psych (version 2.3.6)

fa.diagram: Graph factor loading matrices


Factor analysis or principal components analysis results are typically interpreted in terms of the major loadings on each factor. These structures may be represented as a table of loadings or graphically, where all loadings with an absolute value > some cut point are represented as an edge (path). fa.diagram uses the various diagram functions to draw the diagram. fa.graph generates dot code for external plotting. fa.rgraph uses the Rgraphviz package (if available) to draw the graph. het.diagram will draw "heterarchy" diagrams of factor/scale solutions at different levels. All of these as well as some additional functions may be called by diagram which is a wrapper to the specific functions.


     simple=TRUE, errors=FALSE,g=FALSE,digits=1,e.size=.05,rsize=.15,side=2,
    main,cex=NULL,l.cex=NULL, marg=c(.5,.5,1,.5),adj=1,ic=FALSE, ...) 
        main="Heterarchy diagram",l.cex,gap.size,...) 
    f.cut, e.cut=.1,simple=TRUE,e.simple=FALSE,errors=FALSE,g=FALSE,
    digits=1,e.size=.05,rsize=.15,side=2,main,cex=NULL, e.cex=NULL,
    marg=c(.5,.5,1,.5),adj=1,ic=FALSE, ...)

fa.graph(fa.results,out.file=NULL,labels=NULL,cut=.3,simple=TRUE, size=c(8,6), node.font=c("Helvetica", 14), edge.font=c("Helvetica", 10), rank.direction=c("RL","TB","LR","BT"), digits=1,main="Factor Analysis", ...) fa.rgraph(fa.results,out.file=NULL,labels=NULL,cut=.3,simple=TRUE, size=c(8,6), node.font=c("Helvetica", 14), edge.font=c("Helvetica", 10), rank.direction=c("RL","TB","LR","BT"), digits=1,main="Factor Analysis",graphviz=TRUE, ...)


fa.diagram: A path diagram is drawn without using Rgraphviz. This is probably the more useful function.

fa.rgraph: A graph is drawn using rgraphviz. If an output file is specified, the graph instructions are also saved in the dot language.

fa.graph: the graph instructions are saved in the dot language.



The output of factor analysis, principal components analysis, or ICLUST analysis. May also be a factor loading matrix from anywhere.


Normally not specified (it is is found in the FA, pc, or ICLUST, solution), this may be given if the input is a loadings matrix.


the results of a factor extension analysis (if any)


If it exists, a dot representation of the graph will be stored here (fa.graph)


Variable labels


Loadings with abs(loading) > cut will be shown


factor correlations > f.cut will be shown


extension loadings with abs(loading) > e.cut will be shown


Only the biggest loading per item is shown


Only the biggest loading per extension item is shown


Does the factor matrix reflect a g (first) factor. If so, then draw this to the left of the variables, with the remaining factors to the right of the variables. It is useful to turn off the simple parameter in this case.


If drawing a cluster analysis result, should we treat it as latent variable model (ic=FALSE) or as an observed variable model (ic=TRUE)


A correlation matrix for the het.diagram function


A list of the elements in each level


Should arrows have double heads (in het.diagram)


graph size


sort the factor loadings before showing the diagram


include error estimates (as arrows)


size of ellipses


size of rectangles


on which side should error arrows go?


modify font size


Modify the font size for the Dependent variables, defaults to cex


modify the font size in arrows, defaults to cex


The gap in the arrow for the label. Can be adjusted to compensate for variations in cex or l.cex


sets the margins to be wider than normal, returns them to the normal size upon exit


how many different positions (1-3) should be used for the numeric labels. Useful if they overlap each other.


what font should be used for nodes in fa.graph


what font should be used for edges in fa.graph


parameter passed to Rgraphviz-- which way to draw the graph


Number of digits to show as an edgelable


Graphic title, defaults to "factor analyis" or "factor analysis and extension"


Should we try to use Rgraphviz for output?


other parameters


William Revelle


Path diagram representations have become standard in confirmatory factor analysis, but are not yet common in exploratory factor analysis. Representing factor structures graphically helps some people understand the structure.

By default the arrows come from the latent variables to the observed variables. This is, of course, the factor model. However, if the class of the object to be drawn is 'principal', then reverse the direction of the arrows, and the 'latent' variables are no longer latent, and are shown as boxes. For cluster models, the default is to treat them as factors, but if ic =TRUE, then we treat it as a components model.

fa.diagram does not use Rgraphviz and is the preferred function. fa.graph generates dot code to be used by an external graphics program. It does not have all the bells and whistles of fa.diagram, but these may be done in the external editor.

Hierarchical (bifactor) models may be drawn by specifying the g parameter as TRUE. This allows for an graphical displays of various factor transformations with a bifactor structure (e.g., bifactor and biquartimin. See omega for an alternative way to find these structures.

The het.diagram function will show the case of a hetarchical structure at multiple levels. It can also be used to show the patterns of correlations between sets of scales (e.g., EPI, NEO, BFI). The example is for showing the relationship between 3 sets of 4 variables from the Thurstone data set. The parameters l.cex and gap.size are used to adjust the font size of the labels and the gap in the lines.

extension.diagram will draw a fa.extend result with slightly more control than using fa.diagram or the more generic diagram function.

In fa.rgraph although a nice graph is drawn for the orthogonal factor case, the oblique factor drawing is acceptable, but is better if cleaned up outside of R or done using fa.diagram.

The normal input is taken from the output of either fa or ICLUST. This latter case displays the ICLUST results in terms of the cluster loadings, not in terms of the cluster structure. Actually an interesting option.

It is also possible to just give a factor loading matrix as input. In this case, supplying a Phi matrix of factor correlations is also possible.

It is possible, using fa.graph, to export dot code for an omega solution. fa.graph should be applied to the schmid$sl object with labels specified as the rownames of schmid$sl. The results will need editing to make fully compatible with dot language plotting.

To specify the model for a structural equation confirmatory analysis of the results, use structure.diagram instead.

See Also

diagram, omega.graph, ICLUST.graph, bassAckward.diagram. structure.diagram will convert the factor diagram to sem modeling code.


Run this code

test.simple <- fa(item.sim(16),2,rotate="oblimin")
#if(require(Rgraphviz)) {fa.graph(test.simple) } 
f3 <- fa(Thurstone,3,rotate="cluster")
f3l <- f3$loadings
fa.diagram(f3l,main="input from a matrix")
Phi <- f3$Phi
fa.diagram(f3l,Phi=Phi,main="Input from a matrix")
fa.diagram(ICLUST(Thurstone,2,title="Two cluster solution of Thurstone"),main="Input from ICLUST")

#show how caffiene increases arousal and tension
select <- c("active" , "alert" ,  "aroused", "sleepy",  "tired" , 
          "drowsy" , "anxious", "jittery" ,"nervous",
 "calm" ,   "relaxed" ,"at.ease" ,"gender",  "drug")
 msq.small <- psychTools::msqR[select]
fe <- fa.extend(msq.small, 2,1:12,14)
extension.diagram(fe, ic=TRUE) #drug causes arousal

Run the code above in your browser using DataLab