www.cmm.bristol.ac.uk/forum

I am trying to estimate a growth curve model for a dataset of 9824 units clustered in 102 groups for 7 time points. I am able to estimate a model with the three levels (in order: time, units and groups) of my response variable with respect to a linear trend, Y=f(time). No other covariates are included in the model. When I try to model a first-order autoregressive covariance structure between level 1 errors (i.e. between time within unit), I meet the problem described below. Following the procedure in the MlWiN Supplement Manual (pag. 74-79 “A new method for estimating autocorrelated errors in continuous time”), when I type the command “star” the MlWiN returns the following error message: “”Design vector at level 2 is the wrong length (offs)”. I tried to follow the procedure for a random sub-sample of just 93 units but MlWiN returns always the same error message. Can you help me to understand the nature of my problem?

Thanks in advance!

Please check that your dataset is balanced, the AC function requires that there are no mising data.
Good luck!

moback wrote:Please check that your dataset is balanced, the AC function requires that there are no mising data.
Good luck!

But what if I checked everything and the problem is still there?

Well, unbalanced is what is meant by 'wrong length'. Sometimes everything works out if you press 'clear' and start from scratch. Hope that works, if not I fear I can't come up with any more ideas.