TOPICS
Search

Bartels Rank Test


The Bartels rank test is both a nonparametric test and a rank test for serial correlation in an ordered sample. If R_i is the statistical rank of the ith observation among n observations, its rank von Neumann ratio is

 RVN=(sum_(i=1)^(n-1)(R_i-R_(i+1))^2)/(sum_(i=1)^(n)[R_i-(n+1)/2]^2).

The numerator is the sum of squared differences between successive ranks and therefore measures local change along the given order. The denominator is the total squared deviation from the mean rank (n+1)/2 and measures overall statistical dispersion among the ranks. Thus RVN is the rank version of the ordinary von Neumann ratio, which uses the observations and their sample mean in the two sums. When there are no ties, the denominator is n(n^2-1)/12.

Under the null hypothesis that the observations occur in random order, significance can be assessed from the exact or asymptotic null distribution of RVN, or by a permutation test. Small values indicate that neighboring ranks tend to be similar, as occurs with positive serial correlation or a trend. Large values indicate alternating high and low ranks.


See also

Null Distribution, Null Hypothesis, Permutation Test, Rank Test, Serial Correlation, Statistical Rank, von Neumann Ratio

Explore with Wolfram|Alpha

References

Bartels, R. "The Rank Version of von Neumann's Ratio Test for Randomness." J. Amer. Statist. Assoc. 77, 40-46, 1982. https://doi.org/10.1080/01621459.1982.10477764.

Cite this as:

Weisstein, Eric W. "Bartels Rank Test." From MathWorld--A Wolfram Resource. https://mathworld.wolfram.com/BartelsRankTest.html

Subject classifications