EFAshiny
is an user-friendly application for exploratory factor
analysis (EFA; Bartholomew, Knott, & Moustaki, 2011). The graphical user
interface in shiny (Chang, Cheng, Allaire, Xie, & McPherson, 2017) is
designed to free users from scripting in R by wrapping together various
packages for data management, factor analysis, and graphics.
Easy-to-follow analysis flow and reasonable default settings avoiding
common errors (Henson & Roberts, 2006) are provided. Results of analysis
in tables and graphs are presented on-line and can be exported.
Key features include:
The EFAshiny
application is primarily aimed at behavioral researchers
who want to perform EFA on a set of associated variables (e.g.,
item-level scale dataset). Note that it can also be used to explore
FA-based connectivity analyses (McLaughlin et al., 1992) in instrument
data, such as event related potentials (ERPs) and functional
near-infrared spectroscopy (fNIRS) data. Though the major focus of
EFAshiny
is to perform EFA, it is worth noting that confirmatory
factor analysis (CFA) is an useful future direction for shiny
APP.
To run EFAshiny
on your R,
devtools
and shiny
are required.
install.packages("devtools")
install.packages("shiny")
Install and launch EFAshiny
:
devtools::install_github("PsyChiLin/EFAshiny")
EFAshiny::EFAshiny()
If you want to use the standard version of EFAshiny
, installation is
not required. The application is deployed on shinyapps.io server.
This standard version has all the function except for the Editor
tab
(which is only useful for users who want to code online). Users can
easily explore and analyze their data with this online APP without
worrying about installation. Have fun with EFAshiny
:
https://psychilin.shinyapps.io/EFAshiny/
EFAshiny
adopts exploratory factor analysis (EFA, Bartholomew, Knott,
& Moustaki, 2011), a widely used method to investigate the underlying
factor structure that can be used to explain the correlations in a set
of observed indicators, as the major procedure in the application. EFA
can be useful in lots of situations. For example, it can be used to
conceptualize new constructs, to develop instruments, to select items as
a short form scale, or to organize observed variables into meaningful
subgroups. Major procedures of EFA included correlation coefficients
calculation, number of factors determination, factor extraction, and
factor rotation. In addition to the aforementioned steps of EFA, data
explorations should be conducted before using EFA, and interpreting the
results after using EFA is also an important step. Since that EFA is
helpful to account for the relationship between numerous variables, its
use has permeated fields from psychology to business, education and
clinical domain.
When you open EFAshiny
,
the interface will be shown.
In the Introduction
tab, you can see the main features for EFAshiny
,
a demo figure, and some key references.
The data sets that required the implementations of EFA are typically in
a wide format, i.e., one observation per row. They are composed of
a set of responses in one or more psychometric tests in Likert
scale. In the Data Input
tab, users can upload the data.
If no data is uploaded, EFAshiny
will use the Rosenberg Self-Esteem
Scale
dataset to perform the default demostrations.
After uploading the data, the exploratory data analysis should be
conducted. In Data Summary
tab, three types of explorations are
provided.
plotly
package. In other words, they can be played
dynamically. Try it with some clicks !Correlation Matrix
tab using
corrplot
package, we also provide a ggcorrplot
version. Have fun
with those plots and further get some intuitions.Note that the provided correlation matrix is the basis of EFA, which is a procedure that aim to investigate the underlying structure from the correlations between variables, so either calculating or visualizing the correlation matrix will be really important.
One of the central idea of the EFA is to represent a set of observed
variables by a smaller number of factors. Thus, selecting how many
factors to retain is a critical decision. In Factor Retention
tab, a set of indices to determine numbers of factor are provided.
In addition, Sample Size is another option for users to validate the
results for factor retentions by randomly adjusting different Sample
Size. Although users still have to determine the number of factors
upon their own decisions, EFAshiny
provides users several indices
without worrying on methods implementations.
The major step of EFA is to extract and rotate the factors structure,
further estimating the factor loadings. In Extraction and
Rotation
tab, several factor extraction and rotation methods are
available, and the boostrapping for estimating confidence intervals of
factor loadings is also provided to aide in interpretations.
