Common method variance refers to the spurious covariance being shared among measurement items when common method (i.e. questionnaire survey) is used to collect data (Buckley, Cote
& Comstock 1990). In a survey based study, the subjects respond to the questionnaire items at the same point in time, thereby the data are likely to be susceptible to CMV. It can be assessed by Harman’s single-factor test in EFA (EFA), and goodness-of-fit indices with marker-variable technique in CFA (Malhotra, Kim & Patil 2006).
The results of Harman’s single-factor test (Table 6.8) indicated that total 27 factors were extracted when all of the items were subject to EFA. The first factor accounted for 19.89% of variance explained, which did not explain the majority of the variance among variables.
Therefore, Harman’s single test findings confirmed that common method bias was not substantially exist in the data (Andersson & Bateman 1997). As an alternative to EFA, CFA can be used to test CVM as below.
118 Table 6.8: Total variance explained by Harman’s single factor EFA test (selective screen shot)
In the CFA approach, all the manifested variables were modelled as an indicator of a single factor that assumed to have method effects. The goodness-of-fit of the single factor CFA model is depicted in Figure 6.1. The result shows high chi-square value with significant p-value (i.e.
χ2= 18122.70, p = .000, CMIN/DF = 2.92), and the fit indices are below the threshold values (GFI = .29, AGFI = .27, CFI = .31, TLI = .29, NFI = .23 and RMSEA = .09), and didn’t fit the data. Therefore, common method bias is not an issue. However, EFA and CFA are subject to limitations, thus Marker variable technique was employed.
119 Figure 6.1: All-item CFA with common factor
CMV was also tested using marker-variable technique (Malhotra, Kim & Patil 2006). In this method, CFA of measurement model was conducted first with marker variable (MV), and again without marker variable. The covariance/correlation results are shown in Figures 6.2 to Figure 6.5. The covariance between measurement dimensions and marker variable under each construct is presented in Table 6.9. Note that marker variable comprises of measurement item X1 and X2 since these two items have the least correlation with other measurement items of study variables (Hair et al. 2010). For co-opetition construct, the correlation obtained between management commitment and marker variable was 0.04, between relationship management and marker variable was 0.15, and communication management and marker variable was 0.21.
Thus, the average correlation was estimated as 0.13. For freight consolidation construct, the marker variable was correlated with location dimension (0.29), geographical coverage (0.21) and utilization of transport modes (0.19), yielding average correlation of 0.23. For
120 collaborative freight distribution, the coefficients of partner selection, benefits and risks sharing, and advanced information technology were 0.14, 0.10, and 0.14, respectively, resulting in average correlation of 0.13. For sustainable distribution construct, the correlation of environmental factor, economic factor and social factor with the marker variable was 0.19, 0.18, and 0.10, respectively, which yielded the average correlation(𝑟𝑀) of 0.16 (Table 6.9).
Figure 6.2: CFA without and with marker variable of co-opetition construct
121 Figure 6.3: CFA without and with marker variable of freight consolidation construct
122 Figure 6.4: CFA without and with marker variable of collaborative freight distribution
construct
123 Figure 6.5: CFA without and with marker variable of sustainable distribution construct
124 Table 6.9: Correlation and average correlation between measurement dimension and the
marker variable
Measurement dimensions and marker variable
Correlation (r) Average
correlation (rm) Co-opetition
r(Management commitment, Marker variable) 0.04
0.13 r(Relationship management, Marker variable) 0.15
r(Communication management, Marker variable)
0.21 Freight consolidation
r(Location of freight consolidation centre, Marker variable)
0.29
0.23 r(Geographical coverage, Marker variable) 0.21
r(Utilization of transport modes, Marker variable)
0.19 Collaborative freight distribution
r(Partner selection , Marker variable) 0.14
0.13 r(Benefits and risks sharing , Marker variable) 0.10
r(Advanced Information technologies, Marker variable)
0.14 Sustainable distribution
r(Environmental factor , Marker variable) 0.10
0.16 r(Economics factor, Marker variable) 0.18
r(Social factor, Marker variable) 0.19
Now the average correlation (𝑟𝑀)and original correlation (𝑟𝑈) will be used to calculate a new correlation 𝑟𝐴 (i.e., CMV-adjusted correlation) using the following equation 6.1 proposed by Malhotra et al. (2006).
