🔒 Please log in to use tool features (fill sample / analyze / AI interpretation / export document)
🧪 Normality Test
Does my data look "normal"? Before CpK, t-tests, or Xbar-R charts — pass this gate first
Paste your measurement data as a whole column (separated by space / comma / Tab / newline). Headers, row numbers and other non-numeric text are ignored. ≥8 points recommended; 20–100 is the sweet spot. Good for checking: CpK, t-tests / ANOVA, Xbar-R control charts.
Paste measurement data (one column or one row)
Why check normality first?
· CpK, t-tests, ANOVA, and Xbar control charts all assume data comes from a normal distribution. When data is strongly skewed or heavy-tailed, Cpk can be inflated or distorted, and conclusions become unreliable
· This tool runs two complementary tests: Anderson-Darling (AD) is most sensitive to outliers / heavy tails at both ends and is the industry favorite; K-S (Lilliefors correction) estimates mean and SD from the sample, so it matches real usage better than the "classic K-S". p ≥ 0.05 = cannot reject normality (you're good); p < 0.05 = reject normality (don't force parametric methods on it)
· It also reports skewness g1 / kurtosis g2 as a reference: |z|>1.96 flags marked skewness or leptokurtic / platykurtic shape — skewness is common with one-sided specifications, peakedness often comes from mixed batches
· If normality is rejected, options: ① transform the data (log / Box-Cox / Johnson) then retest; ② switch to non-normal capability analysis (e.g. % out-of-spec rate); ③ investigate mixed batches and outliers before retesting
· This tool runs two complementary tests: Anderson-Darling (AD) is most sensitive to outliers / heavy tails at both ends and is the industry favorite; K-S (Lilliefors correction) estimates mean and SD from the sample, so it matches real usage better than the "classic K-S". p ≥ 0.05 = cannot reject normality (you're good); p < 0.05 = reject normality (don't force parametric methods on it)
· It also reports skewness g1 / kurtosis g2 as a reference: |z|>1.96 flags marked skewness or leptokurtic / platykurtic shape — skewness is common with one-sided specifications, peakedness often comes from mixed batches
· If normality is rejected, options: ① transform the data (log / Box-Cox / Johnson) then retest; ② switch to non-normal capability analysis (e.g. % out-of-spec rate); ③ investigate mixed batches and outliers before retesting
About Normality Test
Normality test online: Anderson-Darling, Shapiro-Wilk and KS tests with histograms, Q-Q plots and p-values to judge whether data are normal.
Learn more about this tool → ← Back to all quality tools