Exploratory factor analysis and Cronbach’s alpha

Questionnaire Validation Workshop, 10/10/2017, USM Health Campus

Wan Nor Arifin ([email protected]), Universiti Sains Malaysia
Website: wnarifin.github.io

©Wan Nor Arifin under the Creative Commons Attribution-ShareAlike 4.0 International License.

Contents
1 Introduction 1

2 Preliminaries 2
2.1 Load libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Load data set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

3 Exploratory factor analysis 3

3.1 Preliminary steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
3.2 Step 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.3 Step 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.4 Step 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

4 Internal consistency reliability 19

5 Results presentation 21

References 22

1 Introduction
In this hands-on session, we are going to explore the validity of a new questionnaire of attitude towards
statistics.
The evidence of internal structure will be provided by
1. Exploratory factor analysis
• Number of extracted factors
• Factor loadings
• Factor correlations (no multicollinearity)
2. Internal consistency reliability
• Cronbach’s alpha

1
2 Preliminaries

2.1 Load libraries

Our analysis will involve psych (Revelle, 2017) and MVN (Korkmaz, Goksuluk, & Zararsiz, 2016) packages.
Make sure you already installed all of them. Load the libraries,
library(foreign)
library(psych) # for psychometrics
library(MVN) # for multivariate normality

2.2 Load data set

Download data set “Attitude_Statistics v3.sav”.

Read the data set as data and import it into a new data frame data1 after removing ID variable. This will
make our analysis easier because the column number = question number.
data = read.spss("Attitude_Statistics v3.sav", use.value.labels = F, to.data.frame = T)
dim(data) # 13 variables

## [1] 150 13
names(data) # list variable names

## [1] "ID" "Q1" "Q2" "Q3" "Q4" "Q5" "Q6" "Q7" "Q8" "Q9" "Q10" "Q11" "Q12"
head(data) # the first 6 observations

## ID Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8 Q9 Q10 Q11 Q12

## 1 1 2 3 3 3 4 4 3 3 3 3 4 2
## 2 2 3 2 3 3 4 4 4 3 3 3 4 2
## 3 3 5 4 5 1 1 1 1 4 4 5 1 4
## 4 4 2 2 2 4 3 2 2 2 2 2 3 3
## 5 5 4 1 4 2 5 1 4 5 5 3 4 4
## 6 6 4 4 4 3 4 4 4 3 4 4 4 4
data1 = data[-1] # remove ID
dim(data1)

## [1] 150 12
names(data1)

## [1] "Q1" "Q2" "Q3" "Q4" "Q5" "Q6" "Q7" "Q8" "Q9" "Q10" "Q11" "Q12"
head(data1)

## Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8 Q9 Q10 Q11 Q12

## 1 2 3 3 3 4 4 3 3 3 3 4 2
## 2 3 2 3 3 4 4 4 3 3 3 4 2
## 3 5 4 5 1 1 1 1 4 4 5 1 4
## 4 2 2 2 4 3 2 2 2 2 2 3 3
## 5 4 1 4 2 5 1 4 5 5 3 4 4
## 6 4 4 4 3 4 4 4 3 4 4 4 4

2
3 Exploratory factor analysis

3.1 Preliminary steps

Descriptive statistics
Check minimum/maximum values per item, and screen for missing values,
describe(data1)

## vars n mean sd median trimmed mad min max range skew kurtosis se
## Q1 1 150 3.13 1.10 3 3.12 1.48 1 5 4 -0.10 -0.73 0.09
## Q2 2 150 3.51 1.03 3 3.55 1.48 1 5 4 -0.14 -0.47 0.08
## Q3 3 150 3.18 1.03 3 3.17 1.48 1 5 4 -0.03 -0.42 0.08
## Q4 4 150 2.81 1.17 3 2.77 1.48 1 5 4 0.19 -0.81 0.10
## Q5 5 150 3.31 1.01 3 3.32 1.48 1 5 4 -0.22 -0.48 0.08
## Q6 6 150 3.05 1.09 3 3.05 1.48 1 5 4 -0.04 -0.71 0.09
## Q7 7 150 2.92 1.19 3 2.92 1.48 1 5 4 -0.04 -1.06 0.10
## Q8 8 150 3.33 1.00 3 3.34 1.48 1 5 4 -0.08 -0.12 0.08
## Q9 9 150 3.44 1.05 3 3.48 1.48 1 5 4 -0.21 -0.32 0.09
## Q10 10 150 3.31 1.10 3 3.36 1.48 1 5 4 -0.22 -0.39 0.09
## Q11 11 150 3.35 0.94 3 3.37 1.48 1 5 4 -0.31 -0.33 0.08
## Q12 12 150 2.83 0.98 3 2.83 1.48 1 5 4 0.09 -0.68 0.08
Note that all n = 150, no missing values. min–max cover the whole range of response options.
% of response to options per item,
response.frequencies(data1)

