Another issue related to path analysis using the manymome data

[weberhw]

library(lavaan)

library(manymome)
data(data_mome_demo_missing)
print(head(data_mome_demo_missing), digits = 3)
x1 x2 m1 m2 m3 y1 y2 w1 w2 c1 c2
1 4.54 4.86 4.44 NA 4.51 7.58 4.01 4.40 6.12 6.37 5.28
2 4.72 6.46 3.09 3.75 4.66 1.56 5.87 4.65 4.23 5.89 5.74
3 4.17 4.90 4.50 5.60 3.53 5.26 3.16 3.07 NA 4.86 6.49
4 6.47 5.75 5.92 4.12 6.92 4.80 5.80 6.19 5.12 3.04 4.40
5 6.36 5.46 6.30 7.83 6.16 19.83 4.63 3.63 9.61 5.19 6.36
6 4.36 3.85 5.20 4.48 1.68 5.09 2.89 6.54 4.70 6.41 3.48
data_centered <- as.data.frame(scale(data_mome_demo_missing, scale = FALSE))
data_centered$w1x1 <- data_centered$w1 * data_centered$x1
data_centered$w2m2 <- data_centered$w2 * data_centered$m2
mod <-

• "
• m1 ~ x1 + w1 + x2 + c1 + c2
• m2 ~ m1 + c1 + c2
• m3 ~ x2 + x1 + c1 + c2
• y1 ~ m2 + w2 + x1 + x2 + m3 + c1 + c2
• y2 ~ m3 + x2 + x1 + m2 + c1 + c2

• Covariances for the error term of m2

• m2 ~~ w2

• Covariances between all exogenous variables

• w2 ~~ x1 + w1 + x2 + c1 + c2
• x1 ~~ w1 + x2 + c1 + c2
• w1 ~~ x2 + c1 + c2
• x2 ~~ c1 + c2
• c1 ~~ c2
• "

library(lavaan)
fit <- sem(mod, data_centered, meanstructure = TRUE,

•        fixed.x = FALSE, missing = "fiml")
  

fit.s = summary(fit, fit.measures = TRUE)
all_p <- all_indirect_paths(fit)
all_p
Call:
all_indirect_paths(fit = fit)
Path(s):
path
1 m1 -> m2 -> y1
2 m1 -> m2 -> y2
3 x1 -> m1 -> m2
4 x1 -> m1 -> m2 -> y1
5 x1 -> m3 -> y1
6 x1 -> m1 -> m2 -> y2
7 x1 -> m3 -> y2
8 w1 -> m1 -> m2
9 w1 -> m1 -> m2 -> y1
10 w1 -> m1 -> m2 -> y2
11 x2 -> m1 -> m2
12 x2 -> m1 -> m2 -> y1
13 x2 -> m3 -> y1
14 x2 -> m1 -> m2 -> y2
15 x2 -> m3 -> y2
16 c1 -> m1 -> m2
17 c1 -> m1 -> m2 -> y1
18 c1 -> m2 -> y1
19 c1 -> m3 -> y1
20 c1 -> m1 -> m2 -> y2
21 c1 -> m2 -> y2
22 c1 -> m3 -> y2
23 c2 -> m1 -> m2
24 c2 -> m1 -> m2 -> y1
25 c2 -> m2 -> y1
26 c2 -> m3 -> y1
27 c2 -> m1 -> m2 -> y2
28 c2 -> m2 -> y2
29 c2 -> m3 -> y2
ind_t <- many_indirect_effects(paths = all_p,

•                              fit = fit,
  

•                              mc_out = mc_med,
  

•                              boot_ci = F,
  

•                              R = 1000,
  

•                              mc_ci = TRUE,
  

•                              standardized_x = TRUE,
  

•                              standardized_y = TRUE,
  

•                              seed = 43143)
  

An error occurred at (function (x, y, m = NULL, fit = NULL, est = NULL, implied_stats = NULL, ...): No path from 'm1' to 'y1' through 'm2'. Please check the arguments x, y, and m.
This error message occurred when I used your data, removed the moderation term, and tested all indirect effects.

