Quick Answer: Anderson Darling Test Result

The Anderson Darling Test checks whether a sample follows a specified theoretical distribution, most commonly the normal distribution. The null hypothesis says the data follow the specified distribution. The alternative hypothesis says the data do not follow the specified distribution. The test statistic is usually written as A². A larger A² statistic means a stronger departure from the reference distribution. The test is particularly sensitive to departures in the lower and upper tails.

In a normality-testing workflow, the Anderson Darling Test should be interpreted with the distribution fit chart, normal Q-Q plot, empirical CDF versus fitted normal CDF chart, tail-weighted contribution chart, AD statistic across variables chart, and group AD statistic comparison chart. These visuals show whether the problem is a general shape mismatch, tail behavior, variable-specific non-normality, or group-specific distribution difference.

What Is the Anderson Darling Test?

The Anderson Darling Test is a statistical goodness-of-fit test used to determine whether sample data follow a specified probability distribution. In applied research, it is most often used as a normality test. It compares the empirical cumulative distribution function of the sample with the theoretical cumulative distribution function of the selected distribution.

The special feature of the Anderson Darling Test is its tail sensitivity. While some normality tests focus mostly on the middle of the distribution or the single largest gap, the Anderson Darling Test gives more weight to differences in the lower and upper tails. This makes it especially useful when the main issue is outliers, heavy tails, light tails, skewed tail behavior, or extreme-score mismatch.

For a full normality analysis, use the Anderson Darling Test with the Q-Q Plot Normality Check, Q-Q Plot Normality Check 2, P-P Plot Normality Check, Histogram Interpretation, and Box Plot Interpretation. These plots show the shape behind the formal test statistic.

Why the Anderson Darling Test Matters

The Anderson Darling Test matters because many statistical methods assume that variables, residuals, or errors are approximately normally distributed. This assumption can affect t tests, ANOVA, regression, confidence intervals and many parametric procedures. However, not all non-normality looks the same. Sometimes the center of the distribution looks acceptable, but the tails are too heavy or too light. The Anderson Darling Test is useful because it is designed to detect these tail problems.

For example, a student performance variable may look roughly bell-shaped in the center but show unusual extreme scores at the low end or high end. A histogram might not make that tail problem obvious, but the Anderson Darling Test and tail-weighted contribution plot can show it clearly. This is why the test is often stronger than relying on one visual alone.

Because the Anderson Darling Test is sensitive, it should be interpreted with sample size and practical shape. In large samples, small deviations may become statistically significant. In small samples, visual diagnostics and additional tests such as the Ryan-Joiner Test can help confirm the conclusion.

Anderson Darling Test Formula

The Anderson Darling Test statistic is usually written as A². It compares sorted sample values with the cumulative probabilities expected under the reference distribution.

Where:

Why the Formula Is Tail Sensitive

The Anderson Darling formula uses logarithmic terms involving F(x) and 1 − F(x). These terms become very important in the lower and upper tails, where cumulative probabilities are close to 0 or 1. That is why the Anderson Darling Test detects tail departures more strongly than many simpler distribution-fit checks.

Null and Alternative Hypotheses for the Anderson Darling Test

The Anderson Darling Test is a formal hypothesis test. For normality testing, the null hypothesis states that the sample comes from a normal distribution. The alternative states that the sample does not come from a normal distribution.

If the normality decision affects a parametric analysis, connect this result with the Central Limit Theorem, Confidence Interval interpretation, Effect Size, and the specific test being used, such as a One-Tailed T Test, One Sample Z Test, or One Proportion Z Test.

Assumptions and Decision Logic

The Anderson Darling Test is used to assess distribution fit, but its interpretation still depends on proper data handling. The variable should be numeric, observations should be independent, and the chosen reference distribution should make sense for the variable. For normality testing, use a variable or residual series where normality is actually relevant.

For a complete assumption-testing workflow, do not confuse normality with variance assumptions. Use tests such as the Levene Test, Brown-Forsythe Test, Cochran C Test, and Goldfeld-Quandt Test when the research question is about equal variance or heteroscedasticity rather than normality.

Dataset and Variables Used

The worked Anderson Darling Test example uses student performance variables. The main normality workflow can be applied to a numeric variable such as G3 final grade, while the cross-variable chart compares Anderson Darling statistics across multiple numeric variables. A group comparison chart also shows whether distribution-fit concerns differ across categories.

Before running the Anderson Darling Test, review each variable with Descriptive Statistics, Frequency Distribution, Histogram Interpretation, Five Number Summary, Coefficient of Variation, and Box Plot Interpretation.

