Do All Roads Lead to Rome? Exploring the Relationship Between Social Referrals, Referral Propensity and Stickiness to Video-on-Demand Websites

With the prevalence of personal communication tools and social sharing buttons on websites, social referrals are increasingly gaining attention (Lis and Neßler 2013). With these tools, consumers can share information with larger audiences without temporal or geographical constraints, just by accessing a content website and by using the integrated social sharing buttons to refer the offer to their friends (Burtch et al. 2014). Figure 1 illustrates the social referral process. Social referrals have been researched in the broader context of electronic WOM (eWOM), which also comprises seller feedback and product reviews on online platforms, such as forums and online communities (Marchand et al. 2016; Cheung and Thadani 2012; Heimbach et al. 2015). The peculiarity of social referrals in an online context is that they involve communication between individuals with social connections to each other (Lis and Neßler 2013; Shi et al. 2013). Social referrals are perceived as more personal than seller feedback or online consumer reviews, which are generally posted publicly (Shi et al. 2013). Consumers also attribute higher credibility to information received from other consumers than to advertisements (Goh et al. 2013; Trusov et al. 2009).

To ensure conceptual clarity, Table 1 provides a brief overview of the different online channels that are part of our analysis – search, social media advertisement, and social referral – and their main characteristics. Following Anderl et al. (2016), we classify the three online channels according to their degree of personalization and contact origin. The degree of personalization reflects whether messages are individualized based on the targeted persons’ characteristics or behaviors. Organic search results are personalized because they are based on individual user queries (Ghose and Yang 2009). Social referrals are personalized because consumers usually share content via personal communication tools with peers who have similar interests. Advertisements on social media can be targeted based on users’ prior browsing behavior or their profiles; however, we examine non-personalized advertisements (Verhoef et al. 2010). As a second dimension, online channels can be classified based on their contact origin. Research distinguishes between customer-initiated channels,¹ in which consumers seek out information themselves, and firm-initiated channels, in which companies initiate marketing communications and push the message to consumers (Li and Kannan 2014; Anderl et al. 2016; Wiesel et al. 2011). In addition to the distinction between customer- and firm-initiated channels, we would like to direct attention to peer-initiated channels, which comprise social referrals. They originate neither from firms nor from interested consumers; they are usually initiated by peers who are already a customer and encourage their peers to consume a certain product or service as well (Burtch et al. 2014).

Most research has investigated the effectiveness of single online channels, such as searching (Ghose and Yang 2009; Yang and Ghose 2010; Chan et al. 2011), online advertisements (Braun and Moe 2013; De Keyzer et al. 2015), and eWOM (Gu et al. 2012; Cheung and Thadani 2012) on consumer decision making. Only a few studies include more than one online channel in their analysis, examining the interplay of firm-initiated and consumer-initiated channels (e.g., Bhatnagar et al. 2016; Xu et al. 2014; Bar-Gill et al. 2018). Furthermore, previous research has analyzed the effect of eWOM, compared to other online channels, on firm-level outcome variables such as restaurant revenue (Villanueva et al. 2008; Lu et al. 2013). The focus of these studies has been almost exclusively on eWOM in terms of online reviews or other public postings, and less on social referrals transmitted through personal communication tools, where referral recipients and referrers know each other. Considering the conceptual differences between these eWOM types (Marchand et al. 2016), two studies examining social referrals are directly related to our study. In the context of social networking sites, Trusov et al. (2009) compared social referrals transmitted via email with offline marketing activities on new network member registration. Their findings indicate that social referrals have a stronger long-term effect on customer acquisition than traditional advertising. Schmitt et al. (2011) examined the economic value of referred consumers compared with non-referred consumers. However, these two studies do not compare online channels at the individual level, and they only report combined channel effects at an aggregated level. Table 2 underlines the conceptual differences between these studies and our study.

Surprisingly little attention has been paid to estimating the impact of social referrals (i.e., peer-initiated channel) compared to other online channels, such as social media advertisements (i.e., firm-initiated channel) or organic search engine results (i.e., consumer-initiated channel). The calls by Trusov et al. (2009) and Lu et al. (2013), who argue that more research is needed, particularly with individual-level data, to assess whether consumers attracted in different ways show differential website usage patterns, have not been answered yet. Such research would enable the development of suitable online channel strategies for content websites. We address this research gap with a large data set from a VOD website and present empirical evidence of the differences in individual usage and referral behavior between consumers acquired through social referrals, organic search, and social media advertising.