[sfcheung]

May you use the code format to paste the code? I

https://docs.github.com/en/get-started/writing-on-github/working-with-advanced-formatting/creating-and-highlighting-code-blocks

It is not easy to reproduce the analysis. As shown on GitHub, some symbols are interpreted incorrectly as markdown format.

I tried to do what you shared but did not get the error:

library(manymome)

data(data_mome_demo_missing)
print(head(data_mome_demo_missing), digits = 3)
#>     x1   x2   m1   m2   m3    y1   y2   w1   w2   c1   c2
#> 1 4.54 4.86 4.44   NA 4.51  7.58 4.01 4.40 6.12 6.37 5.28
#> 2 4.72 6.46 3.09 3.75 4.66  1.56 5.87 4.65 4.23 5.89 5.74
#> 3 4.17 4.90 4.50 5.60 3.53  5.26 3.16 3.07   NA 4.86 6.49
#> 4 6.47 5.75 5.92 4.12 6.92  4.80 5.80 6.19 5.12 3.04 4.40
#> 5 6.36 5.46 6.30 7.83 6.16 19.83 4.63 3.63 9.61 5.19 6.36
#> 6 4.36 3.85 5.20 4.48 1.68  5.09 2.89 6.54 4.70 6.41 3.48
data_centered <- as.data.frame(scale(data_mome_demo_missing, scale = FALSE))
data_centered$w1x1 <- data_centered$w1 * data_centered$x1
data_centered$w2m2 <- data_centered$w2 * data_centered$m2

mod <-
"
m1 ~ x1 + w1 + x2 + c1 + c2
m2 ~ m1 + c1 + c2
m3 ~ x2 + x1 + c1 + c2
y1 ~ m2 + w2 + x1 + x2 + m3 + c1 + c2
y2 ~ m3 + x2 + x1 + m2 + c1 + c2
# Covariances for the error term of m2
m2 ~~ w2
# Covariances between all exogenous variables
w2 ~~ x1 + w1 + x2 + c1 + c2
x1 ~~ w1 + x2 + c1 + c2
w1 ~~ x2 + c1 + c2
x2 ~~ c1 + c2
c1 ~~ c2
"

library(lavaan)
#> This is lavaan 0.6-19
#> lavaan is FREE software! Please report any bugs.
fit <- sem(mod,
           data_centered,
           meanstructure = TRUE,
           fixed.x = FALSE,
           missing = "fiml")

fit.s = summary(fit, fit.measures = TRUE)

all_p <- all_indirect_paths(fit)
all_p
#> Call: 
#> all_indirect_paths(fit = fit)
#> Path(s): 
#>    path                
#> 1  m1 -> m2 -> y1      
#> 2  m1 -> m2 -> y2      
#> 3  x1 -> m1 -> m2      
#> 4  x1 -> m1 -> m2 -> y1
#> 5  x1 -> m3 -> y1      
#> 6  x1 -> m1 -> m2 -> y2
#> 7  x1 -> m3 -> y2      
#> 8  w1 -> m1 -> m2      
#> 9  w1 -> m1 -> m2 -> y1
#> 10 w1 -> m1 -> m2 -> y2
#> 11 x2 -> m1 -> m2      
#> 12 x2 -> m1 -> m2 -> y1
#> 13 x2 -> m3 -> y1      
#> 14 x2 -> m1 -> m2 -> y2
#> 15 x2 -> m3 -> y2      
#> 16 c1 -> m1 -> m2      
#> 17 c1 -> m1 -> m2 -> y1
#> 18 c1 -> m2 -> y1      
#> 19 c1 -> m3 -> y1      
#> 20 c1 -> m1 -> m2 -> y2
#> 21 c1 -> m2 -> y2      
#> 22 c1 -> m3 -> y2      
#> 23 c2 -> m1 -> m2      
#> 24 c2 -> m1 -> m2 -> y1
#> 25 c2 -> m2 -> y1      
#> 26 c2 -> m3 -> y1      
#> 27 c2 -> m1 -> m2 -> y2
#> 28 c2 -> m2 -> y2      
#> 29 c2 -> m3 -> y2

