratum sampled independently using simple random sampling
• Ensures representation from all important subgroups in the population
• Improves precision of estimates compared to simple random sampling
• Reduces sampling error by capturing population diversity
• Requires knowledge of population characteristics for effective stratification

Allocation Methods in Stratified Sampling

• Proportional allocation assigns sample sizes to strata proportional to their size in the population
  • Ensures each stratum represented in proportion to its occurrence in the population
  • Formula: $n_h = n \times \frac{N_h}{N}$ where $n_h$ is sample size for stratum h, n is total sample size, $N_h$ is population size of stratum h, and N is total population size
• Disproportional allocation assigns different sampling fractions to different strata
  • Used when certain strata require oversampling for more precise estimates
  • Allows for cost-effective sampling when some strata more expensive to sample
  • Requires weighting in analysis to account for unequal selection probabilities

Stratification Principles and Effectiveness

• Within-group homogeneity aims for similarity within each stratum
  • Reduces variability within strata, leading to more precise estimates
  • Achieved by selecting stratification variables closely related to the study variables
• Between-group heterogeneity maximizes differences between strata
  • Ensures distinct subgroups captured in the sample
  • Improves overall representation of population diversity
• Sampling error reduced through effective stratification
  • Smaller within-group variance leads to lower overall sampling error
  • Formula for stratified sampling variance: $V(\bar{y}_{st}) = \sum_{h=1}^{L} W_h^2 \frac{s_h^2}{n_h}$ where $W_h$ is the stratum weight, $s_h^2$ is the stratum variance, and $n_h$ is the stratum sample size

Cluster Sampling

Cluster Sampling Methodology

• Cluster sampling selects groups (clusters) of population elements as sampling units
• Clusters typically represent naturally occurring groups (schools, neighborhoods, hospitals)
• All elements within selected clusters included in the sample
• Differs from stratified sampling as heterogeneity within clusters desired
• Useful when individual sampling frame unavailable but cluster-level frame exists
• Often employed in geographically dispersed populations
• Reduces travel and administrative costs in data collection

Cluster Sampling Design and Implementation

• Clusters defined as mutually exclusive and exhaustive groups within the population
• Ideal clusters mirror the overall population characteristics
• Simple random sampling typically used to select clusters
• Sample size determined by number of clusters and average cluster size
• Intraclass correlation coefficient (ICC) measures similarity within clusters
  • Higher ICC indicates greater similarity within clusters, potentially reducing precision
• Design effect quantifies efficiency loss compared to simple random sampling
  • Formula: $DEFF = 1 + (m - 1)\rho$, where m is average cluster size and $\rho$ is ICC

Advanced Cluster Sampling Techniques

• Multi-stage sampling extends cluster sampling to multiple levels
  • First stage selects primary sampling units (PSUs)
  • Subsequent stages select subunits within PSUs
  • Allows for more efficient sampling in large, complex populations
  • Commonly used in national surveys and large-scale studies
• Cost-effectiveness achieved through reduced travel and administrative expenses
  • Fewer locations visited compared to simple random sampling
  • Trade-off between cost savings and potential loss in precision
• Probability proportional to size (PPS) sampling adjusts selection probabilities based on cluster sizes
  • Gives larger clusters higher probability of selection
  • Improves efficiency when cluster sizes vary significantly

Sampling Considerations

Sampling Frame and Coverage

• Sampling frame defines the list or procedure for identifying all elements in the target population
• Comprehensive and accurate sampling frame crucial for valid inference
• Incomplete frames lead to undercoverage bias
  • Systematic exclusion of population subgroups
  • Can result in biased estimates and limited generalizability
• Strategies to improve sampling frame quality include:
  • Regular updates to maintain currency
  • Cross-referencing multiple sources to enhance completeness
  • Employing capture-recapture methods to estimate frame coverage

Precision and Sample Size Determination

• Precision refers to the closeness of sample estimates to the true population parameter
• Influenced by sample size, variability in the population, and sampling design
• Larger sample sizes generally increase precision but also increase costs
• Sample size determination considers:
  • Desired level of precision (margin of error)
  • Confidence level (typically 95% or 99%)
  • Population variability (often estimated from prior studies or pilot data)
  • Expected response rate
• Formula for sample size calculation (simple random sampling): $n = \frac{z^2 \sigma^2}{E^2}$ where z is the z-score for desired confidence level, $\sigma^2$ is population variance, and E is margin of error

Sampling Error and Bias Mitigation

• Sampling error arises from using a sample to estimate population parameters
• Quantified by standard error, which measures variability of the sampling distribution
• Reduced by increasing sample size and employing efficient sampling designs
• Non-sampling errors also impact data quality:
  • Measurement error from inaccurate data collection
  • Non-response bias when sampled units fail to participate
  • Interviewer bias in survey administration
• Strategies to mitigate bias include:
  • Proper training of data collectors
  • Employing standardized measurement instruments
  • Implementing follow-up procedures for non-respondents
  • Using weighting and imputation techniques in analysis