## 1 2 3 4 5 miss
## Q1 0.073 0.220 0.32 0.28 0.107 0
## Q2 0.033 0.093 0.42 0.24 0.213 0
## Q3 0.053 0.180 0.41 0.24 0.113 0
## Q4 0.140 0.280 0.30 0.19 0.093 0
## Q5 0.040 0.167 0.35 0.33 0.113 0
## Q6 0.080 0.233 0.33 0.26 0.093 0
## Q7 0.133 0.267 0.23 0.29 0.080 0
## Q8 0.047 0.100 0.48 0.23 0.147 0
## Q9 0.047 0.093 0.42 0.25 0.187 0
## Q10 0.073 0.107 0.42 0.23 0.167 0
## Q11 0.027 0.153 0.35 0.39 0.087 0
## Q12 0.073 0.327 0.33 0.23 0.033 0
All response options are used with no missing values.
Normality of data
This is done to check for the normality of the data. If the data are normally distributed, we may use maximum
likelihood (ML) for the EFA, which will allow more detailed analysis. Otherwise, the extraction method of
choice is principal axis factoring (PAF), because it does not require normally distributed data (Brown,
2015).
Univariate normality
1. Histograms
par(mfrow = c(3, 4)) # set view to 3 rows & 4 columns
apply(data1, 2, hist)

3
par(mfrow = c(1, 1)) # set to default full view
# multi.hist(data1) # at times, error

Histogram of newX[, i] Histogram of newX[, i] Histogram of newX[, i] Histogram of newX[, i]

40 60
20 40 60

40
40
Frequency

Frequency

Frequency

Frequency

20
20

20
0

0
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5

newX[, i] newX[, i] newX[, i] newX[, i]

Histogram of newX[, i] Histogram of newX[, i] Histogram of newX[, i] Histogram of newX[, i]

40
40
Frequency

Frequency

Frequency

Frequency
40

50
20
20
20

0 20
0

1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5

newX[, i] newX[, i] newX[, i] newX[, i]

Histogram of newX[, i] Histogram of newX[, i] Histogram of newX[, i] Histogram of newX[, i]

60
20 40 60

20 40 60

40
Frequency

Frequency

Frequency

Frequency
40

20
20
0

1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5

newX[, i] newX[, i] newX[, i] newX[, i]

all of which look quite normal.

2. Shapiro Wilk’s test
apply(data1, 2, shapiro.test)

## $Q1
##
## Shapiro-Wilk normality test
##
## data: newX[, i]
## W = 0.91535, p-value = 1.075e-07
##
##
## $Q2

4
##
## Shapiro-Wilk normality test
##
## data: newX[, i]
## W = 0.88321, p-value = 1.656e-09
##
##
## $Q3
##
## Shapiro-Wilk normality test
##
## data: newX[, i]
## W = 0.90785, p-value = 3.76e-08
##
##
## $Q4
##
## Shapiro-Wilk normality test
##
## data: newX[, i]
## W = 0.91347, p-value = 8.225e-08
##
##
## $Q5
##
## Shapiro-Wilk normality test
##
## data: newX[, i]
## W = 0.90615, p-value = 2.986e-08
##
##
## $Q6
##
## Shapiro-Wilk normality test
##
## data: newX[, i]
## W = 0.91619, p-value = 1.214e-07
##
##
## $Q7
##
## Shapiro-Wilk normality test
##
## data: newX[, i]
## W = 0.90559, p-value = 2.768e-08
##
##
## $Q8
##
## Shapiro-Wilk normality test
##
## data: newX[, i]
## W = 0.88115, p-value = 1.301e-09
##

5
##
## $Q9
##
## Shapiro-Wilk normality test
##
## data: newX[, i]
## W = 0.88932, p-value = 3.445e-09
##
##
## $Q10
##
## Shapiro-Wilk normality test
##
## data: newX[, i]
## W = 0.89574, p-value = 7.653e-09
##
##
## $Q11
##
## Shapiro-Wilk normality test
##
## data: newX[, i]
## W = 0.89194, p-value = 4.758e-09
##
##
## $Q12
##
## Shapiro-Wilk normality test
##
## data: newX[, i]
## W = 0.90097, p-value = 1.501e-08
all P-values < 0.05, i.e. not normal.
Multivariate normality
To say the data are multivariate normal:
• z-kurtosis < 5 (Bentler, 2006) and the P-value should be ≥ 0.05.
• The plot should also form a straight line (Arifin, 2015).
Run Mardia’s multivariate normality test,
mardiaTest(data1, qqplot = TRUE)

## Mardia's Multivariate Normality Test

## ---------------------------------------
## data : data1
##
## g1p : 29.99652
## chi.skew : 749.9129
## p.value.skew : 5.923668e-29
##
## g2p : 203.0284
## z.kurtosis : 11.70215
## p.value.kurt : 0
##

6
## chi.small.skew : 767.2563
## p.value.small : 6.235697e-31
##
## Result : Data are not multivariate normal.
## ---------------------------------------

Chi−Square Q−Q Plot

30
25
Chi−Square Quantile

20
15
10
5

0 10 20 30 40 50

Squared Mahalanobis Distance

In our case, z-kurtosis = 11.702 (P < 0.05). The plot looks fairly straight, but with an outlier. Thus,
the data are not normally distributed at multivariate level. Our extraction method PAF can deal with this
non-normality.

