You are on page 1of 6

Analysis of Hoge Religious Motivation Scale by Means of Combined

HAC and PCA Methods

Ana Štambuk
Department of Social Work, Faculty of Law, University of Zagreb,
Nazorova 51, HR-10000 Zagreb, Croatia
E-mail: astambuk@inet.hr

Nikola Štambuk, Paško Konjevoda


1
Ruer Boškovi Institute, Bijenika cesta 54, HR-10002 Zagreb, Croatia
E-mails: stambuk@irb.hr, pkonjev@irb.hr

Abstract. We used a method of combined [1, 2]. However, depending on the dataset
Hierarchical Agglomerative Clustering (HAC) analysed the results of HAC and PCA could
and Principal Components Analysis (PCA) to lead to comparable and complementary
validate Hoge Intrinsic Religious Motivation component extraction [1, 2]. In this
Scale and investigate if this consensus investigation we used a method of combined
procedure provides an efficient technique for HAC and PCA analyses in order to evaluate if
the component extraction. Our results confirm this consensus procedure provides an efficient
the validity of the procedure and suggest that it technique for the component extraction. The
may be useful exploratory technique for the analysis was done on a standard social sciences
data analysis in social sciences. example of the religious motivation
assessment.
Keywords. Principal components, analysis,
hierarchical clustering, social sciences,
religious motivation. 2. Methods

2.1. Dataset
1. Introduction
Hoge Intrinsic Religious Motivation Scale
One of the main goals of the exploratory is a standard instrument for the assessment of
statistical analyses is to identify relevant religious motivation [3-5]. It observes
features and/or structural patterns in the data statements about religious beliefs or
[1, 2]. Principal Components Analysis (PCA) experience [3-5]. The scale is valid for the
is a popular statistical procedure often used in Croatian population since the results do not
Social Sciences for the exploratory statistics, differ from the ones reported for the USA [5].
i.e. to reduce a number of variables of the The responses of 473 participants with no
dataset analysed in order to extract few significant differences in gender and age were
underlying patterns or groups of variables [2]. used for the analysis (man/woman=222/251;
In addition to transforming more variables of age=56.6 ± 21.8 years, range 20-90) [5].
the initial dataset into few main components,
PCA may also help understanding the data Ten items/questions (Q) of the Hoge
structure [1, 2]. Intrinsic Religious Motivation Scale are:

Hierarchical Agglomerative Clustering 1. My faith involves all of my life,


(HAC) procedure may be also used for the 2. Beliefs are less important than living a
exploratory data analyses of main components moral life,

197
th
Proceedings of the ITI 2007 29 Int. Conf. on Information Technology Interfaces, June 25-28, 2007, Cavtat, Croatia
3. One should seek God's guidance when religiosity in Table 5 was done with a free
making important decisions, software Weka 3.4.4 [6].
4. In my life, I experience the presence of
the Divine (God), Weka logistic function classifier is based
5. Refuse to let religion influence on a multinomial logistic regression model
everyday affairs, with a ridge estimator [7]. The algorithm is
6. Faith sometimes restricts my actions, modified to handle the instance weights [7].
7. Nothing is as important as serving God, For the classifier evaluation class attributes
8. Many more important things in life than must be nominal and other variables (Qs) may
religion, be ordinal or interval [6-8].
9. Religious beliefs lie behind my whole
approach to life,
10. Try hard to carry religion over into
life's dealings. Table 1. Clusters of questions obtained by
the analysis of Religious Motivation Scale.
The answers are marked on a 1-5 scale. (1)
denotes the statement that is definitely true and
(5) the statement that is definitely not true for
the participants. Score of 10 indicates high Cluster Size
awareness of spiritual issues and high religious
motivation while a score of 50 indicates no cluster n°1 225
religious/spiritual motivation or understanding
[3, 4].
cluster n°2 154

Participants were also asked to evaluate the


importance of religion for them (religiosity), cluster n°3 94
on a 1-7 scale (1=not important, 7=very
important) [5]. Scores 6-7 were considered as
highly important (A), 3-5 as moderate (B), and
1-2 as of very low or no importance (C).

2.2. PCA and hierarhical Clustering


5

Hierarchical agglomerative clustering


(HAC) and Principal Components Analysis 4

(PCA) of the Hoge scale questions (Q1-Q10)


were done with Tanagra software 1.4.14 3

(http://eric.univ-lyon2.fr/~ricco/tanagra/en/tanagra.html).
2

Tanagra implements the procedure of HAC


known as Hybrid Clustering. First, a low-level 1

clusters are built from fast clustering method


such as K-Means, SOM, then HAC starts form 0

these clusters and builds the dendogram (Fig.


