Hello!
I would like to understand why normality is a crucial assumption of ANOVA
(especially for analyses between-subjects factors).
I understand that in the calculation of mean squares, homogeneity of
variance across groups is important. Am I correct in saying that it is
because the calculation of F involves comparing variances on the dependent
variable across groups? If these variances are not homogeneous the F value
will be biased (and may be significant simply because the variances are
heterogeneous).
Could somebody help with understanding how violations of normality
influence the ANOVA? In particular, if one is comparing effects on a
dependent variable across different groups of subjects, will it be
necessary that the dependent variable is normally distributed for each of
the groups (doing a 1 sample Kolgomorov-Smirnov test on a data file split
by group)? Or will it be sufficient to have normality of the overall
dependent variable?
One more question: can you suggest any test of normality other than K-S
(possibly feasible in SPSS)? I try and look at kurtosis and skewness as
well, but I am not sure of what the convention is for what counts as
mesokurtic and not skewed.
Thanks in advance for any help!
Gaia
*********************************
Gaia Scerif
PhD Student
Neurocognitive Development Unit
Institute of Child Health
University College London
London WC1N 1EH
phone: 020 7 905 2749
e-mail: [log in to unmask]
|