International Journal of Business Management and Administration Vol. 2(5), pp.

xxx-xxx, May 2013

Available online at http://academeresearchjournals.org/journal/ijbma
ISSN 2327-3100 ©2013 Academe Research Journals

References coloured red in the reference section is not cited in the main work.

Full Length Research Paper

Confirmatory factor analysis (CFA) of first order factor

measurement model-ICT empowerment in Nigeria
Baba Hafiz* and Jamal Abdul Nassir Shaari
Faculty of Economics and Business, Universiti Malaysia Sarawak (UNIMAS), 94300 Kota Samarahan, Sarawak,
Malaysia.
Accepted 22 April, 2013

The purpose of confirmatory factor analysis (CFA) of first order factor measurement model is a way of
testing how well measured variables represent in a small construct. Prior to this analysis, Cronbach
Alpha, exploratory factor analysis (EFA) and uni-dimensional (CFA) were performed. After the
Cronbach-Alpha of the first order model, the six variables or factors were further analyzed at the
second order level. The second order analysis was carried out to achieve a valid model fit for the data
obtained as well as theoretical supports behind the developed model.

Key words: Measurement model, constructs, reliability and validity.

INTRODUCTION

Confirmatory factor analysis (CFA) is a confirmatory Barret (2007) view that a good model fit would provide
technique and is theory-driven. Therefore, the planning of an insignificant result of p>0.05 threshold. Thus, the Chi-
the analysis is driven by the theoretical relationships square statistics are often referred to as goodness or
2-
among the observed and unobserved variables. When a badness-of-fit whereby large x values correspond to bad
2
CFA is dis-conducted, the researcher uses a fit and small x - values to good fit. The degree of freedom
2
hypothesized model to estimate a population covariance serve as a standard by which to judge x is small or large
matrix that is compared with the observed covariance (Fitzgerald et al., 1997). Thus, Byrne (2001, 2010) view
matrix. Technically, the researcher wants to minimize the that other model fit such as GFI, AGFI, NNFI and IFI are
difference between the estimate and observed matrices applied to access the goodness of fit of the measurement
(Long, 1983; Hu and Bentler, 1995; Baron and Kendy, model.
1986; Byrne, 1989, 2010; Hair et al., 2010).
In this study, confirmatory factor analysis (CFA) Reliability
measurement model of ICT empowerment first order after
exploratory factor analysis (EFA) and confirmatory factor In assessing the degree of measurement error present in
analysis (CFA) uni-dimensional test, was run using any measure, the research must address two important
Analysis of Moment Structure (AMOS) 16 version to test characteristics of a measure: validity and reliability
the measurement models (James, 2005) on variables of according to Hair et al. (2010) and Byrne (1989, 2010).
ICT empowerment. However, confirmatory factor analysis The researchers‟ goal in this study is to reduce
approach to access uni-dimensionality was adopted measurement error.
because of the existence of single construct underlying a Validity is the degree to which instrument measures
set of measures and as a set of items forming an accurately represent what is supposed to measure, and
instrument that in all measure one thing in common. The the degree to which the “thing” that the instrument
first fit measure in this thesis is the Chi-Square statistic. measures has meaning. Ensuring validity starts with a
Bentler and Hu (1999) view that the Chi-Square statistic thorough understanding of what is to be measured and
is the traditional measure for evaluating the overall model
fit in covariance structure models and it access the
magnitude of discrepancy between the sample and fitted
covariance. *Corresponding author. E-mail: babahafiz@yahoo.com.
then making the measurement as correct and accurate (CFA). This was done as coefficient alpha is not a
as possible. However, accuracy does not ensure validity. sufficient condition to assess uni-dimensionality. For this
If validity is assured still, consideration of reliability of the reason, other authors such as Steenkamp and Van Trijp
measurement is a must (Hair et al., 2010; Pallant, 2010; (1991) and Byrne (2001; 2010) maintain that CFA
Andy, 2009). provides a better estimate of reliability than coefficient
Reliability can be assessed through internal alpha. Hinkin (1995) suggests that CFA approach is able
consistency of the instrument (Zikmund, 2003; Andy, to examine the stability of the factor structure in scale
2009), which is the most appropriate used in this study. construction. Furthermore, assessing reliability by using
The second dimension of reliability is used to assess the CFA is also necessary to ensure that all measures used
reliability of summated scale where several items are in this thesis are reliable, thus providing the researcher
summed to form the total scores (Malhotra, 1996). If they with greater confidence that individual items are
are reliable, the items will show consistency in their consistent in their measurements (Hair et al., 1995,
