Transformation for ratio-interval single-item constructs

[AAljabr]

Hi all,

Beside many latent exogenous constructs, I have three single-item constructs that are measured in a ratio scale in my model. Two of these constructs (Size measured by the number of employees, and cost structure) are single-item exogenous constructs, and one construct (number of cost centers) is a single-item endogenous construct. All of them have moderate to high level of skewness and kurtosis. The output values from SPSS are:
1. Size (skewness= 4.504, kurtosis= 28.318)
2. Cost structure (skewness= 1.166, kurtosis= 2.826)
3. Number of cost centers (skewness= 4.380, kurtosis= 21.851).

I tried the natural log transformation for size and the number of cost centers, and it significantly reduced the magnitude of non-normality (all skewness and kurtosis values become between -1 and +1). Also, the square root transformation helped in reducing the non-normality of the cost structure construct.

My question is: Can I use the values after transformation (loq and square root transformation) instead of the original non-normal values in PLS?
Wouldn't it conflict with the ability of PLS to deal with non-normal data?

Thanks,

[jmbecker]

Reducing the non-normality is also useful in PLS if the interpretation of the transformed variables is still meaningful.
Due to the bootstrapping procedure used for testing the significance of parameters, a moderate level of non-normality should not affect the estimated standard errors and hence, t-values, p-values and confidence intervals, if the bootstrap distribution is not non-normal. The bootstrap distributions can be checked with the histograms.
However, as PLS is fundamentally based on correlations and regression severe non-normality can affect the estimates. Hence, transforming non-normal single-item variables can be useful if interpretation of the results is still meaningful after transformation.

[AAljabr]

Dear Dr. Becker,

Thank you for your advice and answer.