Sage Journals: Discover world-class research

Abstract

Livestream e-commerce has emerged as a novel way to promote and sell products. This channel differs from existing promotion channels like TV/online video advertising because viewers voluntarily consume the content on this channel and are highly engaged due to social interactivity with the livestreamer and other viewers. A key design aspect of product promotions is the duration for which a product is presented during a livestream session. In this article, we empirically study the impact of product presentation duration by analyzing a unique dataset from two of the largest livestream shopping platforms in China. We find that when the product duration is longer, product revenue is higher. However, as the average presentation time increases, the session revenue decreases. The role of presentation duration in driving sales may differ between official (single-brand) livestreams sponsored by brands and third-party (multi-brand) livestreams. On analyzing the heterogeneous effects between official and third-party livestreams, we find that the positive impact on product-level sales is largely driven by third-party livestreams. On the other hand, both official and third-party livestreams can improve session-level sales by reducing the average product duration in a session. Thus, in the context of third-party livestreams, we observe a tension between the incentives of the brands (advertisers) whose goal is to drive an individual product’s sales and third-party livestreamers whose goal is to maximize total sales in a session. Our findings have useful implications for the design of e-commerce livestreams.

Keywords

Livestream E-Commerce Platform Promotion Revenue Management Influencer Marketing

1. Introduction

Livestream shopping is a rapidly growing phenomenon that is providing a huge impetus to e-commerce sales. By 2023, livestream shopping volume was estimated to reach $50 billion (eMarketer, 2024) in the US and $695 billion in China (Statista, 2024). Several livestream platforms have emerged, for example, Taobao Live and Douyin in China, and Facebook Live and Amazon Live in the US. A livestream is essentially a virtual showroom where viewers can watch live presentations by a livestreamer promoting a series of products and purchase these products by clicking on product links. During the presentation, the viewers can interact amongst themselves or with the livestreamer and like, share, and follow the livestreamer. The combination of exhilarating presentations and social interactions entertain viewers and create high engagement levels, which assist sales (Chen and Liao, 2022; Osei-Frimpong and McLean, 2018). Thus, livestream commerce combines product promotions and e-commerce while providing entertainment and real-time social interactions. This unique approach to product promotions calls for a better understanding of how to design a livestream. Despite its promise, the effect of livestream design on sales is under-examined (Chen et al., 2019; Zhang et al., 2024). Television and in-stream video advertising are traditional methods for promoting products through video. A key design decision in these formats is the length of time an advertisement is shown, as previous research has demonstrated that ad duration significantly influences viewer engagement and brand perception (Shapiro et al., 2021; Tuchman, 2019). Similarly, in livestream shopping, the time a product is displayed, referred to as product presentation duration (or product duration for short), is an important design consideration.

Research on video advertising across different media suggests that longer ads can improve message retention but may also negatively impact brand perception, as viewers looking for entertainment often find lengthy promotions distracting. For example, Singh and Cole (1993) found that longer commercials are more effective in enhancing consumer learning and attitudes about products and brands in prime-time TV advertisements. On the other hand, studies by Jeon et al. (2019) and Aslam et al. (2021) show that shorter or skippable ads result in less viewer irritation and higher purchase intentions. Livestream shopping differs from traditional advertising in terms of viewer intent. In traditional media, consumers primarily want to engage with the main content and not the ads. In contrast, livestream shopping viewers are more receptive to promotional content because they participate knowing that product promotion is an integral part of the experience (Goh et al., 2013; Hardy, 2021). While TV and in-stream ads are often seen as intrusive (Li et al., 2002), livestream audiences are more open to marketing messages, meaning that product duration may have a different impact on consumer behavior in this context.

In livestream shopping, real-time social interactions through live chats can keep viewers engaged for longer, increasing both engagement and the likelihood of purchases, as shown by Tan et al. (2019). However, if the product presentation lasts too long, it may lead to reduced viewer engagement, similar to what happens with short-form video ads (Xiao et al., 2023). When viewers lose interest, they are less likely to purchase the products (Todri, 2022). Therefore, our first research objective is to explore the effect of product presentation duration on product sales during a livestream session.¹ The answer to this question helps the brands understand how presentation duration affects their product’s sales in a livestream session.

Prior literature (Berger et al., 2007) shows that a larger product assortment in a store can increase overall sales because consumers are variety-seeking and the inclusion of a larger variety is more likely to satisfy diverse consumer choices. In a livestream session of a given length, an increase in the average product presentation reduces product variety (i.e., the number of products presented) and vice versa. These trade-offs apply only at the session level and not at the individual product level. Therefore, our second research objective is to examine the effect of average product presentation duration on the total sales achieved in a livestream session. The answer to this question helps the livestreamers understand how the average product presentation duration affects the overall sales in a livestream session.

Importantly, we note that there are two types of livestreams-official livestreams that are run by a single brand and showcase items owned by that brand, and third-party livestreams that feature products across multiple brands. We expect that the effect of product presentation duration differs between the two types of livestreams for multiple reasons. Nierobisch et al. (2017) demonstrated that single-brand stores attract a greater proportion of brand-loyal customers. Similarly, official livestreamers may attract more brand-loyal audiences, mitigating the need to promote the brand to the audience. However, official livestreamers may have deep knowledge of the brand’s products and so the marginal benefit of product duration may be higher than in the case of third-party livestreams. Further, third-party livestreams that showcase multiple brands may encourage viewers to engage in cross-brand comparisons which affect their purchase decisions, as seen in multi-brand retail settings (Desmichel and Kocher, 2020; Rahnamaee and Berger, 2013). Therefore, third-party livestreamers may require more time for presenting additional information to enable this comparison. Furthermore, Lou et al. (2019) found that identical ads posted by influencers generated more engagement than those posted by the brand itself, especially for apparel brands. This heterogeneity of response indicates that the effectiveness of information presented may differ across livestreamers, potentially affecting the time they spend on product promotions. Accordingly, our final research objective is to analyze the heterogeneous effects of product presentation duration on sales between official and third-party livestreams at both the product and the session level. The answers to our research questions would help third-party livestreamers and brands design effective livestream shopping sessions and also highlight whether they prefer different designs due to differences in their incentives.

We collect a comprehensive dataset on about 75,000 product presentations from 2,304 livestream shopping sessions conducted on two major platforms during the last quarter of 2021 in China. Using a combination of several high-dimensional fixed effects (FEs) and an instrumental variable (IV) analysis of product-session level data, we find that when product presentation duration increases, product revenue also increases on average. Our heterogeneous effects analyses reveal that the positive relationship between product duration and product sales is entirely driven by third-party livestream sessions. For official livestreams, a longer product presentation is not significantly associated with product revenue. However, at the session level, when the average product presentation duration is longer, the session revenue for both official and third-party livestreams decreases. Notably, our product- and session-level findings for third-party livestreams together highlight a key misalignment between the incentives of a third-party livestreamer and brand owners selling their products through third-party livestreamers. Whereas the former would prefer shorter product presentations, the latter would prefer a longer presentation of their product. For official livestreamers, our findings suggest that while a longer presentation does not seem to have a significant impact on product-level sales, they might prefer to decrease the average product presentation duration to improve session-level sales.

The rest of the article is organized as follows. In Section 2, we present related literature and clarify our positioning. In Section 3, we discuss the context and the data. In Section 4, we present our empirical strategy, followed by Section 5, where we discuss our findings. We conduct several robustness checks and additional analyses in Section 6. Finally, we discuss the implications, and limitations of our results and conclude in Section 7.

2. Related Literature

Livestream is an emerging shopping channel that integrates product advertising with real-time e-commerce and social interactions (Cai et al., 2018; Hamilton et al., 2014). Because of the newness of the channel, existing literature studying livestreams is relatively scant. The design of a livestream is, however, related to the literature on influencer marketing, advertising duration on TV and online videos, and shelf-space planning in traditional stores. Here, we present the relevant literature from all these streams and highlight our key contributions.

2.1. Influencer Marketing and Livestream Shopping

Influencer marketing is an approach to social marketing where brands employ the services of influencers to drive sales (Li et al., 2024). These influencers have a dedicated following on social platforms like Instagram, YouTube, TikTok, etc., because they post content that their viewers like. Brands affiliate with these influencers so that brand messages can be embedded into their content posts.

Existing research has examined several aspects of influencer marketing like the extent of affiliation a brand should seek with influencers (Pei and Mayzlin, 2022), how to select influencers and schedule their posts (Han et al., 2023; Mallipeddi et al., 2022) and how influencers’ characteristics and the nature of content affect consumer engagement (Hughes et al., 2019). Due to the embedded nature of the brand content in posts, consumers are likely to miss the true nature of a post and may be deceived. This issue is especially problematic for content targeted at young children. Van Reijmersdal et al. (2020) studied whether disclosure of the sponsored nature of the content should be made prior to the start of content videos, or concurrently with the start of the videos to ensure that the true nature of the content is recognized by consumers.

Livestream shopping is a specific approach to influencer marketing where the influencer engages in a live presentation of content (Pan et al., 2022). Existing literature recognizes that social presence and synchronicity inherent in a livestream induces consumers to be more engaged than with pre-recorded content (Ang et al., 2018). Further, promotional content (posts and videos) is typically embedded in influencer content. In contrast, in livestream shopping, the entire content is promotional in nature. Due to these reasons, the design of effective livestream sessions gives rise to new types of questions that are not pertinent in the context of standard influencer marketing. We contribute to the broad literature on influencer marketing by focusing on its real-time variety.

