Assumption-Free Statistics

Every classical statistical method carries assumptions. The t-test assumes normality. ANOVA assumes equal variances. Regression assumes Gaussian errors with constant variance. Capability indices assume the process distribution is known. When these assumptions are violated — which is common in manufacturing — the methods produce incorrect results.

The traditional approach to violated assumptions is to test for them (normality tests, variance tests) and then either transform the data, use a different method, or proceed with a caveat. This creates a complex decision tree that requires significant statistical expertise and introduces subjective choices at each branch.

Assumption-free methods eliminate this decision tree. They produce valid results regardless of the underlying distribution, removing a major source of analytical subjectivity and error from the quality workflow.

"Assumption-free" is the defining characteristic of EntropyStat's analytical approach. The Machine Gnostics framework that powers EntropyStat uses gnostic algebra — a deterministic mathematical system based on error geometry and entropy optimization — instead of the probabilistic framework that underlies all classical parametric statistics.

Concretely, this means: the EGDF does not assume your data is normal, Weibull, or any other distribution. The capability indices do not assume normality. The control limits do not assume the Central Limit Theorem holds for your subgroup size. The tolerance intervals do not assume Gaussian tails. Each calculation works directly with the empirical distribution as captured by the EGDF.

This is more than academic. It means a quality team can deploy EntropyStat across all monitored characteristics with a single analytical configuration. There is no need to maintain a mapping of which characteristic uses which distribution family, no need to re-validate distribution assumptions when processes change, and no need for statistical expertise to select and justify the correct method for each case. The assumption-free approach reduces analytical complexity to a single question: "What does the data show?"