We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

stat

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Statistics > Methodology

Title: A Generalized Levene's Scale Test for Variance Heterogeneity in the Presence of Sample Correlation and Group Uncertainty

Abstract: We generalize Levene's test for variance (scale) heterogeneity between $k$ groups for more complex data, which includes sample correlation and group membership uncertainty. Following a two-stage regression framework, we show that least absolute deviation regression must be used in the stage 1 analysis to ensure a correct asymptotic $\chi^2_{k-1}/(k-1)$ distribution of the generalized scale ($gS$) test statistic. We then show that the proposed $gS$ test is independent of the generalized location test, under the joint null hypothesis of no mean and no variance heterogeneity. Consequently, we generalize the recently proposed joint location-scale ($gJLS$) test valuable in settings where there is an interaction effect, but one interacting variable is not available. We evaluate the proposed method via an extensive simulation study, and two genetic association application studies.
Comments: 33 pages (plus 14 pages supplementary)
Subjects: Methodology (stat.ME)
Cite as: arXiv:1605.05715 [stat.ME]
  (or arXiv:1605.05715v1 [stat.ME] for this version)

Submission history

From: David Soave [view email]
[v1] Wed, 18 May 2016 19:37:27 GMT (423kb,D)

Link back to: arXiv, form interface, contact.