3.2 Step 1

Check suitability of data for analysis

1. Kaiser-Meyer-Olkin (KMO) Measure of Sampling Adequacy (MSA)
MSA is a relative measure of the amount of correlation (Kaiser, 1970). It indicates whether it is worthwhile
to analyze a correlation matrix or not. KMO is an overall measure of MSA for a set of items. The following

7
is the guideline in interpreting KMO values (Kaiser & Rice, 1974):

Value Interpretation
< 0.5 Unacceptable
0.5 – 0.59 Miserable
0.6 – 0.69 Mediocre
0.7 – 0.79 Middling
0.8 – 0.89 Meritorious
0.9 – 1.00 Marvelous

KMO of our data,

KMO(data1)

## Kaiser-Meyer-Olkin factor adequacy

## Call: KMO(r = data1)
## Overall MSA = 0.76
## MSA for each item =
## Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8 Q9 Q10 Q11 Q12
## 0.34 0.83 0.64 0.75 0.83 0.81 0.82 0.81 0.68 0.70 0.82 0.68
KMO = 0.76, i.e. middling. In general, > 0.7 is acceptable.
2. Bartlet’s test of sphericity
Basically it tests whether the correlation matrix is an identity matrix1 (Bartlett, 1951; Gorsuch, 2014).
A significant test indicates worthwhile correlations between the items (i.e. off-diagonal values are not 0).
Test our data,
cortest.bartlett(data1)

## R was not square, finding R from data

## $chisq
## [1] 562.3065
##
## $p.value
## [1] 7.851736e-80
##
## $df
## [1] 66
P-value < 0.05, significant.
Determine the number of factors
There are several ways in determining the number of factors, among them are (Courtney, 2013):
1. Kaiser’s eigenvalue > 1 rule.
2. Cattell’s scree test.
3. Parallel analysis.
1a matrix, for example, in case of three variables:
" #
V1 1 0 0
V2 0 1 0
V3 0 0 1

Take note of the zero correlations with other variables.

8
 4. Very simple structure (VSS).
5. Velicer’s minimum average partial (MAP).
1. Kaiser’s eigenvalue > 1 rule.
Factors with eigenvalues > 1 are retained. Eigenvalue can be interpreted as the proportion of the information
in a factor. The cut-off of 1 means the factor contains information = 1 item. Thus it is not worthwhile
keeping factor with information < 1 item.
2. Cattell’s scree test.
“Scree” is a collection of loose stones at the base of a hill. This test is based on eye-ball judgement of an
eigenvalues vs number of factors plot. Look for the number of eigenvalue points/factors before we reach the
“scree”, i.e. at the elbow of the plot.
Obtain the eigenvalues and scree plot,
scree = scree(data1)
print(scree)

Scree plot

PC
FA
3.0
Eigen values of factors and components

2.5
2.0
1.5
1.0
0.5
0.0

2 4 6 8 10 12

factor or component number

9
## Scree of eigen values
## Call: NULL
## Eigen values of factors [1] 2.67 1.78 0.18 0.07 0.02 -0.05 -0.09 -0.15 -0.20 -0.41
## -0.46 -0.70
## Eigen values of Principal Components [1] 3.29 2.66 1.04 0.96 0.79 0.73 0.65 0.52 0.46 0.36
## 0.33 0.22
Based on our judgement on the scree plot and eigenvalues (of factor analysis), the suitable number of factors
= 2.
3. Parallel analysis.
The scree plot based on the data is compared to the scree plot based on the randomly generated data (Brown,
2015). The number of factors is the number of points above the intersection between the plots.
parallel = fa.parallel(data1, fm = "pa", fa = "fa")
print(parallel)

## Parallel analysis suggests that the number of factors = 2 and the number of components = NA
## Call: fa.parallel(x = data1, fm = "pa", fa = "fa")
## Parallel analysis suggests that the number of factors = 2 and the number of components = NA
##
## Eigen Values of
##
## eigen values of factors
## [1] 2.67 1.78 0.18 0.07 0.02 -0.05 -0.09 -0.15 -0.20 -0.41 -0.46 -0.70
##
## eigen values of simulated factors
## [1] 0.71 0.39 0.29 0.19 0.12 0.05 -0.01 -0.08 -0.14 -0.20 -0.27 -0.35
##
## eigen values of components
## [1] 3.29 2.66 1.04 0.96 0.79 0.73 0.65 0.52 0.46 0.36 0.33 0.22
##
## eigen values of simulated components
## [1] NA

10
 Parallel Analysis Scree Plots

FA Actual Data
FA Simulated Data
3

FA Resampled Data
eigen values of principal factors

2
1
0

2 4 6 8 10 12

Factor Number

The parallel-analysis scree plot is also suggestive of 2 factors.

