simple mediation analysis with multiple mediators

[Jordana Zwerling]

Hi, 

I can't figure out why my multiple mediation model has such terrible fit indices. Any suggestions? 

multipleMediation <- '

SIR_Total ~ b1 * IUS_Total + b2 * DTS_Total + b3 * OASM_Insecure + c * R_LSC_S1

IUS_Total ~ a1 * R_LSC_S1
DTS_Total ~ a2 * R_LSC_S1
OASM_Insecure ~ a3 * R_LSC_S1

indirect1 := a1 * b1
indirect2 := a2 * b2
indirect3 := a3 * b3

total := c + (a1 * b1) + (a2 * b2) + (a3 * b3)

'
fit <- sem(model = multipleMediation, data = LSH_R_2, estimator = "MLM")
summary(fit, fit.measures = TRUE, standardized = TRUE)

 Comparative Fit Index (CFI)                    0.565       0.553
  Tucker-Lewis Index (TLI)                      -0.449      -0.489

RMSEA                                          0.380       0.351
  90 Percent confidence interval - lower         0.319       0.295
  90 Percent confidence interval - upper         0.445       0.411
  P-value H_0: RMSEA <= 0.050                    0.000       0.000
  P-value H_0: RMSEA >= 0.080                    1.000       1.000
                                                                 
  Robust RMSEA                                               0.379
  90 Percent confidence interval - lower                     0.313
  90 Percent confidence interval - upper                     0.449
  P-value H_0: Robust RMSEA <= 0.050                         0.000
  P-value H_0: Robust RMSEA >= 0.080                         1.000

Standardized Root Mean Square Residual:

  SRMR                                           0.197       0.197

[Shu Fai Cheung (張樹輝)]

Is the model df equal to 3? If yes, then the misspecificaiton may be due to the three error covariances between the mediators, which are fixed to zero by default. You can verify this by examining the output of summary(). You should see no section for  covariances.

Hope this helps.

-- Shu Fai

[Jordana Zwerling]

If that is the case How would you suggest I fix this then? 

Jordana Zwerling, M.A.

Doctoral Student, Clinical Psychology

Department of Psychology 

St. John's University 

Queens, NY 11439 

--
You received this message because you are subscribed to a topic in the Google Groups "lavaan" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/lavaan/qTtk2i84Jko/unsubscribe.
To unsubscribe from this group and all its topics, send an email to lavaan+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/lavaan/42391864-ef5e-4a52-9951-de976f9ec24en%40googlegroups.com.

[Shu Fai Cheung (張樹輝)]

I found this answer by Terrence at StackExchange to a question which may be related to your case:

https://stats.stackexchange.com/a/569434/64622

There is one issue that you need to know in advance. If you add all three covariances between the error terms, then your model will be a saturated model with zero df. Nothing wrong with a model being saturated. The resutls, including the R-square for the outcome variable (SIR_Total), will be identical to what you get if you use the regression approach (e.g, using PROCESS). However, you may find it difficult to present a 0-df model.

-- Shu Fai