finnigan posted on Wednesday, August 30, 2006 - 7:13 pm
Is it correct to say that stability of a factor mean over time is illustrated by an insignificant slope parametre,i.e if the slope parametre is not significant then the phenomenon under study is not changing hence it is stable.
This seems reasonable for what I assume is a multiple indicator growth model.
finnigan posted on Thursday, August 31, 2006 - 6:56 pm
Yes I am using a multiple indicator growth model
My data set uses four periods and it is nested : observations collected from individuals on occasions who work in teams which exist in an organisation. Is it correct to say that this is a multilevel model with five levels?
I'm using data from team and organisational contexts as both time varying and time invariant predictors of between and within variability of a dependent variable.
Do you have any examples or direction on how MPLUS might handle interaction within five levels using a multiple indicator growth model?
Your model has four levels -- time, person, team, organisation. Mplus can have three levels when one is time. You would need to treat organisation as a fixed effect or control for clustering due to organizaiton using TYPE=COMPLEX TWOLEVEL; Example 9.16 sounds close to what you want.
finnigan posted on Friday, September 01, 2006 - 10:06 am
My predictor variables are organisation and team. I want to examine the impact of these variables on slope and intercept.
How is a fixed effect modeled in MPLUS? Is the variable set to a specific value and if so how is it determined?
Can I for example hold team as a fixed effect and them free organisation? That is, if it makes theoretical sense I may vary the model's predictor variables to see which gives the best statistical fit and makes theoretical sense. Again, all of this presumes measurment invariance across the unit of analysis- individuals over four periods of measurment. Would I also need to ensure MI across team level constructs if it is a time varying covariate and used as a predictor variable? Thanks
If instead you want that slope to be fixed you simply say
y on x;
Do the modeling stepwise with simple models first. In multilevel modeling, "simple" means use only random intercepts, not random slopes. See the examples in the User's Guide for how to set up models.
Yes, you can use team as a fixed effect level, which means that team variables have no random effects (so say y on x, where x is a team variable). Then, organisation would be your "between"-level in Mplus terms. Time and person would be your "within" level.
Measurement invariance across time is needed for growth modeling. MI across teams does not seem necessary.
finnigan posted on Monday, September 11, 2006 - 1:13 pm
Linda and Bengt
I know this may sound like a daft question,but are you aware of how stability can be predicted in a multiple indicator growth model. I am trying to use a set of covariates to predict change in the slope of a growth model. As far as I understand the predictor variable may not have an significant impact and so it does not predict change(insignificant slope parametre value). However is there any way you know about predicting stability in a growth model? For example I expect organisational norms to stabilise behaviour and so may produce an insignificant slope parametre when regressed on the slope. How might I identify this in a growth model? Thanks
Question is if stability is characterized by a slope approaching zero or by less variation around a trajectory mean. In the former case having norms as a time-invariant covariate can show slope as a function of norms, or having norms as a time-varying covariate can show slope changing as a function of changing norms.
finnigan posted on Monday, September 11, 2006 - 7:15 pm
Firstly would changing norms be modeled as a growth process in itself if they change over time? That I have two growth processes. Norms are usually stable in that they may not demonstrate variability hence I would not expect to see a growth process in norms. Based on your first point, I would expect to see that norms would not have a significant relationship when regressed on the slope coefficient. I suspect the latter.