4. Very simple structure (VSS) criterion.
VSS compares the original correlation matrix to a simplified correlation matrix (Revelle, 2017). Look for the
highest VSS value at complexity 1 (vss1) i.e. an item loads only on one factor.
5. Velicer’s minimum average partial (MAP) criterion.
MAP criterion indicates the optimum number of factors that minimizes the MAP value. The procedure
extracts the correlations explained by the factors, leaving only minimum correlations unrelated to the factors.
Obtain these two criteria,

11
vss(data1)

1.0 Very Simple Structure

4
3
4 4 4
4 3 3 2
3 3 3
0.8

2 2 2 2 2 2
Very Simple Structure Fit

1 1
1 1 1
1
0.6

1
0.4
0.2
0.0

1 2 3 4 5 6 7 8

Number of Factors

##
## Very Simple Structure
## Call: vss(x = data1)
## Although the VSS complexity 1 shows 5 factors, it is probably more reasonable to think
## about 3 factors
## VSS complexity 2 achieves a maximimum of 0.83 with 7 factors
##
## The Velicer MAP achieves a minimum of 0.03 with 2 factors
## BIC achieves a minimum of NA with 2 factors
## Sample Size adjusted BIC achieves a minimum of NA with 2 factors
##
## Statistics by number of factors
## vss1 vss2 map dof chisq prob sqresid fit RMSEA BIC SABIC complex eChisq
## 1 0.47 0.00 0.065 54 306.874 4.7e-37 11.9 0.47 0.1814 36 207.2 1.0 622.420
## 2 0.68 0.78 0.029 43 62.250 2.9e-02 4.9 0.78 0.0585 -153 -17.1 1.3 41.527

12
## 3 0.70 0.80 0.048 33 46.613 5.8e-02 4.1 0.82 0.0568 -119 -14.3 1.3 25.305
## 4 0.67 0.80 0.067 24 27.823 2.7e-01 3.7 0.83 0.0385 -92 -16.5 1.4 14.039
## 5 0.72 0.80 0.089 16 19.445 2.5e-01 3.0 0.86 0.0438 -61 -10.1 1.3 8.727
## 6 0.65 0.78 0.119 9 8.585 4.8e-01 2.8 0.87 0.0097 -37 -8.0 1.5 3.975
## 7 0.61 0.83 0.159 3 3.094 3.8e-01 1.9 0.92 0.0261 -12 -2.4 1.5 1.281
## 8 0.62 0.76 0.239 -2 0.082 NA 2.2 0.90 NA NA NA 1.6 0.038
## SRMR eCRMS eBIC
## 1 0.1773 0.196 352
## 2 0.0458 0.057 -174
## 3 0.0357 0.051 -140
## 4 0.0266 0.044 -106
## 5 0.0210 0.043 -71
## 6 0.0142 0.038 -41
## 7 0.0080 0.038 -14
## 8 0.0014 NA NA
VSS indicates 3/5 factors (vss1 largest at 3 and 5 factors), while MAP indicates 2 factors (map smallest at 2
factors).

3.3 Step 2

Run EFA
Our data are not normally distributed, hence the extraction method of choice is principal axis factoring
(PAF), because it does not assume normality of data (Brown, 2015). The recommended rotation method is
oblimin (Fabrigar & Wegener, 2012).
We run EFA by
1. fixing the number of factors as decided from previous step. Two factors are reasonable.
2. choosing an appropriate extraction method. We use PAF, fm = "pa".
3. choosing an appropriate oblique rotation method. We use oblimin, rotate = "oblimin".
fa = fa(data1, nfactors = 2, fm = "pa", rotate = "oblimin")
print(fa, cut = 0.3, digits = 3)
# use `print(fa, digits = 3)` to view FLs < .3

## Factor Analysis using method = pa

## Call: fa(r = data1, nfactors = 2, rotate = "oblimin", fm = "pa")
## Standardized loadings (pattern matrix) based upon correlation matrix
## PA1 PA2 h2 u2 com
## Q1 0.00366 0.996 1.77
## Q2 0.413 0.20708 0.793 1.29
## Q3 -0.339 0.439 0.28192 0.718 1.88
## Q4 0.813 0.65855 0.341 1.00
## Q5 0.584 0.41688 0.583 1.30
## Q6 0.725 0.52512 0.475 1.00
## Q7 0.732 0.53270 0.467 1.01
## Q8 0.655 0.50124 0.499 1.22
## Q9 0.773 0.59830 0.402 1.00
## Q10 0.883 0.77491 0.225 1.01
## Q11 0.528 0.29771 0.702 1.07
## Q12 -0.326 0.17665 0.823 1.98
##
## PA1 PA2
## SS loadings 2.646 2.329

13
## Proportion Var 0.220 0.194
## Cumulative Var 0.220 0.415
## Proportion Explained 0.532 0.468
## Cumulative Proportion 0.532 1.000
##
## With factor correlations of
## PA1 PA2
## PA1 1.000 0.087
## PA2 0.087 1.000
Results
Judge the quality of items.
We must looks at
1. Factor loadings (FL).
2. Communalities.
3. Factor correlations.
1. Factor loadings (pattern coefficients).
Factor loadings (FLs) / pattern coefficients are partial correlation coefficients of factors to items. FLs can be
interpreted as follows (Hair, Black, Babin, & Anderson, 2010):

