5.2.5 P Chart

P Chart Introduction A P Chart (proportion chart) is an attribute control chart used to monitor the proportion of nonconforming units in a process over time when the sample size may vary. It is a core tool for assessing process stability for proportion-type data. Use a P Chart when: - The data are attribute (conforming / nonconforming). - Each item in the sample is classified as defective or not defective. - You track the fraction or percentage nonconforming, not the count. - Sample sizes can be equal or unequal from subgroup to subgroup. Typical applications: - Proportion of orders with errors. - Percentage of invoices with missing information. - Fraction of products failing final inspection. --- Foundations of P Chart Data Attribute Data and Defectives For a P Chart, the basic data structure is: - Each unit is either defective (nonconforming) or non-defective (conforming). - A unit is counted once, regardless of how many defects it has. Key distinctions: - Defective unit: The whole unit is judged unacceptable. - Defect: A specific nonconformity on a unit. P Charts use defectives, not defects. For each subgroup (time period, batch, or lot), you record: - Sample size: nᵢ. - Number of defectives: dᵢ. - Proportion defective: pᵢ = dᵢ / nᵢ. Statistical Basis The P Chart relies on the binomial distribution: - Each unit has two outcomes: defective or non-defective. - Probability of defective: p (assumed constant under stable conditions). - For each subgroup: dᵢ ~ Binomial(nᵢ, p). For large enough sample sizes, the binomial distribution can be approximated by a normal distribution: - Mean of pᵢ: E(pᵢ) = p. - Variance of pᵢ: Var(pᵢ) = p(1 − p) / nᵢ. This normal approximation is used to construct control limits. --- When to Use a P Chart Appropriate Situations Use a P Chart when: - Each sample consists of a number of units, each unit only pass/fail. - The main interest is proportion nonconforming per subgroup. - Subgroup sizes may vary from one time period to another. Examples: - Daily proportion of calls with incorrect routing. - Weekly fraction of shipments with damage claims. - Proportion of patient charts with missing signatures. Comparison with Other Attribute Charts P Chart vs other attribute charts: - P Chart: Proportion defective; varying or constant n; each unit: defective / not. - NP Chart: Count of defectives; constant n only; tracks dᵢ directly. - C Chart: Count of defects; constant area of opportunity; Poisson model. - U Chart: Defects per unit; varying opportunity size; Poisson model. Choose P Chart when: - The metric is a fraction (or percentage) of defective units. - Subgroup sizes are not constant, or you want limits tied to actual nᵢ. --- Constructing a P Chart Step 1: Collect and Organize Data For each subgroup i = 1, 2, …, k: - Record nᵢ: number of units inspected. - Record dᵢ: number of defective units. - Compute pᵢ = dᵢ / nᵢ. Data table structure: - Subgroup (i) - nᵢ (sample size) - dᵢ (defectives) - pᵢ (proportion defective) Ensure: - Subgroups represent logical, time-ordered snapshots (e.g., by shift, day). Step 2: Compute Overall Proportion Calculate the overall proportion defective: - Total defectives: D = Σ dᵢ over all subgroups. - Total inspected: N = Σ nᵢ over all subgroups. - Overall proportion: p̄ = D / N This p̄ is the estimate of the long-term process proportion defective under current conditions. Step 3: Compute Control Limits For each subgroup i, with size nᵢ: - Standard error of pᵢ: σₚᵢ = √[ p̄(1 − p̄) / nᵢ ] Using 3-sigma limits: - UCLᵢ = p̄ + 3 × σₚᵢ - LCLᵢ = p̄ − 3 × σₚᵢ If LCLᵢ is negative, set: - LCLᵢ = 0 If UCLᵢ is greater than 1, set: - UCLᵢ = 1 Because nᵢ can vary, each subgroup has its own control limits, forming a “funnel” shape on the chart. Step 4: Plot the Chart Create three elements: - Center line: horizontal line at p̄. - Points: pᵢ versus subgroup index (or time). - Control limits: UCLᵢ and LCLᵢ for each subgroup. Interpret visually: - Look for points outside control limits. - Look for non-random patterns or trends within the limits. --- Assumptions and Data Requirements Core Assumptions For valid interpretation: - Independent units within each subgroup and across subgroups. - Consistent inspection criteria over time. - Binomial conditions reasonably satisfied: - Each unit can be classified as defective / non-defective. - Probability p is approximately constant when the process is stable. Sample Size Considerations To justify the normal approximation: - A common practical guideline: - For each subgroup i, ensure: - nᵢ × p̄ ≥ 5 and nᵢ × (1 − p̄) ≥ 5 or at least: - nᵢ ≥ 50 and p̄ not extremely close to 0 or 1. If sample sizes are very small: - The normal-approximation-based limits may be inaccurate. - The chart may show excessive false signals or be insensitive. --- Interpreting a P Chart Basic Stability Assessment A process is considered statistically stable when: - All pᵢ values lie within their corresponding UCLᵢ and LCLᵢ. - The pattern of points appears random, with no systematic structure. Signals of special cause variation: - Any point outside UCLᵢ or below LCLᵢ. - Clusters or non-random patterns inside the limits. Common Patterns and Their Meanings Typical patterns to watch: - Point above UCLᵢ: - Sudden increase in proportion defective. - Possible special cause introducing more nonconformities. - Point below LCLᵢ: - Sudden improvement; may indicate an assignable cause. - Run of points on one side of p̄: - Potential shift in process level. - Trend (steady increase or decrease): - Gradual drift in process performance. - Cycles or periodic patterns: - Seasonal or schedule-related effects. Interpretation focus: - Identify whether the process proportion defective is stable. - If unstable, determine whether the shift is beneficial or harmful. - Link signals to real process events or changes. --- Dealing With Varying Sample Sizes Impact of Variable nᵢ When nᵢ changes: - The width of the control limits changes: - Larger nᵢ → smaller σₚᵢ → narrower limits. - Smaller nᵢ → larger σₚᵢ → wider limits. - This is why P Chart limits often appear funnel-shaped. Implications: - Points from large samples provide more precise information. - Large-sample subgroups may more easily show statistically significant deviations. Practical Tips To manage variability in nᵢ: - Try to keep nᵢ reasonably consistent when operationally feasible. - Avoid extremely small nᵢ, which reduce chart sensitivity. - If nᵢ varies widely: - Be cautious when visually comparing points—interpret relative to each point’s own limits. - Recognize that a point from a large sample is more reliable evidence than one from a very small sample. --- P Chart vs Capability and Performance Stability vs Capability A P Chart addresses stability, not capability. - Stability: - Does the proportion defective vary only due to common causes? - Is the process predictable over time? - Capability: - Is the stable process meeting customer or specification requirements? Typical sequence: - First use the P Chart to determine if the process is stable. - For a stable process, use p̄ as the long-term defect proportion. - Compare p̄ (or derived yield) to performance targets or requirements. Linking to Defect Rates From p̄: - Percentage defective: 100 × p̄ % - Yield: 1 − p̄ - Defects per million opportunities (when “defective unit” maps cleanly to “opportunity”): - DPMO ≈ 1,000,000 × p̄ These conversions are valid only when the P Chart indicates the process is stable and p̄ is representative. --- Practical Issues in Using P Charts Data Integrity and Operational Definitions Accurate P Chart analysis depends on consistent definitions: - Clear definition of defective: - Objective criteria on what is counted as a nonconforming unit. - Consistency in inspection: - Same rules applied across operators, shifts, and sites. - Consistent time or subgrouping: - Subgroups should reflect natural process windows where conditions are reasonably homogeneous. If definitions or inspection methods change: - The chart may show apparent shifts that are not real process changes. - Consider starting a new chart after major definition changes. Reconciling Low-Defect Processes When p̄ is very small: - Many subgroups may show pᵢ = 0. - Control limits may be very close to zero. - Detecting small changes becomes challenging. Options within the P Chart framework: - Increase subgroup sizes to accumulate more defectives over time. - Extend the data collection period before finalizing p̄ and limits. --- P Chart Design Choices Subgrouping Strategy Decide how to form subgroups: - By time (e.g., each day, each shift). - By batch or lot. - By transaction count (e.g., per 100 orders). Consider: - Subgroups should capture short-term process conditions. - Avoid mixing very different operating conditions within single subgroups. Trade-offs: - Larger subgroups: - More precise estimates of pᵢ. - Slower detection of short-term shifts. - Smaller subgroups: - Faster responsiveness. - Higher variability in pᵢ. Choose subgrouping that reflects how decisions will be made and how quickly action is needed. Initial vs Ongoing Charts During initial use: - Collect baseline data. - Calculate p̄ and limits using this baseline period. - Confirm that baseline period is reasonably stable. For ongoing monitoring: - Plot new subgroups against fixed baseline limits. - Recalculate limits only when: - The process has been intentionally changed and restabilized. - There is clear evidence of a new stable level. --- Common Pitfalls and Misinterpretations Overreacting to Common Cause Variation Typical mistakes: - Treating every up or down movement as a problem. - Adjusting the process frequently without valid signals. Consequences: - Increased instability. - Over-adjustment and more variability. Use the P Chart signals: - Only act on statistically meaningful patterns (points beyond limits or clear non-random patterns). - Avoid tampering with a stable process based on random fluctuations. Ignoring Special Cause Signals Another error: - Dismissing points outside control limits as “outliers” and not investigating. This leads to: - Lost opportunities for improvement or risk mitigation. - Misunderstanding of real process behavior. Effective use: - Each signal should trigger: - Investigation for root causes. - Appropriate corrective or preventive action. - Documentation of what was learned. --- Summary A P Chart is a control chart for monitoring the proportion of defective units in a process, especially when subgroup sizes vary. It is built on binomial assumptions and uses the overall proportion defective (p̄) and subgroup-specific sample sizes (nᵢ) to calculate 3-sigma control limits for each subgroup. Core steps: - Collect nᵢ and dᵢ, compute pᵢ and p̄. - Calculate subgroup-specific UCLᵢ and LCLᵢ using p̄ and nᵢ. - Plot pᵢ over time with corresponding control limits. - Interpret signals for special causes and assess process stability. A sound P Chart application requires: - Clear definitions of defectives and consistent inspection. - Adequate, reasonably sized subgroups. - Careful interpretation of patterns, avoiding both overreaction and neglect. Used correctly, the P Chart provides a reliable, statistically grounded method to track and understand the stability of proportion nonconforming in attribute data processes.