Multivariate Dispersion Test (PERMDISP)

PERMDISP is a test of multivariate dispersion that computes each observation's distance to its group centre in principal coordinate space and compares those distances among groups with an F statistic and permutations.

PERMDISP는 주좌표 공간에서 각 관측값과 자기 집단 중심 사이의 거리를 구한 뒤, 그 거리를 집단 간에 F 통계량과 순열로 비교하는 다변량 퍼짐 검정입니다.

Is the variability in community composition larger in one group than in another?

한 집단의 군집 구성 변동성이 다른 집단보다 더 큰가요?

PERMDISP measures how spread out each group is by the distance of each sample to its group's spatial median or centroid, after embedding any dissimilarity with principal coordinates. An F test with permuted residuals compares the mean distances, which both answers questions about beta diversity and checks the dispersion assumption of PERMANOVA.

When to use Multivariate Dispersion Test (PERMDISP)

• Comparing multivariate spread or beta diversity among groups
• Checking whether a significant PERMANOVA reflects dispersion rather than location
• Any dissimilarity, including non-Euclidean ones, is used

When to use something else

• You want to test differences in group location: use PERMANOVA
• Data are univariate: use Levene's test

Assumptions and requirements

• Observations are independent and exchangeable under the null hypothesis
• Negative eigenvalues from non-Euclidean dissimilarities are handled by subtracting their axes

What the analysis reports

• Distance to group centre for each observation and group means
• F statistic with parametric and permutation p-values
• Tukey HSD pairwise comparisons of mean distances
• Box plot of distances by group

How to report it

A test of multivariate dispersion (distances to group spatial medians, Bray-Curtis) indicated [no] differences in dispersion among the [k] groups, F([df1], [df2]) = [value], p = [value].

Related methods

• PERMANOVA
• Levene's Test
• ANOSIM

Run Multivariate Dispersion Test (PERMDISP) on Quorant