2.3.2 Bias, Linearity & Stability

Bias, Linearity & Stability Introduction Measurement System Analysis (MSA) evaluates how well a measurement system captures the “true” value of what is being measured. Within MSA, three critical properties are: - Bias - Linearity - Stability These properties determine whether your measurement system is accurate and reliable over the full operating range and over time. Understanding and assessing them is essential before interpreting process data, capability, or improvement results. This article explains what each property means, how it is assessed, and how to interpret and act on the results. --- Bias Concept of Bias Bias is the systematic difference between the average measured value and the reference (true) value for a part or standard. - Positive bias: measurements are consistently higher than the reference. - Negative bias: measurements are consistently lower than the reference. - Zero (or near-zero) bias: measurements are centered around the reference. In formula form: - Bias = Mean of measured values − Reference value Bias reflects accuracy, not spread. A system can be very repeatable (low variation) but still biased (off from the true value). How Bias Is Studied A typical bias study uses: - A known reference standard (or part with a well-established reference value). - Repeated measurements under consistent conditions. General procedure: - Select one part with a known reference value within the normal measurement range. - Have one appraiser measure that part multiple times (often 10 or more). - Calculate: - Mean of the measurements. - Standard deviation of the measurements. - Bias = Mean − Reference. Bias is then evaluated statistically to determine whether it is significantly different from zero. Interpreting Bias Key interpretations: - Magnitude: How large is the bias compared to: - The measurement tolerance. - The process variation. - Statistical significance: Whether the bias differs from zero beyond expected random error. Common criteria and considerations: - If the bias is small relative to the tolerance and not statistically significant, it is usually acceptable. - If the bias is large and statistically significant, the system is systematically miscalibrated. Bias can distort process understanding and decisions: - Underestimation may hide defects or appear to improve performance. - Overestimation may falsely indicate poor process performance. Adjusting for Bias When bias is identified, potential actions include: - Recalibration of the instrument to align with traceable standards. - Adjustment of the measurement method, setup, or environmental conditions. - Mathematical correction (applying a correction factor), used with caution and only if: - Bias is stable and well-characterized. - The organization accepts and documents the correction practice. The goal is not just to detect bias but to ensure the measurement system is accurate enough for its intended use. --- Linearity Concept of Linearity Linearity is the consistency of bias across the entire operating range of the measurement system. A system can have low bias at one level but higher bias at other levels. In other words, linearity answers: - Does the measurement system have the same level of accuracy at low, medium, and high values? If bias changes with the size of the measurement, the relationship between true values and observed values is not linear. How Linearity Is Studied A linearity study evaluates bias at multiple points across the measurement range. Typical steps: - Choose several reference parts or standards: - At least 5 levels spanning the low, mid, and high ends of the operating range. - For each reference: - Record the known reference value. - Measure repeatedly with the same instrument and appraiser. - Compute: - Mean measured value for each reference. - Bias at each reference = Mean measured − Reference. Once bias is calculated at each level, a regression analysis is used: - X-axis: Reference value. - Y-axis: Bias at that reference. The regression line describes how bias changes across the measurement range. Interpreting Linearity Key questions: - Does bias remain roughly constant across the range? - Is there a statistically significant slope in the bias-versus-reference plot? Common interpretations: - Good linearity: - Bias is small at all reference levels. - Regression slope is not significantly different from zero. - The measurement system behaves consistently across the range. - Poor linearity: - Bias increases or decreases as the level changes. - Regression slope is statistically significant. - Errors may be small at one end of the range and large at the other. Practical implications: - If linearity is poor, analyses that depend on measurements at different levels can be distorted. - For example: - Process capability may appear better or worse at certain value ranges. - Control charts may show false shifts at higher or lower levels. Addressing Linearity Issues Actions when linearity is inadequate: - Instrument selection or calibration: - Use an instrument with better linear performance in the required range. - Calibrate specifically across multiple points, not just at a single reference. - Range restrictions: - Restrict the use of the instrument to ranges where linearity is acceptable. - Multiple instruments: - Use different instruments optimized for specific subranges, when practical. Continuous monitoring is needed to ensure improved linearity remains stable over time. --- Stability Concept of Stability Stability is the consistency of measurement accuracy and variation over time. A stable measurement system behaves the same way