The Effects of Autocorrelation on the Curve-of-Factors Growth Model

Abstract

This simulation study examined the performance of the curve-of-factors model (COFM) when autocorrelation and growth processes were present in the first-level factor structure. In addition to the standard curve-of factors growth model, 2 new models were examined: one COFM that included a first-order autoregressive autocorrelation parameter, and a second model that included first-order autoregressive and moving average autocorrelation parameters. The results indicated that the estimates of the overall trend in the data were accurate regardless of model specification across most conditions. Variance components estimates were biased across many conditions but improved as sample size and series length increased. In general, the two models that incorporated autocorrelation parameters performed well when sample size and series length were large. The COFM had the best overall performance.

Keywords:

• autoregression
• autoregressive latent trajectory model
• curve-of-factors model
• latent growth curve model
• structural equation modeling

Notes

^aSimple bias was computed for the zero autocorrelation condition, because the true parameter value of the level was zero.

¹Full tables are available from the first author on request.