unconditional quadratic latent growth curve. there is convergence, but flag indicates standard errors may not be trustworthy due to non-positive def 1st-order derivative matrix; states problem may be due to starting values or noidentification. the model is definitely identified and i've set start values for every estimated parameter in the model but the flag is still there. this is strange, because I actually am dealing with good data for once (i.e. good cov coverage and very normally distributed repeated measures).
The estimates seem correct. any advice? can i comfortably ignore the flag?