today, tomorrow, and in the future under similar conditions. Instability means: - Bias can drift over time. - Variation can increase or decrease. - Decisions made using measurements may be unreliable. Stability is related to time-based performance of the measurement system. How Stability Is Studied A stability study typically uses a known reference part (or standard) measured repeatedly over an extended period. Common structure: - Select one or more reference parts with known values. - Measure them at regular time intervals: - Examples: daily, weekly, or per shift. - For each interval: - Record the measured value(s). - Compute statistics as needed (e.g., mean at each time point). These measurements are then analyzed over time. Tools for assessment: - Time series plot: - Measured value vs. time. - Check for trends, shifts, or cycles. - Control charts (e.g., X̄ chart or Individuals chart): - Determine if measurement behavior is consistent and within expected limits. Interpreting Stability Stable measurement systems show: - No systematic trend upward or downward in measurements over time. - No sudden step changes in levels. - Random variation within predictable limits. Unstable systems may show: - Trend: gradual increase or decrease indicating drift (e.g., wear, aging). - Shift: sudden level change (e.g., recalibration, equipment damage, change of environment). - Cycles: recurring patterns (e.g., temperature cycles, shift changes). Consequences of instability: - Process improvements may appear or disappear artificially due to measurement drift. - Control charts may signal false alarms or miss real changes. - Process capability estimates may become invalid over time. Addressing Stability Issues Common corrective actions include: - Maintenance and calibration: - Schedule regular calibration intervals based on observed drift. - Replace or repair aging components. - Environmental control: - Control factors such as temperature, humidity, or vibration that affect readings. - Standard operating procedures: - Standardize how measurements are taken (setup, warm-up time, handling). - Monitoring plan: - Implement periodic stability checks using reference standards. - Update frequency based on criticality and past performance. The objective is to ensure that any change observed in process data is due to the process, not shifting measurement behavior. --- Integrating Bias, Linearity & Stability Why All Three Matter Together Bias, linearity, and stability must all be adequate for the measurement system to be trusted across its full application: - Bias: Are readings close to the true value at a point? - Linearity: Is that closeness maintained across the entire measurement range? - Stability: Does that performance persist over time? Even if one aspect is acceptable, issues in another can still compromise conclusions: - Low bias but poor linearity: - System appears accurate at one level but misrepresents higher or lower values. - Good linearity but instability: - System is accurate across the range initially but drifts, invalidating historical comparisons. - Stable but biased: - System consistently wrong in one direction, leading to systematic misjudgments. For reliable process analysis, all three properties should be checked and controlled. Typical Assessment Sequence A practical sequence when evaluating a measurement system: - Initial calibration check: - Confirm the instrument is within specification at at least one reference point. - Bias study: - Evaluate the average difference from a known reference value. - Linearity study: - Extend the assessment across multiple reference values to examine changes in bias. - Stability study: - Monitor performance over time to ensure results remain valid. If issues are found at any step, refine the system and re-evaluate before relying on measurement data for process decisions. --- Relationship to Other MSA Characteristics Although the focus here is on bias, linearity, and stability, they interact with other core MSA elements such as repeatability and reproducibility. Key relationships: - Bias and linearity deal with accuracy (systematic error). - Repeatability and reproducibility deal with precision (random error). - Stability concerns both accuracy and precision over time. A complete understanding of a measurement system requires that all these aspects are evaluated, but the three properties discussed here uniquely address systematic and time-based behaviors that can significantly distort process analysis if unrecognized. --- Summary Bias, linearity, and stability define how trustworthy a measurement system is in terms of accuracy across range and time. - Bias: - The average difference between measured and reference values. - Indicates whether measurements are systematically high or low. - Assessed with repeated measurements of a reference part and compared to the true value. - Linearity: - Consistency of bias across the operating range. - Evaluated by measuring multiple reference values and analyzing bias versus reference using regression. - Poor linearity means accuracy varies with measurement size. - Stability: - Consistency of measurement performance over time. - Assessed by repeated measurements of reference standards over extended periods and analyzed with time plots or control charts. - Instability indicates drift, shifts, or cycles that can invalidate long-term conclusions. Robust decision-making requires that bias is acceptably small, linearity is adequate across the required range, and stability is maintained over time. Only then can measurement data be used confidently to describe process behavior and support improvement work.