Hot answers tagged

4

The advantages of SEM over regression are to model multiple outcomes and to model latent variables. If you have either of those in an experiment, then you should use SEM. For example, consider a mediation analysis. You have multiple outcomes (the mediator and the final outcome), and you can model those relationships using a system of simultaneous equation ...


4

Yes, you could do this, but it would be weird. It would be sort of like having a common factor and specific factors in a confirmatory factor analysis, except the factors are growth factors. I predict you are going to run into problems identifying this unless your sample is huge. You're asking every mean and covariance to provide you with a lot of ...


4

Clearly it's possible, because it happened. Typically when you have very low chi-squares, you sometimes have very lower CFI/TLI - that was the first think I looked at, because they indicate lower power. You don't have low power, you just have a well fitting model. This is not a problem.


4

The model is not identified, which means there is no unique solution to the estimation problem. Identification is a challenging topic, one that is often overlooked. First, your graphical model is incorrect. You have manifest variables pointing to the latent variables, when in your model, the manifest variables measure the latent variable. Second, the cause ...


3

It seems me that the interpretation of SEM in econometrics is matters of debate. Pearl defend strongly the causal interpretation of SEM and its parameters. For example you can read: The Causal Foundations of Structural Equation Modeling - Pearl (2012). He consider terms like simultaneous equation model (SIM) as synonym of SEM. In Pearl opinion (pag 3) the ...


3

Curran et al. (2003) write that: It is common to report 90 percent confidence intervals for the RMSEA, primarily because of the resulting direct link to hypothesis testing based on the model test statistic. Three hypothesis tests sometimes reported in the SEM literature are. The test of exact fit, $H_{0}: \epsilon = 0$ and The test of close fit, $...


3

The model with the factor is presumably nested within the model without the factor, so you can just perform a simple likelihood ratio test. Fit the model with the factor and the model without the factor, then use lavTestLRT to compare them. If the test is significant, there is evidence that the model with the factor fits better than the model without it.


3

I think the best walkthrough of how to sample size plan using lavaan and simsem I have ever read was in a very short and very accessible text called Latent Variable Modeling Using R: A Step-by-Step Guide by A. Alexander Beaujean. His examples are admittedly relatively simple, but they should provide an excellent basis to start from when thinking about a ...


3

No. You cannot test for normality only knowing the first two moments of a distribution (correlation, covariance, mean, and SDs are all first-and-second including product moment estimators). You need to know the entire sample to test for goodness of fit for a particular distribution. At the very least, you would need to know skewness and kurtosis to exclude ...


3

I don't know about books, honestly there is not that much into it, but there are some papers that are usually recommended as references Mplus userguide chapter 12, https://www.statmodel.com/ugexcerpts.shtml Although examples are in Mplus, it should be easy to apply them to lavaan Muthen, L. K., & Muthen, B. O. (2002). How to use a Monte Carlo study ...


3

Note that it is very common to have a non-significant Chi-squared fit statistic in CFA testing, as it is heavily sensitive to large sample sizes and higher model complexity (i.e. a number of indicators in your model). In your case, it is very unsurprising that the Chi-squared in non-significant, but both CFI and TLI are large. I will gently echo @JeremyMiles ...


2

Just don’t. Post hoc power is meaningless. However, if you insist on it, you can compute it correctly by noting that power is the probability of obtaining statistical significance. Therefore: If H_0 was rejected, post hoc power = 1.0 Otherwise, post hoc power = 0.0


2

This is likely "whack-a-mole" as my advisor likes to put it. Your model is very clearly misspecified because your exogenous latent variables are uncorrelated with each other and with the exogenous predictors. This is an extremely strong assumption to make, is almost surely not what you intended to have happen, and will cause the misspecification to propagate ...


2

I don't see anybody jumping to answer this, so I'm going to turn my comment into an answer despite not having a complete solution to this. However, the poster is asking about the standard error of measurement in a non-item response theory (IRT) model. Because he or she is referring to a continuous response, I have to assume the indicator variables are ...


2

To add (and then to digress a bit...): selecting a particular marker variable over another can be a reasonable thing to do if one is known to be a high-consensus "gold-standard" indicator of your latent variable of interest (Little, 2013). Imagine you have three tests $x_1$, $x_2$, and $x_3$, that attempt to assess Latent Variable $X$. Perhaps $x_1$ and $x_2$...


