

Simulating heteroscedasticity 

Message/Author 


I am trying to do a monte carlo simulation study of a regression model with heteroscedasticity, and cannot figure out how to simulate the heteroscedasticity. When I model heteroscedasticity in Mplus, I do so using the MODEL CONSTRAINT feature, which is unavailable in specifying the model population. I know that I can use a mixture model to create heteroscedasticity, but would like to more precisely control how the residual variance changes. Is this possible in Mplus? 


See our new book described on our website. Heteroscedasticity scripts are shown for Chapter 1. 


Thank you for your response, I now see that heteroscedasticity can be introduced using a random slopes model. However, in these models, it seems that the heteroscedasticity is always symmetric (large residual variance at the extremes, low residual variance at the center). Is there a way to simulate a population for the model in ex1.15.inp in which the residual variance is a+bx? Thanks again for your help 


The random slopes approach doesn't necessarily have large variances at the extremes if one focuses on a certain range of x values relevant for the application. That's how it is applied in Chapter 1 where the variance is low for low x values and high for high x values. This approach provides an easy way to generate heteroscedastic data. I am not sure how data could be generated using the ex1.15 approach; seems like you would have to generate data for each of a set of distinct x values with different residual variance and then combine. 

Back to top 

