Features

lavaan is reliable, open and extensible

  • by default, lavaan implements the textbook/paper formulas, so there are no surprises

  • lavaan can partially mimic results of the commercial package Mplus via the mimic = "Mplus" argument

  • lavaan is not a black box: you can browse the source code on GitHub

  • lavaan can be extended: see the Related Projects page for extensions and add-ons.

lavaan is easy and intuitive to use

  • the ‘lavaan model syntax’ allows users to express their models in a compact, elegant and useR-friendly way; for example, a typical CFA analysis looks as follows:
library(lavaan)
myData <- read.csv("/path/to/mydata/myData.csv")
myModel <- ' 
    f1 =~ item1 + item2 + item3
    f2 =~ item4 + item5 + item6
    f3 =~ item7 + item8 + item9
'
fit <- cfa(model = myModel, data = myData)
summary(fit, fit_measures = TRUE)

  • you can choose between a user-friendly interface in combination with the fitting functions cfa() and sem() or a low-level interface using the fitting function lavaan() where ‘defaults’ do not get in the way

  • convenient arguments (eg. group_equal="loadings") simplify many common tasks (eg. measurement invariance testing)

  • lavaan outputs all the information you need: a large number of fit measures, modification indices, standardized solutions, and technical information that is stored in a fitted lavaan object

lavaan provides many advanced options

  • full support for meanstructures and multiple groups

  • several estimators are available: ML (and robust variants MLM, MLMV, MLR), GLS, WLS (and robust variants DWLS, WLSM, WLSMV), ULS (ULSM, ULSMV), DLS, (MI)IV, and pairwise ML (PML)

  • standard errors: standard, robust/huber-white/sandwich, bootstrap

  • test statistics: standard, Satorra-Bentler, Yuan-Bentler, Satterthwaite, scaled-shifted, Bollen-Stine bootstrap, the FMG family, Hayakawa

  • missing data: FIML estimation or two.stage (for multiple imputation, see the lavaan.mi package)

  • linear and nonlinear equality and inequality constraints

  • full support for analyzing categorical data: lavaan (from version 0.5 onwards) can handle any mixture of binary, ordinal and continuous observed variables (using ULS(MV), (D)WLS(MV) or PML estimators; but not ML)

  • (from version 0.6 onwards): support for multilevel SEM (two levels only); (from version 0.7 onwards): support for random slopes

  • apart from the start ‘nlminb()’ optimizer, lavaan also offers the EM algorithm for multilevel models, and Gauss-Newton (optim_method = “gn”) alternatives

  • the sam() function provides convenient access to several “structural-after-measurement” (SAM) approaches, including local SAM (LSAM)