Value Interpretation
0.3 to 0.4 Minimally acceptable
≥ 0.5 Practically significant
≥ 0.7 Well-defined structure

The FLs are interpreted based on absolute values, ignoring the +/- signs. We may need to remove items
based on this assessment. Usually we may remove items with FLs < 0.3 (or < 0.4, or < 0.5). But the decision
depends on whether we want to set a strict or lenient cut-off value.
In our output 2 :
## Factor Analysis using method = pa
## Call: fa(r = data1, nfactors = 2, rotate = "oblimin", fm = "pa")
## Standardized loadings (pattern matrix) based upon correlation matrix
## PA1 PA2 h2 u2 com
## Q1 0.00366 0.996 1.77
## Q2 0.413 0.20708 0.793 1.29
## Q3 -0.339 0.439 0.28192 0.718 1.88
## Q4 0.813 0.65855 0.341 1.00
## Q5 0.584 0.41688 0.583 1.30
## Q6 0.725 0.52512 0.475 1.00
## Q7 0.732 0.53270 0.467 1.01
## Q8 0.655 0.50124 0.499 1.22
## Q9 0.773 0.59830 0.402 1.00
## Q10 0.883 0.77491 0.225 1.01
## Q11 0.528 0.29771 0.702 1.07
## Q12 -0.326 0.17665 0.823 1.98
Low FLs? Q1 < .3, Q12 < .4, Q2 & Q3 < .5
Also check for item cross-loading across factors (run the command again as print(fa, digits = 3) without
2 h2 = communality; u2 = error; com = item complexity.

14
cut = .3). A cross-loading is when an item has ≥ 2 significant loading (i.e. > .3/.4/.5) It indicates the
item is not specific to a factor, thus should be removed. The cross-loading can also be judged based on item
complexity (com). An item specific to a factor should have an item complexity close to one (Pettersson &
Turkheimer, 2010).
In our output:
## Factor Analysis using method = pa
## Call: fa(r = data1, nfactors = 2, rotate = "oblimin", fm = "pa")
## Standardized loadings (pattern matrix) based upon correlation matrix
## PA1 PA2 h2 u2 com
## Q1 -0.036 0.052 0.00366 0.996 1.77
## Q2 0.159 0.413 0.20708 0.793 1.29
## Q3 -0.339 0.439 0.28192 0.718 1.88
## Q4 0.813 -0.024 0.65855 0.341 1.00
## Q5 0.584 0.229 0.41688 0.583 1.30
## Q6 0.725 -0.005 0.52512 0.475 1.00
## Q7 0.732 -0.048 0.53270 0.467 1.01
## Q8 0.217 0.655 0.50124 0.499 1.22
## Q9 0.010 0.773 0.59830 0.402 1.00
## Q10 -0.058 0.883 0.77491 0.225 1.01
## Q11 0.528 0.097 0.29771 0.702 1.07
## Q12 -0.326 0.295 0.17665 0.823 1.98
Cross-loadings? Q3 & Q12
2. Communalities (h2).
An item communality3 (IC) is the % of item variance explained by the extracted factors (i.e. by both PA1
and PA2 here). It may be considered as R2 in linear regression.
The cut-off value of what is considered acceptable depends on the researcher; it depends on the amount of
explained variance that is acceptable to him/her.
A cut-off of 0.5 is practical (Hair et al., 2010), i.e. 50% of item variance is explained by all extracted factors.
However, in my practice, it depends on the minimum FL I am willing to accept. Because

Item variance (approximately) = F L2

as if FL = 0 for other factors.

For practical purpose, > 0.25 is acceptable whenever I consider FL > 0.5 as acceptable (because communality
= 0.52 = 0.25).
In our output:
## Factor Analysis using method = pa
## Call: fa(r = data1, nfactors = 2, rotate = "oblimin", fm = "pa")
## Standardized loadings (pattern matrix) based upon correlation matrix
## PA1 PA2 h2 u2 com
## Q1 0.00366 0.996 1.77
## Q2 0.413 0.20708 0.793 1.29
3 The simple formula is,
IC = F LP A1 + F LP A2
for the orthogonally-rotated solution.
For the calculation of communality for the obliquely-rotated solution, we have to include the factor correlation (FC) (Brown,
2015),
IC = F LP A1 + F LP A2 + 2(F LP A1 × F C × F LP A2 )
For example, communality of Q2 = .16ˆ2 + .41ˆ2 + 2*.16*.09*.41 = 0.205508 ≈ 0.2071 in the output. Use print(fa)
instead to view the full results.