indication of concept being measured. To avoid this 2010).
problem, Cronbach‟s (1951) coefficient alpha which is
one of the most common methods in gauging reliability Validity
(Nunnally, 1978; Peter, 1979; Sekaran, 2000), is
considered appropriate. This technique estimates the Validity is the degree to which an instrument measures
degree to which the items in the scale are representative what it intends to measure, and the degree to which the
of the domain of the construct being measured. It is a “thing” that the instrument measures has meaning.
measure of the internal consistency of a set of items, and Validity is viewed by Hair et al. (2010) as the extent to
is considered „absolutely the first measure‟ one should which a measure or set of measures correctly represents
use to assess the reliability of a measurement scale the concept of the study, and the degree to which it is
(Nunnally, 1978; Churchill, 1979). In addition to this, free from any systematic or nonrandom error. Validity is
Cronbach‟s coefficient is important in measuring multi- concerned with how the concept is defined by the
point scale items (that is, 5-point Likert scale used in this measures, whereas reliability relates to the consistency
study) (Sekaran, 2000). Accordingly, this method of of the measures.
internal consistency has been adopted to assess the Reliability alone is not sufficient to consider that an
reliability of the measures in this study. instrument is adequate (Churchill, 1979; Anderson and
Multi-items scales were employed in this study; Gerbing, 1988; Dunn et al., 1994; Hair et al., 1995, 2010).
Cronbach alpha estimate was used as a verification of Therefore, validity is required to validate the constructs of
the reliability of the composite items comprising each this thesis. According to Zikmund (2003), validity means
scale construct. Thus, the constructs of MTC, CGC, PPC, “the ability of a scale to measure what intended to be
RLC, STC and TCC were subjected to such assessment. measured”. Neuman (2003) points out that the better the
In assessing reliability through Cronbach‟s Alpha, some fit between the conceptual and operational definitions, the
authors suggest different levels of acceptance. Nunnally greater the measurement validity. In addition to this,
(1967) recommended that an acceptable alpha is validity represents the relationship between the construct
between 0.50 and 0.60 and in his other book titled and its indicators (Punch, 1998).
“Psychometric Theory”, Nunnally (1978) increased the Nunnally and Bernstein (1994) suggest that there are
level of acceptance and considered that alpha should three important aspects of a valid construct. Firstly, the
exceed the minimum of 0.70 for internal consistency. construct should be seen to be a good representation of
However, Nunnally and Bernstein (1994) suggest a rule the observable domain related to the construct. Secondly,
of thumb level of higher than 0.70, with a level as low as the construct should well represent the alternative
0.60 being accepted for new scale. Other authors such measures. Finally, the construct should be well related to
as Carmines and Zeller (1979) indicate that at least 0.80 other constructs of interest. Taking into account these
is required to establish internal consistency. While considerations, three types of validity, including, content,
different view have been recommended about levels of construct (convergent and discriminant validity) and
acceptance, it is generally agreed that an alpha of 0.70 criterion were examined in this thesis. These are related
and above is acceptable. Therefore, this cut-off point to the internal validity of the scales and their respective
(0.70) was used as the minimum for determining internal items. As for the purpose of the generalizability of the
consistency of scale for this study. research findings, external validity was also investigated.
This study adopted the procedure recommended by
Churchill (1979) and Peter (1979) to develop the scales. Construct validity
The issues of this procedure concerning the item were
determined using Cronbach‟s coefficient alpha and Construct validity was therefore examined in this study by
exploratory factor analysis. Similarly, Anderson and analyzing both convergent and discriminant validity.
Gerbing (1988) expanded the scale to development Convergent validity examines whether or not the
procedure by including confirmatory factor analysis measures of the same construct are correlated highly,
and discriminant validity determines if the measures of a been used as a measure within this thesis, it is therefore
construct have not correlated too highly with other assumed that criterion validity is also accounted for.
constructs (Sekaran, 2000).
A number of methods were suggested for assessing External validity
convergent and discriminant validity: factor analysis,
correlation, and even more advanced procedures The final measure used to validate the measures of this
including CFA existing in SEM. For the purpose of this study is external validity. While the above discussed