mc_med <- do_mc(fit,
                R = 100)
#> Stage 1: Simulate estimates
#> Stage 2: Compute implied statistics

ind_t <- many_indirect_effects(
            paths = all_p,
            fit = fit,
            R = 1000,
            mc_ci = TRUE,
            standardized_x = TRUE,
            standardized_y = TRUE,
            seed = 43143
          )
#> Stage 1: Simulate estimates
#> Stage 2: Compute implied statistics

ind_t
#> 
#> ==  Indirect Effect(s) (Both x-variable(s) and y-variable(s) Standardized)  ==
#>                         std  CI.lo CI.hi Sig
#> m1 -> m2 -> y1        0.173  0.093 0.258 Sig
#> m1 -> m2 -> y2       -0.019 -0.084 0.048    
#> x1 -> m1 -> m2        0.186  0.109 0.264 Sig
#> x1 -> m1 -> m2 -> y1  0.058  0.027 0.094 Sig
#> x1 -> m3 -> y1        0.012 -0.034 0.059    
#> x1 -> m1 -> m2 -> y2 -0.006 -0.030 0.017    
#> x1 -> m3 -> y2        0.177  0.104 0.254 Sig
#> w1 -> m1 -> m2        0.002 -0.070 0.078    
#> w1 -> m1 -> m2 -> y1  0.001 -0.024 0.024    
#> w1 -> m1 -> m2 -> y2 -0.000 -0.006 0.006    
#> x2 -> m1 -> m2        0.092  0.021 0.169 Sig
#> x2 -> m1 -> m2 -> y1  0.029  0.005 0.058 Sig
#> x2 -> m3 -> y1        0.020 -0.061 0.098    
#> x2 -> m1 -> m2 -> y2 -0.003 -0.016 0.009    
#> x2 -> m3 -> y2        0.306  0.215 0.409 Sig
#> c1 -> m1 -> m2       -0.062 -0.140 0.007    
#> c1 -> m1 -> m2 -> y1 -0.019 -0.046 0.002    
#> c1 -> m2 -> y1       -0.007 -0.045 0.030    
#> c1 -> m3 -> y1        0.002 -0.010 0.019    
#> c1 -> m1 -> m2 -> y2  0.002 -0.007 0.012    
#> c1 -> m2 -> y2        0.001 -0.007 0.012    
#> c1 -> m3 -> y2        0.034 -0.034 0.101    
#> c2 -> m1 -> m2       -0.067 -0.148 0.004    
#> c2 -> m1 -> m2 -> y1 -0.021 -0.051 0.001    
#> c2 -> m2 -> y1        0.037 -0.002 0.080    
#> c2 -> m3 -> y1       -0.002 -0.017 0.010    
#> c2 -> m1 -> m2 -> y2  0.002 -0.006 0.013    
#> c2 -> m2 -> y2       -0.004 -0.024 0.011    
#> c2 -> m3 -> y2       -0.027 -0.101 0.045    
#> 
#>  - [CI.lo to CI.hi] are 95.0% Monte Carlo confidence intervals with 1000
#>    replications.
#>  - std: The standardized indirect effects.
#> 

^{Created on 2025-07-06 with reprex v2.1.1}

May you use reprex() to share a reproducible example?

https://reprex.tidyverse.org/

In the simplest case, just copy the code to the clipboard and run reprex() in the console. It will generate the output with the code used that can be pasted directly on GitHub, as I did above.

[sfcheung]

One more thing. The model with moderation (product terms) needs to add many covariances manually because of a behavior in lavaan. This is common in lavaan when we have product terms.

However, if a model has no product term, then this is not recommended to add those covariances manually. If any covriances are added, this overrides lavaan's default behavior and the model fitted may not be the one dsired:

https://groups.google.com/g/lavaan/c/lTnFxlqA_0g/m/rfIUJWHyAAAJ