Verified SPSS Output Interpretation for the Anderson Darling Test

The SPSS output PDF supports the Anderson Darling Test normality workflow for this post. In SPSS-style reporting, the result should be interpreted with the test statistic, p-value or critical-value decision, histogram, Q-Q plot, empirical CDF interpretation, and tail-focused explanation. Because SPSS workflows can vary by extension, syntax, custom calculation or integrated output, the key reporting point is the same: state whether the observed distribution differs from the fitted normal distribution.

How to Read the SPSS Output

SPSS Decision Logic

If the SPSS-supported Anderson Darling p-value is below .05, report that the test found a significant departure from normality. If the p-value is above .05, report that the test did not provide sufficient evidence against normality. If the output uses critical values instead of a p-value, compare the statistic with the critical value at the chosen significance level.

Python Chart-by-Chart Interpretation

The Python charts provide a full Anderson Darling Test visual workflow. They show distribution fit, normal Q-Q plot, empirical CDF versus fitted normal CDF, tail-weighted contributions, AD statistic across variables and group AD statistic comparison.

Python Chart 1: Distribution Fit for Anderson Darling Test

This chart gives the first visual explanation of the Anderson Darling Test. It compares the sample distribution with a fitted normal curve. If the bars or density shape align closely with the fitted curve, normality is more plausible. If the chart shows skewness, heavy tails, outliers, multiple peaks or strong tail mismatch, the Anderson Darling statistic may become larger.

The distribution fit chart is especially useful because the Anderson Darling Test is sensitive to tail behavior. A distribution can look acceptable near the center but still fail because of lower-tail or upper-tail deviations. This chart provides the practical shape context before reviewing the more technical tail-weighted contribution chart.

Decision/reporting conclusion: Use this Python distribution fit chart as the first visual support for the Anderson Darling decision. If the shape visibly departs from the fitted normal curve, report that the chart supports possible non-normality.

Python Chart 2: Normal Q-Q Plot

The Q-Q plot is one of the best visual companions to the Anderson Darling Test. If the points follow the diagonal line, the distribution is close to normal. If the points curve away from the line, especially at the lower or upper ends, the distribution has tail departures. Since Anderson Darling is tail-sensitive, Q-Q plot tail deviations are highly relevant.

A downward or upward curve can suggest skewness. Points far from the diagonal at the ends can suggest heavy tails, light tails or outliers. This chart helps explain why the Anderson Darling statistic may be large even when the histogram looks acceptable.

Decision/reporting conclusion: Use this Q-Q plot to explain whether the Anderson Darling result is driven by center mismatch, lower-tail deviation, upper-tail deviation or extreme observations.

Python Chart 3: Empirical CDF vs Fitted Normal CDF

This chart shows the distribution comparison at the cumulative probability level. The empirical CDF shows the observed cumulative pattern of the sample. The fitted normal CDF shows the expected cumulative pattern if the data were normally distributed. The Anderson Darling Test measures weighted differences between these cumulative functions.

Unlike a simple maximum-gap view, Anderson Darling considers differences across the distribution and weights tail discrepancies more heavily. Therefore, this chart should be read carefully at the lower and upper ends as well as the middle.

Decision/reporting conclusion: Use this ECDF chart to explain where the observed cumulative distribution differs from the fitted normal distribution. Tail separation supports the logic of an Anderson Darling warning.

Python Chart 4: Tail-Weighted Contributions

This is the most distinctive Anderson Darling chart. It shows which parts of the distribution contribute most strongly to the Anderson Darling statistic. Because the test weights tail deviations heavily, the lower and upper tails can make large contributions even when the center looks acceptable.

If the chart shows high contribution in the tails, the analyst should inspect outliers, skewness, heavy tails and possible transformation needs. Tail-weighted contribution is the reason Anderson Darling often detects non-normality missed by a simple visual glance.

Decision/reporting conclusion: Use this chart to explain whether the Anderson Darling result is mainly driven by lower-tail mismatch, upper-tail mismatch or general distribution shape. This is the key chart for tail-sensitive reporting.

Python Chart 5: Anderson Darling Statistic Across Variables

This chart compares Anderson Darling statistics across variables. Larger values indicate stronger departure from the selected reference distribution. This is important because normality is not a property of the whole dataset; it is specific to the variable or residual series being analyzed.

For example, a final grade variable may show different normality behavior than an absences variable. A count-like or skewed variable may produce a much larger Anderson Darling statistic because its distribution has tails or boundaries that differ from normality.

Decision/reporting conclusion: Use this chart to identify which variables need deeper normality review, transformation, robust analysis, or nonparametric alternatives.

Python Chart 6: Group AD Statistic Comparison

This chart compares Anderson Darling statistics across groups. A variable can be close to normal in one group but clearly non-normal in another. This matters for group comparisons, ANOVA, regression subgroup diagnostics and assumption reporting.

If one group has a much larger AD statistic, the analyst should inspect that group’s distribution separately. The issue may be a small subgroup sample size, outliers, ceiling effects, floor effects or true distributional difference.

