|
|
Can't constrain covariances equal acr... |
|
Message/Author |
|
Li Dao posted on Saturday, September 26, 2020 - 2:16 pm
|
|
|
I am trying to build a bivariate latent growth curve model with structured residuals. When I try to constrain residual covariances to be equal at each time point, however, I get a warning that the covariance matrix of the latent variables isn't positive definite. (when checking the covariance matrix, I get negative covariances for slope and intercept) The error disappears if I constrain slope and intercept covariances to be zero. The model also runs without warnings if I let all residual covariances vary or if I constrain all but the first time point to be zero. Does anyone know why this could be happening? Why would not constraining the first time point make the model run without warnings? Thanks for your help! |
|
|
There is no general answer to this. It depends on what the true model is. If these covariances are significant, I would not eliminate them and if holding them equal across time causes the warning, I would not hold them equal. |
|
Back to top |
|
|