Generating Correlated, Non-normally Distributed Data Using a Non-linear Structural Model
- 523 Downloads
An approach to generate non-normality in multivariate data based on a structural model with normally distributed latent variables is presented. The key idea is to create non-normality in the manifest variables by applying non-linear linking functions to the latent part, the error part, or both. The algorithm corrects the covariance matrix for the applied function by approximating the deviance using an approximated normal variable. We show that the root mean square error (RMSE) for the covariance matrix converges to zero as sample size increases and closely approximates the RMSE as obtained when generating normally distributed variables. Our algorithm creates non-normality affecting every moment, is computationally undemanding, easy to apply, and particularly useful for simulation studies in structural equation modeling.
KeywordsNon-normal multivariate data Structural equation modeling Simulation
This research was supported in parts by a grant from the Baden-Württemberg foundation to the second author
- Foldnes, N., & Grønneberg, S. (in press). How general is the Vale-Maurelli simulation approach? Psychometrika. doi: 10.1007/s11336-014-9414-0.
- Headrick, T. C. (2010). Statistical simulation: Power method polynomials and other transformations. Boca Raton, FL: Chapman & Hall/CRC.Google Scholar
- Hodis, F. A., Headrick, T. C., & Sheng, Y. (2012). Power method distributions through conventional moments and L-moments. Applied Mathematical Sciences, 6, 2159–2193.Google Scholar
- Mattson, S. (1997). How to generate non-normal data for simulation of structural equation models. Multivariate Behavioral Research, 32, 355–373. doi: 10.1207/s15327906mbr3204_3.
- Yuan, K.-H., & Bentler, P. M. (1999). On normal theory and associated test statistics in covariance structure analysis under two classes of nonnormal distributions. Statistica Sinica, 9, 831–853.Google Scholar