Correlation Metrics for Cross-Site Test Alignment

When scaling hardware production across multiple locations, aligning measurements between the R&D lab (OEM) and the assembly line (Supplier) is vital. To establish a correlation matrix between stations, engineers must choose mathematical correlation metrics.

This guide compares three primary correlation metrics—Pearson, Spearman, and Mutual Information—highlighting their mathematical bounds, assumptions, and application to optical metrology.

1. Pearson Product-Moment Correlation ($r$)

Pearson’s correlation coefficient measures the linear relationship between two continuous variables. It is calculated as:

$$r = \frac{\sum (x_i - \bar{x})(y_i - \bar{y})}{\sqrt{\sum (x_i - \bar{x})^2 \sum (y_i - \bar{y})^2}}$$

Key Characteristics

• Range: $[-1, 1]$, where $1$ is a perfect positive linear relationship, $-1$ is a perfect negative linear relationship, and $0$ indicates no linear correlation.
• Primary Assumption: The relationship is linear, and the data is normally distributed without significant outliers.
• Metrology Application: Standard gauge-to-gage comparison where both instruments are calibrated and share a linear scaling factor (e.g., comparing two identical spectrophotometers measuring optical density).
• Limitation: Highly sensitive to outliers. A single corrupted measurement can inflate or deflate $r$ significantly. It cannot detect non-linear dependencies.

2. Spearman Rank Correlation ($\rho$)

Spearman’s rank correlation is a non-parametric metric that measures the monotonic relationship between two variables. It operates on the ranks of the data rather than the raw values:

$$\rho = 1 - \frac{6 \sum d_i^2}{n(n^2 - 1)}$$

Where $d_i$ is the difference between the ranks of corresponding values of $x$ and $y$, and $n$ is the number of samples.

Key Characteristics

• Range: $[-1, 1]$.
• Primary Assumption: The relationship is monotonic (as $x$ increases, $y$ either consistently increases or consistently decreases, not necessarily at a constant linear rate). It does not require normal distributions.
• Metrology Application: Comparing measurements where one system has a non-linear but consistent response (e.g., comparing a sensor’s raw pixel intensity response against a calibrated photodiode reading before gamma correction is applied).
• Limitation: It ignores the actual physical spacing between data points, focusing only on their relative order.

3. Mutual Information ($I(X; Y)$)

Mutual Information (MI) is an information-theory metric that quantifies the amount of information shared between two variables. It detects any relationship, whether linear, non-linear, or complex:

$$I(X; Y) = \iint p(x,y) \log \frac{p(x,y)}{p(x)p(y)} ,dx,dy$$

Key Characteristics

• Range: $[0, \infty)$, where $0$ means the variables are completely independent. The upper bound depends on the entropy of the variables.
• Primary Assumption: No assumption of functional form (linear, quadratic, exponential, etc.).
• Metrology Application: High-dimensional cross-site diagnostics, such as correlating early component-level optical wavefront aberrations with final system-level imaging MTF, where the relationship is dictated by complex optical physics.
• Limitation: Requires significantly larger datasets to estimate probability density functions ($p(x, y)$) accurately. Calculating MI on small sample sizes (e.g., $n < 30$) can yield biased results.

4. Engineering Metric Selection Guide (Synthetic Scenarios)

Below is a diagnostic guide for selecting the correlation metric:

Summary Caveat

When correlating manufacturing lines, always plot the raw scatter-plot ($X$ vs $Y$) first. A high correlation coefficient does not prove station alignment; it only indicates a predictable mathematical relationship. You must still calculate and apply calibration offsets and gains to align the actual measurements.