Measurement model and construct validity

To assess the adequacy of the measurement model, a confirmatory factor analysis (CFA) was carried out using EQS for Windows 5.7b. All constructs, except for partner criticality, were included in the CFA. Through a measurement purification process, items with their loading less than 0.5 were eliminated to increase convergent validity. Furthermore, using the Lagrange multiplier (LM) test, some items that were cross-linked to more than one constructs are dropped off the model to improve discriminant validity. The purification process left at least three items for IT adoption, firm coordination, and market performance. However, there are two items for partner coordination included in the final CFA model. The CFA model revealed fit indexes of x2 — 139.760 on 71 degrees of freedom (df), Tucker-Lewis index (TLI) — 0.950, comparative fit index (CFI) — 0.961, standardized root mean-square residual (SRMR) of 0.048, and root mean-square error of approximation (RMSEA) of 0.071 as shown in Table I. These fit indexes indicate an excellent fit of the model with the empirical covariances (Hu and Bentler, 1999; Shook et al., 2004).

With the results of CFA, we assess the unidimensionality of constructs. For convergent validity, we examined the standardized loading and its significance. For an adequate level of convergent validity, items must load on the respective constructs significantly with their loading greater than 0.5. According to the CFA results, all items load significantly on the respective factor (p < 0.01) and all loadings are greater than 0.5 as shown in Table I. Thus, the CFA results indicate an adequate level of convergent validity of each construct (Bagozzi and Yi, 1988). Second, we examined factor correlations that are expected to be significantly less than unity, and average variance extracted that should be greater than shared variances of each construct for an adequate level of discriminant validity (Fornell and Larcker, 1981). The results indicate that all correlations between constructs differed significantly from one (Bagozzi et al. , 1991; Burnkrant and Page, 1982), ranging from 0.23 to 0.71 as reported in the lower triangle of Table II. In addition, we calculated average variance extracted (AVE) using the formula suggested by Fornell and Larcker (1981). As shown in Table I, AVE ranged from 0.67 to 0.74 while shared variances among constructs are between 0.05 and 0.50 as shown in the upper triangle of Table II. These reveal a good level of discriminant validity between constructs used in the study. As a final step to assess the unidimensionality of each construct,

Table II Intercorrelations and shared variances

Partner firm coordination (F3) 0.23 0.71 - 0.12

Market performance (F4) 0.37 0.42 0.34

Note: The correlations are included in the lower triangle of the matrix. The shared variances are included in the upper triangle of the matrix

Table I Measures, reliabilities, and average variance extracted




IT adoption for SCCS

Firm coordination

Partner coordination

Market performance

Partner criticalityc

My BU uses the most advanced IT for SCCS

Our IT for SCCS is always state-of-the-art technology

My BU is very proactive in adopting or developing advanced IT for SCCS

My BU is always first to use IT for SCCS in our industry

My BU is more efficient in coordination activities with our partner than are our competitors with theirs

My BU conducts transaction follow-up activities more efficiently with our partner than do our competitors with theirs

My BU spends less time coordinating transactions with our partner than do our competitors with theirs

My BU can conduct the coordination activities at less cost than our competitors

Our partner spends less time searching for information about our products than its major competitors do for the information of their own partner's products

Our partner has reduced product-searching costs more than its competitors

My BU performs much better than competitors in sales growth

My BU performs much better than competitors in market share

My BU performs much better than competitors in market development

My BU performs much better than competitors in product development

Please circle the number that best reflects your agreement with the following statements regarding your primary partner:

Our partner is important for meeting customer requirements

Our partner is critical for our BU's long-term benefit

Our partner is important for our BU's core competency



Notes: CFA goodness of fit indices: chi-square = 139.760 on 71 df; TLI = 0.950; CFI = 0.961; SRMR = 0.048; RMSEA = 0.072. a For all measures, seven-point Likert scales were used in the questionnaire (1 = strongly disagree and 7 = strongly agree). b CR = composite reliability, AVE = average variance extracted. c Partner criticality, the moderator, was not included in the CFA, as the moderating effect was assessed with a nested model and, subsequently, reliability reported is Cronbach's alpha and loadings are item-to-total correlations we calculated composite reliabilities (Fornell and Larcker, 1981) and Table I presents them with the standardized parameters of measurement items. All composite reliabilities ranged from 0.83 to 0.92, which is far above the generally acceptable level of 0.70 (Nunnally, 1978). The good reliabilities along with adequate convergent and discriminant validities suggest the validity of constructs adopted in the study.

To investigate the moderating effect of partner criticality effectively using structural equation modeling (SEM), we carried out a two-group analysis (Bentler, 1989). Before the two-group analysis, we conducted a two-group CFA. We separated the sample into two groups (e.g. high criticality vs low criticality) using the mean of the composite index of the three scales as the cutoff value. The process assigned 111 cases to the high criticality group and 73 cases to the low criticality group. The results of the two-group CFA provide fit indexes of x2 = 242.375 on 142 df, LTI = 0.928, CFI = 0.944, SMR = 0.063, and RMSEA = 0.062. Although most fit indexes decreased from the first CFA model, these indexes still suggest a very good fit of the two-group CFA model with the empirical covariances from the two groups (Hu and Bentler, 1999; Shook et al., 2004). Subsequently, we assess the validity of constructs using the same procedure we used for the first CFA. No issues on convergent and discriminant validities, and composite reliabilities emerged. Thus, we proceeded to estimate structural models for hypothesis testing.

0 0

Post a comment