I am attempting to fit a latent growth curve model with one continuous measure assessed at 9 time points. When the model is estimated with an intercept and linear slope, fit is marginally acceptable and there are no heywood cases. Adding a quadratic growth factor to the model improves model fit substantially.
However, when a quadratic growth factor is added to the model, there are several numbers in the stdyx columns of the linear slope and the quadratic slope that are greater than 1. There are no R-square values greater than 1, and no negative residual variances. Would you recommend sticking with the intercept + linear slope only model in this case? I would like to keep the quadratic factor if possible because of its benefit to model fit and its conceptual importance.