Decision/reporting conclusion: Use this chart to decide whether normality concerns are general across the dataset or concentrated in particular groups.

R Chart-by-Chart Validation

The R charts validate the Python and SPSS Anderson Darling Test workflow using a separate analytical environment. The same six visual diagnostics are provided: distribution fit, normal Q-Q plot, empirical CDF versus fitted normal CDF, tail-weighted contributions, AD statistic across variables and group AD statistic comparison.

R Chart 1: Distribution Fit for Anderson Darling Test

The R distribution fit chart validates the Python view of the observed sample shape. It shows whether the data follow the fitted normal curve or whether there are visible departures in the center, spread or tails. This provides a visual foundation for interpreting the Anderson Darling statistic.

Decision/reporting conclusion: Use this R chart to confirm whether the distribution fit pattern is reproducible across software workflows.

R Chart 2: Normal Q-Q Plot

The R Q-Q plot validates the tail and quantile interpretation. If the points follow the diagonal line, the normality assumption looks more reasonable. If the points depart strongly at the ends, the Anderson Darling Test may detect non-normality because of tail mismatch.

Decision/reporting conclusion: Use this R Q-Q plot to confirm whether tail departures support the Anderson Darling result.

R Chart 3: Empirical CDF vs Fitted Normal CDF

The R ECDF chart confirms the cumulative-distribution comparison. It shows how the observed cumulative distribution tracks or deviates from the fitted normal cumulative distribution. This chart is the bridge between visual distribution checking and the formal Anderson Darling formula.

Decision/reporting conclusion: Use this chart to explain whether the empirical cumulative pattern supports or contradicts the fitted normal reference.

R Chart 4: Tail-Weighted Contributions

The R tail-weighted contribution chart confirms which portions of the distribution influence the Anderson Darling statistic most strongly. If the tails contribute heavily, the normality issue may involve outliers, heavy tails or skewed extremes rather than only central shape.

Decision/reporting conclusion: Use this chart to explain whether the Anderson Darling result is tail-driven and whether follow-up checks should focus on extreme values.

R Chart 5: Anderson Darling Statistic Across Variables

The R cross-variable chart validates which variables show the strongest departure from normality. Variables with higher AD statistics should receive closer attention in reports, especially if they are used in parametric tests, regression residual checks or confidence interval calculations.

Decision/reporting conclusion: Use this chart to prioritize which variables require transformation, robust methods or additional distribution diagnostics.

R Chart 6: Group AD Statistic Comparison

The R group comparison chart validates whether normality problems differ by subgroup. This is important when the next analysis compares groups or includes group predictors. If one group has a much larger Anderson Darling statistic, report that normality concerns may be group-specific.

Decision/reporting conclusion: Use this chart to decide whether normality is acceptable across all groups or whether subgroup-specific caution is needed.

SPSS, Python, R and Excel Workflows for the Anderson Darling Test

A complete Anderson Darling Test workflow includes a formal statistic and multiple charts. The analyst should not rely on one output table only. The best workflow includes descriptive statistics, distribution fit, Q-Q plot, empirical CDF comparison, tail-weighted contributions, and final p-value or critical-value decision.

SPSS Workflow

Python Workflow

R Workflow

Excel Workflow

SPSS, Python, R and Excel Code for the Anderson Darling Test

SPSS Syntax for Anderson Darling Supporting Workflow

* Anderson Darling Test supporting normality workflow in SPSS.
* Replace G3 with your selected numeric variable.
* SPSS may require an extension, custom procedure, or R/Python integration for direct AD p-values.

TITLE "Anderson Darling Test Normality Workflow".

EXAMINE VARIABLES=G3
  /PLOT BOXPLOT HISTOGRAM NPPLOT
  /COMPARE GROUPS
  /STATISTICS DESCRIPTIVES
  /CINTERVAL 95
  /MISSING LISTWISE
  /NOTOTAL.

FREQUENCIES VARIABLES=G3
  /STATISTICS=MEAN MEDIAN STDDEV SKEWNESS SESKEW KURTOSIS SEKURT MINIMUM MAXIMUM
  /HISTOGRAM NORMAL
  /ORDER=ANALYSIS.

* Optional: save standardized values for external Anderson Darling calculation.
DESCRIPTIVES VARIABLES=G3
  /SAVE
  /STATISTICS=MEAN STDDEV MIN MAX.

OUTPUT EXPORT
  /CONTENTS EXPORT=VISIBLE
  /PDF DOCUMENTFILE="Anderson-Darling-Test-SPSS-Output.pdf".

