Estimation may take a few seconds to minutes depending on the dataset.
Data
Input values must be separated by tabs. Copy and paste from Excel/Numbers.
Specifying the model
See lavaan for the syntax.
Estimator
ML: Use when observed variables are continuous and approximately normally distributed; results are comparable to conventional ML SEM software such as Amos.
MLR: Use when continuous variables may deviate from multivariate normality; it provides robust standard errors and a scaled test statistic.
MLM: Use for complete continuous data when multivariate normality may be violated; it provides robust standard errors and a Satorra-Bentler scaled test statistic.
WLSMV: Use for ordered categorical or ordinal questionnaire items such as Likert-type responses; this app treats all pasted variables as ordered categorical variables when WLSMV is selected.
Basic statistics
Correlation
Scatter plot matrices
Normality tests
Univariate normality is tested with the Shapiro-Wilk test for each numeric variable. Multivariate normality is checked with Mardia's skewness and kurtosis tests using complete cases. A small p-value suggests departure from normality.
Estimator suggestion
SEM result
Modification indices
Modification indices are sorted from largest to smallest. A modification index estimates how much the model chi-square would decrease if a currently fixed parameter were freely estimated.
Use modification indices as diagnostic suggestions, not as automatic instructions for changing the model.
- As a practical screening rule, MI >= 10 is often treated as a noteworthy value, but a change should be considered only when it is theoretically defensible.
- For example, adding a residual covariance should usually reflect a clear reason such as similar item wording, shared method effects, or overlapping content.
- If model fit is poor, also inspect factor loadings and consider whether weak indicators should be removed or revised before adding new paths.
- Make changes one at a time, rerun the model, and report any post hoc modifications transparently. Avoid overfitting the sample.
| MI value | Approx. p-value | Interpretation |
|---|---|---|
| > 10.83 | p < .001 | Very strong statistical signal |
| > 6.63 | p < .01 | Strong statistical signal |
| > 3.84 | p < .05 | Statistically significant signal |
The p_approx column below is not a default lavaan column; it is calculated from the MI value using a chi-square distribution with 1 degree of freedom. In lavaan output, epc is the expected parameter change. The standardized EPC columns help judge the likely size of the change on a standardized scale.
Path diagram (Unstandardized estimates)
Path diagram (Standardized estimates)
R session info
This web application is developed with Shiny.
List of Packages Used
library(shiny)
library(shinyAce)
library(psych)
library(lavaan)
library(semPlot)
Code
Source code for this application is based on "The handbook of Research in Foreign Language Learning and Teaching" (Takeuchi & Mizumoto, 2012).
The code for this web application is available at GitHub.
If you want to run this code on your computer (in a local R session), run the code below:
library(shiny)
runGitHub("sem","mizumot")
Citation in Publications
Mizumoto, A. (2015). Langtest (Version 1.0) [Web application]. Retrieved from http://langtest.jp
Article
Mizumoto, A., & Plonsky, L. (2015). R as a lingua franca: Advantages of using R for quantitative research in applied linguistics. Applied Linguistics, Advance online publication. doi:10.1093/applin/amv025
Recommended
To learn more about R, I suggest this excellent and free e-book (pdf), A Guide to Doing Statistics in Second Language Research Using R, written by Dr. Jenifer Larson-Hall.
Also, if you are a cool Mac user and want to use R with GUI, MacR is defenitely the way to go!
Author
Atsushi MIZUMOTO,
Ph.D.
Professor of Applied Linguistics
Faculty of Foreign Language Studies /
Graduate School of Foreign Language Education and Research,
Kansai University, Osaka, Japan
