Growth model interpretation question

33 views

fitfreeinggrowth

Skip to first unread message

Brandon Weiss

unread,

Apr 6, 2020, 9:59:34 AM4/6/20

to lavaan

Hi all,

Appreciate your help. I've attached some output and just had some questions about it. Code is right below here:

Mod <- '

# intercept and slope with fixed coefficients

i =~ 1*ET1 + 1*ET2 + 1*ET3

s =~ 0*ET1 + 1*ET2 + NA*ET3

#variances

#s~~.01*s

fit<-growth(Mod,data,missing='fiml',estimator="MLR",control=list(iter.max=1000000))

1. My data is longitudinal (3 timepoints). Data is characterized by a pattern of upward slope from T1 to T2 and then a falling off by T3. As such, a perfectly linear model involving "s =~ 0*ET1 + 1*ET2 + 2*ET3" did not fit the data. I instead freed the third loading. My question is: should the .842 loading on s (see output) influence my interpretation of the slope mean (i.e., .221), and what does this slope characterize. Is this the predicted slope of best fit through the three timepoints.

2. I notice that the variance of s is not significant. Is it still worthwhile to predict variance in the slope with an exogenous moderator?

3. I have some models that will not run unless - in addition to freeing the third timepoint - I constrain a particular parameter. Note the "#s~~.01*s" in the output. In the past I have looked to constrain a parameter that isn't significant. What parameters are fair game to constrain that do not distort the key results (i.e., mean of slope, covariance of i & s)? I've received direction from my advisor to constrain the slope's variance when it's nonsignificant, but are there other things. I'd like to avoid constraining variance of s if I can because I'd like to test moderation by another variable.

Thanks very much for your help with this, and if anything I've said isn't kosher, I invite that feedback as well.

All the best,

Brandon

example_output.png

Terrence Jorgensen

unread,

Apr 9, 2020, 7:28:29 PM4/9/20

to lavaan

should the .842 loading on s (see output) influence my interpretation of the slope mean (i.e., .221)

Yes. If change were linear, your estimated final loading would be 2 (1 unit more than the second loading, just like the second loading is 1 unit more than the first). A value < 2 means there is less change between the second and third occasions than between the first and second. The fact that the third loading is less that the second means that the expected value decreases from occasion 2 to 3.

what does this slope characterize. Is this the predicted slope of best fit through the three timepoints.

These parameters are a regression equation. Just use them to calculate expected values, or ask lavaan for them:

fitted(fit)