Page 118 - 臺大管理論叢第32卷第1期

Internet Celebrity Economy: Exploring the Value of Viewers’ Comment Features and Live Streamers’
Marketing Strategies in Forecasting Revenue

The coefficient estimates (i.e., the mean of the posterior distribution for each
coefficient) for the HB model are shown in Table 4. When the posterior estimates are
compared to the posterior standard deviations, some coefficients of the metrics (e.g., total
or positive comments; see bold numbers in Table 4), discrete emotions (e.g., complaining

comments), and streamers’ behaviors (e.g., chatting with viewers or sharing food features)
are “significant” at α = .05 using a one-sided test (Allenby et al., 1996). Although only
a few coefficients of the specific variable (i.e., positive comments) appear to have a

significant, consistent effect on the viewers’ gifting behavior across 10 live-streaming data,
this result shows that the heterogeneity of the streamers had a cross-level effect on these
relationships.
Table 5 shows that estimates of the hyperparameters (i.e., gender, physical
attractiveness, and streamers’ hosting styles) provides a useful summary of the

comment metrics, discrete emotions, and streamers' behaviors. In other words, streamer
heterogeneity has cross-level effect on the relationships between the viewers’ comment
metrics and discrete emotions and between streamers’ behaviors and viewers’ gift-sending

behavior. Past research (e.g., Wohn and Freeman, 2020; Yu et al., 2018; Zhou et al., 2019)
has only discussed the use of the OLS method, which may have led to biased results
because they all ignore the significant impact of streamer heterogeneity. Specifically, in
terms of gender effect, this study finds that compared to female streamers, the more male
9
streamers chat with the viewers (M = 1.245, SD = .620), the higher the viewers’ gift-
sending behavior.
Second, the streamer with higher outward beauty has more excited comments (M
= 1.221, SD = .606), and the more she/he chats with the viewers (M = 1.203, SD = .601)

and responds to questions (M = 1.178, SD = .587), the higher the viewers’ gift-sending
behavior.
Third, in terms of streamers’ hosting styles, a sociable streamer has more total (M
= 1.232, SD = .608), negative (M = 1.468, SD = .598), and complaining (M = 1.409, SD
= .597) comments, and the more he/she responds to questions (M = 1.211, SD = .603)

and shows the cooking process (M = 1.522, SD = .617), the higher the viewers’ gift-

9 M is an abbreviation for the posterior mean, and SD is an abbreviation for the posterior standard
deviation. These abbreviations are also used hereafter.

110