Anderson–Darling test

In statistics, the Anderson–Darling test, named after Theodore Wilbur Anderson (born
1918) and Donald A. Darling (born 1915), who invented it in 1952[1], is a statistical test of
whether there is evidence that a given sample of data did not arise from a given probability
distribution. In its basic form, the test assumes that there are no parameters to be estimated in
the distribution being tested, in which case the test and its set of critical values is distribution-
free. However, the test is most often used in contexts where a family of distributions is being
tested, in which case the parameters of that family need to be estimated and account must be
taken of this in adjusting either the test-statistic or its critical values.

When applied to testing if a normal distribution adequately describes a set of data, it is one of
the most powerful statistical tools for detecting most departures from normality.[2][3]

In addition to its use as a test of fit for distributions, it can be used in parameter estimation as
the basis for a form of minimum distance estimation procedure.

K-sample Anderson–Darling tests are available for testing whether several collections of
observations can be modelled as coming from a single population, where the distribution
function does not have to be specified.

Contents
[hide]

 1 The single-sample test

o 1.1 Basic test statistic
 2 Tests for families of distributions
o 2.1 Test for normality
o 2.2 Tests for other distributions
 3 Non-parametric k-sample tests
 4 See also
 5 External links
 6 References

The single-sample test

The Anderson-Darling and Cramér-von Mises statistics belong to the class of quadratic EDF
statistics (tests based on the empirical distribution function).[3] If the hypothesized
distribution is F, and empirical (sample) cumulative distribution function is Fn, then the
quadratic EDF statistics measure the distance between F and Fn by
where w(x) is a weighting function. When the weighting function is w(x)=1, the statistic is
the Cramér-von Mises statistic. The Anderson-Darling (1954) test[4] is based on the distance

which is obtained when the weight function is . Thus,

compared with the Cramér-von Mises distance, the Anderson-Darling distance places more
weight on observations in the tails of the distribution.

Basic test statistic

The Anderson–Darling test assesses whether a sample comes from a specified distribution. It
makes use of the fact that, when given a hypothesized underlying distribution and assuming
the data does arise from this distribution, the data can be transformed to a uniform
distribution. The transformed sample data can be then tested for uniformity with a distance
test (Shapiro 1980). The formula for the test statistic A to assess if data
(note that the data must be put in order) comes from a distribution with cumulative
distribution function (CDF) F is

where

The test statistic can then be compared against the critical values of the theoretical
distribution. Note that in this case no parameters are estimated in relation to the distribution
function F.

Tests for families of distributions

Essentially the same test statistic can be used in the test of fit of a family of distributions, but
then it must be compared against the critical values appropriate to that family of theoretical
distributions and dependent also on the method used for parameter estimation.

Test for normality

2
In comparisons of power, Stephens[2] found A to be one of the best Empirical distribution
function statistics for detecting most departures from normality. The only statistic close was
2
the W Cramér–von Mises test statistic. It may be used with small sample sizes n ≤ 25. Very
large sample sizes may reject the assumption of normality with only slight imperfections, but
industrial data with sample sizes of 200 and more have passed the Anderson–Darling
test.[citation needed]
The computation differs based on what is known about the distribution:[5]

2
 Case 1: The mean μ and the variance σ are both known.
2
 Case 2: The variance σ is known, but the mean μ is unknown.
2
 Case 3: The mean μ is known, but the variance σ is unknown.
2
 Case 4: Both the mean μ and the variance σ are unknown.

The n observations Xi, for , of the variable X that should be tested are sorted
from low to high and the notation in the following assumes that Xi represent the ordered
observations. Let

The values Xi are standardized to create new values Yi, given by

2
With the standard normal CDF Φ, A is calculated using

An alternative expression in which only a single observation is dealt with at each step of the
summation is:

A modified statistic is calculated using

*2
If A exceeds a given critical value, then the hypothesis of normality is rejected with some
significance level. The critical values are given in the table below (valid for ).[2]
 2
Note 1: If = 0 or any Φ(Yi) = (0 or 1) then A cannot be calculated and is undefined.

Note 2: The above adjustment formula is taken from Shorak & Wellner (1986, p239). Care is
required in comparisons across different sources as often the specific adjustment formula is
not stated.

Note 3: Stephens[2] notes that the test becomes better when the parameters are computed from
the data, even if they are known.

