Not normally distributed data

#1
Good afternoon,
I am comparing scales of optimism, subjective vitality and hope between two nations- so i need to do t-test and ANOVA.
My data are not normally distributed with negative skewness so I tried to transform the data through log10(max value+1-variable) and root square transformation SQRT(max value+1-variable) and for some scale i got the Saphiro-Wilk test of normality sig>0.05 after using root square transformation and for some after log10 transformation.
Maybe a stupid question but do I have to choose only way of transformation or I can work with both of them? F.e. compare SQRT_LOTR_CZ and LOG10_LOTR_SPAIN?
Thanks for any reaction. Anna
 

noetsi

Fortran must die
#2
I don't know of any requirement that you do it one way, normally you do it in the way that makes the data most normal. Why would you want to analyze more than one type of transformation?

I recommend using a QQ plot to test for normality. The Saphiro-Wilk test has well known issues.
 
#3
Thanks for reaction. So lets say I use the log10 data transformation and I still do not have a normal distribution in all the scales...can I do t-test anyway?
I not really experienced user of SPSS so thank you for patience.
 

noetsi

Fortran must die
#4
I don't know SPSS well either, my answers are just about the statistics (I use SAS).

The rule is if you do a transformation and it does not make the data normal you try a more extreme transformation. There are list of which are more powerful in this regard. Or, even better but less simply, you can do a Box-Cox transformation which is a family of transformations that you can chose from. There is a lot on that, and macros to do them, on line.

Remember after you do this to interpret the results you have to untransform the data back.