2
$\begingroup$

Background

Introduction

I have a data set consisting of data collected from a questionnaire that I wish to validate. I have chosen to use confirmatory factor analysis to analyse this data set.

Instrument

The instrument consists of 11 subscales. There is a total of 68 items in the 11 subscales. Each item is scored on an integer scale between 1 to 4.

Confirmatory factor analysis (CFA) setup

I use the sem package to conduct the CFA. My code is as below:

cov.mat <- as.matrix(read.table("http://dl.dropbox.com/u/1445171/cov.mat.csv", sep = ",", header = TRUE))
rownames(cov.mat) <- colnames(cov.mat)

model <- cfa(file = "http://dl.dropbox.com/u/1445171/cfa.model.txt", reference.indicators = FALSE)
cfa.output <- sem(model, cov.mat, N = 900, maxiter = 80000, optimizer = optimizerOptim)

Warning message:
In eval(expr, envir, enclos) : Negative parameter variances.
Model may be underidentified.

Straight off you might notice a few anomalies, let me explain.

  • Why is the optimizer chosen to be optimizerOptim?

ANS: I originally stuck with the default optimizerSem but no matter how many iterations I run, either I run out of memory first (8GB RAM setup) or it would report no convergence Things "seemed" a little better when I switched to optimizerOptim where by it would conclude successfully but throws up the error that the model is underidentified. Upon closer inspection, I realise that the output shows convergence as TRUE but iterations is NA so I am not sure what is exactly happening.

  • The maxiter is too high.

ANS: If I set it to a lower value, it refuses to converge, although as mentioned above, I doubt real convergence actually occurred.

Problem

So by now I guess that the model is really underidentified so I looked for resources to resolve this problem and found:

I followed the 2nd link quite closely and applied the t-rule:

  • I have 68 observed variables, providing me with 68 variances and 2278 covariances between variables = 2346 data points.
  • I also have 68 regression coefficients, 68 error variances of variables, 11 factor variances and 55 factor covariances to estimate making it a total of 191 parameters.
  • Since I will be fixing the variances of the 11 latent factors to 1 for scaling, I would remove them from the parameters to estimate making it a total of 180 parameters to estimate.
    • My degrees of freedom is therefore 2346 - 180 = 2166, making it an over identified model by the t-rule.

Questions

  1. Is the low variance of some of my items a possible cause for the underidentification? I asked a previous question on items with zero variance which led me to think about items which are very close to zero. Should they be removed too? Confirmatory factor analysis using SEM: What do we do with items with zero variance?
  2. After reading much, I surmise that the underidentification might be a case of empirical underidentification. Is there a systematic way of diagnosing what kind of underidentification it is? And what are my options to proceed with my analysis?

I have more questions but let's take it at these 2 for now. Thanks for any help!

$\endgroup$
  • 1
    $\begingroup$ Two of the factors are defined with only a single indicator, which if I'm not mistaken would make the CFA under-identified. Do you have any luck when removing these two factors and re-estimating? $\endgroup$ – philchalmers Feb 12 '13 at 16:39
  • $\begingroup$ @philchalmers Thanks! The CFA works when I removed the 2 items with only a single factor but for model comparison purposes, I would like to keep it there. I read from David Kenny's page "Single indicator constructs are best handled by fixing their loading to one, forcing their error variance to zero, and leaving their variances free to be estimated. Of course, the assumption of zero error variance must be justified theoretically." What are the possible justifications for fixing the error variance to zero? $\endgroup$ – RJ- Feb 15 '13 at 9:01
  • 1
    $\begingroup$ When error variances are constrained to 0 in the normal SEM case this basically means that the regression is perfect, and you wish to test the very constrained hypothesis that the variable can be predicted perfectly. However, often when a manifest variable is represented by a single latent variable (as you described) the end goal is to use the latent variable as a 'phantom' variable. These have traditionally been used to trick SEM programs into doing interesting parameter constraints that once were difficult to specify directly. David Kenny has a page for this topic as well I believe. $\endgroup$ – philchalmers Feb 17 '13 at 4:23
2
+50
$\begingroup$

Ken Bollen and I wrote about negative variance estimates (aka Heywood cases). You might want to take a look for some insights. For this huge model, God only knows how model misspecifications are going to show up, but in my experience, Heywood cases are typical outlets for the model to let the steam out when something is not fitting right.

That having said, I would try different diagnostics: first fit all of the submodels with 6 or so indicators, and see if there's anything wrong with them. In the CFA context, I would imagine that underidentification would arise only if some variables have zero coefficient/covariance with the factors they are supposed to measure. You should be able to catch that with the analysis of subscales.

Finally, for the Likert scales with 4 categories, you really should use polychoric correlations (polycor package). For one thing, the categorical nature of the data would yield the likelihood ratio tests unreliable (as if I would trust that 900 observations could give rise to 2166 independent degrees of freedom, anyway).

$\endgroup$
  • $\begingroup$ Hi @StasK, thanks for your answer. I will would study your paper carefully. I have actually removed variables with 0 variance. My question is, wouldn't items with very low variance lead to underidentification as well? If so, how low should be considered low? My second question is that if I were to use polychoric correlations, can this be used as an input to the sem package? (I've checked the help file but it only allows input to be covariance matrix or the full dataset) Thanks so much for your time and help. $\endgroup$ – RJ- Feb 15 '13 at 9:05
2
$\begingroup$

In the interest of completness of discussion, I'm posting Professor John Fox's response to this question:

http://r.789695.n4.nabble.com/Troubleshooting-underidentification-issues-in-structural-equation-modelling-SEM-td4658018.html

Your model is underidentified by virtue of two of the factors having only one observed indicator each. No SEM software can magically estimate this model as it stands. Beyond that, I won't comment on the wisdom of what you're doing, such as computing covariances between ordinal variables -- but see what I discovered below.

Removing these two variables and the associated factors produces the following model:

model <- cfa(reference.indicators=FALSE)
1: F01: I01, I02, I03
2: F02: I04, I05, I06, I07, I08, I09, I10, I11, I12, I13
3: F03: I14, I15, I16, I17, I18, I19, I20, I21, I22, I23, I24, I25, I26
4: F04: I27, I28, I29, I30, I31, I32, I33, I34
5: F05: I35, I36, I37, I38, I39, I40, I41, I42, I43
6: F07: I46, I47, I48, I49, I50, I51
7: F08: I54, I55, I56, I57, I58, I59, I60, I61, I62, I63, I64
8: F09: I65, I66, I67
9: F11: I69, I70, I71

cfa.output <- sem(model, cov.mat, N = 900)

sem() ran out of iterations, but the summary output is revealing:

summary(cfa.output)

Several of the observed variables have R^2s that round to 0 and many more are very small.

I don't have your original data, but I did look at the input covariance matrix. Here are the standard deviations of the observed variables:

Some of the standard deviations are very small, suggesting that the corresponding variables must have been close to invariant in your data set.

If you haven't already done so, I think that you might back up and look more closely at your data, and perhaps seek some competent local help.

I hope that this helps, John

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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