15
## Q3 -0.339 0.439 0.28192 0.718 1.88
## Q4 0.813 0.65855 0.341 1.00
## Q5 0.584 0.41688 0.583 1.30
## Q6 0.725 0.52512 0.475 1.00
## Q7 0.732 0.53270 0.467 1.01
## Q8 0.655 0.50124 0.499 1.22
## Q9 0.773 0.59830 0.402 1.00
## Q10 0.883 0.77491 0.225 1.01
## Q11 0.528 0.29771 0.702 1.07
## Q12 -0.326 0.17665 0.823 1.98
Low communalities? Q1 < Q12 < Q2 < .25 (.004 / .177 / .207 respectively)
3. Factor correlations.
In general, correlations of < 0.85 between factors are expectable in health sciences. If the correlations are
> 0.85, the factors are not distinct from each other (factor overlap, or multicollinearity), thus they can be
combined (Brown, 2015). In EFA context, this can be done by reducing the number of extracted factors.
In our output:
## With factor correlations of
## PA1 PA2
## PA1 1.000 0.087
## PA2 0.087 1.000
PA1 ↔ PA2 = .087 < .85

3.4 Step 3

In Step 2, we found a number of poor quality items. These must be removed from the item pool.
Repeat
Repeat Step 2 every time an item is removed. Make sure that you remove only ONE item at each repeat
analysis. Make decisions based on the results.
Stop
We may stop once we have
• satisfactory number of factors.
• satisfactory item quality.
We proceed as follows,
Remove Q1? Low communality and FL:
fa1 = fa(data1[-1], nfactors = 2, fm = "pa", rotate = "oblimin")
print(fa1, cut = 0.3, digits = 3)

## Factor Analysis using method = pa

## Call: fa(r = data1[-1], nfactors = 2, rotate = "oblimin", fm = "pa")
## Standardized loadings (pattern matrix) based upon correlation matrix
## PA1 PA2 h2 u2 com
## Q2 0.412 0.207 0.793 1.29
## Q3 -0.337 0.438 0.280 0.720 1.88
## Q4 0.813 0.658 0.342 1.00
## Q5 0.584 0.417 0.583 1.30
## Q6 0.726 0.526 0.474 1.00

16
## Q7 0.733 0.534 0.466 1.01
## Q8 0.653 0.499 0.501 1.22
## Q9 0.774 0.601 0.399 1.00
## Q10 0.886 0.779 0.221 1.01
## Q11 0.529 0.298 0.702 1.07
## Q12 -0.325 0.175 0.825 1.98
##
## PA1 PA2
## SS loadings 2.645 2.327
## Proportion Var 0.240 0.212
## Cumulative Var 0.240 0.452
## Proportion Explained 0.532 0.468
## Cumulative Proportion 0.532 1.000
##
## With factor correlations of
## PA1 PA2
## PA1 1.000 0.086
## PA2 0.086 1.000
Remove Q12? Low communality & FL
fa2 = fa(data1[-c(1, 12)], nfactors = 2, fm = "pa", rotate = "oblimin")
print(fa2, cut = 0.3, digits = 3)

## Factor Analysis using method = pa

## Call: fa(r = data1[-c(1, 12)], nfactors = 2, rotate = "oblimin", fm = "pa")
## Standardized loadings (pattern matrix) based upon correlation matrix
## PA1 PA2 h2 u2 com
## Q2 0.420 0.211 0.789 1.24
## Q3 -0.313 0.409 0.239 0.761 1.87
## Q4 0.841 0.702 0.298 1.01
## Q5 0.595 0.426 0.574 1.26
## Q6 0.707 0.501 0.499 1.00
## Q7 0.732 0.531 0.469 1.01
## Q8 0.661 0.507 0.493 1.19
## Q9 0.780 0.608 0.392 1.00
## Q10 0.892 0.789 0.211 1.01
## Q11 0.529 0.298 0.702 1.06
##
## PA1 PA2
## SS loadings 2.554 2.256
## Proportion Var 0.255 0.226
## Cumulative Var 0.255 0.481
## Proportion Explained 0.531 0.469
## Cumulative Proportion 0.531 1.000
##
## With factor correlations of
## PA1 PA2 PA1 1.0 0.1
## PA2 0.1 1.0
Remove Q2? Low communality & FL
fa3 = fa(data1[-c(1, 2, 12)], nfactors = 2, fm = "pa", rotate = "oblimin")
print(fa3, cut = 0.3, digits = 3)

## Factor Analysis using method = pa

17
## Call: fa(r = data1[-c(1, 2, 12)], nfactors = 2, rotate = "oblimin",
## fm = "pa")
## Standardized loadings (pattern matrix) based upon correlation matrix
## PA1 PA2 h2 u2 com
## Q3 -0.307 0.395 0.229 0.771 1.88
## Q4 0.842 0.705 0.295 1.01
## Q5 0.604 0.438 0.562 1.27
## Q6 0.705 0.496 0.504 1.00
## Q7 0.730 0.528 0.472 1.01
## Q8 0.630 0.465 0.535 1.22
## Q9 0.796 0.635 0.365 1.00
## Q10 0.908 0.819 0.181 1.01
## Q11 0.529 0.295 0.705 1.05
##
## PA1 PA2
## SS loadings 2.531 2.080
## Proportion Var 0.281 0.231
## Cumulative Var 0.281 0.512
## Proportion Explained 0.549 0.451
## Cumulative Proportion 0.549 1.000
##
## With factor correlations of
## PA1 PA2
## PA1 1.000 0.089
## PA2 0.089 1.000
Remove Q3? Low communality & FL. High item complexity indicates cross-loading.
fa4 = fa(data1[-c(1, 2, 3, 12)], nfactors = 2, fm = "pa", rotate = "oblimin")
print(fa4, cut = 0.3, digits = 3)