𝑟𝐴 = 𝑟1−𝑟𝑈−𝑟𝑀
𝑀
Equation 6.1: Common method variance estimation
Where, rM= average correlation between marker variable and measurement dimensions, rU = the actual correlation, rA = adjusted correlation
125 As can be seen from the results presented in Table 6.10, the adjusted correlation (𝑟𝐴) between management commitment and relationship management is reduced from 0.18 to 0.06, while 𝑟𝐴 between management commitment and communication management is reduced from 0.26 to 0.15; and between relationship management and communication management is reduced from 0.69 to 0.65. The 𝑟𝐴between location and geographical coverage is reduced from 0.84 to 0.79;
between location and utilization of transport modes is reduced from 0.77 to 0.70 and so on. The correlation between geographical coverage and utilization of transport modes dimension reduced from 0.85 to 0.81. The correlation between partner selection dimension and benefits and risks sharing dimension reduced from 0.62 to 0.57. The correlation between partner selection dimension and advanced information technology dimension reduced from 0.61 to 0.55. The correlation between benefits and risks sharing dimension and advanced information technology dimension reduced from 0.80 to 0.78. The correlation between environmental dimension and economic dimension reduced from 0.69 to 0.63. The correlation between environmental dimension and social dimension reduced from 0.64 to 0.57. Finally, the correlation between economic/organizational dimension and social dimension reduced from 0.71 to 0.66. Thus, these values indicate that the difference between the original correlation 𝑟𝑈 and the CMV-adjusted correlation 𝑟𝐴 , is relatively small (i.e., ∆r < 0.12).
Table 6.10: Changes in correlation between measurement items
Measurement items Original
Correlation
𝑟𝑈
CMV-adjusted correlation 𝑟𝐴 ∆r
Co-opetition
r( Management commitment , Relationship management )
0.18 0.06
0.12 r( Management commitment , Communication
management )
0.26 0.15
0.11 r( Relationship management , Communication
management )
0.69 0.65
0.04 Freight consolidation
r( Location of freight consolidation centre , Geographical coverage )
0.84 0.79
0.05 r( Location of freight consolidation centre , 0.77 0.70 0.07
126 The chi-square difference test was conducted to ensure that the difference between original and CMV-adjusted correlation was not substantial and the common method variance introduced small bias to the data set. Table 6.11 presents the chi-square and degrees of freedom (df) from CFA of each measurement construct, both with and without the marker variable. For co- opetition construct, introduction of the marker variable changed the chi-square value from 1749.69 to 1806.35, and increased the degrees of freedom from 249 to 293. Change of chi- square and degree of freedom values for other constructs are presented in Table 6.11. The chi- square difference test was conducted by subtracting the chi-square value without marker variable from the chi-square value with marker variable, as well as the differences in respective degrees of freedom (Bollen 1998). Results indicate that the changes in the chi-square values (∆χ2) with associated changes in the degrees of freedom (∆df) are non-significant at the 0.05 significance level (p > 0.05). In other words, the CFA with and without the marker variable of each measurement construct were not significantly different at the 0.05 level of significance.
Therefore, it can be concluded that common method variance does not introduce substantial bias to the data set (Malhotra, Kim & Patil 2006).
Table 6.11: Chi-square difference test
Chi-square difference test χ2 ∆χ2 df ∆df Chi-Square Critical Values;
p =0.05
P value
Co-opetition
CFA without marker variable 1749.69 56.66 249 44 Non-significant 0.095 Utilization of transport )
r( Geographical coverage , Utilization of transport ) 0.85 0.80 0.04 Collaborative freight distribution
r( Partner selection , Benefits and risks sharing ) 0.62 0.57 0.05 r( Partner selection , Advanced Information
technologies)
0.61 0.55
0.06 r(( Benefits and risks sharing , Advanced
Information technologies)
0.80 0.78
0.03 Sustainable distribution
r( Environmental factor , Economics factor ) 0.69 0.63 0.06
r( Environmental factor , Social factor ) 0.64 0.57 0.07
r( Economics factor , Social factor ) 0.71 0.66 0.05
127 CFA with marker variable 1806.35 293
Freight consolidation
CFA without marker variable 784.07 62.99 220 49 Non-significant 0.086 CFA with marker variable 847.06 269
Collaborative freight distribution
CFA without marker variable 1201.26 67.23 431 58 Non-significant 0.190 CFA with marker variable 1268.49 489
Sustainable distribution
CFA without marker variable 2171.19 66.12 557 66 Non-significant 0.473 CFA with marker variable 2238.69 623