Non Normal Data

2 replies [Last post]
K Ram's picture
Joined: 08/21/2012


Some of the variables in my dataset do not satisfy the assumption that the data should be normally distributed. So I am unable to rationally run parametric SEM analyses.

I have attempted to transform the data to approximate a normal distribution using cumulative distribution functions, log10 and square root transformations, but with no success. These transformations slightly shifts the skewness or squeezes the data together, however the data remains immune to normality.

1. What steps can be taken when the normality assumptions cannot be met?

2. Are there any openMx scripts designed for non-normal data?

3. Any other suggestions??

Many thanks

tbates's picture
Joined: 07/31/2009
try box-cox

> Some of the variables do not satisfy the assumption of being normally distributed.

Might not matter, but...

> I have attempted log10 and square root transformations

Box Cox is a great normalizer

estLambda = car::powerTransform(data$GPA) # estimate lambda
data$GPAbc = car::bcPower(data$GPA, coef(estLambda, round=TRUE))

trivia: Joan Fisher married George Box

K Ram's picture
Joined: 08/21/2012
Box Cox

Thank you for your helpful reply! Box Cox worked wonderfully for 19 of the 46 variables I'm interested in.

PS: interesting to know George Box married Ronald Fisher's daughter.