## Factor Analysis using method = pa

## Call: fa(r = data1[-c(1, 2, 3, 12)], nfactors = 2, rotate = "oblimin",
## fm = "pa")
## Standardized loadings (pattern matrix) based upon correlation matrix
## PA1 PA2 h2 u2 com
## Q4 0.818 0.664 0.336 1.00
## Q5 0.614 0.447 0.553 1.21
## Q6 0.704 0.494 0.506 1.00
## Q7 0.747 0.549 0.451 1.02
## Q8 0.634 0.471 0.529 1.19
## Q9 0.849 0.717 0.283 1.00
## Q10 0.861 0.733 0.267 1.01
## Q11 0.533 0.299 0.701 1.04
##
## PA1 PA2
## SS loadings 2.441 1.933
## Proportion Var 0.305 0.242
## Cumulative Var 0.305 0.547
## Proportion Explained 0.558 0.442
## Cumulative Proportion 0.558 1.000
##
## With factor correlations of
## PA1 PA2
## PA1 1.000 0.121

18
## PA2 0.121 1.000
We are satisfied with the item quality and factor correlation. Please also note the Proportion Var row; the
values indicate the amount of variance explained by each factor (i.e. remember R2 multiple linear regression?).
PA1 explains 30.5%, and PA2 explains 24.2% of the variance in the items. In total, the extracted factors
explain 54.7% of the variance.

3.5 Summary

PA1: Q4, Q5, Q6, Q7, Q11

PA2: Q8, Q9, Q10
Name the factor based on the content of the remaining items/factor (look at the PDF file/printed copy of
the questionnaire).
PA1 – Affinity
PA2 – Importance

4 Internal consistency reliability

Cronbach’s alpha
Next, we want to determine the internal consistency reliability of the factors extracted in the EFA. This will
be done by Cronbach’s alpha. We determine the reliability of each factor separately by including the selected
items per factor.
List all items in data1,
names(data1)

## [1] "Q1" "Q2" "Q3" "Q4" "Q5" "Q6" "Q7" "Q8" "Q9" "Q10" "Q11" "Q12"
Then, we group the items in PA1 and PA2 factors into R objects.
PA1 = c("Q4", "Q5", "Q6", "Q7", "Q11")
PA2 = c("Q8", "Q9", "Q10")

We can now analyze by the subsets of data1.

In the results, we specifically look for:
1. Cronbach’s alpha
The Cronbach’s alpha indicates the internal consistency reliability. The interpretation is detailed as follows
(DeVellis, 2012, pp. 95–96):

Value Interpretation
<0.6 Unacceptable
0.60 to 0.65 Undesirable
0.65 to 0.70 Minimally acceptable
0.70 to 0.80 Respectable
0.8 to 0.9 Very good
>.90 Consider shortening the scale (i.e. multicollinear4 ).

4 See Streiner (2003).

19
 2. Corrected item-total correlation
There are four item-total correlations provided in psych. We consider these two:
• r.cor = Item-total correlation, corrected for item overlap (Revelle, 2017). This is recommended by
Revelle (2017).
• r.drop = Corrected item-total correlation, i.e. the correlation between the item with total WITHOUT
the item. This is reported in SPSS.
Ideally must be > 0.5 (Hair et al., 2010)
3. Cronbach’s alpha if item deleted
raw_alpha under Reliability if an item is dropped: heading is the Cronbach’s alpha if the item is
deleted.
This indicates the effect of removing the item on the Cronbach’s alpha. If there is a marked improvement in
Cronbach’s alpha, removing the item is justified. Keep the item whenever it results in a reduction in the
alpha, or the improvement is very minimal.
PA1
alpha.pa1 = alpha(data1[PA1])
print(alpha.pa1, digits = 3)

##
## Reliability analysis
## Call: alpha(x = data1[PA1])
##
## raw_alpha std.alpha G6(smc) average_r S/N ase mean sd
## 0.817 0.815 0.791 0.469 4.41 0.0231 3.09 0.824
##
## lower alpha upper 95% confidence boundaries
## 0.771 0.817 0.862
##
## Reliability if an item is dropped:
## raw_alpha std.alpha G6(smc) average_r S/N alpha se
## Q4 0.748 0.748 0.704 0.426 2.97 0.0332
## Q5 0.792 0.788 0.749 0.482 3.72 0.0268
## Q6 0.776 0.776 0.736 0.464 3.46 0.0292
## Q7 0.771 0.771 0.726 0.457 3.36 0.0299
## Q11 0.810 0.809 0.768 0.514 4.22 0.0249
##
## Item statistics
## n raw.r std.r r.cor r.drop mean sd
## Q4 150 0.836 0.825 0.782 0.709 2.81 1.172
## Q5 150 0.723 0.737 0.635 0.569 3.31 1.011
## Q6 150 0.772 0.765 0.687 0.624 3.05 1.092
## Q7 150 0.795 0.777 0.710 0.640 2.92 1.190
## Q11 150 0.660 0.687 0.557 0.500 3.35 0.935
##
## Non missing response frequency for each item
## 1 2 3 4 5 miss
## Q4 0.140 0.280 0.300 0.187 0.093 0
## Q5 0.040 0.167 0.347 0.333 0.113 0
## Q6 0.080 0.233 0.333 0.260 0.093 0
## Q7 0.133 0.267 0.227 0.293 0.080 0
## Q11 0.027 0.153 0.347 0.387 0.087 0