Some literature that specifically studies livestream shopping does exist. One of the first research objectives of researchers studying livestream shopping was to ascertain whether this channel adds value by providing additional sales. Chen et al. (2019) showed that sellers on Taobao who host livestreams on Weibo to promote their products generate more sales than the sellers who do not. Possible reasons why livestream shopping increases sales are its ability to acquire new customers and retain existing ones (Wongkitrungrueng and Assarut, 2020). Some recent work has explored the features of the livestream that increase consumer engagement. Thus, Park and Lin (2020) showed that video content attractiveness and livestreamer trustworthiness improve engagement, and Sun et al. (2019) find that interactivity between the livestreamer and viewers influences viewers’ engagement. Zhang et al. (2024) studied the nuanced short- and long-term impacts of the adoption of a lucky draw during a livestream session results on sales improvement.

We contribute to this nascent literature on livestream shopping research by studying the pace of product presentation as a key design feature in livestreams. We show that appropriately choosing the pace helps in improving viewer engagement and sales. We differentiate between the official and third-party livestreams and recognize that the appropriate pace of product presentation may differ across the two types of livestreams. Accordingly, we analyze the potential heterogeneous effects between official and third-party livestreams.

2.2. Advertising Duration on TV and Online Videos

Similar to TV and video advertising, consumers are exposed to products during a livestream session, and the purpose of product exposure in all of these formats is to drive sales. However, the mechanism responsible for driving sales across these formats is very different. Consumers exposed to a product on TV/online video advertising may not buy it right away. Therefore, a major objective of advertising in these channels is to drive long-term recall, leading to studies that explore the connection between ad duration and recall (Clarke, 1976; Goldstein et al., 2015; Peters and Bijmolt, 1997). However, the objective of product exposure during a livestream session is different because managers are interested in driving immediate sales. Due to the difference in objectives across these formats, the impact of product presentation duration on consumer response may differ.

Second, TV advertising is a distraction for consumers who are exposed to advertising while watching TV shows. Thus, overexposure may irritate them and make the ads counterproductive (Goldstein et al., 2014; Todri et al., 2020; Wilbur, 2008). The interests of consumers watching livestreams are aligned with exposure to multiple products because they explicitly watch the livestreams for that purpose. Due to these reasons, the effect of presentation duration in TV/video ads may be significantly different from that in the livestream format.

Third, an individual TV/video ad is relatively short (usually, lasting from 5 seconds to 60 seconds) in practice. Several of such ads (typically 5–7) are presented during a commercial break on TV which lasts 3–4 minutes. It is well-known that more competition (i.e., more number of ads) in a commercial break reduces ad effectiveness (Peters and Bijmolt, 1997). Hence, there is a tradeoff between the duration of ads and the number of ads in a break. In contrast, a livestream session lasts for several hours and allows the presenter to introduce a relatively large number of products to the consumers. Consequently, the extent of competition between the products is very different and the format allows for product promotions that can be several minutes long unlike TV or video ads. Therefore, it is likely that the effect of ad duration in livestream sessions may substantially differ from TV/video ads.

These differences between the presentation duration in livestreams and the standard advertising channels call for an in-depth examination of product presentation duration in livestreams. Further, heterogeneous effects based on the nature of livestreams would also be useful to understand. Our article is one of the first ones that do such an analysis and contributes broadly to the literature on the design of product promotion strategy in the context of livestreams.

Last, we briefly compare the livestream shopping context to teleshopping² where there has been a serious lack of quantitative empirical work (Alcañiz et al., 2006; Blas et al., 2006). Like in livestream shopping, a host presents products on the television home shopping channel, and viewers can call or use a link shown on the television screen to make a purchase. Despite some similarities, there are important differences between the two contexts. Typically, a presenter in the home shopping channel promotes a given product for a very long time (several hours), unlike in a livestream setting where the product presentation duration is much shorter. This difference makes the product presentation duration a salient design feature for livestreams while it is not an important decision in the context of home shopping TV channels. Further, the scale and variety of real-time social interactions possible in the livestream shopping channel are quite limited in TV channels. Consequently, it is much harder to engage consumers on home shopping channels. A recent article (Pymnts, 2016) confirms that the home shopping channels are watched mainly by women in the age group of 35–64 years. Thus, livestream channel caters to a very different demographic (much younger) compared to teleshopping channels and hence the design features of this channel are also likely to be unique.

2.3. Shelf-Space Planning

We also relate our research to the shelf-space planning literature. Brick-and-mortar stores do traditional shelf space planning to utilize the limited physical space available to them to maximize their revenue. Increasing the product assortment in a limited space increases the risk of inventory stock-outs, requiring frequent replenishment and lost sales (Hübner et al., 2013). On the other hand, a smaller product assortment is likely to miss the demand from the excluded products, thus reducing sales (Akçay and Tan, 2008). The shop’s revenue conditioned on a given assortment size can be increased by limiting the amount of inventory on the shelves and keeping the additional inventory in the backroom (Hübner et al., 2020), or allowing each product to be displayed on multiple shelves (Geismar et al., 2015). To summarize, the theme in the literature on store shelf space planning is to study the judicious utilization of space in stores.

In contrast to the physical space constraint, livestreamers must deal with the constraint of limited time availability. They decide the duration of the session and how much time to spend promoting each product with the aim of maximizing sales under the limited duration of the livestream session. In choosing a product presentation duration, livestreamers must consider several trade-offs. Choosing a larger number of products (implying a shorter product presentation duration) has a higher chance of meeting the diverse preferences of consumers, thus driving sales (Besbes and Sauré, 2016; Draganska et al., 2009). However, there is a potential negative impact of a relatively large product assortment, often referred to as the “over-choice effect,” which contends that the higher cognitive cost on consumers in deciding what to buy from a wider choice set may lead to a paralysis in decision making and reduce sales. This effect has been documented both theoretically and empirically (Dhar, 1997; Tversky and Shafir, 1992).

Another pertinent factor is that the time spent on presenting a product may influence viewer engagement and consequently affect sales (Kumar et al., 2020). A longer presentation duration can be disengaging (Brechman et al., 2016) and reduce the viewers’ purchasing intention, while a shorter presentation duration may not be compelling to the viewers to make a purchase. Therefore, the selection of a suitable presentation duration, which is part of the session design of livestream sessions, is crucial to maximizing the livestreamers’ revenue. To the best of our knowledge, our work is the first one that addresses the question of how livestreamers could plan the pace of display of their product assortment to maximize sales. Therefore, our main interested variables are both product sales and session sales.

3. Data

3.1. Empirical Setting

Livestream shopping involves different stakeholders: livestreamers, consumers/viewers, product brand owners, and the platform (see Figure 1). Livestreamers are central figures in livestream shopping, serving as hosts and shaping the direction of sessions. They can host multiple sessions at different times through their own channel. Livestream sessions conducted by official livestreamers exclusively promote products from a single brand, while those hosted by third-party livestreamers³ promote products from multiple brands. Brand owners can opt for dedicated official livestreamers, third-party options, or both. For instance, in Figure 1, brand owner 1 exclusively presents through its official livestreamer (Livestreamer 1), while brand owner 2 utilizes both official and third-party livestreamers (Livestreamers 2 and 3). Conversely, brand owners 3 and 4 solely rely on third-party livestreamers (Livestreamers 3 and 4). The platform facilitates livestreamer interactions and viewer purchases, playing a crucial role in facilitating transactions.

Figure 1.

Relationship between stakeholders.

The flow of a certain livestream session is illustrated in Figure 2. Before starting a livestream session, livestreamers must choose the total session duration, accordingly, the products to feature and their presentation durations, and tailor their livestream style to match both their approach and the nature of the products. After each session, livestreamer performance is evaluated using metrics such as total sales, average watch time, and conversion rate. Between sessions, livestreamers maintain their fan groups by addressing logistics, answering questions, and posting promotional updates. Therefore, the decisions made by livestreamers play a crucial role for all stakeholders on the livestream shopping platform.

Figure 2.

Flowchart of a livestream session.

Consumers, on the other hand, have the flexibility to join livestream sessions at various points in time, leading to diverse engagement scenarios. In Figure 2, we demonstrate distinct patterns among them. Consumer 1 enters the session before the presentation of product $n$ starts and exits during the presentation of product $n$ . In contrast, Consumer 2 has a shorter engagement, entering and leaving during the presentation of product $n$ . Lastly, Consumer 3 exhibits a more extended stay, joining the session before the presentation of product $n$ begins and departing after its presentation is over. This variability in viewer entry and exit points adds a dynamic dimension to the livestreaming experience.

A typical user interface that users see when they join a session is shown in Figure 3. The left panel of the figure demonstrates the user interface (UI) of a typical livestreaming session hosted by a livestreamer. While the livestreamer introduces the product, consumers can interact with the livestreamer by liking, sharing, subscribing, and asking questions or commenting through the comment box. Besides the current presenting products, consumers can also use the item pocket to view the product list, which lists all the products being presented in the section (as shown in the right panel of the figure).⁴ The associated links are automatically deactivated when a product sells out during a session. Despite this, livestreamers are obligated to continue presenting until the pre-determined time expires. Moreover, it falls upon the livestreamer to manage the presentation tempo in sync with the pre-determined time slots for each product. This ensures a seamless transition between products with minimal time gaps, optimizing the overall efficiency of the livestream session. Due to the real-time nature of livestreaming sessions, consumers lack the option to playback while a session is active.⁵

Figure 3.

