Solved – SPSS: Log-transformation of data that is not normal distributed

data transformationnormality-assumptionspss

I work on my thesis and use SPSS to analyze the data.

Because some of my data is not normal distributed, I would like to log-transform the data to see, if this changes the distribution.

I differ between two groups and in one group there is a normal distribution but in the other one there is not. I tried two drifferent log-transformations (ln and lg10):

COMPUTE LG10FCQT_sum=LG10(FCQT_sum). 
EXECUTE.

EXAMINE VARIABLES=LG10FCQT_sum BY YFAS_FA 
/PLOT BOXPLOT HISTOGRAM NPPLOT 
/COMPARE GROUPS 
/STATISTICS DESCRIPTIVES 
/CINTERVAL 95 
/MISSING LISTWISE 
/NOTOTAL.


COMPUTE LnFCQT_sum=LN(FCQT_sum). 
EXECUTE.

EXAMINE VARIABLES=LnFCQT_sum BY YFAS_FA 
/PLOT BOXPLOT HISTOGRAM NPPLOT 
/COMPARE GROUPS 
/STATISTICS DESCRIPTIVES 
/CINTERVAL 95 
/MISSING LISTWISE 
/NOTOTAL.

When I then looked at the distribution, I noticed that in both groups there is no normal distribution. Did I do something wrong or how can this outcome be explained?

My goal is to transform the data to obtain a normal distribution in both groups.

Thank you!

Best Answer

Welcome to CV. First, questions about SPSS (or any programming language) are off topic here, but you question seems to be about statistics, not SPSS.

Second, just because a distribution is not normal does not mean that the log of it will be normal. And, if the log base 10 does not make it normal, neither will log base e.

Third, there are lots of other ways to transform variables to make them normal. Look into the Box Cox transformations, for instance, however:

Fourth: It usually is not necessary and often not advisable to transform your data this way. You haven't told us what you are doing, but, usually, there are alternate methods that do not assume normality (and regression, for one, does not assume normality of the data).

Related Question