1). The advantage of HAC is that the user can
visualize the tree and guess the right
partitioning and prune the tree between the Figure 1. Hierarchical agglomerative
clustering of the subjects into 3 main
nodes. Following this PCA procedure is done
groups.
based on the results of HAC. This enables
explanation of HAC subgroups using PCA
factors (Tables 2-4, Fig. 2).

Logistic regression based classification


with respect to class attributes gender, age and

198
Table 2. Principal component analysis of Table 3. Factor loadings (communality
the Hoge Religious Motivation Scale. estimates) of PCA analyses

Axis 1 Axis 2 Axis 3


Axis Eigenvalues % variance % cumulative
% % %
1 4.93 49.29% 49.29% Corr. (Tot.) Corr. (Tot.) Corr. (Tot.)
2 1.66 16.61% 65.89% Q1 73 % 0% 0%
3 0.81 8.10% 73.99% 0.85 (73 %) -0.05 (73 %) 0.04 (73 %)
4 0.58 5.84% 79.83% Q2 11 % 25 % 61 %
5 0.53 5.27% 85.10% -0.33 (11 %) 0.50 (35 %) 0.78 (97 %)
6 0.40 3.97% 89.07% Q3 66 % 0% 2%
7 0.32 3.19% 92.26% 0.81 (66 %) -0.02 (66 %) 0.15 (68 %)
8 0.31 3.06% 95.32% Q4 73 % 2% 0%
9 0.25 2.51% 97.83% 0.86 (73 %) 0.12 (75 %) -0.03 (75 %)
10 0.22 2.17% 100.00% Q5 3% 63 % 8%
Tot. 10 - - 0.17 (3 %) 0.79 (66 %) -0.28 (74 %)

Q6 52 % 2% 2%
0.72 (52 %) 0.15 (54 %) -0.13 (56 %)

Q7 73 % 2% 4%
0.85 (73 %) -0.15 (75 %) 0.20 (80 %)

Q8 1% 69 % 1%
Correlation scatterplot (PCA_1_Axis_1 vs. PCA_1_Axis_2) -0.08 (1 %) 0.83 (70 %) -0.11 (71 %)
1
0,9
Q8
0,8 Q5 Q9 71 % 2% 0%
0,7 0.84 (71 %) 0.15 (74 %) -0.05 (74 %)
0,6
Q2
0,5 Q10 70 % 0% 2%
0,4 0.84 (70 %) -0.07 (71 %) 0.14 (73 %)
0,3
Var.
0,2 Q6 Q9
Q4 49 % 17 % 8%
PCA_1_Axis_2

0,1 Expl.
0 Q3Q1 4.93 (49 %) 1.66 (66 %) 0.81 (74 %)
Q10
-0,1 Q7
-0,2
-0,3
-0,4
-0,5
3. Results and Discussion
-0,6
-0,7
-0,8 HAC procedure of unsupervised learning,
-0,9
-1
based on Qs scores, extracted 3 clusters of
-1 -0,9 -0,8 -0,7 -0,6 -0,5 -0,4 -0,3 -0,2 -0,1 0 0,1 0,2
PCA_1_Axis_1
0,3 0,4 0,5 0,6 0,7 0,8 0,9 1
subjects with different religious motivation
(Table 1, Fig. 1). Following this, PCA analysis
identified questions that discriminate the
clusters of subjects identified by HAC (Tables
2-4, Fig. 2).
Figure 2. Analysis of Religious Motivation
Scale by means of two dimensional PCA The first group of questions explained the
(three clusters). intrinsic religious motivation. It consisted of
Q1, Q3, Q4, Q6, Q7, Q9 and Q10 (Fig. 2,
Tables 3-4). The second group was
characterized by Q5 and Q8 and the third
group by Q2 (Fig. 2, Tables 3-4).

