| < Day Day Up > |
|
Surveys were collected over ten days with a response of 2,094 questionnaires. To ensure the respondents' reliability, one question was asked twice in a reverse direction. Of the total responses, 186 questionnaires did not show consistency between these questions, and 21 questionnaires showed the same mark for more than 20 questions in row. Both were considered as having low reliability and were removed from the sample. Thus, after deleting 207 questionnaires (about 10%), 1,887 questionnaires were used for further analysis.
The demographics were as follows. The respondents were 57.6% female and 42.4% male. The occupations of the respondents fell into the following categories: 49.3% were white-collar workers, 37.9% students and 12.8% housewives. The average age was 26.8, with a range from 11 to 60. The average period of participation per member was 4.9 months.
Factor analysis with varimax rotation was performed to ascertain the discriminant validity of each construct. Table 3 shows that among the managing strategies, rule and role as well as event and ritual could not be discriminated. Thus, it was decided to reduce the number of managing strategies to four, based upon these results, by merging rule and role into one factor, as well as event and ritual into another.
Component | ||||
---|---|---|---|---|
1 | 2 | 3 | 4 | |
Rule1 | .837 | |||
Rule2 | .708 | .244 | ||
Role2 | .571 | .201 | ||
Role1 | .523 | .239 | .354 | |
Ritual1 | .809 | |||
Event2 | .715 | .226 | ||
Ritual2 | .267 | .616 | .205 | |
Event1 | .342 | .444 | .226 | |
Purpose2 | .905 | |||
Purpose1 | .905 | |||
Subgroup1 | .873 | |||
Subgroup2 | .860 |
Table 4 shows that system and information quality could be discriminated except for system quality 4 (functionality). In prior research (Liu & Arnett, 2000; Kim & Ahn, 1998), this was used for system quality, and the question hardly could be interpreted as information quality. Thus, it was decided to follow the original construct [1].
Component | ||
---|---|---|
1 | 2 | |
Information Quality 5 | .799 | |
Information Quality 1 | .788 | |
Information Quality 4 | .756 | |
Information Quality 2 | .754 | |
Information Quality 3 | .736 | .218 |
System Quality 4 | .610 | .361 |
System Quality 1 | .784 | |
System Quality 2 | .729 | |
System Quality 3 | .298 | .696 |
System Quality 5 | .326 | .653 |
Sense of community resulted in three discriminant factors instead of four. Table 5 shows that emotional connection has no discriminant validity. Two questions of this went to integration and fulfillment of needs and one other went to membership. Prior research has found that sense of community could show the different structure of factors by the types of communities or the features of members (Chipuer & Pretty, 1999). Thus, based upon the results of factor analysis, sense of community was composed of three distinct factors such as membership, influence, and integration and fulfillment of needs. Meanwhile, visit and participation show high construct validity, as shown in Table 6.
Component | |||
---|---|---|---|
1 | 2 | 3 | |
Integration and Fulfillment of Needs 2 | .841 | ||
Integration and Fulfillment of Needs 1 | .785 | .229 | |
Integration and Fulfillment of Needs 3 | .700 | ||
Emotional Connection 3 | .564 | .316 | .317 |
Emotional Connection 1 | .511 | .424 | .366 |
Membership 1 | .863 | ||
Membership 2 | .325 | .765 | |
Emotional Connection 2 | .211 | .653 | .341 |
Membership 3 | .412 | .606 | .285 |
Influence 2 | .227 | .835 | |
Influence 3 | .250 | .814 | |
Influence 1 | .425 | .517 |
Component | ||
---|---|---|
1 | 2 | |
Participation 3 | .844 | |
Participation 4 | .811 | |
Participation 1 | .750 | |
Participation 2 | .744 | .243 |
Visit 1 | .937 | |
Visit 2 | .250 | .900 |
In order to ensure that the variables comprising each proposed research construct were internally consistent, reliability assessment was carried out using Cronbach's alpha. As Table 7 shows, all Cronbach's alphas were greater than 0.64, satisfying a minimum requirement of 0.6 (Van de van & Ferry, 1980).
Construct | # of Questions | Cronbach's Alpha |
---|---|---|
Purpose | 2 | 0.8582 |
Rule | 4 | 0.6675 |
Event | 4 | 0.6407 |
Subgroup | 2 | 0.7749 |
System Quality | 5 | 0.7492 |
Information Quality | 5 | 0.8521 |
Visit | 2 | 0.8601 |
Participation | 4 | 0.8139 |
Integration and fulfillment of needs | 5 | 0.8172 |
Membership | 4 | 0.8222 |
Influence | 3 | 0.6894 |
To avoid being dominated by a single community, we gathered data from eight different communities. However, the numbers of respondents from each community were quite diverse. The recommended sample size for LISREL is around 200; if the sample size greatly exceeds 200, the chi-square statistic can become too sensitive to its sample size difference and overall model fitness can be decreased (Hair et al., 1998). Therefore, we randomly selected 30 cases from each of the eight communities for a total of 240 cases to be analyzed by LISREL.
The LISREL test results are presented in Figure 3, showing the significant paths (p-value < .05) in solid lines. The goodness of fit was 0.98 and the normed fit index was 0.96. The root mean square residual was 0.033. These fitness indices show that the model quite well explains the relationships between variables.
Figure 3: Path Estimates of Structural Model (N=240).
Among the hypotheses regarding managing strategy, hypotheses 1 and 2 were accepted, but hypothesis 3 was rejected. The LISREL results showed that managing strategy has a positive influence on visit (g=0.18) and sense of community (g=0.39), while it has no influence on participation. The visit was positively influenced by IS quality (g=0.26). Thus, hypothesis 4 was accepted. Moreover, visit was positively related with sense of community (b=0.32), and sense of community was positively related with participation (b=0.53). Thus, hypotheses 5 and 6 were also accepted.
Because the result of the hypothesis test cannot show the detailed implication for community manager, a LISREL analysis for the detailed factors of each construct was conducted. Although overall fitness was marginally acceptable (Ratio: 11.76 = chi-square 188.24/degree of freedom 16, GFI: 0.88, RMR: 0.086, NFI: 0.78), it was deemed sufficient to proceed to show the detailed relationships.
As in Figure 4, among the three constructs of sense of community only membership (b=0.52) had an influence on participation. Membership was influenced by purpose (g=0.16), subgroups (g=0.13), and visit (b=0.20). Meanwhile, subgroups (g=0.18) and information quality (g=0.31) had an influence on visit.
Figure 4: Path Estimates of Detailed Structural Model (N=240).
To specify the result of test by the type of community, the LISREL analysis was performed twice. As in Figure 5 (nonprofit communities), membership (b=0.22) and integration and fulfillment of needs (b=0.45) had an influence on participation. Integration and fulfillment of needs was influenced by purpose (g=0.30), rule (g=0.22) and visit (b=0.36). Membership was influenced by subgroups (g=0.20) and visit (b=0.33). Meanwhile, only purpose (g=0.23) had an influence on visit.
Figure 5: Path Estimates of Non-Profit Community (N=120).
As in Figure 6 (profit-oriented communities), among the three constructs of sense of community only membership (b=0.63) had an influence on participation. Membership was influenced by visit (b=0.19). Meanwhile, event (g=-0.20) and information quality (g=0.55) had an influence on visit.
Figure 6: Path Estimates of Profit-Oriented Community (N=120).
[1]The result of using system quality 4 as information quality is the same for using that of the original construct.
| < Day Day Up > |
|