Statistical Methods
Quantify One-Dimensional Distribution Displacement
Define the task
Use the first Wasserstein distance to describe how far probability mass must move between two one-dimensional distributions. Inputs must refer to the same measurement and units. Unlike a difference between means, this distance can detect changes in spread or shape. It is a descriptive comparison, not a probability of a difference or a significance test. The local tool treats each observation within a sample as equally weighted; it does not implement survey weights or paired change analysis.
Carry out the check
Sort both samples and evaluate the absolute difference between their empirical cumulative distributions over each interval between successive pooled observations. Multiply each difference by its interval width and sum. Unequal sample sizes are permitted because each sample is normalized separately. Retain ties and use a prespecified missing-data rule. Record sample sizes, measurement units and collection periods. A larger distance can reflect a change in case mix, measurement practice or selection rather than an intervention effect. Sampling uncertainty requires a separate design-aware method.
Worked hypothetical example
For hypothetical samples A=(0,2) and B=(1,3), both with equal masses, the distance is 1 measurement unit: moving each A value one unit gives B. For A=(0,2) and B=(1,1), the means are equal but the distance remains 1. This second example shows why a mean comparison alone misses the concentration change. The distance from a sample to itself is zero, and exchanging A and B leaves it unchanged; retain these checks in a reproducibility record.
Method checked 16 November 2026 against SciPy documentation: wasserstein_distance.