Solved – Unable to estimate standard error after freeing first indicator in SEM model – Why is it so

structural-equation-modeling

I'm new to SEM + posting on this forum; do let me know if I'm being unclear in any way, and I'll do my best to clarify.

Background

I'm working on a SEM assignment to estimate the fit of a model, with 6 indicators loading on to a latent variable. I'm using the following packages for the assignment:

require(lavaan)

require(semPlot)

My dataset is loaded into a dataframe named my.df

The model that I'm specifying is as follows – this model automatically fixes the first factor loading of GeneralMotivation to x1 to the value of 1.0:

my.model1 <- 'GeneralMotivation =~ x1 + x2 + x3 + x4 + x5 + x6'

I know there's no need to do so, but for the sake of better understanding how SEM works, I specified the following model as well, freeing the first indicator.

problematicmy.model1 <- 'GeneralMotivation =~ NA*x1 + x2 + x3 + x4 + x5 + x6'

Problem

I then ran sem on the two models, as shown below:

my.fit1 <- sem(my.model1, data=my.df)
problematicmy.fit1 <- sem(problematicmy.model1, data=my.df)

When I specify the model using the default parameters on lavaan in my.model1, where the first indicator of the model is fixed to 1.0, there weren't any problems. The issue comes in problematicmy.model1, where I see the following error:

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.

I've also attached the output for the offending model:

lavaan (0.5-17) converged normally after  14 iterations

  Number of observations                           400

  Estimator                                         ML
  Minimum Function Test Statistic              112.214
  Degrees of freedom                                 8
  P-value (Chi-square)                           0.000

Model test baseline model:

  Minimum Function Test Statistic              360.443
  Degrees of freedom                                15
  P-value                                        0.000

User model versus baseline model:

  Comparative Fit Index (CFI)                    0.698
  Tucker-Lewis Index (TLI)                       0.434

Loglikelihood and Information Criteria:

  Loglikelihood user model (H0)              -3181.787
  Loglikelihood unrestricted model (H1)      -3125.680

  Number of free parameters                         13
  Akaike (AIC)                                6389.574
  Bayesian (BIC)                              6441.463
  Sample-size adjusted Bayesian (BIC)         6400.213

Root Mean Square Error of Approximation:

  RMSEA                                          0.180
  90 Percent Confidence Interval          0.152  0.211
  P-value RMSEA <= 0.05                          0.000

Standardized Root Mean Square Residual:

  SRMR                                           0.111

Parameter estimates:

  Information                                 Expected
  Standard Errors                             Standard

                   Estimate  Std.err  Z-value  P(>|z|)   Std.lv  Std.all
Latent variables:
  GeneralMotivation =~
    x1                0.826                               0.765    0.672
    x2                0.571                               0.528    0.534
    x3                0.829                               0.767    0.694
    x4                0.191                               0.176    0.215
    x5                0.301                               0.278    0.308
    x6                0.295                               0.273    0.322

Variances:
    x1                0.709                               0.709    0.548
    x2                0.701                               0.701    0.715
    x3                0.632                               0.632    0.518
    x4                0.640                               0.640    0.954
    x5                0.740                               0.740    0.905
    x6                0.643                               0.643    0.896
    GeneralMotvtn     0.856                               1.000    1.000

I've also attached the graphical model below for problematic.myfit1:

Offending model

Steps taken to understand the error

I first thought "okay, maybe the model is underidentified", and calculated the pieces of information I have + the number of parameters to be estimated.

Correct me if I'm wrong: There should be 21 pieces of information (6 variables, therefore [(6)(7)]/2 = 21).

However, I cannot, for the p <.05 love of all things statistics, understand why the model is underidentified if I'm simply freeing the first indicator x1. From what I'm understanding, I'm only estimating a total of 13 parameters (6 residuals for the observed variables x1 to x6, 6 factor loadings, and the variance of the latent variable GeneralMotivation). Shouldn't my model be overidentified in this case?