Livestreaming shopping interface.

Several popular platforms across different countries provide livestream shopping services, such as Taobao Live, Facebook Live, and TikTok. In this study, we explore data from two platforms in China. One primarily oriented towards e-commerce and the other centered around entertaining short videos. Figure 4 illustrates the UIs of one identical official channel on both platforms. While esthetic differences, such as varying color choices and icon placements, may exist, the core functionalities are the same.

Figure 4.

User interfaces on the two platforms.

3.2. Sampling

We obtain our data from two major livestream shopping platforms in China.⁶ The business model of one of these platforms is short-video streaming (we call it Platform A), while the other is an e-commerce platform (we call it Platform B). These two platforms are the two most popular platforms with millions of daily active users. Together, they account for more than 60% of the livestreaming market by gross merchandise value sold in 2021. The data contains details of 74,997 product presentations during 2,304 livestreaming sessions by 163 livestreamers across the two platforms from September 1 to December 31, 2021.⁷ We acquire data at both the product and session levels. For each livestreaming session, we have information about the date and time of the day a session starts⁸ , the livestreamer, the platform where the session was conducted, the average price of all in the session, and the livestreaming design features, such as the number of products in that session and the total session duration. For each product-session, we collect the following information: unit sales and revenue in Chinese Yuan (CNY), the product price, the date of product presentation, the time spent presenting each product (product duration), and the duration for which a product was available for sale from the time product presentation begins (on-shelf duration). If a product is sold out before the presentation ends, the on-shelf duration is shorter than the product duration. In contrast, the on-shelf duration is longer than the product duration if the livestreamer keeps the purchase link active even after the presentation for that product is over. The definitions of all the variables can be found in Table 1.

Table 1.
Variable definitions.

Variable Definition

Panel A: Product-session level variables

Product revenue Revenue in CNY of a product in a session

Product duration Time livestreamers spent on promoting a product in a session (in minutes)

On-shelf duration Duration of a product purchase link being active

Product price Product price after discount (if any)

Product date Date on which the product was livestreamed

Product time of day Time of the day at which a product livestream started

Panel B: Session level variables

Session revenue Total revenue of products presented in a session

Session unit sales Total unit sales of products presented in a session

Session duration Duration of the livestreaming session (in minutes)

Number of products Number of products presented in a session

Session date Date on which a session was livestreamed

Session time of day Time of the day at which a session started

Avg product duration Session duration divided by the number of products presented

Informative style Binary variable indicating if a session’s presentation style is labeled as informative

Average price Average price of the products presented in a session

Platform Binary variable indicating if a session is on Platform A

Variable	Definition
Panel A: Product-session level variables
Product revenue	Revenue in CNY of a product in a session
Product duration	Time livestreamers spent on promoting a product in a session (in minutes)
On-shelf duration	Duration of a product purchase link being active
Product price	Product price after discount (if any)
Product date	Date on which the product was livestreamed
Product time of day	Time of the day at which a product livestream started
Panel B: Session level variables
Session revenue	Total revenue of products presented in a session
Session unit sales	Total unit sales of products presented in a session
Session duration	Duration of the livestreaming session (in minutes)
Number of products	Number of products presented in a session
Session date	Date on which a session was livestreamed
Session time of day	Time of the day at which a session started
Avg product duration	Session duration divided by the number of products presented
Informative style	Binary variable indicating if a session’s presentation style is labeled as informative
Average price	Average price of the products presented in a session
Platform	Binary variable indicating if a session is on Platform A

3.3. Variables and Descriptive Statistics

Table 2 presents the descriptive statistics of the variables of interest. We show distributions for product-session level data of the full sample (74,997). There are 27,448 product-session observations for official livestreams while 47,549 product-session observations for third-party livestreams. The products belong to 22 categories and 103 subcategories. A product category is a broad classification, such as clothing, cosmetics, and food and beverages, while a product subcategory sorts the products into a more detailed class, such as women’s clothing, women’s accessories, perishable food, seafood, etc. Each session belongs to a single category, whereas each product belongs to a single subcategory. We then show distributions for session-level data of the full sample (2,304), the subset of official livestreaming sessions(1,203), and third-party livestream sessions (1,101).

Table 2.
Summary statistics.

Full sample Official Third-party

Mean SD Mean SD Mean SD

Product-session level ( $N = 74,997$ )

Product revenue 125,063 732,654 117,789 782,809 137,966 633,817

Product unit sales 1,009 3,914 810 3,419 1,353 4,628

Product duration (in minutes) 16.4 7.3 18.5 7.9 15.8 6.5

Product on-shelf duration 26.0 43.2 41.1 42.5 17.2 36.5

Product price 996 1,221 1,130 1,457 964 1,060

Session level ( $N = 2,304$ )

Session revenue 4,070,898 1,526,137 2,691,425 990,182 5,958,352 1,658,436

Session unit sales 32,844 178,456 18,481 148,985 58,432 97,852

Session duration (in minutes) 533 296 442 287 582 489

Number of products 32.6 29.8 22.8 24.1 36.8 22.6

Avg product duration (in minutes) 16.3 7.8 19.4 12.9 15.8 9.8

Informative style 0.41 0.49 0.60 0.49 0.23 0.42

Average price 1,020 4,581 1,190 6,589 869 992

Platform A 0.53 0.50 0.48 0.496 0.60 0.49

Official livestreamer 0.52 0.31

	Full sample	Official	Third-party
Product-session level ( $N = 74,997$ )
Product revenue	125,063	732,654	117,789	782,809	137,966	633,817
Product unit sales	1,009	3,914	810	3,419	1,353	4,628
Product duration (in minutes)	16.4	7.3	18.5	7.9	15.8	6.5
Product on-shelf duration	26.0	43.2	41.1	42.5	17.2	36.5
Product price	996	1,221	1,130	1,457	964	1,060
Session level ( $N = 2,304$ )
Session revenue	4,070,898	1,526,137	2,691,425	990,182	5,958,352	1,658,436
Session unit sales	32,844	178,456	18,481	148,985	58,432	97,852
Session duration (in minutes)	533	296	442	287	582	489
Number of products	32.6	29.8	22.8	24.1	36.8	22.6
Avg product duration (in minutes)	16.3	7.8	19.4	12.9	15.8	9.8
Informative style	0.41	0.49	0.60	0.49	0.23	0.42
Average price	1,020	4,581	1,190	6,589	869	992
Platform A	0.53	0.50	0.48	0.496	0.60	0.49
Official livestreamer	0.52	0.31

3.3.1. Product-Session Level Data

The duration for which the products are presented during a livestream session ( $P r o d u c t d u r a t i o n$ ) is the main variable for our research question at the product level. At the product-session level, the average revenue is 125,063 CNY and the main unit sale is 1,009. The mean product duration is 16.4 minutes, while the mean on-shelf duration is 26.0 minutes. Compared to third-party livestreamers, official livestreamers spend a longer time presenting each product but generate fewer sales on average.

3.3.2. Session-Level Data

At the livestream session level for the full sample, the average revenue is 4,070,898 CNY, and the average unit sales is 32,844. The mean session duration is about 533 minutes. At the session level, our main independent variable is $A v g p r o d u c t d u r a t i o n$ , which equals the total duration of the session divided by the total number of products presented during the session. If the number of products per unit of time is larger, the average time spent on each product is lower. The mean of $A v g p r o d u c t d u r a t i o n$ is 16.3 minutes in the full sample. This figure is slightly lower for the third-party livestreamers when compared to the official livestreamers. The average spending per purchaser in each session is 563 CNY. We have roughly balanced panels between the two platforms, with 53% from Platform A. Compared to third-party livestreamers, official livestreamers generate lower sales per session, spend longer time presenting each product, and present fewer products in a session.

4. Empirical Strategy

In this section, we discuss our empirical strategies to estimate the product- and session-level effects of product presentation duration.

4.1. Product Level

Our primary research question examines the relationship between product presentation duration and livestreaming revenue. At the product-session level (hereinafter product level), we examine whether a product’s presentation duration impacts that particular product’s revenue in the session. Therefore, for the product level analysis, our main independent variable is given by the duration for which product $j$ is presented during session $i$ ( $P r o d u c t d u r a t i o n_{i j}$ ). Our main dependent variable is the revenue of product $j$ during session $i$ for product-level analysis.

We can estimate this effect by regressing the dependent variable on $P r o d u c t d u r a t i o n_{i j}$ by exploiting the variation across different product sessions. However, this estimate is likely biased because of multiple selection issues.

First, we need to account for the unobserved heterogeneity across different livestreamers. For example, a livestreamer who promotes products at a greater pace could also have a higher ability to positively influence the viewers, generating higher sales. Second, the estimate may be driven by selection across product types. For instance, subcategories that generate higher average product sales may also need shorter presentation time. Third, we also need to ensure that the effect is not marred by selection issues related to a session’s timing. It is plausible that sessions on certain days or during certain times of the day not only generate higher sales but also have more products presented per unit of time.