study, convergent and discriminant validity were validity relates to the internal validity of the scales and
assessed by performing CFA. To demonstrate their respective items, external validity is concerned with
convergent validity, magnitude of the direct structural establishing the extent to which the study findings can be
relationship between the item and latent construct (or generalized to other subjects or groups (that is, other
factor) should be statistically different from zero (Holmes- youths in Nigeria) (Zikmund, 2003). In more specific
Smith et al., 2006). In other words, the final items (not terms, external validity is related to the generalizability of
including deleted items) should be loaded highly on one the cause-effect relationships of the research findings
factor (Anderson and Gerbing, 1988), with a factor (Yin, 1994). Hence, evidence on external validity for this
loading of 0.50 or greater (Hair et al., 1995). Furthermore, thesis has been obtained by employing a representative
average variance extracted (AVE) was used as an sample (that is, 389 youths within the ages of 18-35
indicator for supporting convergent validity (Fornell and years from the three locations in the area of study)
Larcker, 1981). As for discriminant validity, two methods (Leedy and Ormrod, 2001; Zikmund, 2003).
have been employed in this study. The first method For the purpose of this study, four types of validity,
checks the estimated correlations between the factors including content, construct, criterion and external, were
(Kline, 2005). adopted.
This is consistent with the above discriminant validity
definition of Sekaran (2000). That is, if the two factors are METHODOLOGY
highly correlated, redundant items that show a lack of
discriminant validity are deleted (Kline, 2005). The The research was conducted among 386 employed and
second method of assessing discriminant validity unemployed youths between 18-35 years old in Bauchi
examines pattern structure coefficient to determine State, Nigeria from the three geo-political zones present
whether factors in measurement models are empirically in the location of the study areas of the state (South -
distinguishable (Thompson, 1997). Pattern coefficient is Bauchi with a population estimated as 493,810 and total
2
the standardized factor loading derived from AMOS land mass of 3,687 km ; North - Katagum whose
analysis. In addition to these restrictive assessments of population‟s estimate was 375,970 and total landmass of
2
convergent and discriminant validity, construct validity in 4,625 km ; and Central - Ningi population was estimated
this thesis was enhanced by assuring that the model at 295,970 with a total landmass of 1,436 local
(through goodness-of-fit results obtained from CFA) fits to government (Figure 1) (NPC, 2006; FOS, 1987).
the data adequately (Hsieh and Hiang, 2004). The questionnaire items have 65 indicators/constructs
with the exception of demographic questions reflecting
Criterion validity the relevant variables adapted from previous studies. The
variables are divided into 6 broad categories related to,
Criterion validity is the third measure of validity material cognitive, perceptual, relational, and
demonstrated within this study. It refers to the ability of technological and status changes. The questionnaire was
measures to correlate with other standard measures of developed using 5-point Likert Scale, ranging from (1=
the same construct (Zikmund, 2003). It can be classified strongly disagree) to (5=strongly agree).
as concurrent validity or predictive validity (Sekaran, After data had been collected from the respondents,
2000), depending on the time sequence in which the new editing of the data was undertaken in order to ensure the
measurement scale and the criterion measure are omission, completeness and consistency of the data
correlated (Zikmund, 2003). The former, for example, is (Zikmund, 2003; Sekaran, 2000). Out of 500
established when a new measure is taken at the same questionnaires distributed during the survey, 420 were
time as criterion and is shown to be valid, while the latter returned, which shows 84% of response rate. 389
is established when a new measure predicts a future respondents were identified and selected as
event. According to Peter (1981), criterion validity was representative sample size after series of screening.
commonly used in earlier research. However, its
popularity has vanished with the increased use of Assessment of goodness of fit
construct validity. This is because criterion validity is
synonymous with convergent validity, and thus The confirmatory factor analysis approach to assessing
assessment of the latter would mean that the former was measurement model of ICT empowerment is adopted
satisfied (Zikmund, 1994). Since convergent validity has because of the existence of single constructor trait
 Figure 1. Bauchi State map indicating the three zones of the study areas, that is, North (Katagum), Central (Ningi) and South
(Bauchi). Source: http//www.speakersoffice.gov.ng/constituencies_bauchi.htm (2013).

