Chapter 10: Incomplete and Non-normal Data

10.1. At the end of Example 8.2 I pointed out that although indirect and total effects could be estimated by ‘ordinary’ ML (or any of the other estimation methods available), these methods would not report standard errors and P-values for such effects. We have to fall back on bootstrapping for testing.

The input data in Example 8.2 was a covariance matrix as will often be the case if your analysis is based on journal data. Have a look at the raw data (‘Thomsen2.sav’). As you will observe, the data are far from normal. Most manifest variables are severely skewed – luckily they are all skewed to the same side (which makes the problem a bit less serious), and that bootstrapping perhaps would solve that problem too. So, estimate the model from Example 8.2 by bootstrapping and test for indirect and total effects.

10.2 Experiment with the data (‘Thomsen2.sav). Try various ways of coping with the non-normality of the data – e.g. compare various transformations and estimation methods. You are also welcome to check for outliers and other data problems.