We leverage the large scale and the high granularity of our data to tackle the above challenges. We use linear models with many levels of FEs since they can accommodate various factors that potentially bias our estimates. As we observe multiple sessions of a livestreamer, we include livestreamer FEs to account for unobserved heterogeneity in ability across various livestreamers. To solve the category-level selection issue, we include subcategory FEs. Finally, we include date-time FEs, that is, the interaction between date and time of the day, Date $\times$ Time of Day, that account for any unobserved heterogeneity across different days or times of any given day.

Specifically, we estimate the following regression model for our product-level analysis:

\begin{aligned} Y_{i j} & = β_{0} + β_{1} P r o d u c t d u r a t i o n_{i j} + X_{i j} γ + S_{i} θ + L S e r_{i} δ \\ + P r o d u c t D a t e T i m e_{i j} ζ + S u b c a t_{j} η + ϵ_{i j}, \end{aligned}

(1)

where

Y_{i j}

is the dependent variable, denoting the logarithm of the revenue of product

j

in session

i

L S e r_{i}

denotes a vector of livestreamer dummies;

P r o d u c t D a t e T i m e_{i j}

denotes a vector of date-time dummies;

S u b c a t_{j}

represents a vector of subcategory dummies; and

ϵ_{i j}

denotes the error term. Note that the livestreamer, date-time, and subcategory dummies essentially act as FEs.

X_{i j}

denotes a vector of product-session level covariates, such as the on-shelf duration and the price of the product; and

S_{i}

denotes a vector of session-level covariates, such as the livestreamer type, the style of the presentation, the platform on which the session took place, the duration of the session, the number of products presented in the session, and the average price of all products presented in the session.

We include on-shelf duration as a covariate because this is the time during which viewers can buy the product, making it potentially correlated with both product presentation duration and product sales. Product price is included as a covariate because we are interested in estimating the impact on revenue independent of the price. We control for the style (informative or persuasive) of the presentation as it is likely to influence the time the presenter spends on the presentation and also the sales of the product. As the mindset of viewers of the two platforms is potentially different, the time spent on product presentation could vary with the platform. Further, because of the inherent popularity difference between the platforms, product sales could also be affected by the platform type. Therefore, we control for the platform. The rationale behind considering session duration is that the effectiveness of a certain product presentation can significantly differ based on the overall duration of the livestream sessions. To account for further differences across sessions, we include the number of products presented and the average price of the products presented as covariates.

4.2. Session Level

At the session level, we examine whether the average product duration influences session revenue. For session-level analysis, our main independent variable is given by the total duration of the session divided by the number of products presented during the session ( $A v g p r o d u c t d u r a t i o n_{i}$ ). Our main dependent variable is the total revenue of all the presented products in session $i$ for the session-level analysis. We can estimate these effects by regressing the dependent variable on $A v g p r o d u c t d u r a t i o n_{i}$ by exploiting their variation across different livestream sessions. Specifically, we model the relationship between our key explanatory variable and session-level revenue in the following way:

\begin{aligned} Y_{i} & = β_{0} + β_{1} A v g p r o d u c t d u r a t i o n_{i} + S_{i} θ + L S e r_{i} δ \\ + S e s s i o n D a t e T i m e_{i} ζ + C a t_{i} η + ϵ_{i}, \end{aligned}

(2)

where

Y_{i}

is the dependent variable, denoting the logarithm of the total revenue during session

i

;

C a t_{i}

represents a vector of category dummies;

S_{i}

denotes the same vector of session-level covariates as before except the number of products; and

ϵ_{i}

denotes the error term. Note that

A v g p r o d u c t d u r a t i o n

is the ratio of session duration to the number of products. We include session duration as a covariate because it is possible that a longer session can have longer product presentations (i.e., higher average product duration) and is also likely to be associated with higher sales. As we control for session duration in our regression specification, an increase in

A v g p r o d u c t d u r a t i o n

can be interpreted as a decrease in the number of products presented, given the session duration. In Section 7, we also estimate the alternate specification where, instead of session duration, we include the number of products as a covariate. In such a case, an increase in

A v g p r o d u c t d u r a t i o n

can be interpreted as an increase in session duration, given the number of products presented in the session.

4.3. Instrumental Variables

Finally, to resolve any remaining endogeneity issues, we resort to IV analysis for both the product-session-level and the session-level analyses. First, we discuss our IV for the product-session level analysis, followed by a discussion of the IV that we use for the session-level analysis.

4.3.1. Product-Level IV

At the product level, for any focal product $j$ in session $i$ , we use the average product duration of products in session $i$ from subcategories other than that of product $j$ (i.e., non-focal subcategories) as an instrument for $P r o d u c t d u r a t i o n_{i j}$ . The IV satisfies the relevance assumption because the average duration spent on presenting products of other subcategories is highly correlated with the duration spent presenting the focal product, as a longer duration for the former leaves less time for the latter.

The exclusion restriction of the instrument requires that the average product duration of non-focal subcategories does not affect the sales of the focal product, except through its influence on the product duration of the focal product. The idea behind choosing the average product duration of non-focal subcategories as the IV is that the demand for the focal product is unlikely to be directly influenced by the pace at which products in the other subcategories are presented. The only way it should influence the demand for the focal product is through its effect on the time it leaves for the focal product presentation.

However, livestreamers may choose to present complementary products from different subcategories in a session. If this is the case, the time spent on products of other subcategories may influence the sales of the focal product. To mitigate this concern, we re-construct our IV based on two relevant facts in our data. First, the mean product presentation duration is about 16 minutes. Second, on average, viewers watch a livestream session for 13.6 minutes with a standard deviation of 9.3 minutes ( $σ$ ). This implies that a user is highly unlikely to watch a session for longer than 32.2 minutes (13.6+2 $σ$ ). In our sample, the total duration of any four consecutive product presentations exceeds 32.2 minutes 99% of times. Putting these facts together, if we exclude four product presentations before and after a focal product to construct the IV, we would ensure that very few users who watch the products included in the IV have also watched the focal product presentation. Therefore, most users would be unaffected by the complementarity between products. Such an IV is likely to be independent of the sales of the focal product. Hence, to calculate our IV, we exclude four products presented immediately before and after the focal product presentation (a total of eight products were excluded per focal product presentation). Further, note that, conditional on the covariates, the date-time, livestreamer, and product subcategory fixed effects take into account many of the potential session- and product-level confounders, ensuring the exogeneity of the instrument. We find that this IV is strongly correlated with the key explanatory variable, as shown in the first row of Table 1 in E-Companion Section 1, which confirms the relevance assumption.

To further mitigate any concerns with respect to the validity of the exclusion restriction assumption, in E-Companion Section 2, we perform inference using the plausibly exogenous IV framework (Conley et al., 2012). The results from this analysis indicate that our results are robust even when the exclusion restriction of the IV may not hold.

The monotonicity of the product-level instrument will be violated if there exists a subpopulation of product-session observations where a longer average product duration of non-focal subcategories would increase the focal product duration, and vice versa (these are called defiers). The defiers may exist, for example, if the average product duration of non-focal subcategories is positively correlated with the product duration of some focal-subcategory products. This situation is more likely in relatively long livestreams with few product presentations (i.e., sessions with very high average product presentation duration), as such livestreams may allow for allocating more time to products from both non-focal and focal product subcategories. Conversely, such a situation is also likely in relatively short livestreams with many product presentations (i.e., sessions with very low average product presentation duration). To check whether such potential defiers are driving our results, we repeat our analysis after splitting sessions with long, medium, and short average product durations. In Table 4 of the E-Companion, we observe that the results for livestreams with relatively high and low average product durations are qualitatively similar to the medium ones, indicating that the presence of defiers is unlikely.

Further, if at all there exist defier products, these are likely to be the most important products in the focal subcategory, because, controlling for session duration, an increase in the average duration of non-focal product subcategories leaves less time for the focal subcategory. In such a situation, some less important products from the focal subcategory may be dropped, leaving more time to present such a focal product. To address this possibility, we take the product presented for the longest duration in a subcategory as the most important product in that subcategory. Based on our argument, such a product is most likely to be a defier. Next, we regress the average duration of top products (i.e., products presented for the longest) in the focal subcategory on our IV which is the average time spent presenting products of non-focal subcategories, excluding six adjacent products of the focal product, while including session and subcategory FE. The results reported in Table 5 of the E-Companion clearly suggest a strong negative relationship not only between the average product duration of the focal subcategory products and the IV but also between the longest product duration in the focal subcategory and the IV, indicating that the presence of defiers due to the aforesaid reason is unlikely.

4.3.2. Session-Level IV

Now, we turn to discuss the IV we use at the session level. For the impact on session-level revenue, our main treatment variable is $A v g p r o d u c t d u r a t i o n_{i}$ , which is the average time spent presenting products in session $i$ . We instrument our treatment variable with the mean of the average product duration of all the past sessions belonging to the same category as that of session $i$ . First, this IV is strongly correlated with the focal explanatory variable (see the second row in Table 1 of E-Companion Section 1 for the correlation between the key explanatory variable and IV results)—relevance assumption. Second, the average duration for which the products are promoted in a category’s past livestreaming session is unlikely to affect the sales of the focal session directly. However, there is a possibility that the average sales of past sessions could be correlated with the sales of the focal session, as they belong to the same category. To account for such a possibility, we combine two tests later in Section 7.2, where we control for category-level time trends and past session sales from the same category. Further, in E-Companion Section 2, we present the effect estimates using the plausibly exogenous IV framework to relax the instrument exclusion restriction for our session-level IV (Conley et al., 2012).

