In dr-JT/semoutput: SEM Output

knitr::opts_chunk$set(out.width = "200%", error = TRUE, message = FALSE, warning = TRUE)

Example Output {.tabset .tabset-pills}

Last updated: r format(Sys.Date(), "%B %d %Y")

Results

Setup

Required Packages

library(readr)
library(lavaan)
library(semPlot)
library(sjPlot)

# Packages developed by Jason Tsukahara. To install, devtools::install_github("dr-JT/packagename")
library(semoutput) # For displaying lavaan output
library(resultsoutput) # Optional: For displaying descriptives (you can also just use whatever package/functions you prefer to display descriptives)

Import Data

Data file: r params$data

## Import Data
data <- PoliticalDemocracy

For this example I am using the PoliticalDemocracy data set from the lavaan package.

## How to handle missing data. "ML" is a full-information maximum likelihood. Alternatively, "listwise" for listwise deletion
missing <- "ML"
## Do you want to mimic the results of another statistical software? If not, then set to "lavaan"
mimic <- "lavaan"

---

---

Descriptives

# Prints basic descriptive statistics
sem.descriptives(data)

---

Correlation Matrix

# Uses sjPlot to print a nice looking correlation table
sjt.corr(data, na.deletion = "pairwise", digits = 2)

---

CFA {.tabset .tabset-pills}

# factors is a list of the latent factor names. To be used for printing lavaan output to nice looking tables. Not needed to run actual lavaan model
factors <- c("ind60", "dem60", "dem65")
# Specify the model parameters. Must be incased within single or double quotes. Can add labels to organize parameter specifications. I provided just some example labels, they are not required though and you can use whatever labels you want. You need to know how lavaan works! See http://lavaan.ugent.be/tutorial/index.html for more information on how to specify the model parameters
model <- '
# latent factors
ind60 =~ x1 + x2 + x3
dem60 =~ y1 + y2 + y3 + y4
dem65 =~ y5 + y6 + y7 + y8

# correlated errors
y1 ~~ y5
y2 ~~ y4 + y6
y3 ~~ y7
y4 ~~ y8
y6 ~~ y8

# constraints

'
# You might want to specify other arguments. But this is pretty standard. 
#std.lv = TRUE means that the latent factor variances are set to 1. If False, then the path for the first variable on the factor specificed in the model will be set to 1.
fit <- cfa(model=model, data=data, mimic = mimic, missing=missing, std.lv = TRUE)

Missing Procedure: r missing

Mimic: r mimic

Summary Output

sem.sig(fit)
sem.fitmeasures(fit)
sem.factorloadings(fit)
sem.factorcor(fit, factors = factors)
sem.factorvar(fit, factors = factors)
sem.rsquared(fit)

Diagram Output

semPaths(fit, whatLabels = "std", edge.label.cex = .5, layout = "tree2", rotation = 2, style = "lisrel", intercepts = FALSE, residuals = TRUE, curve = 1, curvature = 3, nCharNodes = 8, sizeMan = 6, sizeMan2 = 3, optimizeLatRes = TRUE, edge.color = "#000000", latents = factors)

Residual Correlation Matrix

sem.residuals(fit)

Full Output

summary(fit, fit.measures=TRUE, standardized=TRUE)
varTable(fit)
modificationIndices(fit, sort.=TRUE, minimum.value=3)

---

SEM {.tabset .tabset-pills}

# factors is a list of the latent factor names. To be used for printing lavaan output to nice looking tables. Not needed to run actual lavaan model
factors <- c("ind60", "dem60", "dem65")
# Specify the model parameters. Must be incased within single or double quotes. Can add labels to organize parameter specifications. I provided just some example labels, they are not required though and you can use whatever labels you want. You need to know how lavaan works! See http://lavaan.ugent.be/tutorial/index.html for more information on how to specify the model parameters
model <- '
# latent factors
ind60 =~ x1 + x2 + x3
dem60 =~ y1 + y2 + y3 + y4
dem65 =~ y5 + y6 + y7 + y8

# correlated errors
y1 ~~ y5
y2 ~~ y4 + y6
y3 ~~ y7
y4 ~~ y8
y6 ~~ y8

# regressions
dem60 ~ ind60
dem65 ~ ind60 + dem60

# latent covariances

# latent variances

'
# You might want to specify other arguments. But this is pretty standard. 
#std.lv = TRUE means that the latent factor variances are set to 1. If False, then the path for the first variable on the factor specificed in the model will be set to 1. For an SEM, it is better to set as FALSE. And if you want a latent factor variance set to 1 then specify it in the model and include NA* before the first variable on the factor. See http://lavaan.ugent.be/tutorial/index.html for more details
fit <- sem(model=model, data=data, mimic = "EQS", missing=missing, std.lv = FALSE)

Missing Procedure: r missing

Mimic: r mimic

Summary Output

sem.sig(fit)
sem.fitmeasures(fit)
sem.factorloadings(fit)
sem.paths(fit)
sem.factorcor(fit, factors = factors)
sem.rsquared(fit)

Diagram Output

semPaths(fit, whatLabels = "std", edge.label.cex = .5, layout = "tree2", rotation = 2, style = "lisrel", intercepts = FALSE, residuals = TRUE, curve = 1, curvature = 3, nCharNodes = 8, sizeMan = 6, sizeMan2 = 3, optimizeLatRes = TRUE, edge.color = "#000000", latents = factors)

Residual Correlation Matrix

sem.residuals(fit)

Full Output

summary(fit, fit.measures=TRUE, standardized=TRUE)
modificationIndices(fit, sort.=TRUE, minimum.value=3)

---

---

Session Info

citation("lavaan")
citation()
sessionInfo()

dr-JT/semoutput documentation built on March 25, 2024, 9:18 p.m.