underlying a set of measures and a set of items forming square statistic. Hair et al. (2010), Byrne (2010), Pallant
an instrument that all measure one thing in common. The (2010) and Barret (2007) stated that the Chi-square
first fit measure to be reported in this study is the Chi- statistic is the traditional measure for evaluating overall
 Table 1. Assessment of goodness of fit.

Types of constructs Cut-off

2
Chi-square (x ) Smaller to 0
Degree of freedom (df)
Probability (P) ≥ 0.05
CMIN/df ≤ 2-5
Goodness of fit index (GFI) ≥ 0.90
Adjusted goodness of fit index (AGFI) ≥ 0.90
Comparative fit index (CFI) ≥ 0.95
Root mean squared error of approximation (RMSEA) ≤ 0.08
Source: Hair et al. (1998, 2010), Byrne (2001, 2010), and Ernest et al. (2008).

Table 2. Summary of goodness of fit for ICT empowerment (CFA of first order).

Test Marker Chi-sq df P CMIN/df GFI AGFI CFI NFI RMSEA

MM - 573.979 309 0.000 1.858 0.904 0.883 0.956 0.911 0.047
MM1 PPC2 541.860 284 0.000 1.908 0.905 0.882 0.951 0.903 0.048
MM2 TCC5 462.502 260 0.000 1.779 0.916 0.895 0.958 0.910 0.045
MM3 PPC6 388.115 237 0.000 1.638 0.924 0.904 0.967 0.920 0.041
MM4 MTC7 335.134 215 0.000 1.559 0.930 0.910 0.971 0.924 0.038
MM5 RLC5 286.144 194 0.000 1.475 0.937 0.918 0.976 0.931 0.035
MM6 STC3 234.300 174 0.002 1.347 0.946 0.929 0.984 0.940 0.030
MM7 STC1 195.641 155 0.015 1.262 0.952 0.935 0.989 0.947 0.026
MM8 CGC3 160.931 137 0.079 1.175 0.958 0.942 0.993 0.954 0.021
Source: Developed from Quantitative Field Report (2013).