The monotonicity of the instrument will be violated if there exists a subpopulation of sessions where a longer average product duration of past sessions of the same category would decrease the average product duration of the current session and vice versa. Such defiers could exist when a livestreamer chooses not to follow the category’s average product duration trend and experiments by instead hosting a session with a very long or a short average product duration. Therefore, sessions with a very long or a short average product duration are likely to be defiers. This would imply the relationship between the IV and the treatment variable could be negative (opposite of expected) for sessions with a very short or long average product duration. To check if this is indeed the case, we repeat our analyses after splitting our full sample into three subsamples: sessions in the lowest, highest, and middle quartiles of average product duration (see Table 6) in the E-Companion. The first-stage results in all three cases indicate a strong positive correlation between our session-level IV and the average product duration, mitigating the concern related to the lack of monotonicity.

5. Results

In this section, we present the results from estimating the empirical models described above. We first present the results of the effect on product revenue in a livestreaming session. Next, we discuss the results of the session-level outcomes. We also investigate any heterogeneous effects between official and third-party livestreams along with the two levels of analysis.

5.1. Impact of Product Duration on Product Revenue

We first focus on the impact of the product duration on product revenue during a livestreaming session. Table A.1 in the Appendix presents the regression results (without IV) for the full sample, showing the impact of product presentation duration on product revenue using different models. The first column does not include any FEs but includes all the controls. In the subsequent columns, we add livestreamer, product subcategory, and date-time FEs sequentially. The positive and statistically significant coefficient for $P r o d u c t d u r a t i o n$ in all specifications confirms when the product duration is longer, product revenue is higher.

Our main specification addresses endogeneity concerns by employing instrumental variables, as shown in column (1) of Table 3.⁹ The coefficient for $P r o d u c t d u r a t i o n$ remains positive and statistically significant, suggesting that when the presentation is longer, product revenue increases, reinforcing the robustness of our previous findings. We now explore the heterogeneous effects between official and third-party livestreams through the interaction term, Product duration $\times$ Official. The results in column (2) show that when product duration increases, product revenue in third-party livestreams increases. However, for official livestreams, the negative and statistically significant interaction term indicates that longer product presentations do not yield the same benefit. This finding underscores the critical difference in how product duration relates to sales across livestream types, marking it as a key contribution of our study.

Table 3.
Impact of product duration on product revenue.

(1) (2)

IV-2SLS Log(Product revenue) Log(Product revenue)

Product duration 0.1725 $^{* * }$ 0.1532 $^{ * }$

(0.0472) (0.0243)

Product duration $\times$ Official $- {0.1783}^{ * *}$

(0.0441)

Product- and session-level controls Yes Yes

Livestreamer FE Yes Yes

Product subcategory FE Yes Yes

Date $\times$ Time of day FE Yes Yes

Observations 57,528 57,528

	(1)	(2)
Product duration	0.1725 $^{* * *}$	0.1532 $^{* * *}$
	(0.0472)	(0.0243)
Product duration $\times$ Official		$- {0.1783}^{* * *}$
		(0.0441)
Product- and session-level controls	Yes	Yes
Livestreamer FE	Yes	Yes
Product subcategory FE	Yes	Yes
Date $\times$ Time of day FE	Yes	Yes
Observations	57,528	57,528

Notes: IV-2SLS = instrumental variable with two-stage least squares; FE = fixed effect. Standard errors in parentheses.

$^{*} p < 0.05$ , $^{* *} p < 0.01$ , $^{* * *} p < 0.001$ .

Table 4.

Impact of average product duration on session revenue.

	(1)	(2)
IV-2SLS	Log(Product revenue)	Log(Product revenue)
Avg product duration	$- {0.1164}^{* *}$	$- {0.1086}^{* * *}$
	(0.0305)	(0.0261)
Avg product duration $\times$ Official		$- 0.0289$
		(0.0379)
Session-level controls	Yes	Yes
Livestreamer FE	Yes	Yes
Product category FE	Yes	Yes
Date $\times$ Time of day FE	Yes	Yes
Observations	2,099	2,099

Notes: IV-2SLS = instrumental variable with two-stage least squares; FE = fixed effect. Standard errors in parentheses.

$^{*} p < 0.05$ , $^{* *} p < 0.01$ , $^{* * *} p < 0.001$ .

5.2. Impact of Average Product Duration on Session Revenue

Here, we analyze the impact of average product duration on session revenue while controlling for session duration, in Table A.2 in the Appendix. The coefficient for $A v g p r o d u c t d u r a t i o n$ is negative and statistically significant, indicating that when the average presentation time increases, the session revenue decreases.

Next, we investigate the potential heterogeneous effect between official and third-party livestreams at the session level using the interaction term Avg product duration $\times$ Official in our session-level regression. The results are presented in column (2) of Table 4. We find that the coefficient of $A v g p r o d u c t d u r a t i o n$ is negative and statistically significant, but the coefficient of Avg product duration $\times$ Official is not significant. This implies that both types of livestreamers experience a negative relationship between session revenue and the average time spent on presenting products. Therefore, there is a misalignment between the incentives of brand owners and third-party livestreamers. This misalignment occurs because, at the product level, the brand owner of a product benefits as the livestreamer increases their product presentation time since they can generate higher revenue for their product. However, third-party livestreamers who aim to generate higher session revenue would prefer to spend less time on each product.

6. Robustness Checks

In this section, we include several checks and additional analyses to establish the robustness of our main findings.

6.1. Product-Level Analyses

6.1.1. Alternative IV

To test the robustness of our results, we introduce an alternative IV. For the product-level analysis, we use the average product presentation duration in other non-focal subcategories in the same session as the IV for the focal product. Besides the employment of this IV, to mitigate the endogeneity concerns, we also control for the product sales of other subcategories in the same session. This is to ensure that the demand for the focal product is not directly influenced by the pace at which products in the other subcategories are presented. The results are presented in Table 5. Our main results remain robust.

Table 5.
Impact on product revenue using alternative IV.

(1) (2)

Log(Product revenue) IV-2SLS IV-2SLS

Product duration 0.0410 $^{* * }$ 0.0651 $^{ * }$

(0.0046) (0.0104)

Product duration $\times$ Official $- {0.108}^{}$

(0.0437)

Product- and session-level controls Yes Yes

Livestreamer FE Yes Yes

Product subcategory FE Yes Yes

Date $\times$ Time of day FE Yes Yes

Observations 69,262 69,262

	(1)	(2)
Product duration	0.0410 $^{* * *}$	0.0651 $^{* * *}$
	(0.0046)	(0.0104)
Product duration $\times$ Official		$- {0.108}^{*}$
		(0.0437)
Product- and session-level controls	Yes	Yes
Livestreamer FE	Yes	Yes
Product subcategory FE	Yes	Yes
Date $\times$ Time of day FE	Yes	Yes
Observations	69,262	69,262

Notes: IV-2SLS = instrumental variable with two-stage least squares; FE = fixed effect. Standard errors in parentheses.

$^{*} p < 0.05$ , $^{* *} p < 0.01$ , $^{* * *} p < 0.001$ .

6.1.2. Non-Sold-Out Products

During a particular session, the products may get sold out before the end of the live streamer’s presentation. To ensure that our results are not driven by those sold-out products, we do a subsample analysis focusing on products that did not sell out.¹⁰ These “non-sold-out” products are characterized by having an on-shelf duration longer than their presentation duration. The results of this analysis are presented in Table 6. We find that our main conclusion remains valid. This suggests that our findings are not solely influenced by products that quickly sell out and have implications for all products in general.

Table 6.
Sub-sample analysis with only non-sold-out products.

(1) (2)

Log(Product revenue) IV-2SLS IV-2SLS

Product duration 0.1652 $^{* * }$ 0.1943 $^{ * }$

(0.0321) (0.0324)

Product duration $\times$ Official $- {0.186}^{ * *}$

(0.0339)

Product- and session-level controls Yes Yes

Livestreamer FE Yes Yes

Product subcategory FE Yes Yes

Date $\times$ Time of day FE Yes Yes

Observations 52,512 52,512

	(1)	(2)
Product duration	0.1652 $^{* * *}$	0.1943 $^{* * *}$
	(0.0321)	(0.0324)
Product duration $\times$ Official		$- {0.186}^{* * *}$
		(0.0339)
Product- and session-level controls	Yes	Yes
Livestreamer FE	Yes	Yes
Product subcategory FE	Yes	Yes
Date $\times$ Time of day FE	Yes	Yes
Observations	52,512	52,512

Notes: IV-2SLS = instrumental variable with two-stage least squares; FE = fixed effect. Standard errors in parentheses.

$^{*} p < 0.05$ , $^{* *} p < 0.01$ , $^{* * *} p < 0.001$ .

6.1.3. Position of the Product in the Session

One of the potential confounders between product sales and the duration is the position at which the product is presented during the session. It is possible that the product’s position (rank) in a session is correlated with both the time spent on the product presentation and the product’s sales. To mitigate such concerns, we repeat our main analysis after including the product’s rank/position FEs in a session and find similar results (see Table 7).

