Mixed ANOVA or repeated measures ANOVA w/ transformed variable

#1
Hi,

I would like to run a 2x3 mixed ANOVA on SPSS. I have two groups and would like to see if the groups differ on their responses to 3 likert scale questions (of the same scale). One of the likert scale questions had a substantial negative skew so i transformed the variable using a reflect and log transformation. I then re-reflected the variable after transformation to restore the original meaning of the scores.

My question is: is it ok to run a mixed ANOVA or a repeated measures ANOVA with a transformed variable (within-subjects)? I'm wondering because only one of the likert scale questions is transformed and the other two are not.

Any advice would be greatly appreciated!
 

Karabiner

TS Contributor
#2
Usually you should not perform transformations only to fulfil some model assumptions.
It must make sense. What is the meaning of a "reflected, log transformed response"?
It is uninterpretable.

Moreover, in the present case, to transform 1 variable and 2 variables not, will not only
destroy the meaning of the tranformed variable, but it will also make all comparisons
between the transformed and untransformed variables meaningless. You said you wanted
to analyse their differences! How will you interpret any statistically significant difference
between the reflected, log transformed response variable and an untransformed
response variable?

You did not mention the most crucial aspect, which is sample size. If sample size is
large enough, then analyses of variances are considered robust against non-normality
(non-normality of the residuals, not of the response variable! if you did test the response
variable and not the residuals [alternatively the variable in each group separately], then
you tested the wrong thing anyway).

With kind regards

Karabiner
 
Last edited: