3.2.2 Sampling Techniques & Uses

Sampling Techniques & Uses Purpose of Sampling in Process Improvement Sampling is used to draw reliable conclusions about a process or population without measuring every item or event. It balances cost, speed, and accuracy. Sampling supports: - Estimating process performance (defects, means, variation) - Comparing groups, shifts, or conditions - Monitoring stability and detecting change - Validating improvements with reasonable effort To use sampling correctly, it is essential to understand when and how to sample, which technique to choose, and how sampling affects statistical validity. --- Key Concepts: Population, Sample, and Error Population, Sample, and Parameter vs. Statistic - Population: The complete set of items, units, or observations of interest (all produced units in a month, all transactions, all patients in a period). - Sample: A subset selected from the population for measurement or analysis. - Parameter: A true but usually unknown numerical characteristic of the population (true mean, true proportion defective). - Statistic: A numerical summary calculated from the sample (sample mean, sample standard deviation, sample proportion). Statistics are used to estimate parameters. Good sampling methods increase the accuracy and reliability of these estimates. Sampling Error and Bias - Sampling error: Random difference between a sample statistic and the true population parameter, arising purely because only part of the population is measured. - Sampling bias: Systematic deviation from the true population value due to a flawed sampling method (e.g., only inspecting easy-to-reach units). Key points: - Sampling error cannot be eliminated, but it can be reduced with larger and better-designed samples. - Sampling bias must be avoided through proper technique and adherence to a defined sampling plan. --- When and Why to Use Sampling Reasons to Sample Instead of Measuring All Units Sampling is preferred when: - Measurement is destructive, expensive, or time consuming - The population is large or continuously produced - Faster decision-making is needed - Resources (people, instruments, time) are limited Examples: - Sampling products for tensile strength tests that destroy the item - Sampling call recordings for quality evaluation - Sampling medical records for compliance checks Trade-offs in Sampling Sampling decisions trade off: - Accuracy and precision (smaller error vs. more uncertainty) - Cost and time (larger samples cost more and take longer) - Operational practicality (accessibility of units, scheduling of measurements) Designing a sampling plan involves selecting a technique and sample size that meet accuracy needs at acceptable cost and effort. --- Types of Sampling: Overview Sampling techniques fall into two major groups. - Probabilistic sampling: Each unit has a known, non-zero chance of selection. - Supports valid statistical inference - Required for confidence intervals and hypothesis tests - Non-probabilistic sampling: Selection based on judgment, convenience, or other non-random rules. - Faster and easier - Limited statistical validity; mainly exploratory or preliminary For rigorous analysis and decision-making, probabilistic sampling is preferred. --- Probabilistic Sampling Techniques Simple Random Sampling Simple random sampling gives every unit in the population the same known chance of being selected. - How it works: - Define the population and ensure every unit is listed or accessible. - Use random numbers or randomization tools to select units. - Every subset of a given size has equal probability of selection. - Uses: - Estimating the process mean or proportion defective - General-purpose data collection when population is relatively homogeneous - Advantages: - Statistically straightforward - Easy to analyze and explain - Limitations: - Requires a good sampling frame (list or method to access all units) - May be inefficient if the population has strong subgroups or strata Systematic Sampling Systematic sampling selects units at a regular interval from a sequence. - How it works: - Determine the sampling interval k (e.g., inspect every 10th unit). - Randomly choose a starting point within the first interval. - Select each k-th unit thereafter. - Uses: - Production lines with continuous or high-volume flow - Administrative processes with sequential records - Advantages: - Easier to implement on the floor than pure random sampling - Ensures coverage across time or sequence - Risks: - If there is a periodic pattern in the process that matches k, results can be biased. - Mitigate by: - Randomizing the starting point - Choosing k carefully - Verifying the absence of strong periodic behavior Stratified Sampling Stratified sampling divides the population into non-overlapping subgroups (strata) and samples from each. - Strata are groups that are internally similar but different from each other (e.g., shifts, machine types, regions, product grades). - How it works: - Define relevant strata (e.g., day shift vs. night shift). - Decide sample size per stratum: - Proportional allocation: Sample sizes proportional to stratum sizes. - Disproportional allocation: Over-sample key strata (e.g., high-risk or highly variable ones). - Within each stratum, select units by random or systematic sampling. - Uses: - Comparing performance across groups (machines, shifts, locations) - Improving precision of overall estimates