Factor Structure of SOVC

To identify the factor structures of SOVC, the samples for the two scenarios were combined and subjected to exploratory factor analysis (EFA) with maximum likelihood estimation. Promax rotation was used to consider the inter-factor correlations (McMillan &

Chavis, 1986).³ The Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy (Kaiser, 1970) is 0.936 larger than the value of 0.9. In addition, Bartlett sphericity test indicates that

2(190, N = 850) is 9458.7 (p < .001). These results show that the sample is marvelous for factoring (Kaiser & Rice, 1974).

Three factors with eigenvalues exceeding one were extracted after six iterations, while two items (sovc1 and sovc18) were deleted for having factor loadings below .40 (sovc18 was deleted after first-round EFA, and sovc1 was deleted after second-round EFA). The results of EFA are shown in Appendix C. These three factors can explain 53.76% of total variance for twenty items. All items whose factor loadings are above 0.4, and respectively loaded on single factor where nine items are loaded on the first factor, seven items are loaded on the second factor, and four items are loaded on the third factor. For correlation matrix among these three factors, all correlations are positive where correlation between factor 1 and factor 2 has the highest value of 0.74, and correlation between factor 1 and factor 2 is 0.29 and correlation between factor 2 and factor 3 is 0.30.

3 McMillan and Chavis (1986) defined sense of community (SOC) as having four key elements:

membership, influence, integration and fulfillment of needs, and shared emotional connection. Meanwhile, -reinforcing circ Blanchard and Markus (2004) extended the

Blanchard (2007) further devised the scale of SOVC based on McMillan and Chavis (1986) and their observations of online multiple sports newsgroups. Thus, the three SOVC factors are expected to correlate.

In order to confirm model fitness for factor structure of SOVC in positive eWOM scenario and negative eWOM scenario, the remaining 20 items were used to implement confirmatory factor analysis (CFA) using the factor structures obtained from EFA. Table 4-2 presented four items (sovc6, sovc7, sovc9, and sovc10) are deleted owing to their standardized factor loadings falling below .60 in both scenarios.⁴ The final CFA in Table 4-2 shows that nine items (sovc8, sovc11, sovc12, sovc14, sovc15, sovc16, sovc20, sovc21, sovc22) are loaded on first factor, three items (sovc13, sovc17, sovc19) are loaded on second factor, and four items (sovc2, sovc3, sovc4, sovc5) are loaded on the third factor.

Regarding the meanings of the three SOVC factors, according to the loaded items of each factor, the first SOVC factor is labeled the second is Compared to the model of sense of community proposed by McMillan and Chavis (1986), the meaning of Emotional Linkages resembles the concept of Shared Emotional Connection, which interprets the affective component related to community consciousness. The meaning of Anticipated Support resembles that of Reinforcement of Needs, which explains the feeling that a member hopes that other community members will solve their problems. Finally, the meaning of Membership & Influence combines the concept of Membership and Influence into a single construct, and explains the feelings of belonging to, identifying with, and mattering to a community. These factors are identical to the findings of Blanchard and Markus (2004) regarding the origins of sense of virtual community (i.e., recognition of members, exchange of

4 Tabachnica and Fidell (2007) proposed a factor loading of .55 (30% of variance) as an appropriate boundary for a scale in social science research. This study used a higher value of 0.6, i.e., 36% of variance, as the value of item-deleted decision.

support, emotional attachment, and personal relationships with members, sense of obligation, and self identity and identification with others).

Table 4-2 lists the results of standardized factor loadings of the three-factor SOVC model. All factor loadings for the two scenarios are significant (p < .001), with values ranging from .63 to .87 for emotional linkages, .71 for anticipated supports, .62 to .82 for membership & influence in the positive eWOM scenario and with values ranging from .66 to .83 for emotional linkages, .68 to .74 for anticipated supports, .60 to .88 for membership &

influence in the negative eWOM scenario.

In the subsequent structural equitation model (SEM) analysis, the scores of three SOVC factors are averaged using the standardized factor loadings of respective items as weights.

Additionally, all Cronbach

acceptable reliabilities (range from .74 to .93).⁵ Furthermore, the convergent validity can be considered adequate while the average variance extractions (AVE) exceed the suggested minimum value of .50 (Fornell and Larcker, 1981) except for anticipated supports in the negative scenario which has AVE of only .49.

5 The concept of composite reliability resembles

reliability, and is calculated based on the factor loadings and error variances of loaded indicators (Fornell &

Larcker, 1981). Bagozzi and Yi (1988) suggested .60 as the minimum value of composite reliability.

Table 4-2 Standardized Estimated Factor Loadings of Three-Factor SOVC Model

--M&I 0.33/0.35 0.34/0.26

--sovc1

AVE 0.58/0.56 0.51/0.49 0.57/0.59

CR 0.92/0.92 0.76/0.75 0.84/0.85

Note: SOVC=senseofvirtualcommunity;EL=emotionallinkages;AS=anticipatedsupport;M&I=

membership& Influence;AVE=averagevarianceextractions;CR =compositereliability. The numbers to the right of the slash are the results of the positive eWOM scenario, and those to the left of the slash are the results of the negative eWOM scenario. All factor loadings are statistically signif . Six items, i.e., sovc1, sovc6, sovc7, sovc9, sovc10, and sovc18, are excluded because their factor loadings do not satisfy the suggested minimum values in EFA (.40) and CFA (.60).

)

Var( _i represents standardized .

Model A in Table 4-3 shows the model fitness of the three-factor SOVC model, and reveals good fit in both scenarios. The three-factor SOVC model fits well in the two scenarios in that the ratios of chi square and degree of freedom are both less than 3 (2.41 for positive eWOM and 2.89 for negative eWOM), and the values of RMSEA are both smaller than 0.08 (0.058 for positive eWOM and 0.066 for negative eWOM) (McDonald & Ho, 2002).

All fit indices for GFI, AGFI, NFI, NNFI, CFI, IFI, RFI, PGFI, PNFI, and CN satisfy the suggested criteria, so the model fit is acceptable (Doll, Xia, & Torkzadeh, 1994;Jöreskog &

Sörbom, 1993;Marcoulides & Schumacker, 1996;Reisinger & Turner, 1999).

Correlations among the three SOVC factors can be employed to examine their discriminant validity. The correlations between EL and AS (r(415) = .76, p < .001), EL and M&I (r(415) = .33, p < .001), and AS and M&I (r(415) = .34, p < .001) in the positive eWOM scenario, and between EL and AS (r(431) = .53, p < .001), EL and M&I (r(431) = .35, p

< .001), and AS and M&I (r(431) = .26, p < .001) in the negative eWOM scenario are all significantly smaller than one, indicating discriminant validity.⁶ To summarize, all results suggest that the three-factor SOVC model exhibits good reliability and validity.