Wald-Wolfowitz Runs Test
Also known as: Wald-Wolfowitz test, runs test for randomness, Runs Testi (Wald-Wolfowitz)
The Wald-Wolfowitz runs test is a nonparametric hypothesis test that determines whether a sequence of observations — coded as a series of binary symbols — follows a random pattern or contains systematic structure. Introduced by Abraham Wald and Jacob Wolfowitz in 1940, the test counts the number of uninterrupted runs of identical symbols and asks whether that count is consistent with random arrangement.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use the runs test when you need to verify that a sequence of observations is randomly ordered before applying methods that assume independence — such as regression residual checks, quality-control charts, or survey response order checks. The test requires at least approximately 10 observations and that each observation can be assigned to one of two categories. It applies to time-series data, cross-sectional ordered records, and quality-control streams. It does not require normality or any distributional assumption.
Strengths & limitations
- Completely nonparametric: no distributional assumption is required.
- Applicable to both genuinely binary data and continuous data dichotomised at the median or another threshold.
- Simple to interpret: the run count directly exposes whether the sequence clusters or alternates.
- Sensitive to how continuous data are dichotomised; different cut-points can change the verdict.
- The normal approximation is unreliable for very small samples (n₁ or n₂ below 5); exact tables should be used.
- Tests only for non-randomness of ordering, not for any specific type of dependence or trend.
Frequently asked
How do I apply the runs test to continuous data?
Dichotomise the series at a natural threshold — most commonly the sample median. Observations above the median are coded with one symbol and observations at or below it with the other. The runs test then proceeds on this binary sequence. The choice of threshold matters: using the median is conventional, but another theoretically motivated cut-point may be more appropriate in specific contexts.
When should I use the exact distribution rather than the normal approximation?
When either symbol count (n₁ or n₂) is smaller than about 10 the normal approximation can be inaccurate. In that case use exact critical values tabulated for the joint counts (n₁, n₂). Most statistical software switches to the exact method automatically for small samples.
What does a significant result actually tell me?
A significant runs test tells you that the ordering of the sequence is unlikely to be random. It does not tell you what the structure is. Too few runs indicate clustering (positive autocorrelation or trend); too many indicate alternation (negative autocorrelation). Examining the run count relative to its expected value points toward which type of non-randomness is present.
Is the runs test the same as the Wald-Wolfowitz two-sample test?
They share the same theoretical framework and authors, but they address different questions. The one-sample runs test examines whether a single sequence is randomly ordered. The two-sample Wald-Wolfowitz test combines two samples into one ranked sequence and counts runs to test whether both samples come from the same population — analogous in spirit to the Mann-Whitney U test.
Sources
- Wald, A. & Wolfowitz, J. (1940). On a test whether two samples are from the same population. Annals of Mathematical Statistics, 11(2), 147–162. DOI: 10.1214/aoms/1177731909 ↗
How to cite this page
ScholarGate. (2026, June 1). Wald-Wolfowitz Runs Test. ScholarGate. https://scholargate.app/en/statistics/runs-test
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Durbin-Watson TestEconometrics↔ compare
- Kolmogorov-Smirnov TestStatistics↔ compare
- Ljung-Box TestEconometrics↔ compare
- Mann-Whitney U testStatistics↔ compare
- Sign TestStatistics↔ compare