Hi Andrés:
The answer is not easy. Take a look at this paper:
www.psych.nyu.edu/cohen/eps12dr1.pdf
I usually prefer Welch if the design is heavily unbalanced, and
Brown-Forsythe when it is approximately balanced.
HTH,
Marta GG
El 15/06/2011 23:08, ANDRES ALBERTO BURGA LEON escribió:
Hello to everybody:
I want to know in which
conditions one
of these corrections is recommended.
I've discarded the variance
homogeneity
hypothesis given the Levene results, but when I run both
corrections, I
get significant results (at .05) with Brown-Forsythe, but
non-significant
with Welch.
Also the Games-Howell post hoc
shows
no significant differences in any of the paired comparisons.
Kindly
Andrés
Mg. Andrés Burga León
Coordinador de Análisis e Informática
Unidad de Medición de la Calidad Educativa
Ministerio de Educación del Perú
Calle El Comercio s/n (espalda del Museo de la Nación)
Lima 41
Perú
Teléfono 615-5840
Locked
