MerQuR+Statistics Glossary

Statistics Glossary → Asymptotic (normal) approximation

Asymptotic (normal) approximationconcept

Approximating the sampling distribution of a test statistic or estimate by the normal (or chi-square) distribution that holds in large samples.

The exact sampling distribution of many statistics is complicated or tedious to compute. Results such as the central limit theorem say that, as the sample grows, these distributions approach a normal or chi-square distribution. An asymptotic p value or confidence interval is computed from that approximate distribution; for example, the Mann-Whitney U statistic is standardized by its mean and variance and evaluated as a z score.

How good the approximation is depends on the sample size, ties and the shape of the distribution. In small samples or very sparse tables an asymptotic p value can mislead, and an exact test, a Monte Carlo or a permutation p value is preferred. For discrete statistics (counts, ranks) a continuity correction can improve the move to the continuous normal distribution.

An everyday example

To see whether a die is loaded after 10 throws, the probability of each possible number of sixes has to be worked out one by one (the exact route). After 6000 throws, however, the number of sixes is approximately normal (mean 1000, standard deviation about 29), and that normal curve shows at once how surprising 1100 sixes would be.

In MerQur

The 'Normal-approximation CI for d' and 'Normal-approximation CI for d_z' options of the one-sample and paired t-tests, the 'Large-sample Z' option of the sign test, the asymptotic p reported next to the exact p in the Mann-Whitney U and Wilcoxon tests, and the normality or randomization choice under 'Global inference' in Moran's I.

Analyses where this concept appears

Normality TestsOne-Sample t-TestIndependent-Samples t-TestPaired t-TestMann-Whitney UWilcoxon Signed-RankKruskal-WallisFriedmanSign TestRuns TestChi-Square Goodness of FitMcNemar TestCrosstabCochran's Q TestLinear Mixed Models (LMM)Confusion Matrix MetricsVariance Components Analysis (VARCOMP)Moran's I (Spatial Autocorrelation)

Related concepts

Exact test and exact p-valueContinuity correction (Yates)