By providing plenty of factor extraction methods, rotation methods, and
useful interval estimations of factor loadings, EFAshiny
is not only
helpful for EFA newbies, but also flexible for EFA users with many
experiences.
For EFA results, the fundamental visualizations is plotting the
relationship between factors and indicators. In Diagram
tab, the
path diagram representation is provided by using psych
R package
(Revelle, 2017). It has the structure that all factors and
indicators are represented as a bigger or smaller node, and all loadings
with absolute values greater than some thresholds (e.g. 0.3) are
represented as a line. Through the graphical representations with
flexible plotting options, users can easily understand the factor
structure.
In Factor Loadings
tab, EFAshiny provides useful visualization of
factor loadings to facilitate proper interpretations of extracted
factors.
In addition to providing a table of loadings for EFA results, users can automatically get the whole picture of the EFA results through these visualizations.
We summarize, in six concrete steps, our provided flow in EFAshiny
for
performing EFA.
To see the tutorial in vignettes:
browseVignettes("EFAshiny")
By following this analysis flow in EFAshiny
, users without any
knowledge of programming are able to perform EFA and obtain great
understandings for their own studies.
In addition to the GUI, we also provide an Editor
tab with several
code demonstrations in the Github version of EFAshiny
. In this
Editor
mode (see figure below), we already present some quick examples
allowing users to perform similar analyses in EFAshiny
GUI. Users can
also write their own R code here. With this feature users might have the
possibility to use EFAshiny
within a script pipeline. In general, this
cool feature allow users to learn R, understand the code underlying
analyses in EFAshiny
or automate the analyses in the future.
Note that this feature can also allow the use of lavaan
R package to
perform confirmatory factor analysis (CFA), which is also a widely used
method but not the main focus of EFAshiny
. Simply input
require(lavaan)
should work (see lavaan
package for details). Another useful tool is
the showcase
version of shiny
when running the APP ( definitely, you
can directly see the code in server.R
and ui.R
).
In summary, Users who want to further understand EFAshiny
or learn R
can (1) see the code in Editor
tab of github version EFAshiny
GUI
(as shown in figure), (2) download the R markdown file similar to the
code in editor mode
here,
(3) see the same R markdown file in this public
link, (4) use showcase
function in shiny
, and (5) directly see the code in server.R
and
ui.R
.
The dataset for demonstration is the 10-items Rosenberg Self-Esteem
Scale (RSE; Rosenberg, 1965) via an online platform for psychological
research. The RSE was
recorded in 1 to 4 Likert scale, where higher scores indicated higher
agreements for the items (1=strongly disagree, 2=disagree, 3=agree, and
4=strongly agree). Previous studies suggested that the RSE could be
treat as a one factor un-dimensional scale, which simply assessed a
positive self-evaluation construct, or a two factor bi-dimensional
scale, where one factor is proposed to assess positive self-esteem
(e.g. I feel that I have a number of good qualities) with another
measuring negative self-esteem (e.g. At times I think I am no good at
all). EFAshiny
already implements a 256 participants RSE data as a
built-in dataset, but
RSE.csv
with
codebook
can also be directly
downloaded.
bootnet
(Epskamp,
2017)corrplot
(Taiyun & Viliam,
2017)EFAutilities
(See Zhang, 2014 for
detail)reshape2
(Wickham, 2014)EGA
(Golino & Epskamp,
2017)ggplot2
(Wickham,
2016)ggcorrplot
(Kassambara,
2016)gridExtra
(Auguie,
2017)igraph
(Csardi & Nepusz,
2006)moments
(Komsta & Novomestky, 2013)plotly
(Sievert, et al., 2017)psych
(Revelle,
2017)psycho
(Makowski,
2018)qgraph
(Epskamp, et al., 2012)shiny
(Chang, Cheng, Allaire, Xie, & McPherson,
2017)shinytheme
(Chang, 2016)Chi-Lin Yu : Department of
Psychology, National Taiwan University, Taiwan Ching-Fan
Sheu : Institute of Education,
National Cheng Kung University, Taiwan If you have a
question, comment, concern or code contribution about EFAshiny
, please
send us an email at psychilinyu@gmail.com.
Please cite as:
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.