5
$\begingroup$

I created following structural equation model from iris data set using lavaan package in R:

enter image description here

How do I interpret these numbers. The output (of sem() function of lavaan package) is given below. It did not give any P values: 

lavaan (0.5-18) converged normally after  64 iterations

  Number of observations                           150

  Estimator                                         ML
  Minimum Function Test Statistic                   NA
  Degrees of freedom                                -4
  Minimum Function Value               0.0000000000000

Parameter estimates:

  Information                                 Expected
  Standard Errors                             Standard

                   Estimate  Std.err  Z-value  P(>|z|)
Latent variables:
  sepf =~
    Sepal.Length      1.000
    Sepal.Width      -0.469
  petf =~
    Petal.Length      1.000
    Petal.Width       0.507
  lenf =~
    Petal.Length      1.000
    Sepal.Length     -0.177
  widf =~
    Sepal.Width       1.000
  strf =~
    sepf              1.000
    petf              2.084
  bulkf =~
    lenf              1.000
    widf              0.579

Regressions:
  strf ~
    Species           0.842
  bulkf ~
    Species           0.290

Covariances:
  strf ~~
    bulkf             0.065

Variances:
    Sepal.Length      0.361
    Sepal.Width       0.129
    Petal.Length      0.231
    Petal.Width       0.047
    sepf             -0.120
    petf             -0.220
    lenf             -0.179
    widf              0.084
    strf              0.053
    bulkf            -0.025
-----------------------------------------------
Warning messages:
1: In lav_data_full(data = data, group = group, group.label = group.label,  :
  lavaan WARNING: unordered factor(s) with more than 2 levels detected in data: Species
2: 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.

3: In lavaan::lavaan(model = model, data = mydf, model.type = "sem",  :
  lavaan WARNING: some estimated variances are negative
4: In lavaan::lavaan(model = model, data = mydf, model.type = "sem",  :
  lavaan WARNING: covariance matrix of latent variables is not positive definite; use inspect(fit,"cov.lv") to investigate.
5: In sqrt(ETA2) : NaNs produced
6: In sqrt(ETA2) : NaNs produced
7: In sqrt(ETA2) : NaNs produced
>

Do I just take large estimates as signficant? Thanks for your insight.

Improve this question
$\endgroup$
7
  • $\begingroup$ Might be wrong, but propably the negative degrees of freedoms indicate that your model is underfitted and thus, the proposed model cannot be estimated/ tested. Therefore you cannot make any conclusions based on your model. If this is wrong, please correct me, statistics pros. $\endgroup$
    – Jens
    Commented Jun 19, 2015 at 19:19
  • 1
    $\begingroup$ @Jens you're completely correct, this model is vastly over-identified and hence it is not unique (therefore, it's likelihood surface has no curvature, and no standard errors can be computed). The model converges to some location, but given some different starting values it will almost certainly converge to an entirely different location that fits equally well. Therefore, do not interpret this model at all, as it is largely meaningless, and lavaan surely printed a warning message saying that the model is probably not identified. $\endgroup$
    – philchalmers
    Commented Jun 19, 2015 at 20:08
  • $\begingroup$ I have added the warnings that came with the output. Can we say if there is a particular latent factor or a particular association is completely wrong and the model should be retried after its removal? Also, there is warning that there are more than 2 levels in a factor (Spc or Species variable). So do I take that more than 2 factors are not allowed and I have to create dummy variables for such factor variables? $\endgroup$
    – rnso
    Commented Jun 20, 2015 at 0:16
  • $\begingroup$ You really can't say anything at all about this model. On top of being completely unidentified it is badly broken (negative variance terms abound). So I would throw this model out and try something completely different. $\endgroup$
    – philchalmers
    Commented Jun 20, 2015 at 0:23
  • 1
    $\begingroup$ The dashed lines appear to be fixed coefficients, which are often used for identifying latent variables (setting their metric). $\endgroup$
    – philchalmers
    Commented Jun 21, 2015 at 14:23

1 Answer 1

Reset to default
7
$\begingroup$

Moving comments above to an answer:

You really can't say anything at all about this model. On top of being completely unidentified it is badly broken (negative variance terms abound). So I would throw this model out and try something completely different.

Edit: This model is vastly over-identified and hence it is not unique (therefore, it's likelihood surface has no curvature, and no standard errors can be computed). The model converges to some location, but given some different starting values it will almost certainly converge to an entirely different location that fits equally well. Therefore, do not interpret this model at all, as it is largely meaningless, and lavaan surely printed a warning message saying that the model is probably not identified

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.