A First Look at Lavaan
Content
www.structuralequations.org presents!
A First Look at the lavaan R Package for Classical Structural Equation Modeling
last updated: July 31, 2013
1
Objective of this tutorial (Tutorial #1)
The objective in this tutorial is to introduce first-time lavaan users to the basic resources for the package. In addition to providing some links to on-line resources, this tutorial illustrates a few of the most basic lavaan commands and output. This tutorial requires preexisting knowledge of R, but the lavaan syntax is both familiar and largely stand-alone, so one does not have to be an expert to start using lavaan quickly.
2
A first look at lavaan
“lavaan” is an R package for classical structural equation modeling. The basic description of the package at the R website can be found at: http://cran.r-project.org/web/packages/lavaan/lavaan.pdf The developer’s website for lavaan is: http://lavaan.ugent.be/?q=node/2 At the developer’s website you will find a number of useful resources. In addition to those items, there is a Google Group where you can usually get quick answers to your coding issues. https://groups.google.com/forum/#!forum/lavaan
3
A first look at lavaan syntax
In addition to this tutorial, you will want to download R code and data for tutorial #1 at the following url: http://www.structuralequations.com/LavaanTutorials.html Our first model:
age firesev
The lavaan equation for this model is: ‘cover ~ age + firesev’
cover
ec
4
“A_First_Look_at_lavaan.R*”
*note: you can open .R files (or .r files) with any text reader.
### A FIRST LOOK AT LAVAAN # from www.structuralequations.org # This code accompanies tutorial "Lavaan_FirstLook.pdf" # # # # DATA and example used in this demonstration from Grace and Keeley (2006) Ecol. Apps. 16:503-514. download:(http://www.werc.usgs.gov/OLDsitedata/seki/pdfs /k2006_grace_sem_ea.pdf)
# Note: comments are in regular font, commands in bold. # Set your working directory setwd("F:/Documents/LavaanTutorials") # Load data and name file “k.dat” k.dat<-read.csv("./Keeley_rawdata_select4.csv") # Examine contents of keeley data file names(k.dat) head(k.dat)
5
“A_First_Look_at_lavaan.R” (cont.)
R code continued . . .
### Load Libraries library(lavaan) ### Lavaan syntax for a single-equation ################ # Write lavaan code for this single equation model mod1 <- 'cover ~ age + firesev' # Fit the model (i.e. est. parameters) mod1.fit <- sem(mod1, data=k.dat)
# Output a summary of the computed results summary(mod1.fit, rsq=T) # rsq=T means output the r-sqr
########################################################
Note: exogenous variables are automatically allowed to freely intercorrelate in lavaan.
6
Output from single-equation model
R output
Warning message: In getDataFull(data = data, group = group, group.label = group.label, : lavaan WARNING: some observed variances are (at least) a factor 100 times larger than others; please rescale
# Fix the data summary(k.dat$cover); summary(k.dat$age); summary(k.dat$firesev) k.dat$age <- k.dat$age/100 # scale age variable k.dat$firesev <- k.dat$firesev/10 # scale firesev # then rerun the above model
7
Output from single-equation model
lavaan (0.5-12) converged normally after Number of observations Estimator Minimum Function Test Statistic Degrees of freedom P-value (Chi-square) Estimate Regressions: cover ~ age firesev Variances: cover R-Square: cover Std.err Z-value 1 iterations 90 ML 0.000 0 1.000 P(>|z|)
-0.483 -0.672
0.264 0.201
-1.833 -3.353
0.067 0.001
0.078
0.012
0.220
8
Output from single-equation model (annotated in red)
lavaan (0.5-12) converged normally after Number of observations 1 iterations 90
Estimator ML Minimum Function Test Statistic (i.e. chi-sqr) 0.000 Degrees of freedom (model is saturated) 0 P-value (Chi-square) 1.000 Estimate Std.err Z-value P(>|z|) (raw est.) (ML t-value) -0.483 -0.672 0.264 0.201 -1.833 -3.353 0.067 0.001
Regressions: cover ~ age firesev
Variances:(for endogenous vars, these are error variances) cover 0.078 0.012 R-Square: cover
0.220
9
A second look at lavaan syntax
Here we look at a model that requires two equations, one for each endogenous variable. Our second model:
age
The lavaan equation for this model is: 'cover ~ firesev firesev ~ age'
firesev
ec
cover
ec 10
A second look at lavaan syntax
### Lavaan multi-equation model ######################### # Lavaan model: separate line for each equation; # whole model is enclosed in quotes. # Model code mod2 <- 'cover ~ firesev firesev ~ age' # Fit model mod2.fit <- sem(mod2, data=k.dat)
# Output results summary(mod2.fit, rsq=T)
11
Output from two-equation model (fit measures)
lavaan (0.5-12) converged normally after Number of observations Estimator Minimum Function Test Statistic Degrees of freedom P-value (Chi-square) 19 iterations 90 ML 3.297 1 0.069
12
Output from two-equation model (parameter estimates)
Parameter estimates: Information Standard Errors Estimate Regressions: cover ~ firesev firesev ~ age Variances: cover firesev Std.err Z-value Expected Standard P(>|z|)
-0.839 0.597
0.182 0.124
-4.611 4.832
0.000 0.000
0.081 0.021
0.012 0.003
R-Square:
cover firesev 0.191 0.206
13
End of First Look.
