Normality and Goodness of Fit

Anderson-Darling Test Table: Critical Values of A²

Critical values of the Anderson-Darling statistic, which gives extra weight to the tails of the distribution.

  • Normality and specified distributions
  • n = 4 to 100
  • Test your data

Anderson-Darling Table

Two tables: normality with estimated mean and SD, and a fully specified distribution. Cells: critical A². Reject if A² ≥ the cell.

Testing normality, mean and SD estimated from the data

Testing normality, mean and SD estimated from the data
α →
n ↓
0.150.10.050.0250.010.005
40.43480.48310.55990.62970.69790.7343
50.46380.51570.59980.68510.79680.8718
60.48060.53710.63260.72570.84450.9343
70.49310.55250.65240.75100.88080.9764
80.50250.56380.66750.77060.90521.0074
90.50960.57220.67850.78410.92291.0274
100.51540.57920.68710.79470.93641.0436
110.51960.58400.69360.80320.94761.0574
120.52340.58860.69930.80950.95541.0659
130.52650.59220.70380.81560.96301.0743
140.52890.59520.70750.81980.96881.0804
150.53110.59770.71070.82350.97331.0865
160.53330.60030.71410.82810.97911.0937
170.53480.60190.71600.83020.98091.0965
180.53660.60380.71860.83290.98451.0992
190.53780.60540.72030.83520.98771.1040
200.53900.60650.72200.83760.99021.1058
210.54000.60790.72370.83940.99301.1104
220.54090.60890.72470.84090.99441.1111
230.54200.61020.72630.84260.99651.1137
240.54260.61090.72710.84370.99811.1163
250.54330.61180.72810.84451.00021.1189
260.54390.61240.72890.84601.00131.1183
270.54430.61300.73000.84681.00251.1206
280.54510.61390.73110.84811.00371.1212
290.54610.61490.73210.84931.00531.1238
300.54630.61510.73210.84941.00551.1243
350.54810.61710.73480.85291.00991.1308
400.54970.61920.73740.85581.01271.1330
450.55080.62020.73870.85751.01531.1359
500.55180.62160.74060.85961.01731.1382
600.55310.62280.74180.86161.02061.1419
700.55410.62410.74360.86361.02271.1440
800.55480.62490.74450.86461.02421.1460
900.55520.62530.74470.86501.02451.1453
1000.55580.62600.74590.86601.02571.1460

Fully specified distribution (no parameters estimated)

Fully specified distribution (no parameters estimated)
α →
n ↓
0.150.10.050.0250.010.005
41.63161.95942.54433.15603.98724.6330
51.62851.95272.53363.14003.96434.5970
61.62721.94912.52603.12603.94434.5796
71.62591.94692.52123.11913.93434.5662
81.62561.94462.51733.11543.93574.5695
91.62521.94432.51563.11333.92574.5507
101.62521.94242.51033.10503.91604.5504
111.62471.94142.51043.10383.91364.5393
121.62451.94052.50783.10123.91354.5387
131.62331.93982.50823.10123.90494.5314
141.62321.93952.50623.09653.90924.5361
151.62281.93802.50443.09803.90624.5331
161.62341.93882.50603.09823.90704.5305
171.62301.93842.50573.09663.90574.5290
181.62371.93862.50393.09353.90004.5212
191.62351.93822.50313.09333.90024.5268
201.62291.93712.50143.09353.90094.5254
211.62281.93732.50083.09093.90144.5229
221.62271.93752.50173.09113.89544.5185
231.62241.93712.50113.09213.90084.5273
241.62281.93712.50183.09413.90164.5219
251.62291.93702.50133.08943.89334.5146
261.62251.93602.49943.08913.89654.5244
271.62281.93652.50053.08713.89084.5155
281.62231.93462.49863.08693.88924.5187
291.62241.93702.50093.09013.89304.5181
301.62351.93742.50093.09003.89564.5180
351.62271.93602.49793.08413.88944.5189
401.62161.93432.49603.08493.88864.5117
451.62141.93452.49493.08233.88594.5093
501.62101.93342.49403.08073.88154.5046
601.62111.93392.49443.08103.88584.5118
701.62171.93522.49593.08483.88644.5034
801.62261.93462.49663.08253.89124.5159
901.62081.93332.49553.08203.88254.5019
1001.62171.93512.49513.08093.88074.5005

Tip: click any value to highlight its row and column.

How to read this table

  1. Sort the data and compute zᵢ = F(x₍ᵢ₎), using the normal CDF with the sample mean and SD (or the fully specified distribution).
  2. Compute A² = −n − (1/n) Σ (2i − 1)[ln zᵢ + ln(1 − z₍ₙ₊₁₋ᵢ₎)].
  3. Use the first table when the mean and SD were estimated from the data, and the second when the distribution was specified in advance.
  4. Reject if A² is greater than or equal to the table value.

Worked example

A normality check on 20 residuals gives A² = 0.80. Row 20, α = 0.05 of the first table gives 0.7220 and the α = 0.025 value is 0.8376, so 0.025 < p < 0.05.

Click the underlined links to highlight the value in the table.

Calculator

For values between the rows and columns of the table. Results update as you type.

Formula and how the values were computed

A² = −n − (1/n) Σᵢ (2i − 1) [ln zᵢ + ln(1 − z₍ₙ₊₁₋ᵢ₎)]  ·  Stephens' adjusted A* = A²(1 + 0.75/n + 2.25/n²)

Critical values are percentiles of A² from 10 million simulated samples for each n. The fully specified case converges to Anderson and Darling's asymptotic values (1.933, 2.492 and 3.857 at α = 0.10, 0.05 and 0.01); the normal case converges to Stephens' values (0.631, 0.752 and 1.035).

Frequently asked questions

What is Stephens' A*?

A small-sample adjustment that lets one set of asymptotic critical values (0.752 at α = 0.05) be used for any n. This table gives exact percentiles of the unadjusted A² for each n instead.

Why use Anderson-Darling?

It weights the tails more heavily than the Kolmogorov-Smirnov statistic, so it is better at detecting heavy tails and outliers.

Related calculators and tutorials