Skip to first unread message
Ivett Pipoly
Apr 30, 2019, 6:41:34 AM4/30/19
to lavaan
Dear all,
I have created a path for modelling the effects of rare weather events on bird reproduction, so the distribution of some of the variables are highly skewed (lot of 0s). Additionally, some variables are highly correlated (e.g. mean temperature with number of heat days), but I allow the correlations with '~~'. I always have one "dependent" variable (bird reproduction) where all the relationships goes (this observed variable has a lot of connections). When I fit the path, I got bad model fit measures (I have large >600 number of observations and I also have enough free parameters and df --> 39 and 30 respectively. I would like to ask is it possible to get bad model fit because of the original non-normally distributed variables in the regressions? If it is relevant in this case, then what can I do to reach better model fit? Or, somewhere I've read that if model fit is problematic, the standardized residual matrix can help to find out the problem. If it is true, how can I inspect the standardised residual matrix to see what variable(s) cause the problem(s)?
My original plan was to make a multi group analysis comparing the effects in two different habitat types (so habitat would be the grouping variable), but could it be a problem if year is in the path as dummy coded exogenous variables and I have some years when I haven't had any observations from one of the habitats?
Sorry to asking a lot, and thank you in advance for any help with these issues.
Bests, Ivett
Terrence Jorgensen
May 8, 2019, 5:22:19 AM5/8/19
to lavaan
somewhere I've read that if model fit is problematic, the standardized residual matrix can help to find out the problem. If it is true, how can I inspect the standardised residual matrix to see what variable(s) cause the problem(s)?
resid(fit, type = "cor")Your remaining questions are not about lavaan, and you can probably find more help on SEMNET:
When you post your question there, make sure you describe your variables more concretely. It sounds like the first variables you mentioned are not continuous, but counts of rare events, in which case it is not an issue of "skew" but of "zero-inflated count data", which lavaan cannot handle (but I think Mplus can).
Terrence D. Jorgensen
Assistant Professor, Methods and Statistics
Research Institute for Child Development and Education, the University of Amsterdam
Reply all
Reply to author
Forward
0 new messages