Christy posted on Thursday, May 19, 2005 - 9:44 am
Hello, I am using latent growth curve modeling to look at changes in alcohol use among college students with three waves of data. I am getting a negative error variance when i run an unconditional model (i.e., intercept and slope only). The univariates on my drinking variable show a potential floor effect at all 3 timepoints, yet only the error variance at wave 3 is negative. Any suggestions on how to interpret this and how to correct it?
Negative residual variances often occur when you have strong floor effects. There are two possible solutions: (1) hold the residual variances equal across time and (2) use two-part growth modeling as shown in Example 6.16 in the Mplus User's Guide.