Michelle posted on Thursday, June 09, 2011 - 1:34 am
Hi, I have been running a series of univariate conditional latent growth curve models on a number of separate outcomes when participants were aged 10-16. On assessing the covariances/correlations between the growth factors, Ive noted a number of inconsistent findings. Ive found that on quite a number of occasions some covariance’s were not significant, however on inspection of the standardised output these relationships which were not significant in the unstandardised models, are in fact significant in the standardised output (correlations). For ease of interpretation I was planning on reporting the standardised results as correlations, but was hoping to get some input into why these inconsistencies might be occurring? Does this appear to be problematic, and if so if you could offer any suggestions for improving such issues it would be greatly appreciated. Many thanks Michelle
I am working on an unconditional LGM and have noticed a difference between the significance values for standardized and unstandardized estimates for the slope mean. The model includes 6 time points of IRT math scores. The fit for the linear model appears good and the model estimation terminated normally. The Tech 4 output suggests a linear relationship with the estimated means increasing at each time point. I also do not see any correlations over 1 in the estimated correlation matrix.
Unfortunately, I have only 65 cases, as I am awaiting data from a second cohort. I was anxious to analyze the data. Is the problem a result of the extremely small sample size or is there something else I should consider? In the event the standardized and unstandardized results differ, which should be reported?
The reason the standard errors are different for standardized and unstandardized coefficients is because their sampling distributions are not the same. You should report what is normally reported for your discipline.