20
PA2
alpha.pa2 = alpha(data1[PA2])
print(alpha.pa2, digits = 3)

##
## Reliability analysis
## Call: alpha(x = data1[PA2])
##
## raw_alpha std.alpha G6(smc) average_r S/N ase mean sd
## 0.826 0.825 0.771 0.611 4.71 0.0246 3.36 0.904
##
## lower alpha upper 95% confidence boundaries
## 0.777 0.826 0.874
##
## Reliability if an item is dropped:
## raw_alpha std.alpha G6(smc) average_r S/N alpha se
## Q8 0.840 0.841 0.725 0.725 5.28 0.0260
## Q9 0.715 0.717 0.559 0.559 2.54 0.0462
## Q10 0.708 0.708 0.548 0.548 2.43 0.0476
##
## Item statistics
## n raw.r std.r r.cor r.drop mean sd
## Q8 150 0.807 0.816 0.645 0.596 3.33 1.00
## Q9 150 0.882 0.881 0.807 0.726 3.44 1.05
## Q10 150 0.892 0.885 0.815 0.732 3.31 1.10
##
## Non missing response frequency for each item
## 1 2 3 4 5 miss
## Q8 0.047 0.100 0.48 0.227 0.147 0
## Q9 0.047 0.093 0.42 0.253 0.187 0
## Q10 0.073 0.107 0.42 0.233 0.167 0
For both PA1 and PA2,
• the Cronbach’s alpha values > 0.7.
• the corrected item-total correlations > 0.5.
• deleting any of the items will result in reductions of the alpha values.
We may conclude that the factors are reliable and we must keep all items.

5 Results presentation
In the report, you must include a number of important statements and results pertaining to the EFA,
1. The extraction and rotation methods.
2. The KMO and Bartlett’s test of sphericity results.
3. The number of extracted factors, based on the applied methods e.g. scree plot, parallel analysis, MAP
etc.
4. Details about the cut-off values of the FLs, communalities and factor correlations.
5. Details about the repeat EFA, i.e. item removed, reduction/increase in the number of factors etc.
6. The percentage of variance explained (in the final solution).
7. The cut-off value of the Cronbach’s alpha.
8. Summary table, which includes FLs, communalities, Cronbach’s alpha, and factor correlations.
Factor loadings and reliability in the EFA.

21
 Factor Item Factor loading Communality Cronbach’s alpha
Affinity Q4 0.818 0.664 0.817
Q5 0.614 0.447
Q6 0.704 0.494
Q7 0.747 0.549
Q11 0.533 0.299
Importance Q8 0.634 0.471 0.826
Q9 0.849 0.717
Q10 0.861 0.733
Factor correlation:
- Affinity ↔ Importance r = 0.121.

References
Arifin, W. N. (2015). The graphical assessment of multivariate normality using spss. Education in Medicine
Journal, 7 (2), e71–e75.
Bartlett, M. S. (1951). The effect of standardization on a χ 2 approximation in factor analysis. Biometrika,
38 (3/4), 337–344.
Bentler, P. M. (2006). EQS 6 structural equations program manual. Encino, CA: Multivariate Software, Inc.
Brown, T. A. (2015). Confirmatory factor analysis for applied research. New York: The Guilford Press.
Courtney, M. G. R. (2013). Determining the number of factors to retain in efa: Using the spss r-menu v2. 0
to make more judicious estimations. Practical Assessment, Research & Evaluation, 18 (8), 1–14.
DeVellis, R. F. (2012). Scale development: Theory and applications (3rd ed). California: Sage publications.
Fabrigar, L., & Wegener, D. (2012). Exploratory factor analysis. New York: Oxford University Press.
Gorsuch, R. L. (2014). Exploratory factor analysis. New York: Routledge.
Hair, J. F., Black, W. C., Babin, B. J., & Anderson, R. E. (2010). Multivariate data analysis. New Jersey:
Prentice Hall.
Kaiser, H. F. (1970). A second generation little jiffy. Psychometrika, 35 (4), 401–415.
Kaiser, H. F., & Rice, J. (1974). Little jiffy, mark iv. Educational and Psychological Measurement, 34 (1),
111–117.
Korkmaz, S., Goksuluk, D., & Zararsiz, G. (2016). MVN: Multivariate normality tests. Retrieved from
https://CRAN.R-project.org/package=MVN
Pettersson, E., & Turkheimer, E. (2010). Item selection, evaluation, and simple structure in personality data.
Journal of Research in Personality, 44 (4), 407–420.
Revelle, W. (2017). Psych: Procedures for psychological, psychometric, and personality research. Retrieved
from https://CRAN.R-project.org/package=psych

22