Help with non normality distribution. Transformations didn't work. Help!

#1
Hi,

new to this forum. I'm writing up my MsC in psychology and I've got problems with my data.
The study involved administering questionnaires to 157 young carers. The dependent variables I've got are: emotional symptoms, peer problems, prosocial behaviour, hyperactivity, difficulties, anxiety, depression and quality of life. My goal is looking at the impact of age, gender and caring circumstances on these variables.
Problem is that the data are non normally distributed on most of these variables (except on quality of life and difficulties). I tried all sorts of transformations but it didn't help. Shall I treat all data as non normal? or can I perform one way ANOVAs anyways? Is the sample size big enough for doing that?

Thanks a lot for your help (in advance)

Martina
 

Miner

TS Contributor
#2
What type of data (e.g., ordinal, Likert scale, etc.)? Your variables appear to be some type of ordinal rating at first glance.

Normality tests presume continuous data, so if you have discrete data, it would not be surprising to fail a normality test.
 

noetsi

Fortran must die
#3
I assume you are using a method like regression. If so it is the normality of the residuals that matters not the normality of the predictors or the dependent variable.
 
#4
Miner: the data I got are on Likert scale. They are continuous.

Noetsi: I wanted to use ANOVAs actually, I didn't think about regression but I guess I might use that...thanks for the tip. I will have a look at the normality of residuals:)

Thanks for replying
 

Miner

TS Contributor
#5
Miner: the data I got are on Likert scale. They are continuous.

Noetsi: I wanted to use ANOVAs actually, I didn't think about regression but I guess I might use that...thanks for the tip. I will have a look at the normality of residuals:)

Thanks for replying
Likert scales are not continuous, they are ordinal, and would not pass a normality test. You should use an analysis method for ordinal data.
 

noetsi

Fortran must die
#7
I think, I am not sure, that both regression and ANOVA are in the General Linear Models family. Regardless the rule that the residuals has to be normal applies equally to ANOVA and regression.

If you have a likert scale with enough levels the results could be roughly normal although it is unlikely you would have that many levels ever. It might be noted that likert scale data is commonly treated as "interval like" and used in regression and ANOVA. I wonder now how that is possible given the issue of normality.
 

Miner

TS Contributor
#8
Typically, if Poisson data is large enough to use the Normal approximation, you can treat it as pseudo-continuous.