2

As Patrick mentioned in the comments, choosing a scaling indicator sets the scale of the latent variable. If you want that variable to play a role in another model, it makes sense for that variable to be on the same scale as the construct you are using it to represent. For example, let's say you want to measure someone's income, but you believe their ...


2

Welcome gipicci! You have already fit your two competing models--in your case, the configural and loading invariant models. These models are "nested", in that they contain the same observed indicators, and the "nested" model (loading invariant) is a more parsimonious/constrained version of the more complicated "full" or "parent" model. Subsequently, the ...


2

Of course you can use. Not restricted to polynomial regression, depending on your data, you can use any suitable transformation. The question is about if that transformation fits your data well, or not. By using your model, you assume that your DV and IVs have the following relation: $$\log y=a+b\log{x_1}+c\log{x_2}+d\log{x_1}\log{x_2}+e(\log{x_1})^2+f(\log ...


2

Since you mention feedback I would suggest you view your system as evolving in time, namely a time series. In this case I would at least consider Sugihara's S-map and CCM methods and forget about the final diagnosis for a while. The advantage of these methods is that they are specifically aimed at data arising from dynamical systems. CCM tells you which ...


2

If you don't have the original data, your approach makes sense. You would need to take the parameters from the original model, then constrain each of the model parameters in the new model. The most strict test would be to constrain all model parameters to what the original paper found. This would be everything: factor loadings, residual variances, factor ...


2

Thanks for the help Jeremy. I found the solution in Bollen (1989), p. 111. Assume again the three-wave version of the model above, $x_{it} = \alpha_i + \delta_{it}$ and, for the sake of demonstration, assume the observed covariance matrix $S$ is $\begin{align} S & = \begin{bmatrix} x_{1}^{2} & & \\ x_{2}x_{1} & x_{2}^{2} & \\ x_{3}...


2

You can just run as.data.frame() on the sempreds object to turn it into a data.frame. Note that this isn't actually prediction; you're estimating factor scores. This covered in any SEM textbook (e.g., Bollen 1989). Factor scores are imperfect estimates of the latent variable that can often be used in subsequent analyses or for descriptive purposes. That ...


2

The multilevel model allows you to simultaneously model within- and between-plot variation. Consider a simplified multilevel model, with the Xs representing within-plot variables and W representing a between-plot variable. The within-plot model: $y_{ij}$ = $\beta0_j$ + $\beta1_jX_{1ij}$ + $\beta2X_{2ij}...$ + $e_{ij}$ Here you can include as many X ...


1

There are several reasons you might add more than one outcome (endogenous) variable to SEM. You can examine the partial correlation. You can carry out multivariate tests of two (or more parameters simultaneusly). You can compare the coefficients predicting each outcome. As well as the reason you suggest.


1

The key distinction in my view between SEM as psychologists use the term and structural estimation/modeling as economists use the term is economists' focus on specifying a full economic model, solving for its equilibrium, and using restrictions implied by the equilibrium to estimate the parameters. (FWIW, my impression is that modern practice is somewhat ...


1

A confirmatory factor analysis (CFA) is a structural equation model (SEM); as you said, it's the measurement part of the SEM. If the CFA has poor fit, then a SEM that adds variables but doesn't affect the measurement model will also have poor fit. Indeed, if there is misfit in the measurement model and the other parts of the larger SEM are estimated together ...


1

A latent variable doesn't have values, so it can't appear in a data set. If you estimated values of the latent variables, called factor scores, then you're no longer dealing with latent variables. Factor scores are faulty estimates of the latent variable values and there are several problems with using them in regression and SEMs. The two biggest are that ...


1

Well, the PROCESS macro can't handle latent variables, and you have a latent variable, so I'm not sure why you'd even consider it when there is another clear option available. See Muthen & Asparouhov (2015) for an introduction to (causal) mediation with latent variables along with Mplus code. The sem function in Stata is also good for performing ...


1

You will want to ensure an adequate sample size when variability of a variable is unequal across the range of values of a second variable that predicts it. If a regression model is able to consistently predict across all values a smaller sample size is possible, where the predictions are poor at one end or the other (because it's ordinal) then a larger ...


1

The point of an SEM is to explain the correlations among the observed variables (i.e., indicators in this case) using a latent variable model. A bad fit indicates that your latent variable model doesn't do a good job of explaining why certain indicators are correlated with each other. That is, even after including your model specification, the model could be ...


Only top voted, non community-wiki answers of a minimum length are eligible