14
A First Look at the lavaan R Package for Classical Structural Equation Modeling
last updated: July 31, 2013
1
Objective of this tutorial (Tutorial #1)
The objective in this tutorial is to introduce first-time lavaan users to the basic resources for the package. In addition to providing some links to on-line resources, this tutorial illustrates a few of the most basic lavaan commands and output. This tutorial requires preexisting knowledge of R, but the lavaan syntax is both familiar and largely stand-alone, so one does not have to be an expert to start using lavaan quickly.
2
A first look at lavaan
“lavaan” is an R package for classical structural equation modeling. The basic description of the package at the R website can be found at: http://cran.r-project.org/web/packages/lavaan/lavaan.pdf The developer’s website for lavaan is: http://lavaan.ugent.be/?q=node/2 At the developer’s website you will find a number of useful resources. In addition to those items, there is a Google Group where you can usually get quick answers to your coding issues. https://groups.google.com/forum/#!forum/lavaan
3
A first look at lavaan syntax
In addition to this tutorial, you will want to download R code and data for tutorial #1 at the following url: http://www.structuralequations.com/LavaanTutorials.html Our first model:
age firesev
The lavaan equation for this model is: ‘cover ~ age + firesev’
cover
ec
4
“A_First_Look_at_lavaan.R*”
*note: you can open .R files (or .r files) with any text reader.
### A FIRST LOOK AT LAVAAN # from www.structuralequations.org # This code accompanies tutorial "Lavaan_FirstLook.pdf" # # # # DATA and example used in this demonstration from Grace and Keeley (2006) Ecol. Apps. 16:503-514. download:(http://www.werc.usgs.gov/OLDsitedata/seki/pdfs /k2006_grace_sem_ea.pdf)
# Note: comments are in regular font, commands in bold. # Set your working directory setwd("F:/Documents/LavaanTutorials") # Load data and name file “k.dat” k.dat<-read.csv("./Keeley_rawdata_select4.csv") # Examine contents of keeley data file names(k.dat) head(k.dat)
5
“A_First_Look_at_lavaan.R” (cont.)
R code continued . . .
### Load Libraries library(lavaan) ### Lavaan syntax for a single-equation ################ # Write lavaan code for this single equation model mod1 <- 'cover ~ age + firesev' # Fit the model (i.e. est. parameters) mod1.fit <- sem(mod1, data=k.dat)
# Output a summary of the computed results summary(mod1.fit, rsq=T) # rsq=T means output the r-sqr
########################################################
Note: exogenous variables are automatically allowed to freely intercorrelate in lavaan.
6
Output from single-equation model
R output
Warning message: In getDataFull(data = data, group = group, group.label = group.label, : lavaan WARNING: some observed variances are (at least) a factor 100 times larger than others; please rescale
# Fix the data summary(k.dat$cover); summary(k.dat$age); summary(k.dat$firesev) k.dat$age <- k.dat$age/100 # scale age variable k.dat$firesev <- k.dat$firesev/10 # scale firesev # then rerun the above model
7
Output from single-equation model
lavaan (0.5-12) converged normally after Number of observations Estimator Minimum Function Test Statistic Degrees of freedom P-value (Chi-square) Estimate Regressions: cover ~ age firesev Variances: cover R-Square: cover Std.err Z-value 1 iterations 90 ML 0.000 0 1.000 P(>|z|)
-0.483 -0.672
0.264 0.201
-1.833 -3.353
0.067 0.001
0.078
0.012
0.220
8
Output from single-equation model (annotated in red)
lavaan (0.5-12) converged normally after Number of observations 1 iterations 90
Estimator ML Minimum Function Test Statistic (i.e. chi-sqr) 0.000 Degrees of freedom (model is saturated) 0 P-value (Chi-square) 1.000 Estimate Std.err Z-value P(>|z|) (raw est.) (ML t-value) -0.483 -0.672 0.264 0.201 -1.833 -3.353 0.067 0.001
Regressions: cover ~ age firesev
Variances:(for endogenous vars, these are error variances) cover 0.078 0.012 R-Square: cover
0.220
9
A second look at lavaan syntax
Here we look at a model that requires two equations, one for each endogenous variable. Our second model:
age
The lavaan equation for this model is: 'cover ~ firesev firesev ~ age'
firesev
ec
cover
ec 10
A second look at lavaan syntax
### Lavaan multi-equation model ######################### # Lavaan model: separate line for each equation; # whole model is enclosed in quotes. # Model code mod2 <- 'cover ~ firesev firesev ~ age' # Fit model mod2.fit <- sem(mod2, data=k.dat)
# Output results summary(mod2.fit, rsq=T)
11
Output from two-equation model (fit measures)
lavaan (0.5-12) converged normally after Number of observations Estimator Minimum Function Test Statistic Degrees of freedom P-value (Chi-square) 19 iterations 90 ML 3.297 1 0.069
12
Output from two-equation model (parameter estimates)
Parameter estimates: Information Standard Errors Estimate Regressions: cover ~ firesev firesev ~ age Variances: cover firesev Std.err Z-value Expected Standard P(>|z|)
-0.839 0.597
0.182 0.124
-4.611 4.832
0.000 0.000
0.081 0.021
0.012 0.003
R-Square:
cover firesev 0.191 0.206
13
End of First Look.
14