Significance Case 1 Case 2* Case 3 Case 4

15% 1.610 ? ? 0.576
10% 1.933 0.908 1.760 0.656
5% 2.492 1.105 2.323 0.787
2.5% 3.070 1.304 2.904 0.918
1% 3.857 1.573 3.690 1.092

(*) For the case 2, the values are for the asymptotic distribution.

Alternatively, for case 3, D'Agostino (1986, pp. 372-373)[5] gives the adjusted statistic

*2
and normality is rejected if A exceeds 0.631, 0.752, 0.873, 1.035, or 1.159 at 10%, 5%,
2.5%, 1%, and 0.5% significance levels, respectively; the procedure is valid for sample size
at least n=8.

Tests for other distributions

Above, it was assumed that the variable Xi was being tested for normal distribution. Any
other family of distributions can be tested but the test for each family is implemented by
using a different modification of the basic test statistic and this is referred to critical values
specific to that family of distributions. The modifications of the statistic and tables of critical
values are given by Stephens (1986)[3] for the exponential, extreme-value, Weibull, gamma,
logistic, Cauchy, and von Mises distributions. Tests for the (two-parameter) log-normal
distribution can be implemented by transforming the data using a logarithm and using the
above test for normality. Details for the required modifications to the test statistic and for the
critical values for the normal distribution and the exponential distribution have been
published by Pearson & Hartley (1972, Table 54). Details for these distributions, with the
addition of the Gumbel distribution, are also given by Shorak & Wellner (1986, p239).
Details for the logistic distribution are given by Stephens (1979). A test for the (two
parameter) Weibull distribution can be obtained by making use of the fact that the logarithm
of a Weibull variate has a Gumbel distribution.

Non-parametric k-sample tests

Scholz F.W. and Stephens M.A. (1987) discuss a test, based on the Anderson-Darling
measure of agreement between distributions, for whether a number of random samples with
possibly different sample sizes may have arisen from the same distribution, where this
distribution is unspecified.

See also
 Kolmogorov–Smirnov test
 Kuiper's test
 Shapiro–Wilk test
 Smirnov–Cramér–von-Mises test
 Jarque–Bera test
 Goodness of fit

External links
 US NIST Handbook of Statistics

References
1. ^ Anderson, T. W.; Darling, D. A. (1952). "Asymptotic theory of certain "goodness-of-fit"
criteria based on stochastic processes". Annals of Mathematical Statistics 23: 193–212.
doi:10.1214/aoms/1177729437.
2. ^ a b c d Stephens, M. A. (1974). "EDF Statistics for Goodness of Fit and Some Comparisons".
Journal of the American Statistical Association 69: 730–737. doi:10.2307/2286009.
3. ^ a b c M. A. Stephens (1986). "Tests Based on EDF Statistics". In D'Agostino, R.B. and
Stephens, M.A.. Goodness-of-Fit Techniques. New York: Marcel Dekker. ISBN 0-8247-
7487-6.
4. ^ Anderson, T.W. and Darling, D.A. (1954). "A Test of Goodness-of-Fit". Journal of the
American Statistical Association 49: 765–769.
5. ^ a b Ralph B. D'Agostino (1986). "Tests for the Normal Distribution". In D'Agostino, R.B.
and Stephens, M.A.. Goodness-of-Fit Techniques. New York: Marcel Dekker. ISBN 0-8247-
7487-6.

 Corder, G.W., Foreman, D.I. (2009).Nonparametric Statistics for Non-

Statisticians: A Step-by-Step Approach Wiley, ISBN 978-0-470-45461-9
 Pearson E.S., Hartley, H.O. (Editors) (1972) Biometrika Tables for
Statisticians, Volume II. CUP. ISBN 0-521-06937-8.
 Shapiro, S.S. (1980) How to test normality and other distributional
assumptions. In: The ASQC basic references in quality control: statistical
techniques 3, pp. 1–78.
 Shorak, G.R., Wellner, J.A. (1986) Empirical Processes with Applications to
Statistics, Wiley. ISBN 0-471-86725-X.
 Stephens, M.A. (1979) Test of fit for the logistic distribution based on the
empirical distribution function, Biometrika, 66(3), 591-5.
 Scholz F.W., Stephens M.A. (1987), K-sample Anderson-Darling Tests,
Journal of the American Statistical Association, 82, 918–924.

Above is from Wikipedia