Python Code for the Anderson Darling Test

import pandas as pd
import numpy as np
from scipy import stats

df = pd.read_csv("dataset.csv")

x = pd.to_numeric(df["G3"], errors="coerce").dropna().to_numpy()

# Anderson-Darling test for normality
ad_result = stats.anderson(x, dist="norm")

print("Anderson-Darling statistic:", ad_result.statistic)
print("Critical values:", ad_result.critical_values)
print("Significance levels:", ad_result.significance_level)

# Decision by critical values
for sig, crit in zip(ad_result.significance_level, ad_result.critical_values):
    decision = "Reject normality" if ad_result.statistic > crit else "Do not reject normality"
    print(f"{sig}% level: critical value={crit:.4f}; {decision}")

# Empirical CDF and fitted normal CDF for plotting
mu = np.mean(x)
sigma = np.std(x, ddof=1)

x_sorted = np.sort(x)
n = len(x_sorted)

empirical_cdf = np.arange(1, n + 1) / n
fitted_cdf = stats.norm.cdf(x_sorted, loc=mu, scale=sigma)

# Avoid exact 0 or 1 probabilities for log terms
eps = 1e-10
fitted_cdf_clipped = np.clip(fitted_cdf, eps, 1 - eps)

# Manual Anderson-Darling statistic for learning
i = np.arange(1, n + 1)
A2 = -n - np.mean((2 * i - 1) * (
    np.log(fitted_cdf_clipped) +
    np.log(1 - fitted_cdf_clipped[::-1])
))

print("Manual A-squared:", A2)

R Code for the Anderson Darling Test

# Anderson Darling Test in R

# install.packages("nortest")
library(nortest)

df <- read.csv("dataset.csv")

x <- as.numeric(df$G3)
x <- na.omit(x)

# Anderson-Darling normality test
ad_result <- ad.test(x)
print(ad_result)

# Fitted normal parameters
mu <- mean(x)
sigma <- sd(x)

# Empirical CDF and fitted normal CDF
x_sorted <- sort(x)
n <- length(x_sorted)

empirical_cdf <- (1:n) / n
fitted_cdf <- pnorm(x_sorted, mean = mu, sd = sigma)

# Manual AD statistic for explanation
eps <- 1e-10
fitted_cdf <- pmin(pmax(fitted_cdf, eps), 1 - eps)

i <- 1:n
A2_manual <- -n - mean((2 * i - 1) * (
  log(fitted_cdf) + log(1 - rev(fitted_cdf))
))

print(A2_manual)

# Q-Q plot
qqnorm(x)
qqline(x, col = "blue", lwd = 2)

Excel Formulas for an Anderson Darling-Style Manual Check

Assume sorted values are in A2:A650.

Sample size:
=COUNT(A2:A650)

Mean:
=AVERAGE(A2:A650)

Standard deviation:
=STDEV.S(A2:A650)

Fitted normal CDF in B2:
=NORM.DIST(A2,$Mean_Cell$,$SD_Cell$,TRUE)

Reverse fitted normal CDF:
Use the fitted CDF value from the opposite end of the sorted list.

Log term for row i:
=LN(B2)+LN(1-Reversed_CDF)

Weighted term:
=(2*i-1)*Log_Term

Anderson-Darling statistic:
=-n-(1/n)*SUM(Weighted_Terms)

Decision:
Compare A² with Anderson Darling critical values or use SPSS, Python, or R for a formal p-value.

APA Reporting Wording for the Anderson Darling Test

APA reporting for the Anderson Darling Test should include the test name, variable, statistic, p-value or critical-value decision, and interpretation. Because the Anderson Darling Test is tail-sensitive, a strong report should also mention Q-Q plot or tail-contribution evidence.

If the Anderson Darling Test Is Significant

If the Anderson Darling Test Is Not Significant

Student-Friendly Reporting Sentence

Common Mistakes in Anderson Darling Test Interpretation

When to Use the Anderson Darling Test

Use the Anderson Darling Test when you need a formal distribution-fit or normality test and tail behavior matters. It is especially useful when Q-Q plots suggest tail deviation, when outliers may affect normality, when the sample distribution appears close in the center but questionable in the tails, or when you want a stronger normality diagnostic than a basic distribution plot.

If the Anderson Darling Test rejects normality, review sample size, tails, outliers, skewness, kurtosis, and the planned analysis. Possible follow-ups include transformation, robust methods, nonparametric methods, bootstrap confidence intervals, or practical reliance on the Central Limit Theorem when justified. For skewed variables, a guide such as Reciprocal Transformation may help.

Downloads and Resources for the Anderson Darling Test

The SPSS output PDF below supports the Anderson Darling Test workflow used in this guide. The Python and R charts provide visual evidence for distribution fit, Q-Q plot behavior, empirical CDF comparison, tail-weighted contributions, variable-level AD statistics and group-level AD statistics.

Related Internal Guides

Use these related Salar Cafe guides to connect the Anderson Darling Test with normality checks, distribution-fit tests, descriptive statistics, transformations, variance assumptions, regression diagnostics and reporting.