Cantyo Dannis
Cantyo Dannis

Reputation: 1

Longitudinal Mediation Analysis with Binary DV

So, I am testing a longitudinal mediation analysis, and here is my model:

Model

JOV and NFS are continuous variables, but SS is a binary variable (0 v 1). So, I put this script in R:

modelz <- '

# Autoregressive paths

T2_JoV.23_Tot ~ T1_JoV.23_Tot

T3_JoV.23_Tot ~ T2_JoV.23_Tot

T3_JoV.23_Tot ~ T1_JoV.23_Tot

 

T2_SNeed1_Tot ~ T1_SNeed1_Tot

T3_SNeed1_Tot ~ T2_SNeed1_Tot

T3_SNeed1_Tot ~ T1_SNeed1_Tot

 

T2_SS1_Prev_2 ~ T1_SS1_Prev_2

T3_SS1_Prev_2 ~ T2_SS1_Prev_2

T3_SS1_Prev_2 ~ T1_SS1_Prev_2

 

# Cross-lagged paths

T3_SNeed1_Tot ~ T2_JoV.23_Tot

T2_SS1_Prev_2 ~ T1_SNeed1_Tot

 

# intercorrelation for control

T1_SNeed1_Tot ~~ T1_JoV.23_Tot

T1_SS1_Prev_2 ~~ T1_SNeed1_Tot

T1_SS1_Prev_2 ~~ T1_JoV.23_Tot

 

T2_SNeed1_Tot ~~ T2_JoV.23_Tot

T2_SS1_Prev_2 ~~ T2_SNeed1_Tot

T2_SS1_Prev_2 ~~ T2_JoV.23_Tot

 

T3_SNeed1_Tot ~~ T3_JoV.23_Tot

T3_SS1_Prev_2 ~~ T3_SNeed1_Tot

T3_SS1_Prev_2 ~~ T3_JoV.23_Tot

 

# Direct effect of X on Y (c path)

T3_SS1_Prev_2 ~ c*T1_JoV.23_Tot

 

# Effect of X on M (a path)

T2_SNeed1_Tot ~ a*T1_JoV.23_Tot

 

# Effect of M on Y (b path)

T3_SS1_Prev_2 ~ b*T2_SNeed1_Tot

 

#label effects with :=

direct := a

indirect := a*b

total := c + (a*b)

'

 

Fitz1 <- sem(modelz, data = jov4_sem, ordered = c("T2_SS1_Prev_2", "T3_SS1_Prev_2"), estimator = "WLSMV")

summary(fitz1, standardized = TRUE, fit.measures = TRUE, rsquare = TRUE)

parameterEstimates(fitz1, standardized = TRUE)

But it always says this warning and when I print the summary, it does not show the SE, z-value, and p value. It also shows the same warning if I put T1_SS1_Prev_2 as the ordered variable.

Warning message:

In lavaan::lavaan(model = modelz, data = jov4_sem, ordered = c("T2_SS1_Prev_2",  :

  lavaan WARNING:

    the optimizer warns that a solution has NOT been found!

However, this is not a problem if I do not specify the ordered variables:

fitz2 <- sem(modelz, data = jov4_sem)

summary(fitz2, standardized = TRUE, fit.measures = TRUE, rsquare = TRUE)

I wonder what is wrong with the first model (i.e., fitz1), and if I can do the second approach (i.e., fitz2; without specifying which are the ordered variables and WLS estimator? Any idea how should I address this?

Upvotes: 0

Views: 54

Answers (0)

Related Questions