when strata differ - Advantages: - More precise estimates than simple random sampling for the same total sample size - Ensures representation of important subgroups - Key points: - Strata must be clearly defined and non-overlapping. - Analysis must weight strata correctly if disproportional sampling is used. Cluster Sampling Cluster sampling divides the population into clusters, then randomly selects clusters and usually measures all or many units inside them. - Clusters are natural groupings of units (e.g., pallets, batches, branches, days). - How it works: - Define clusters (e.g., a shift, a pallet, a batch). - Randomly select a subset of clusters. - Sample within selected clusters (sometimes all units, sometimes a subset). - Uses: - When it is logistically difficult or costly to sample units spread across many locations - Audits of branches, sites, or days - Advantages: - Reduces travel or setup effort - Practical for geographically or operationally dispersed populations - Limitations: - Units in the same cluster are often similar, so information is redundant. - For the same total number of measured units, cluster sampling is usually less statistically efficient than simple random sampling. - Requires careful analysis that accounts for clustering to avoid underestimating variability. Multi-Stage Sampling Multi-stage sampling combines techniques in multiple steps. - Example structure: - Stage 1: Randomly sample plants. - Stage 2: Within each selected plant, randomly sample lines. - Stage 3: Within each line, systematically sample units. - Uses: - Large, complex populations with several levels (regions, sites, lines, shifts) - Balancing cost and precision step by step - Key idea: - Each stage uses a probabilistic method. - Analysis must consider the design (e.g., clusters and strata) when interpreting results. --- Non-Probabilistic Sampling Techniques Convenience Sampling Convenience sampling selects units that are easiest to access. - Examples: - Inspecting items that are on top of a pallet - Collecting data from staff who are available at the moment - Uses: - Preliminary exploration - Quick checks or pilot measurements - Risks: - High potential for bias - No valid basis for statistical inference about the entire population Judgment (Purposive) Sampling Judgment sampling relies on expert choice of which units to sample. - Examples: - Selecting “worst-case” conditions - Choosing high-risk machines for focused checks - Uses: - Identifying potential problem areas - Stress-testing a process under known challenging conditions - Risks: - Results cannot be generalized statistically to the whole population - Expert judgment may be biased or incomplete --- Designing a Sampling Plan Defining the Objective and Population Before selecting a sampling method, clarify: - Objective: - Estimate a mean or proportion - Detect a difference between groups or conditions - Verify conformance to a requirement - Population definition: - Time frame (e.g., all units produced this quarter) - Scope (e.g., all transactions in region X) - Inclusion and exclusion rules (e.g., exclude units under rework) A vague population definition makes the sample ambiguous and weakens conclusions. Choosing the Sampling Frame and Unit - Sampling frame: The operational way you access or list population units (e.g., production schedule, transaction database, physical lots). - Sampling unit: The element that will be selected and measured (e.g., a part, a batch, a transaction, a patient visit). Ensure that: - The frame covers the entire defined population with minimal omissions or duplications. - The sampling unit matches the measurement strategy (e.g., if variability is within batch, then the sampling unit may need to be the individual item rather than the batch). Deciding When to Sample (Time Basis) Sampling across time affects representativeness. - Options: - Random times across the period of interest - Systematic times (e.g., every hour, every 100 units) - Stratified by time segments (e.g., morning vs. evening, weekdays vs. weekends) Important considerations: - Cover different operating conditions (shifts, product mix, environmental conditions). - Avoid clustering samples in “easy” times that do not represent typical or worst-case behavior. --- Sample Size Considerations Factors Influencing Sample Size Sample size depends on: - Required precision (width of confidence interval) - Desired confidence level (e.g., 95%) - Expected variability (standard deviation for continuous data, p(1−p) for proportions) - Population size (often negligible if population is large compared to sample) - Statistical test or estimation method (e.g., estimation vs. hypothesis testing) Higher precision, higher confidence, and higher variability all increase required sample size. Practical Approaches to Sample Size For rigor, formulas or software are typically used, but the core ideas are: - For a mean (continuous data): - Larger standard deviation requires larger n. - Narrower desired margin of error requires larger n. - For a proportion (attribute data): - When the true proportion is unknown, using p = 0.5 gives the largest required n (worst case). Practical steps: - Use prior data or pilot samples to estimate variability. - Start with a preliminary sample, estimate variability, then refine required n. - Balance statistical needs with cost and feasibility. Finite Population Correction (FPC) When the sample is a significant