Page 125 - 臺大管理論叢第32卷第1期

NTU Management Review Vol. 32 No. 1 Apr. 2022

would need to focus on chatting with the viewers, responding to viewers’ questions, and
keeping viewers excited. A sociable streamer could focus on responding to the viewers’
questions, showing the cooking process, and stimulating the viewers to leave a message
even if the message is negative. A persuasive streamer could share food features and try to

elicit viewers’ praise. A comical streamer could encourage the viewers to leave messages
whether the message is positive or negative and continue to chat with the viewers.
Furthermore, to avoid measurement errors, this study suggests that the marketer should

precisely correct the coding of the viewers’ comments and gift-sending variables based on
the streamer behavior and gift values confirmed by the live streaming platform.

7.4 Limitations and Future Research
There are a number of avenues for future work to extend the proposed HB model

in live-streaming marketing. First, in practice, the normal distribution used in this study
may not be appropriate to some live-streaming contexts and may need to be relaxed. The
normal distribution implies a symmetrical relationship, which may be overly restrictive
in some live-streaming environments. Second, in reality, the forms of live streaming on

various Internet or mobile platforms (e.g., Facebook, YY platform, YouTube, and Twitch)
are diverse, which may produce different effects on revenue. This study suggests that
future research needs to consider the effects of the heterogeneity of different live-streaming
platforms on streamer revenues. Similarly, research on live-streaming revenue will be

affected by heterogeneity in different product types (e.g., food, sports, games). Future
research can also investigate the impact of the heterogeneity of varying product types.
Third, as the number of viewers increases or decreases over time during a live stream, if
the prediction takes place before or at the start of the live stream, the number of viewers

cannot be included as a predictor. This study recommends that future scholars who adopt
a purely predictive approach include the number of viewers appearing at time t during the
live stream to predict the streamer’s final revenue. Even more useful is the time series of
the number of viewers from the start of the live stream until time t and the revenue curve

until time t. In addition, it should be noted that the predictive models for streamer revenue
cannot provide direct causal explanations.

117