Understanding consumer online shopping behaviour from the perspective of transaction costs

4.8 COMMON METHOD VARIANCE 
As the study used a self-report questionnaire for the purposes of data collection from a single 
informant, the potential for common method variance (also known as common method bias) 
may cause concern (Bagozzi and Yi 1991, Gupta and Kim 2007). Common method variance, 
as described by Fiske (1982), refers to variance that is attributable to the measurement 

---

194
method rather than to the constructs the measures represent. Such a variance may occur as a 
result of factors such as social desirability, halo effect and selective memory brought about 
by the self-reporting method, and it can threaten the internal validity of conclusions about the 
predictive relationships between measures (Campbell and Fiske 1959, Howard 1994, Spector 
1994). As suggested by Kaynak (1997), a researcher therefore should plan how to overcome 
common method variance.
As confirmed in the literature, one of the techniques for minimizing common method 
variance is to carefully design the questionnaire and survey procedures. Specifically, 
assurances were given that the data provided by respondents would be held in strict 
confidence, the analysis would be done at the aggregate level, and no respondent would be 
identified individually. All of this information was stated clearly on the project information 
sheet provided to each informant. These procedures also were aimed at reducing respondents
’
evaluation apprehension, so making them less likely to edit their responses to be more 
socially desirable (Gupta and Kim 2007). In addition, the measurement scales in the survey 
were arranged so that the measures of independent variables preceded the dependent 
variables and items on constructs which have the same scale poles (e.g., TCs, e-service 
quality, etc.) were distributed in a non-sequential order (Salancik and Pfeffer 1977). 
In the behavioural sciences, there have been a number of published techniques which assist 
with the assessment of common method variance, for example, partial correlation procedures, 
Harman
’
s one-factor test, multiple method factors test, etc. However, no test is without its 
disadvantages (Podsakoff et al., 2003).

---

195
A statistical technique widely used by scholars to determine the influence of common method 
variance is Harman
’
s one-factor (or single-factor) test (Podsakoff and Organ 1986). 
Researchers using this technique traditionally load all variables in their study into an 
exploratory factor analysis (EFA), and examine the unrotated factor solution to determine the 
number of factors that are necessary to account for variance in the variables (Organ and 
Greene 1981, Andersson and Bateman 1997, Aulakh and Gencturk 2000). It is assumed that 
if only one factor emerges from the unrotated factor solution as accounting for most of the 
variance observed in the data, it is likely that common method variance is the primary source 
(Podsakoff and Organ 1986).
As an alternative to EFA, confirmatory factor analysis (CFA) can be used when 
implementing Harman
’
s single-factor test (Podsakoff and Organ 1986). Specifically, in the 
CFA approach, all of the manifested items are modelled as the indicators of a single factor 
that represents method effects. Common method bias are assumed to be substantial if the 
single-factor model fits the data significantly better than the proposed model with many 
factors (Podsakoff and Organ 1986). In this study, Harman
’
s one-factor test was performed 
through both EFA and CFA (reported in the next chapter) in order to detect the severity of 
common method variance in the current data.