Individual varying times of observation
Message/Author
 Daniel Rodriguez posted on Wednesday, March 13, 2013 - 10:33 am
Hello, I ran a growth curve model with individually varying times of observation and am wondering how to interpret the results in terms of units of time. My time variable is hours from birth and my dependent variable is attention/orientation. Since I am not setting any specific time unit, could I say that hours from birth is my unit such that an hour increase in birth results in a beta change in attention?
 Linda K. Muthen posted on Wednesday, March 13, 2013 - 6:00 pm
Yes.
 Yaoyue Hu posted on Thursday, December 12, 2013 - 11:59 am
Hi,

I am running a growth curve model of 4 time-point data with individually varying time of observation. In this case, chi-square cannot be calculated as the covariance matrices could be held constant across individuals.

However, given loglikelohood, AIC and BIC, how could I tell which model fits the data better, the linear growth curve model or the non-linear model (add a quadratic growth factor)?

Many thanks!
 Linda K. Muthen posted on Friday, December 13, 2013 - 10:22 am
See if the quadratic mean is significant. If so, go with the quadratic model.
 Yaoyue Hu posted on Sunday, December 15, 2013 - 2:43 am
Dear Linda

Given my data, I have more questions about the way to fit the growth curve model:

1. I have a four time-point data of cognitive function from participants aged 45-70 at the first measurement occasion, as well as individually varying time of follow-up.
As suggested by Mehta & West (2000), I am planing to carry out my analysis centering the age at 59£¨mean age at baseline£©at every measurement occasion. Therefore, the interept factor is the mean cognitive function at 59 years old and the slope factor is the rate of change in cognitive funtion per year.
Does the age in my data have a too wide span, which might not be suitable to use the centering age at a specific age technique?

2.When I fitted an unconditional non-linear growth curve model, the linear slope is not significant but the quadratic slope is. Further adding covariates into the non-linear model, neither linear slope nor quadratic slope is significant any more.
Should I choose the model, the linear or non-linear, based on unconditional model or conditional one?

Again, many thanks for your help!
 Bengt O. Muthen posted on Sunday, December 15, 2013 - 11:01 am
1. That centering should be fine. The wide age range has more impact on the growth function - whether it can be said to be linear or e.g. quadratic.

2. When you add covariates, your output shows the intercepts, not the means, of the growth factors. The means are seen in TECH4.