model fit in covariance structure models and it assess the Hamilton, 1996). This approach was applied to examine
magnitude of discrepancy between the sample and fitted the dimensionality of each variable or factor and also to
covariance. However, they stated further that a good test the model fit of the six dimensions or factor of
model fit would show a significant result of p>0.05. The empowerment step by step (Table 2). Figures 2 and 3
Chi-square statistics are often referred to as either a present the results of the measurement model fit of the
2-
goodness or badness-of-fit measure, whereby large x six dimensions or factor of empowerment.
2-
values correspond to bad fit and small x values The list of items removed or deleted from the model
correspond to good fit. The degree of freedom serves as through the iteration process is shown in Table 3. The
2
a standard by which to access x is small or large. Thus, iteration process changed the number of factors under
as stated by Byrne (2001, 2009), Hair et al. (1998, 2010) post-EFA construct of empowerment. Two factors,
and Ernest et al. (2008), other model fits such as GFI, ICTTR7 and GSK3, were removed from the analysis.
AGFI, CMIN/df, P, CFI and RMSEA are employed to Thus, eight factors were resulted from CFA of first order
access the goodness of fit of the measurement model. factor model for empowerment. Table 3 presents
To evaluate the result of goodness of fit, in this thesis, summary of items removed or deleted from the analysis.
the rule of thumb of cutoff point was adopted as related to Cronbach Alpha scores for the six factors from CFA of
the model only. Evaluating the goodness of fit using first order factor model (the iterated model) were
cutoff point (Byrne, 2001, 2010; Ernest et al., 2008; Hair measured, and the dimensions show acceptable
et al., 1998, 2010), Table 1 presents assessment of reliabilities as suggested by Nunnally (1978) with scores
goodness of fit. that exceed the required > = 0.70. Table 4 shows
summaries of the results for each Cronbach-Alpha of the
ANALYSIS AND DISCUSSION dimensions level of empowerment CFA of first order
model iterated.
The model was subjected to first order confirmatory factor
analysis using AMOS 16.0. The 27 items of Conclusion
empowerment were examined using confirmatory factor
analysis (CFA) on the basis of EFA results (Gerbing and After the Cronbach-Alpha of the first order model, the six
 Figure 2. First order CFA model for empowerment - Measurement model.

Figure 3. CFA model ICT empowerment - Final measuring model 8.

 Table 3. Summary of items removed or deleted from the analysis.

Code Construct Item(s) statements Removed at

e7 PPC2 Gaining ICT skills and knowledge enhances self esteem Iteration 1 (e7)
e26 TCC5 ICT skills and knowledge contributes to research and development Iteration 2 (e26)
e5 PPC6 Beneficial of ICT skill and knowledge can benefits their immediate communities Iteration 3 (e5)
e1 MTC7 ICT skills and knowledge change one‟s life Iteration 4 (e1)
e15 RLC5 ICT skills and knowledge enhance one‟s belief Iteration 5 (e15)
e12 STC3 ICT skills and knowledge change one‟s social status Iteration 6 (e12)
e13 STC1 ICT skills and knowledge increase the status from unemployed to employed Iteration 7 (e13)
e22 CGC3 ICT skills and knowledge promote one‟s mental process Iteration 8 (e22)
Source: Developed from Quantitative Field Report (2013).

Table 4. Cronbach-Alpha of the dimension level of empowerment-CFA of first order factor model (Iterated).

Factor Description Number of items (after iteration) Cronbach-Alpha

1. Material change (MTC) 3 0.882
2. Perceptual change (PPC) 2 0.826
3. Status change (STC) 3 0.741
4. Relational change (RLC) 4 0.800
5. Cognitive change (CGC) 4 0.752
6. Technological change (TCC) 3 0.870
Source: Developed from Quantitative Field Report (2013).