Table 7.
Impact on product revenue after controlling product’s position in a session.

(1) (2)

Log(Product revenue) IV-2SLS IV-2SLS

Product duration 0.187 $^{* * }$ 0.161 $^{ * }$

(0.0426) (0.0265)

Product duration $\times$ Official $- {0.1827}^{}$

(0.0867)

Product- and session-level controls Yes Yes

Livestreamer FE Yes Yes

Product subcategory FE Yes Yes

Date $\times$ Time of day FE Yes Yes

Product Position FE Yes Yes

Observations 57,508 57,508

	(1)	(2)
Product duration	0.187 $^{* * *}$	0.161 $^{* * *}$
	(0.0426)	(0.0265)
Product duration $\times$ Official		$- {0.1827}^{*}$
		(0.0867)
Product- and session-level controls	Yes	Yes
Livestreamer FE	Yes	Yes
Product subcategory FE	Yes	Yes
Date $\times$ Time of day FE	Yes	Yes
Product Position FE	Yes	Yes
Observations	57,508	57,508

Notes: IV-2SLS = instrumental variable with two-stage least squares; FE = fixed effect. Standard errors in parentheses.

$^{*} p < 0.05$ , $^{* *} p < 0.01$ , $^{* * *} p < 0.001$ .

6.1.4. Propensity Score Matching

In the main model, we use linear models with many levels of FEs since they can accommodate various factors that potentially bias our estimates. However, certain selection-induced biases could remain despite including those FEs. To resolve such issues further, we use propensity score matching (PSM) to address the endogeneity issue.

We redefine the treatment as a binary indicator that equals 1 if the product duration is above the median product duration of each livestreamer $\times$ product subcategory pair, otherwise, 0. We first compute the propensity score $p (X_{n})$ for each observation $n$ using a Logistic regression model. This model includes product- and session-level variables ( $X_{n}$ ) like the on-shelf duration, price of the product, the style of presentation, the platform on which the product is livestreamed, the session duration, the number of products in a session and the average price of the session. After finding the propensity score of each observation, we find a control observation with the most similar propensity score for each observation in the treatment group.¹¹

After this matching process, we check whether the treatment and control groups have similar characteristics. The results are reported in Table 7 in the E-Companion. The two groups appear to be fairly balanced in terms of the covariates after PSM. Next, we run our main regression based on the matched sample. Table 8 reports the results. We observe that our primary findings at the product level remain robust.

Table 8.
Impact on product revenue using PSM.

Log(Product revenue) (1) (2)

Product duration (Above Median) 0.280 $^{* * }$ 0.165 $^{ * }$

(0.0155) (0.0193)

Product duration (Above Median) $\times$ Official $- {0.244}^{ * *}$

(0.0308)

Product- and session-level controls Yes Yes

Livestreamer FE Yes Yes

Product subcategory FE Yes Yes

Date $\times$ Time of day FE Yes Yes

Observations 71,805 71,805

Log(Product revenue)	(1)	(2)
Product duration (Above Median)	0.280 $^{* * *}$	0.165 $^{* * *}$
	(0.0155)	(0.0193)
Product duration (Above Median) $\times$ Official		$- {0.244}^{* * *}$
		(0.0308)
Product- and session-level controls	Yes	Yes
Livestreamer FE	Yes	Yes
Product subcategory FE	Yes	Yes
Date $\times$ Time of day FE	Yes	Yes
Observations	71,805	71,805

Notes: PSM = propensity score matching; FE = fixed effect. Standard errors in parentheses.

$^{*} p < 0.05$ , $^{* *} p < 0.01$ , $^{* * *} p < 0.001$ .

6.1.5. Non-Linear Relationship

In equation (1), we use a linear model specification and find that product duration has a significant negative effect on official livestream product revenue while a positive effect on third-party livestream product revenue. However, by allowing a non-linear model specification, the link between product duration and revenue can be much better explained. We use a quadratic model to fit our data. The results are reported in Table 9.¹² In column (1), the coefficient of $P r o d u c t d u r a t i o n$ is positive and statistically significant, whereas the squared term coefficient is negative and statistically significant. This implies that an increase in product presentation duration initially benefits the product revenue, but as the product duration increases, the effect starts to decrease.¹³

Table 9.
Non-linear impact of product duration on product revenue.

(1) (2)

Log(Product Log(Session

revenue) revenue)

Product duration 0.5060 $^{* * }$

(0.1424)

Product duration $^{2}$ $- {0.0040}^{ }$

(0.0014)

Avg product duration $- {0.2830}^{ * }$

(0.0999)

Avg product duration $^{2}$ 0.0024 $^{}$

(0.0012)

Product- and session-level controls Yes No

Session-level controls Yes Yes

Livestreamer FE Yes Yes

Product subcategory FE Yes No

Product category FE No Yes

Date $\times$ Time of day FE Yes Yes

Observations 57,528 2,099

	(1)	(2)
Product duration	0.5060 $^{* * *}$
	(0.1424)
Product duration $^{2}$	$- {0.0040}^{* *}$
	(0.0014)
Avg product duration		$- {0.2830}^{* * *}$
		(0.0999)
Avg product duration $^{2}$		0.0024 $^{*}$
		(0.0012)
Product- and session-level controls	Yes	No
Session-level controls	Yes	Yes
Livestreamer FE	Yes	Yes
Product subcategory FE	Yes	No
Product category FE	No	Yes
Date $\times$ Time of day FE	Yes	Yes
Observations	57,528	2,099

Notes: FE = fixed effect. Standard errors in parentheses.

$^{*} p < 0.05$ , $^{* *} p < 0.01$ , $^{* * *} p < 0.001$ .

Table 10.

Product revenue with product subcategory week FE.

	(1)	(2)
Log(Product revenue)	IV-2SLS	IV-2SLS
Product duration	0.2029 $^{* * *}$	0.1066 $^{* * *}$
	(0.0542)	(0.0251)
Product duration $\times$ Official		$- {0.2207}^{* * *}$
		(0.0463)
Product- and session-level controls	Yes	Yes
Livestreamer FE	Yes	Yes
Product subcategory $\times$ Week FE	Yes	Yes
Date $\times$ Time of day FE	Yes	Yes
Observations	57,348	57,348

Notes: IV-2SLS = instrumental variable with two-stage least squares; FE = fixed effect. Standard errors in parentheses.

$^{*} p < 0.05$ , $^{* *} p < 0.01$ , $^{* * *} p < 0.001$ .

6.1.6. Accounting for Subcategory-Level Unobserved Temporal Shocks

One of the potential concerns regarding our IV construction is that there could exist time-varying unobserved variables, for example, previous ad spending, that may affect both the sales and the time allocation of the presentation of different products in the current period. Hence, our IV (average product duration of products from other subcategories) and the focal product’s sales might not be independent of each other. To resolve such a concern, one can analyze a model that includes the product-week FEs. Including these interaction FEs into our regression ensures that any unobserved temporal shock to a product does not bias our estimates. However, we note that, as the same product is not often presented multiple times in any given week in our data (about 90% of observations get dropped), we instead include subcategory-week level FEs, where a subcategory includes multiple products. We believe the analysis with subcategory-week FE serves as a reasonable proxy because the unobserved variables at the individual product level are likely correlated across products within a subcategory. Another reason is that our IV is based on the average product duration of products across (non-focal) subcategories and not on the product duration of an individual product. Table 10 presents the results. Our main results remain robust.

6.2. Session-Level Analyses

6.2.1. Alternative Session Controls

Earlier, in our main analysis, we analyzed the impact of average product duration on session revenue, after controlling for session duration. As average product duration is the ratio of session duration to the number of products presented, this implies that we capture the variation in the number of products across sessions. Alternatively, one could also analyze the impact of average product duration after controlling for the number of products so that average product duration varies due to the variation in session duration. Therefore, instead of session duration, we include the number of products as a covariate in our regression.¹⁴ In such a case, an increase in Avg product duration can be interpreted as an increase in session duration, given the number of products presented in the session. Table 11 presents the results. Our main results remain robust.

Table 11.
Impact on session revenue controlling for number of products in a session.

(1) (2)

Log(Session revenue) IV-2SLS IV-2SLS

Avg product duration $- {0.159}^{* * }$ $- {0.145}^{ * *}$

(0.0274) (0.0289)

Avg product duration $\times$ Official $-$ 0.0217

(0.0510)

Session-level controls Yes Yes

Livestreamer FE Yes Yes

Product category FE Yes Yes

Date $\times$ Time of day FE Yes Yes

Observations 2,099 2,099

	(1)	(2)
Avg product duration	$- {0.159}^{* * *}$	$- {0.145}^{* * *}$
	(0.0274)	(0.0289)
Avg product duration $\times$ Official		$-$ 0.0217
		(0.0510)
Session-level controls	Yes	Yes
Livestreamer FE	Yes	Yes
Product category FE	Yes	Yes
Date $\times$ Time of day FE	Yes	Yes
Observations	2,099	2,099

Notes: IV-2SLS = instrumental variable with two-stage least squares; FE = fixed effect. Standard errors in parentheses.

$^{*} p < 0.05$ , $^{* *} p < 0.01$ , $^{* * *} p < 0.001$ .