My guess is that

  1. Although the graphical model doesn't say this, I'm actually estimating the covariances of between the residual of indicators (i.e. x1 ~~ x2, x1 ~~ x6 etc.). If x1 is fixed at 1.0, I'm actually trying to estimate 21 parameters (5 residuals from x2 to x6, 10 residual covariances from x2 to x6, 5 residual variances from x2 to x6, 5 factor loadings from GeneralMotivation to x2 – x6, and one variance of GeneralMotivation), making the model just identified (df = 0). By freeing up x1, I have to estimate an additional 7 parameters (residual of x1, residual covariances of x1 ~~ x2 to x1 ~~ x6 and the factor loading from GeneralMotivation to x1), resulting in an underidentified model
  2. The issue isn't underidentification, but something else altogether
  3. SEM and RStudio hates me – not likely, but I'm not ruling it out.

Closing

Can anyone help me understand why the error from lavaan is popping up? Please let me know if you need more information from me.

Thank you!

Best Answer

As Maarten points out, your problem is that you have not set the scale of the second model. True, you have more observed variances/covariances than what you need to identify your model, but you still need to provide a point of reference from which other model parameters can be calculated (Brown, 2015).

You can set the scale using one of three methods:

  1. Marker variable: one factor loading per latent variable is fixed to 1
  2. Fixed factor: each latent variable's variance is fixed to 1
  3. Effects-coding: factor loadings for each latent variable are constrained to average 1

Code for each approach (using the lavaan package's HolzingerSwineford1939 dataset) is presented below. The latent variable I've created is nonsensical/poor-fitting, but it has the same number of indicators as your model, so the example will hopefully be more transferable to your situation.

library(lavaan)

#marker-variable; first factor loading fixed to 1 by default
marker.variable<-'f1=~ x1+x2+x3+x4+x5+x6'
summary(output.marker<-cfa(marker.variable, data=HolzingerSwineford1939), fit.measures=TRUE)

#fixed-factor method; manually free first factor loading/fix latent variance to 1
fixed.factor<-'f1=~ NA*x1+x2+x3+x4+x5+x6
          f1~~1*f1'
summary(output.fixed<-cfa(fixed.factor, data=HolzingerSwineford1939), fit.measures=TRUE)

#effects coding; manually free first loading/constrain loadings to average 1
effects.coding<-'f1=~ NA*x1+a*x1+b*x2+c*x3+d*x4+e*x5+f*x6
          a+b+c+d+e+f==6'
summary(output.effects<-cfa(effects.coding, data=HolzingerSwineford1939), fit.measures=TRUE)

Note that model fit is identical, regardless of which method of scale-setting that you use; the fit in all three models is $\chi^2 (df = 9) = 103.23, ~p < .001$.

Which method you should use largely depends on the nature of your data and your research goals. The marker variable method is a highly arbitrary method of scale-setting. Like Maarten stated, your latent variables will take on the units of their respective marker variables, so this approach is only informative to the extent that your marker variables are especially meaningful, or perhaps represent some "gold standard" indicator of your latent construct.

The fixed factor method, alternatively, is easy to specify, and essentially standardizes your latent variables (if you're examining mean structures, you would fix the latent means to zero as well). Since we standardize variables all the time, this is a highly intuitive and widely acceptable form of scale-setting for latent variables, though the resultant scaling is not inherently meaningful. Even so, it's probably the best method to "default" to, unless you have a strong imperative to use one of the other methods.

Effects-coding is a relative new-comer to methods of scale-setting (see Little, Slegers, & Card, 2006, for a thorough discussion). It's greatest advantage is when you are modeling latent means. When doing so, you would also constrain item intercepts to average 0. The effect of these constraints is that your latent variables will be on the exact same scale as your original items. For example, if the average of your indicators was "5", your latent mean would also be "5", though your latent variance would be smaller than you observed variance. Because the constraints on the loadings and intercepts can be more computationally demanding, especially in more complicated models, and occasionally result in convergence errors, effects-coding is probably not worth it unless you plan to examine latent means. But for the particular purpose of examining latent means, it's great.

References

Brown, T. A. (2015). Confirmatory factor analysis for applied research (2nd Edition). New York, NY: Guilford Press.

Little, T. D., Slegers, D. W., & Card, N. A. (2006). A non-arbitary method of identifying and scaling latent variables in SEM and MACS models. Structural Equation Modeling, 13, 59-72.

Related Question