R language Access Menu

Item Cluster Analysis

iclust() of psych package is used to show clustering of variables (items), rather than cases (individuals or rows). This is similar to factor analysis where grouping and relationships between different variables is produced. Moreover, the cluster produced is hierarchical.

code:

> library(psych)
> iclust(birthwt)

ICLUST (Item Cluster Analysis)
Call: iclust(r.mat = birthwt)

Purified Alpha:
  C6   C7   C1 
0.63 0.34 0.51 

G6* reliability:
  C6   C7   C1 
0.21 1.00 0.25 

Original Beta:
  C6   C7   C1 
0.42 0.17 0.51 

Cluster size:
C6 C7 C1 
 4  4  2 

Item by Cluster Structure matrix:
       O  P    C6    C7    C1
low   C6 C6 -0.78 -0.15  0.02
age   C7 C7  0.09  0.37  0.12
lwt   C7 C7  0.28  0.53  0.11
race  C1 C1 -0.17 -0.31 -0.54
smoke C1 C1 -0.26 -0.08  0.53
ptl   C6 C6 -0.31 -0.09  0.17
ht    C7 C7 -0.08  0.20 -0.01
ui    C6 C6 -0.37 -0.29  0.01
ftv   C7 C7  0.10  0.26  0.07
bwt   C6 C6  0.83  0.14  0.00

With eigenvalues of:
  C6   C7   C1 
1.67 0.67 0.61 

Purified scale intercorrelations
 reliabilities on diagonal
 correlations corrected for attenuation above diagonal: 
      C6   C7    C1
C6  0.63 0.31 -0.08
C7  0.14 0.34  0.20
C1 -0.05 0.08  0.51

Cluster fit =  0.48   Pattern fit =  0.97  RMSR =  0.07

The graph clearly shows hierarchical relationship between different variables: 

                           

These relations can be subjected to confirmatory factor analysis using structural equation modeling as discussed in a later section. 


    Comments & Feedback