Wald-Wolfowitz Runs Test I

A run is a series of consecutive identical symbols or values in a sequence. For example, in the sequence 0, 0, 1, 1, 1, 0, the first run is [0, 0] and the second run is [1, 1, 1]. Runs must be mutually exclusive—consecutive different values cannot form a single run. The test counts total runs to assess whether data ordering is random.

The test rejects randomness when the observed number of runs is extremely low or extremely high compared to expected values under randomness. A dataset with too few runs suggests clustering or patterns, while too many runs may indicate alternation. The test provides a quantitative measure at a specified significance level, such as 0.05, to evaluate whether the sequence is random.

The runs test applies to binary, categorical, and numerical data. Binary data includes sequences like wins and losses. Categorical data includes DNA sequences. For numerical data, such as leaf sizes, values are converted to binary signs by comparing each value to the mean or median, assigning + for values above and - for values below.

The runs test is nonparametric because it applies to any ordered data regardless of the population and sample distribution, even with larger sample sizes. It does not assume a specific probability distribution and works with any sequence of two different element types. This makes it flexible for diverse data types and distributions.

To analyze numerical data with the runs test, calculate the mean or median of the dataset. Assign a + sign to every value higher than the mean or median and a - sign to every value lower. This creates a binary sequence of signs that can then be analyzed for runs, converting continuous numerical data into categorical form.

An extreme number of runs—either too low or too high—indicates the data is less random. Too few runs suggest values cluster together in groups, showing a pattern or trend. Too many runs suggest excessive alternation between values. Moderate run counts closer to expected values under randomness indicate the sequence is more random.

The runs test alone does not provide a clear quantitative measure of how random a dataset is. While it determines whether randomness should be rejected at a significance level, the magnitude of randomness remains qualitative and requires interpretation based on the nature of the data—whether it is binary, categorical, or numerical.