Show Options Set by lavaan

R
lavaan
semhelpinghands
Author

Shu Fai Cheung

Published

September 26, 2022

lavaan is a convenient tool for doing structural equation modelling in R (Rosseel, 2012). One of its strength is having “prepackaged” estimators, which are shortcuts to a set of options, such as “ML”, “MLR”, “MLMVS”, and others (Savalei & Rosseel, 2022). It also tries to set default values for options based on the model and data.

However, probably due to my not-so-good memory, I sometimes forgot what the settings are for a model. Therefore, in the package semhelpinghands, I wrote the function show_more_options() to show some of the settings of the output of lavaan() and its wrappers, such as sem() and cfa().1

The function show_more_options() is very easy to use … because it accepts only one argument, the output of lavaan().

This is an example based on the example of lavaan::cfa():

library(lavaan)
This is lavaan 0.6-13
lavaan is FREE software! Please report any bugs.
HSmodel <-
"
visual  =~ x1 + x2 + x3
textual =~ x4 + x5 + x6
speed   =~ x7 + x8 + x9
"
fit <- cfa(HSmodel,
           data = HolzingerSwineford1939)

To show the major options, just pass the output to show_more_options():

library(semhelpinghands)
show_more_options(fit)
 Options                             Call    Actual  
 Estimator(s)                        default ML      
 Standard Error (SE)                 default standard
 Model Test Statistic(s)             default standard
 How Missing Data is Handled         default listwise
 Information Matrix (for SE)         default expected
 Information Matrix (for Model Test) default expected
 Mean Structure                      default No      
 'x' Fixed                           default FALSE

The column Call shows whether the default setting is used for each row, based of the call used when fitting the model. However, it is not always clear to me what the default values are.

The column Actual shows the values extracted by lavaan::lavInspect() or from the Options slot. These are what the default values stand for in the fitted model.

Many of the entries are either (a) already available in the output of summary(), or (b) can be deduced from the output. However, I would like to have a table for quick reference, hence I wrote this function.

Suppose "MLR" is used as the estimator:

fit_MLR <- cfa(HSmodel,
               data = HolzingerSwineford1939,
               estimator = "MLR")
show_more_options(fit_MLR)
 Options                             Call    Actual            
 Estimator(s)                        MLR     ML                
 Standard Error (SE)                 default robust.huber.white
 Model Test Statistic(s)             default yuan.bentler.mplus
 How Missing Data is Handled         default listwise          
 Information Matrix (for SE)         default observed          
 Information Matrix (for Model Test) default observed          
 Mean Structure                      default No                
 'x' Fixed                           default FALSE

The output shows the exact names of the options (e.g., "robust.huber.white" and "yuan.bentler.mplus"). They can complement the more readable output of summary() if we need to manually set these options, or want to know which values these options refer to when consulting the help page.

For example, summary() reports that "Sandwich" is the method used for standard errors, and show_more_options() shows that the exact name in the option is "robust.huber.white". This is useful because the word "sandwich" does not appear in the help page of lavOptions(), while the word "robust.huber.white" does. Some users may not know what "Sandwich" stands for.

This is a dataset for a path model, with missing data:

set.seed(8745315)
n <- 100
x <- rnorm(n)
m <- 5 + .2 * x + rnorm(n, 0, .8)
y <- 10 + .3 * m + rnorm(n, 0, .8)
path_data <- data.frame(x, m, y)
path_data[c(1, 3, 5, 7), "m"] <- NA
path_data[c(1, 3, 6, 8), "y"] <- NA

Suppose we use only the default options to fit a path model:

path_model <-
"
m ~ a * x
y ~ b * m
ab := a * b
"
fit_path <- sem(path_model,
                data = path_data)
show_more_options(fit_path)
 Options                             Call    Actual  
 Estimator(s)                        default ML      
 Standard Error (SE)                 default standard
 Model Test Statistic(s)             default standard
 How Missing Data is Handled         default listwise
 Information Matrix (for SE)         default expected
 Information Matrix (for Model Test) default expected
 Mean Structure                      default No      
 'x' Fixed                           default TRUE

The output shows that, by default, the mean structure is not modelled, listwise selection is used to handle missing data, and x variables (exogenous covariates, x in this example) are treated as fixed. This can be verified by the parameter estimates, in which the variance of x is a fixed parameter and hence has no standard error and no p-value:

parameterEstimates(fit_path)
  lhs op rhs label   est    se     z pvalue ci.lower ci.upper
1   m  ~   x     a 0.075 0.078 0.966  0.334   -0.077    0.228
2   y  ~   m     b 0.248 0.119 2.081  0.037    0.014    0.482
3   m ~~   m       0.550 0.080 6.856  0.000    0.393    0.708
4   y ~~   y       0.745 0.109 6.856  0.000    0.532    0.958
5   x ~~   x       0.967 0.000    NA     NA    0.967    0.967
6  ab := a*b    ab 0.019 0.021 0.876  0.381   -0.023    0.060

Suppose we set missing to "FIML":

fit_path_fiml <- sem(path_model,
                     data = path_data,
                     missing = "FIML")
show_more_options(fit_path_fiml)
 Options                             Call    Actual  
 Estimator(s)                        default ML      
 Standard Error (SE)                 default standard
 Model Test Statistic(s)             default standard
 How Missing Data is Handled         FIML    ml      
 Information Matrix (for SE)         default observed
 Information Matrix (for Model Test) default observed
 Mean Structure                      default Yes     
 'x' Fixed                           default TRUE

x variables are still treated as fixed, but now mean structure is modelled (required for FIML, full information maximum likelihood), even though I did not explicitly ask for it.

Only options I think are likely needed (by me) are included in the output.2 More may be added in the future. In any case, if other options are needed, they can be retrieved by lavaan::lavInspect() or from the Options slot of the output. In most cases I myself encountered, all I want is a simple function that is easy to remember and no need to set any arguments other than the lavaan output. If I need something else, I will just extract the information myself.

This function was inspired by a script I wrote to enumerate the options set by the prepackaged shortcuts. Interested readers can read this thread at the Google Group for lavaan and this gist, to check how options will be set for different combinations of estimator, data, and some other options.

References

Rosseel, Y. (2012). lavaan: An R package for structural equation modeling. Journal of Statistical Software, 48(2), 1–36. https://doi.org/10.18637/jss.v048.i02
Savalei, V., & Rosseel, Y. (2022). Computational options for standard errors and test statistics with incomplete normal and nonnormal data in SEM. Structural Equation Modeling: A Multidisciplinary Journal, 29(2), 163–181. https://doi.org/10.1080/10705511.2021.1877548

Footnotes

  1. To be precise, any object of the class lavaan.↩︎

  2. The information matrices are technical but I occasionally need them.↩︎