6.2.2. Accounting for Category-Level Unobserved Temporal Shocks

It is plausible that our results are driven by rising/declining time trends or temporal shocks related to the sales of products in a given category. To check if our estimates are indeed biased by such temporal movements in a category’s sales, we further include the interaction (cross) fixed effects between a category and week. We find that our results remain qualitatively similar even after controlling for category-level time trends (see Table 12). Importantly, this analysis also allows us to account for any trends associated with the correlation between sales across weeks in a given category. Furthermore, there is a related potential concern that our session-level IV (mean average product duration of past sessions of the focal category) is correlated with the average sales of the past sessions which, in turn, could be correlated with the sales of the focal session, as they belong to the same category. To assuage this concern, we also control for the sales (per unit time) from sales of the past sessions of the same category. Estimates in Table 12 suggest that our main results remain unaltered.

Table 12.
Impact on session revenue after controlling for category-level time trends.

(1) (2)

Log(Session revenue) IV-2SLS IV-2SLS

Avg product duration $- {0.242}^{* * }$ $- {0.238}^{ * *}$

(0.0388) (0.0381)

Avg product duration $\times$ Official $-$ 0.0365

(0.0497)

Session-level controls Yes Yes

Livestreamer FE Yes Yes

Product category $\times$ Week FE Yes Yes

Date $\times$ Time of day FE Yes Yes

Observations 1,973 1,973

	(1)	(2)
Avg product duration	$- {0.242}^{* * *}$	$- {0.238}^{* * *}$
	(0.0388)	(0.0381)
Avg product duration $\times$ Official		$-$ 0.0365
		(0.0497)
Session-level controls	Yes	Yes
Livestreamer FE	Yes	Yes
Product category $\times$ Week FE	Yes	Yes
Date $\times$ Time of day FE	Yes	Yes
Observations	1,973	1,973

Notes: IV-2SLS = instrumental variable with two-stage least squares; FE = fixed effect. Standard errors in parentheses.

$^{*} p < 0.05$ , $^{* *} p < 0.01$ , $^{* * *} p < 0.001$ .

Table 13.

Impact on session revenue with PSM.

Log(Session revenue)	(1)	(2)
Avg product duration (Above Median)	$- {0.3150}^{* *}$	$- {0.1033}^{* * *}$
	(0.1103)	(0.0352)
Avg product duration (Above Median) $\times$ Official		0.5889
		(0.3389)
Session-level Controls	YES	YES
Livestreamer FE	YES	YES
Product Category FE	YES	YES
Date $\times$ Time of day FE	YES	YES
Observations	1,657	1,657

Notes: PSM = propensity score matching; FE = fixed effect. Standard errors in parentheses.

$^{*} p < 0.05$ , $^{* *} p < 0.01$ , $^{* * *} p < 0.001$ .

6.2.3. Propensity Score Matching

Analogous to product-level analysis, we use PSM to address the endogeneity issue for the session-level analyses. We redefine the treatment as a binary indicator that equals 1 if the Avg product duration is above the median average product duration of each livestreamer $\times$ product category pair; otherwise, 0. We first compute the propensity score $p (X_{n})$ for each observation $n$ using a Logistic regression model and session-level variables ( $X_{n}$ ) like the style of presentation, the average price of the session, the platform on which the product is livestreamed, and the session duration in which the product is presented. After finding the propensity score of each observation, we find a control observation with the most similar propensity score for each observation in the treatment group.¹⁵

After this matching process, we check whether the treatment and control groups have similar characteristics. The results are reported in Table 8 in the E-Companion. The two groups appear to be fairly balanced in terms of the covariates after PSM. Next, we run our main session-level regression based on the matched sample. Table 13 reports the results. We observe that our primary findings at the session level remain robust.

7. Conclusion

To the best of our knowledge, our work is the first to study a key design aspect of the livestream shopping channel, which is a new type of online channel for product promotions that crucially relies on real-time interaction with consumers. This design aspect is the product promotion duration which is likely to influence livestream sales. We empirically uncover an interesting heterogeneity in the way the product duration is related to sales in official and third-party livestreams—while an increase in product duration reduces product sales for third-party livestreams, it has no significant effect on official livestreams. We also discover that, for third-party livestreams, when product duration increases, session-level sales reduce whereas product sales increase.

Our findings provide practical implications for both livestreamers and brand owners. First, the finding that longer product presentations increase product revenue has practical significance for brand owners. In third-party livestreams, where multiple products are featured, brands must negotiate more airtime for their products. Longer presentations may allow for a deeper dive into product features and unique selling points, which enhances consumer understanding, potentially leading to higher conversion rates. To secure this extended time, brands may need to offer additional incentives to livestreamers, such as higher commissions, or bonuses tied to sales milestones. Brand owners can also aim to build long-term relationships with successful third-party livestreamers. By nurturing these relationships, brands can negotiate better terms for product presentations. Second, the tension between the goals of brand owners and third-party livestreamers stems from their differing motivations. Brand owners focus on improving the sales of their products in third-party livestreams, which may require longer presentations. In contrast, third-party livestreamers, who showcase multiple products from various brands in one session, are incentivized to keep the presentation for each product shorter. By featuring a broader range of products, they may be able to retain the attention of a diverse audience and increase overall session-level sales. This dynamic creates a potential conflict, as what benefits one party (longer presentation) might detract from the other’s goal (session sales).

Our article is not without its limitations. First, as in any observational study based on the IVs, we cannot completely rule out all potential confounders. Particularly, we do not have access to product-level time-varying variables such as advertisement spending and percentage of discount, which could have potentially influenced product duration and sales. Second, we do not have access to very granular user-level data to observe how they are engaging with each product presentation during livestream sessions, which could have helped us explore the mechanism behind our results in greater detail.

Supplemental Material

sj-pdf-1-pao-10.1177_10591478251314455 - Supplemental material for Designing E-commerce Livestreams: How Product Presentation Duration Affects Sales?

Supplemental material, sj-pdf-1-pao-10.1177_10591478251314455 for Designing E-commerce Livestreams: How Product Presentation Duration Affects Sales? by Si Xie, Siddhartha Sharma and Amit Mehra in Production and Operations Management

Footnotes

A. Appendix

Acknowledgments

The authors would like to thank the participants of the research seminar at the University of Connecticut, as well as attendees of the Conference on Information Systems and Technology (CIST 2023) and the Production and Operations Management Society Annual Conference (POMS 2022), for their valuable feedback and insightful discussions.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iDS

Si Xie

Siddhartha Sharma

Amit Mehra

Supplemental Material

Supplemental material for this article is available online ().

Notes

How to cite this article

Xie S, Sharma S and Mehra A (2025) Designing E-commerce Livestreams: How Product Presentation Duration Affects Sales? Production and Operations Management 34(12): 4079–4096.

References

Akçay

Tan

(2008) On the benefits of assortment-based cooperation among independent producers. Production and Operations Management 17(6): 626–640.

Alcañiz

Blas

Torres

(2006) Dependency in consumer media relations: An application to the case of teleshopping. Journal of Consumer Behaviour: An International Research Review 5(5): 397–410.

Ang

Wei

Anaza

(2018) Livestreaming vs pre-recorded: How social viewing strategies impact consumers’ viewing experiences and behavioral intentions. European Journal of Marketing 52(9/10): 2075–2104.

Aslam

Farhat

Arif

(2021) Skippable advertisement versus full-length advertisement: An empirical analysis in a developing economy. Journal of Interactive Advertising 21(1): 49–67.

Berger

Draganska

Simonson

(2007) The influence of product variety on brand perception and choice. Marketing Science 26(4): 460–472.

Besbes

Sauré

(2016) Product assortment and price competition under multinomial logit demand. Production and Operations Management 25(1): 114–127.

Blas

García

Economía

, et al. (2006) Development of a sector with a bright future: Teleshopping. Esic Market 6: 597–624.

Brechman

Bellman

Robinson

Rask

Varan

(2016) Limited-interruption advertising in digital-video content: An analysis compares the effects of “midroll” versus “preroll” spots and clutter advertising. Journal of Advertising Research 56(3): 289–298.

Cai

Wohn

Mittal

Sureshbabu

(2018) Utilitarian and hedonic motivations for live streaming shopping. In: Proceedings of the 2018 ACM international conference on interactive experiences for TV and online video, pp.81–88.

10.

Chen

Hong

(2019) Everyone can be a star: Quantifying grassroots online sellers’ live streaming effects on product sales. In: Proceedings of the annual hawaii international conference on system sciences. Hawaii International Conference on System Sciences.

11.

Chen

Liao

(2022) Antecedents of viewers’ live streaming watching: A perspective of social presence theory. Frontiers in Psychology 13: 839629.

12.

Clarke

(1976) Econometric measurement of the duration of advertising effect on sales. Journal of Marketing Research 13(4): 345–357.

13.

Conley

Hansen

Rossi

(2012) Plausibly exogenous. Review of Economics and Statistics 94(1): 260–272.

14.

Desmichel

Kocher

(2020) Luxury single- versus multi-brand stores: The effect of consumers’ hedonic goals on brand comparisons. Journal of Retailing 96(2): 203–219.

15.

Dhar

(1997) Consumer preference for a no-choice option. Journal of Consumer Research 24(2): 215–231.