199
The advantage of combined HAC and PCA
methods is that multidimensional data can be Table 4. Eigen vectors - factor scores of
visualized as two-dimensional maps. PCA.
Moreover, different subgroups (e.g. gender,
age, religious attitudes, etc.) can be, as shown Attribute Mean SD Axis 1 Axis 2 Axis 3
in Fig. 3, also displayed by means of different
graphical patterns which makes their Q1 2.6 1.4 0.38 -0.04 0.05
comparison easier. Q2 2.7 1.4 -0.15 0.39 0.87
Q3 2.7 1.5 0.37 -0.01 0.17
Q4 2.3 1.4 0.39 0.10 -0.03
a) Q5 3.0 1.4 0.08 0.62 -0.31
(X1) PCA_1_Axis_1 vs. (X2) PCA_1_Axis_2 by (Y) GENDE_AB
3
Q6 2.7 1.4 0.33 0.11 -0.14
2 Q7 3.0 1.4 0.38 -0.12 0.23
Q8 2.9 1.5 -0.04 0.65 -0.13
1

Q9 2.4 1.4 0.38 0.12 -0.06


0

Q10 2.9 1.4 0.38 -0.05 0.16


-1

-2

-3 -2 -1 0 1 2 3 4

A B Table 5. The results of logistic regression


analysis for variables gender, age and
b) religiosity.

(X1) PCA_1_Axis_1 vs. (X2) PCA_1_Axis_2 by (Y) AGE_ABC


3
% correct % ten-fold
2
classification CV
Man 42.3 38.7
1 Woman 71.3 68.1
Overall 57.7 54.3
0

Age 20-34 12.9 12.9


-1
Age 35-64 46.2 41.8
Age ≥65 67.2 64.1
-2
Overall 48.4 45.5
-3 -2 -1 0 1 2 3 4

C A B
Very religious 85.2 84.8
Religious 50.0 46.0
Not religious 73.6 71.7
c) Overall 72.7 71.7
(X1) PCA_1_Axis_1 vs. (X2) PCA_1_Axis_2 by (Y) REL_ABC
3

The results of logistic regression analysis


1
(Table 5) confirm the validity of the Hoge
0
scale [3, 5] and show that intrinsic and
extrinsic religious motivation are dependent on
-1
the persons’ religiosity, mainly for very
-2
religious persons and non-religious
-3 -2 -1 0 1 2 3 4
individuals, while the group of moderately
A B C religious persons tends to be misclassified,
with present set of questions (Q1-Q10).

Figure 3. Analysis of Religious Motivation Logistic regression analysis (Table 5)


Scale by means of two dimensional PCA additionally showed that variables gender and
(three clusters). age do not affect the answers of the subjects

200
(Qs), however, the visual output of HAC and 5. References
PCA is more intuitive (Fig. 3).
[1] Gentle JE. Elements of Computational
Hoge Intrinsic Religious Motivation Scale Statistics. New York: Springer-Verlag;
observes statements about religious beliefs [3- 2004.
5]. The percentage of explained variance using [2] Everitt BS, Dunn G. Applied Multivariate
three extracted principal components is Data Analysis. London:Arnold; 2001.
sufficiently high (73.99% cumulative, Table 2) [3] Hoge DR. (1972). A validated Intrinsic
and explains the dataset variation better then Religious Motivation Scale. Journal for the
general factors like gender and age (Fig. 3, Scientific Study of Religion 1972; 11: 369-
Table 5). The second component consists of 76.
Q5 and Q8 and the third one of Q2 only (Fig. [4] King M, Speck P, Thomas A. The Royal
2, Tables 3-4). However, they contribute Free Interview for Spiritual and Religious
considerably to the percentage of explained Beliefs:Development and Validation of a
dataset variance (24.70% cumulative, Table 2). Self-report Version. Psychological
Medicine 2001; 31: 1015-23.
Combined method of HAC and PCA [5] Štambuk A. Stavovi Starijih Osoba Prema
exploratory data analysis provides useful Smrti i Umiranju. PhD thesis: University
information regarding subsequent formal of Zagreb; 2004.
statistical procedures, since it enables the [6] Witten IH, Frank E. Data Mining. San
identification of factors important for further Francisco: Morgan Kaufmann; 2005.
statistical modeling based on the supervised [7] Le Cessie S, Van Houwelingen JC. Ridge
learning methods. Estimators in Logistic Regression.
Applied Statistics 1992; 41: 191-201.
[8] Siegel S, Castellan NJ. Nonparametric
4. Acknowledgements Statistics for the Behavioral Sciences.
Singapore: McGraw-Hill; 1988.
The work was supported in part by the
Croatian Ministry of Science Education and
Sport (N. Štambuk and P. Konjevoda; Grant Correspondence to: Nikola Štambuk
No. 098-0982929-2524).

201
202

You might also like