53 / 372
臺大管理論叢
第
26
卷第
2
期
53
4.1 Sample Characteristics
The entire sample comprised 54.9% women (45.1% men), indicating that Facebook
users in Taiwan are predominantly women (
t
= 18.97). Respondents ranged from age 15 to
above 50 years, with 46.2% of the respondents falling in the 21~25 age range. More than
half the respondents were students (50.3%); the majority were either undergraduates or
postgraduate degree holders (89.8%). In total, our Facebook user sample consisted of mostly
young adults who were relatively well-educated and knowledgeable. On the behavioral side,
most respondents were registered Facebook users (ranging from 1.5 to 3.5 years of
membership) at the time of study, and 20.3% spent 45~60 minutes on Facebook every day. A
high proportion of respondents (89.7%) reported that their close relatives or friends were
aware of their Facebook membership; 65.3% of the respondents used Facebook to maintain
relationships with old friends (either often or always), whereas only 14.6% used Facebook to
meet new friends. The sample profiles indicate that users visit Facebook mainly for
maintaining current relationships already established offline rather than for meeting new
friends online. Moreover, more than half the respondents (56.3%) either often or always
interact with friends on Facebook, providing evidence of high-level interactions among
Facebook users.
4.2 Evaluation of Measurement Models
Table 1 summarizes the constructs and their corresponding measurement items
(including sources of scale), item loadings, and Cronbach’s alpha values. As the table shows,
the standardized factor loadings for all items were significant, and Cronbach’s alpha values
for all constructs were greater than 0.7. Before estimating our structural model, we
performed a CFA to check the reliability, validity, and uni-dimensionality of measurements
used in this study. The results of CFA, as summarized in Table 2, show that the measurement
models fit the data well. The average variances extracted (AVEs) for all constructs were
significantly higher than the stipulated criteria (50%). These figures thus demonstrated
convergent validity of the measurement model. Moreover, all the inter-construct correlations
in Table 2 were lower than the square root of the AVE, providing evidence of discriminant
validity. Consequently, the measurement models passed both reliability and convergent/
discriminant validity checks.