16.

Draganska

Mazzeo

Seim

(2009) Beyond plain vanilla: Modeling joint product assortment and pricing decisions. QME 7(2): 105–146.

17.

eMarketer (2024) Livestreaming commerce sales in the United States between 2022 and 2026 (in billion U.S. dollars). https://www-statista-com.mutex.gmu.edu/statistics/1276120/livestream-e-commerce-sales-united-states/ (accessed 3 January 2025).

18.

Geismar

Dawande

Murthi

Sriskandarajah

(2015) Maximizing revenue through two-dimensional shelf-space allocation. Production and Operations Management 24(7): 1148–1163.

19.

Goh

K-Y

Heng

C-S

Lin

(2013) Social media brand community and consumer behavior: Quantifying the relative impact of user-and marketer-generated content. Information Systems Research 24(1): 88–107.

20.

Goldstein

McAfee

Suri

(2015) Improving the effectiveness of time-based display advertising. ACM Transactions on Economics and Computation (TEAC) 3(2): 1–20.

21.

Goldstein

Suri

McAfee

Ekstrand-Abueg

Diaz

(2014) The economic and cognitive costs of annoying display advertisements. Journal of Marketing Research 51(6): 742–752.

22.

Hamilton

Garretson

Kerne

(2014) Streaming on twitch: Fostering participatory communities of play within live mixed media. In: Proceedings of the SIGCHI conference on human factors in computing systems, pp.1315–1324.

23.

Han

Wang

Fan

(2023) Cost-effective social media influencer marketing. INFORMS Journal on Computing 35(1): 138–157.

24.

Hardy

(2021) Branded Content: The Fateful Merging of Media and Marketing. Routledge.

25.

Hübner

Schäfer

Schaal

(2020) Maximizing profit via assortment and shelf-space optimization for two-dimensional shelves. Production and Operations Management 29(3): 547–570.

26.

Hübner

Kuhn

Sternbeck

(2013) Demand and supply chain planning in grocery retail: An operations planning framework. International Journal of Retail & Distribution Management 41(7): 512–530.

27.

Hughes

Swaminathan

Brooks

(2019) Driving brand engagement through online social influencers: An empirical investigation of sponsored blogging campaigns. Journal of Marketing 83(5): 78–96.

28.

Jeon

Son

Chung

Drumwright

(2019) Temporal certainty and skippable in-stream commercials: Effects of ad length, timer, and skip-ad button on irritation and skipping behavior. Journal of Interactive Marketing 47(1): 144–158.

29.

Kumar

Tan

Wei

(2020) When to play your advertisement? Optimal insertion policy of behavioral advertisement. Information Systems Research 31(2): 589–606.

30.

Edwards

Lee

J-H

(2002) Measuring the intrusiveness of advertisements: Scale development and validation. Journal of Advertising 31(2): 37–47.

31.

(2024) Retail platform analytics: Practice, literature, and future research. Production and Operations Management. DOI: https://doi.org/10.1177/10591478241238972.

32.

Lou

Tan

S-S

Chen

(2019) Investigating consumer engagement with influencer- vs. brand-promoted ads: The roles of source and disclosure. Journal of Interactive Advertising 19(3): 169–186.

33.

Mallipeddi

Kumar

Sriskandarajah

Zhu

(2022) A framework for analyzing influencer marketing in social networks: Selection and scheduling of influencers. Management Science 68(1): 75–104.

34.

Nierobisch

Toporowski

Dannewald

Jahn

(2017) Flagship stores for fmcg national brands: Do they improve brand cognitions and create favorable consumer reactions? Journal of Retailing and Consumer Services 34: 117–137.

35.

Osei-Frimpong

McLean

(2018) Examining online social brand engagement: A social presence theory perspective. Technological Forecasting and Social Change 128: 10–21.

36.

Pan

Feng

Zhao

(2022) Fly with the wings of live-stream selling–channel strategies with/without switching demand. Production and Operations Management 31(9): 3387–3399.

37.

Park

Lin

(2020) The effects of match-ups on the consumer attitudes toward internet celebrities and their live streaming contents in the context of product endorsement. Journal of Retailing and Consumer Services 52: 101934.

38.

Pei

Mayzlin

(2022) Influencing social media influencers through affiliation. Marketing Science 41(3): 593–615.

39.

Peters

Bijmolt

(1997) Consumer memory for television advertising: A field study of duration, serial position, and competition effects. Journal of Consumer Research 23(4): 362–372.

40.

Pymnts (2016) TV Home Shopping Is Alive and Well, Thanks to Grandmas- or Teenagers. https://www.pymnts.com/news/retail/2016/tv-home-shopping-alive-well-thanks-to-grandmas-or-teenagers/ (accessed 3 January 2025).

41.

Rahnamaee

Berger

(2013) Investigating consumers’ online purchasing behavior: Single-brand e-retailers versus multi-brand e-retailers. Journal of Marketing Analytics 1: 138–148.

42.

Shapiro

Hitsch

Tuchman

(2021) TV advertising effectiveness and profitability: Generalizable results from 288 brands. Econometrica 89(4): 1855–1879.

43.

Singh

Cole

(1993) The effects of length, content, and repetition on television commercial effectiveness. Journal of Marketing Research 30(1): 91–104.

44.

Statista (2024) Market size of live streaming e-commerce in China from 2019 to 2023 with estimates until 2026 (in trillion yuan). https://www-statista-com.mutex.gmu.edu/statistics/1127635/china-market-size-of-live-commerce/ (Accessed 3 January 2025).

45.

Sun

Shao

Guo

Nie

(2019) How live streaming influences purchase intentions in social commerce: An it affordance perspective. Electronic Commerce Research and Applications 37: 100886.

46.

Tan

Wang

Tan

(2019) Impact of live chat on purchase in electronic markets: The moderating role of information cues. Information Systems Research 30(4): 1248–1271.

47.

Todri

(2022) Frontiers: The impact of ad-blockers on online consumer behavior. Marketing Science 41(1): 7–18.

48.

Todri

Ghose

Singh

(2020) Trade-offs in online advertising: Advertising effectiveness and annoyance dynamics across the purchase funnel. Information Systems Research 31(1): 102–125.

49.

Tuchman

(2019) Advertising and demand for addictive goods: The effects of e-cigarette advertising. Marketing Science 38(6): 994–1022.

50.

Tversky

Shafir

(1992) Choice under conflict: The dynamics of deferred decision. Psychological Science 3(6): 358–361.

51.

Van Reijmersdal

Rozendaal

Hudders

Vanwesenbeeck

Cauberghe

Van Berlo

(2020) Effects of disclosing influencer marketing in videos: An eye tracking study among children in early adolescence. Journal of Interactive Marketing 49(1): 94–106.

52.

Wilbur

(2008) A two-sided, empirical model of television advertising and viewing markets. Marketing Science 27(3): 356–378.

53.

Wongkitrungrueng

Assarut

(2020) The role of live streaming in building consumer trust and engagement with social commerce sellers. Journal of Business Research 117: 543–556.

54.

Xiao

Zhang

(2023) Exploring the factors influencing consumer engagement behavior regarding short-form video advertising: A big data perspective. Journal of Retailing and Consumer Services 70: 103170.

55.

Zhang

Liu

Ming

Cheng

(2024) Express: The sales impacts of traffic acquisition promotion in live-streaming commerce. Production and Operations Management, DOI: https://doi.org/10.1177/10591478231224938.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.21 MB

	Full sample		Official		Third-party
	Mean	SD	Mean	SD	Mean	SD
Product-session level ( $N = 74,997$ )
Product revenue	125,063	732,654	117,789	782,809	137,966	633,817
Product unit sales	1,009	3,914	810	3,419	1,353	4,628
Product duration (in minutes)	16.4	7.3	18.5	7.9	15.8	6.5
Product on-shelf duration	26.0	43.2	41.1	42.5	17.2	36.5
Product price	996	1,221	1,130	1,457	964	1,060
Session level ( $N = 2,304$ )
Session revenue	4,070,898	1,526,137	2,691,425	990,182	5,958,352	1,658,436
Session unit sales	32,844	178,456	18,481	148,985	58,432	97,852
Session duration (in minutes)	533	296	442	287	582	489
Number of products	32.6	29.8	22.8	24.1	36.8	22.6
Avg product duration (in minutes)	16.3	7.8	19.4	12.9	15.8	9.8
Informative style	0.41	0.49	0.60	0.49	0.23	0.42
Average price	1,020	4,581	1,190	6,589	869	992
Platform A	0.53	0.50	0.48	0.496	0.60	0.49
Official livestreamer	0.52	0.31

	(1)	(2)
	Log(Product	Log(Session
	revenue)	revenue)
Product duration	0.5060 $^{* * *}$
	(0.1424)
Product duration $^{2}$	$- {0.0040}^{* *}$
	(0.0014)
Avg product duration		$- {0.2830}^{* * *}$
		(0.0999)
Avg product duration $^{2}$		0.0024 $^{*}$
		(0.0012)
Product- and session-level controls	Yes	No
Session-level controls	Yes	Yes
Livestreamer FE	Yes	Yes
Product subcategory FE	Yes	No
Product category FE	No	Yes
Date $\times$ Time of day FE	Yes	Yes
Observations	57,528	2,099