variables or factors were further analyzed at the second variable distinction in social psychological research:
order level. The analysis was carried out to achieve a Conceptual, strategic and statistical consideration.
valid model fit for the data obtained as well as theoretical Barret P (2007). Structural equation modeling: Adjudging
supports behind the developed model. The test of the first model fit. University of Auckland, Management and
order implies that six factors/variables were fit and Employee Relation Department, Commerce Building C.
represented in small construct. These factors are: Brace N, Kemp R, Snelgar R (2006). A guide to data
material change, perceptual change, status change, analysis using SPSS for windows third edition.
relational change, cognitive change and technological Palgrave Macmillan Houndmills, Basingstoke,
change. Hampshire RG21 6XS and 175 Fifth Avenue, New
York, N. Y. 10010.
REFERENCES Byrne BM (1989). Measurement of statistical models in
the study: psychometric measures for statistical
Anderson JC, Gerbing DW (1982). Structural equation hypothesis.
modeling in practice: A review and recommended two- Byrne BM (2001). Structural equation modeling with
step approach. Psychol. Bull. 1986, 103(3): 401-423 AMOS: Basic concepts, applications and programming.
Anderson JC, Gerbing DW (1984). The effects of Lawrence Erbaum Associates, Inc., 10 Industrial
sampling error on convergence, improper solutions and Avenue Mahwa, New Jessey U.S.A.
goodness-of-fit indices. Byrne BM (2010). Structural equation modeling with
Anderson JC, Gerbing DW (1988). Structural equation AMOS: Basic concepts, application and programming
nd
modeling in practice: A review and recommended two- 2 Edition. Taylor & Francis Group 270 Madison
step approach. Psychol. Bull., 103(3): 411-423. Avenue New York, NY 10016.
Anderson JC, Gerbing DW (1991). An Analysis of Carmines E, Zeller RA (1979). Reliability and validity
variance approach of content validity organizational assessment. Beverly-Hills, CA Saga Publications
research methods 2(2). A Saga Publications. Churchill GA (1999). Marketing research: Methodological
Andy F (2009). Discovering statistics using SPSS third foundation. USA: The Dryden Press.
edition, SAGE Publication limited 10 Oliver‟s Yard 55 Cronbach A (1951). A useful coefficient for assessing
city Road London EC1Y 1SP. consistency. Retrieved from
Baron RM, Kenny DA (1986). The moderator-mediator http:/www.bmj.com/content/314/7080/572
Ernest L, Heriyadi (2008). Basic analysis: A guide for Hu L, Bentler PM (1999). Structural equation modeling:
student and researchers. Lee Miing Press Sdn Bhd, Guidelines for determining model fit.
Kuching Srawak. Nunnally JC (1967). Psychometric Theory. New York:
Fornell C, Larcker DF (1981). Evaluating Structural McGraw-Hill.
Equation Models with Unobservable Variables and Nunnally JC, Bernstein IH (1994). Psychometric Theory
Measurement error. Journal of Marketing Research, (3rd ed.). New York:McGraw-Hill.
18(1): 39-50. Nunnally JL (1978). Psychometric theory NY:McGraw-Hill
Hair JF (1998). Multivariate data analysis. New York, NY: Pallant J (2011). A step by step guide to data analysis
McGraw-Hill ISBN-10:0138948585. using SPSS program survival manual 4th edition. Allen
Hair JF, Anderson RE, Tatham RL, Black WC (1995). & Unwin 83 Alexander Street Crow‟s Nest NSW 2065
th
Multivariate data analysis with readings (4 ed.). Australia.
Englewood Cliffs, NJ: Prentice Hall. Pohlmann J (2004). Reporting Structural Equation
Hair JF, Bush RB, Ortinau DJ (2003). Marketing research Modeling on Confirmatory Factor Analysis Results: A
within a changing information environment. New York, Review. The Journal of Education.
NY: McGraw-Hill. Punch KF (1998). Introduction to social research:
Hair JF, William CB, Barry JB, Rolph EA (2006). Quantitative and qualitative approaches. London: Sage.
Assessing structural equation modeling model validity. Spearman C (1904). Rank correlation, a non-parametric
Hair JF, William CB, Barry JB, Rolph EA (2010). version of the conventional Pearson…3: 114-118.
th
Multivariate data analysis 7 Edition. Spearman C (1927). Introduction to factor analysis: Their
Hu L, Bentler PM (1995). Evaluating model fit in: nature and measurement.
Structural equation modeling concepts, issues and
applications.
Kline RB (1998). Principles and practice of structure
st
equation modeling (1 ed.). New York: The Guildford
Press.
Kline RB (2005). Principles and practice of structural
nd
equation modeling (2 ed.). New York: The Guilford
Press.