fraction of the population (often > 5–10%), sampling without replacement reduces variability, and a finite population correction can reduce the required sample size. Key points: - FPC is relevant mainly when populations are small. - For large populations, FPC has negligible effect and is often ignored. --- Minimizing Bias and Ensuring Representativeness Sources of Sampling Bias Common causes of bias: - Selection bias: - Choosing only accessible or “good” units - Excluding difficult areas or times - Periodic or patterned sampling: - Sampling at a fixed interval that coincides with process cycles - Nonresponse or missing data: - Failure to measure selected units, replacing them with convenient ones Controls to Reduce Bias To maintain validity: - Use clearly documented, repeatable selection rules. - Randomize selection wherever possible. - Avoid substituting units unless predefined rules are in place. - Train those collecting data on the sampling plan. - Audit adherence to the plan, especially during early implementation. --- Uses of Sampling in Process Analysis Estimating Process Performance Sampling supports estimating: - Means and variation: - Average cycle time, average diameter, standard deviation of measurements - Proportions and counts: - Proportion defective, defect rate per unit, error frequency Typical uses: - Baseline assessment before changes - Post-improvement confirmation of performance - Ongoing monitoring to ensure stability Comparing Groups or Conditions Sampling enables comparisons between: - Shifts or work teams - Machines or lines - Locations or branches - Before vs. after an intervention Key considerations: - Use appropriate sampling design (often stratified or balanced sampling across groups). - Ensure comparable conditions except for the factor being compared. - Use equal or appropriately weighted sample sizes for fair comparison. Control and Monitoring Beyond initial studies, sampling is used to: - Feed control charts with regularly sampled data - Detect trends or shifts over time - Verify that process performance remains within control limits Sampling for monitoring should: - Be consistent over time in method and frequency - Cover typical and critical operating conditions - Use a plan robust to minor operational disruptions --- Special Considerations for Continuous Processes Time-Based and Product-Based Sampling For continuous flow processes (chemicals, utilities, high-speed manufacturing), sampling decisions include: - Time-based sampling: - Collecting samples at fixed time intervals (e.g., every hour) - Quantity-based sampling: - Sampling per amount produced (e.g., per 10,000 units) Choice depends on: - Speed of process changes - Measurement cost and time - Storage and transport of samples Handling Autocorrelation Measurements taken close together in time can be correlated (autocorrelation). Implications: - Effective information content may be lower than sample size suggests. - Standard analysis methods that assume independence can underestimate variability. Mitigation: - Increase time between sampled units to reduce correlation. - Use sampling spread across the production period. - Be cautious in interpreting overly smooth data trends. --- Practical Steps to Implement a Sampling Plan Step 1: Clarify the Question Define what needs to be learned: - Estimate a level (e.g., average, proportion) - Detect a difference or change - Confirm compliance or conformance Step 2: Define Population, Frame, and Unit Specify: - Exact population (scope and time) - Sampling frame (where/how samples will be drawn) - Sampling unit (what exactly is selected and measured) Step 3: Choose Sampling Technique Select: - Simple random or systematic for homogeneous processes - Stratified when key subgroups must be represented or compared - Cluster or multi-stage when accessibility or cost is a concern - Non-probabilistic methods only for exploratory or limited purposes Step 4: Decide Sample Size and Timing - Estimate variability where possible. - Decide required precision and confidence. - Distribute sample collection across relevant times, locations, or conditions. Step 5: Document and Train Prepare a clear sampling plan: - Objectives and scope - Selection rules and intervals - Handling of missing or unusable samples - Responsibilities for data collection and recording Ensure everyone involved understands and can follow the plan. Step 6: Monitor and Adjust During execution: - Check adherence to the plan. - Note deviations or issues. - Adjust method or frequency if operational realities require changes, documenting impacts on interpretation. --- Summary Sampling techniques and their uses center on obtaining reliable information about a process or population without measuring everything. Effective sampling: - Starts with a clear definition of the objective, population, and sampling unit. - Relies primarily on probabilistic methods (simple random, systematic, stratified, cluster, multi-stage) to support valid inference. - Recognizes that sample size, variability, and desired precision are tightly linked. - Minimizes bias through careful design, documentation, and adherence to a sampling plan. - Supports estimation, comparison, and monitoring of process performance in a practical and resource-efficient way. A solid grasp of these principles enables sound decisions based on sampled data while understanding the limitations and strengths of the conclusions drawn.