Help!lavaan WARNING: could not compute standard errors!

[Xiaochi Zhang]

Dear all,
I tried to run a path model with lavaan and want to compare the paths between 2 countries:

Model3<-"dass_dep_t2~age_t1+gender+eq_a_t1+eq_4_t1+stress_t1+angst_t1+dep_t1+social_sup_all_t1+resilience_all_t1+pmh9_all_t1+FASII+rhy10_all_t1+childwisht1

dep_t1~angst_t1+stress_t1+social_sup_all_t1+resilience_all_t1+pmh9_all_t1

angst_t1~dep_t1+stress_t1+social_sup_all_t1+resilience_all_t1+pmh9_all_t1

stress_t1~dep_t1+angst_t1+social_sup_all_t1+resilience_all_t1+pmh9_all_t1

pmh9_all_t1~dep_t1+angst_t1+stress_t1+social_sup_all_t1+resilience_all_t1

resilience_all_t1~dep_t1+angst_t1+stress_t1+social_sup_all_t1+pmh9_all_t1

social_sup_all_t1~dep_t1+angst_t1+stress_t1+resilience_all_t1+pmh9_all_t1"

fit1<-sem(Model3,data=Land2,group= "land",estimator="ML",std.lv = TRUE)

Warning message:
In lav_model_vcov(lavmodel = lavmodel, lavsamplestats = lavsamplestats,  :
  lavaan WARNING: could not compute standard errors!
  lavaan NOTE: this may be a symptom that the model is not identified.

fitMeasures(fit1, c("chisq","df","cfi", "rmsea", "srmr"))

 chisq       df      cfi    rmsea     srmr 
4534.628   54.000    0.834    0.132    0.107 

But " fit2<-cfa(Model3,data=Land2,group= "land",group.equal="regressions",estimator="ML",std.lv = TRUE)" can compute standard errors.

my questions are:
1.Does it mean, that Model3 is meaningless?
2. How can I get the standard errors for fit1?

Thank U all!

Chi

[Edward Rigdon]

There is a statistical identification problem. You have many sets of reciprocal relations (where the same mode includes A -> B and B -> A). For each of these sets, at the barest minimum, at least one of the two reciprocating variavles (A or B) must have an unshared predictor--a variable that predicts the one variable but does not predict the other one.. See, e.g.,:

https://www.researchgate.net/publication/247503395_A_Necessary_and_Sufficient_Identification_Rule_for_Structural_Models_Estimated_in_Practice

Better would be that both of the reciprocating variables have their own unshared predictor, or "instrument."

Why does the model converge with stardard errors when regression slopes are held equal across groups? That gets complicated, but understand that lavaan's judgment on the matter is based on the "information matrix," the matrix of second order derivatives of the discrepancy function with respect to the free parameters of the model. In the constrained case, lavaan found this matrix to be positive definite, while in the former cae, the matrix or a submatrix was found to be not positive definite, triggering the message.

--
You received this message because you are subscribed to the Google Groups "lavaan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lavaan+unsubscribe@googlegroups.com.
To post to this group, send email to lav...@googlegroups.com.
Visit this group at https://groups.google.com/group/lavaan.
For more options, visit https://groups.google.com/d/optout.

[Xiaochi Zhang]

Dear Edward,
Thank U so much for the help!
in my model, the variables from the following 6 regressions are all correlated with each other.  will it help to compute standard error, if I try to correlate all variables with "_t1" insted of the double headed paths(A->B, B->A)? since no unshared predictor is available here.
dep_t1~angst_t1+stress_t1+social_sup_all_t1+resilience_all_t1+pmh9_all_t1

angst_t1~dep_t1+stress_t1+social_sup_all_t1+resilience_all_t1+pmh9_all_t1

stress_t1~dep_t1+angst_t1+social_sup_all_t1+resilience_all_t1+pmh9_all_t1

pmh9_all_t1~dep_t1+angst_t1+stress_t1+social_sup_all_t1+resilience_all_t1

resilience_all_t1~dep_t1+angst_t1+stress_t1+social_sup_all_t1+pmh9_all_t1

social_sup_all_t1~dep_t1+angst_t1+stress_t1+resilience_

all_t1+pmh9_all_t1


Chi

To unsubscribe from this group and stop receiving emails from it, send an email to lavaan+un...@googlegroups.com.

[Edward Rigdon]

If you replace each reciprocal relation (A -> B and B -> A, estimating 2 free parameters) with correlated residuals (A ~~ B, estimating 1 free parameter), that will make the model identified. I don't know if that will suit your research purposes--that is a much broader question.

To unsubscribe from this group and stop receiving emails from it, send an email to lavaan+unsubscribe@googlegroups.com.

[Xiaochi Zhang]

Dear Edward,
Thank U again for the rapid answer!! I now tried the model with correlated residuals and the model was identified.
So, does it mean, that lavaan can identify models with reciprocal relations? if it not, which package or program can handle this problem?
Regards,
Chi

[Edward Rigdon]

This is not a lavaan problem. It is a matter of algebra. You can read the article I referenced, or any basic book on structural  models. Not every model that can be specified can be estimated. Some models can be evaluated using techniques like confirmatory tetrad analysis which cannot be evaluated by methods that actually estimate the parameters.

To unsubscribe from this group and stop receiving emails from it, send an email to lavaan+unsubscribe@googlegroups.com.