Demers and Lev (2001) identified three factors that capture the most relevant dimensions of website performance: website stickiness, reach, and customer loyalty. Website stickiness measures the average time spent on the website per visit and the average number of pages viewed per visit. The reach factor encompasses the extent to which the website attracts unique visitors, and customer loyalty is driven primarily by the average number of visits to the website per unique visitor per period.

Website stickiness has been recognized as a key factor for profitability (Trueman et al. 2000; Danaher et al. 2006). As websites’ purposes vary considerably, performance measures must be adjusted (Alpar et al. 2001). E-commerce websites, for example, which focus on direct monetization from selling products, have to balance the desire to keep consumers on their website with the need to encourage them to complete transactions, while media and entertainment websites that rely on advertising revenue may have different goals (Bucklin and Sismeiro 2009). These websites usually strive to increase consumers’ attachment to the website and seek to maximize the number of page views in order to increase the number of advertising exposures (Benlian 2015). In fact, Danaher et al. (2006) and Bhatnagar and Ghose (2004) argue that longer visits help to increase advertising revenue and can increase the likelihood of consumers revisiting the website. The number of pages viewed might also increase advertising exposure (Demers and Lev 2001). In this study, we examine a VOD website. These websites usually display advertisements directly before or after a website visitor starts to consume the content. Therefore, this study incorporates the number of videos a specific consumer views as an additional measure of consumers’ stickiness to a website.

As noted, reach is an important performance measure for VOD websites. Research in information systems (IS) and marketing has dealt with the outcomes of consumer referral behavior at the firm level and with the drivers of referral behavior at the individual level (De Matos and Rossi 2008; Garnefeld et al. 2013). Different content characteristics have been identified as having an impact on whether people share it. Content that evokes more interest (Berger and Schwartz 2011; Heath et al. 2001), arouses high emotion, or contains more useful information (Berger 2014) is shared more often. Factors such as self-enhancement, identity signaling, extreme satisfaction, and customer commitment have been identified as important motivators for consumer referrals (De Matos and Rossi 2008; Berger 2014). Moreover, consumers’ perceptions of information value influence consumers’ referral propensity (Pihlström and Brush 2008; Koch and Benlian 2015). Other researchers have claimed that individuals are concerned with whether their actions will impair or enhance their social image and whether this concern will affect their decision to make referrals (Zhang et al. 2014). Social benefits have also been suggested as an important motivator for consumers to share referrals (Hennig-Thurau et al. 2004). The most important reason why consumers engage in referrals may be social capital, defined as “the sum of the actual and potential resources embedded within, available through, and derived from the network of relationships possessed by an individual or social unit” (Nahapiet and Ghoshal 1998, p. 243). It governs relations among individuals, making its creation and maintenance relevant (Coleman 1988). Also, social referral incentive systems and standard promotional tactics have been proposed as motives that lead consumers to engage in referrals (Shi et al. 2013; Koch and Benlian 2015).

Several studies have shown that the information gained through informal, interpersonal exchanges makes such recommendations more persuasive and even more influential than traditional marketing efforts (Trusov et al. 2009). Consumers are flooded by traditional forms of advertising, such as e-mail marketing, which decreases the latter’s effectiveness (Porter and Golan 2006). Against this backdrop, many companies are increasingly investing in advertising on social media platforms such as Facebook in order to get the attention of potential consumers (Goh et al. 2013; Rishika et al. 2013). In fact, a large amount of traffic to websites can be attributed to social media (Wong 2014). As social referrals are commonly disseminated for personal and non-commercial purposes and only shared with personal connections when a certain threshold of satisfaction is passed, referral recipients will attribute high credibility to this information. They will more likely trust the referred website (Biyalogorsky et al. 2001; Burtch et al. 2014). Supporting this notion, Lin (2007) argues that the more a consumer trusts a website, the more easily the consumer will stick to the website. Principles of triadic balance imply that existing consumers will be more likely to bring in others who they believe would like the company’s offer (Schmitt et al. 2011). In addition, social enrichment is the phenomenon that the relationship between the referral receiver and the company is enriched because a family member or friend is a customer of the same firm (Castilla 2005). In short, consumers acquired through social referrals are likely to have a stronger sense of attachment to the website than consumers acquired through social media advertisements. As a result, social referral recipients will be more likely to stick to a website than consumers who received the message directly from the company (Koch and Benlian 2015; Lis and Neßler 2013). Thus, we argue that consumers acquired through social media advertising will terminate their visit earlier and will have a lowered likelihood to view more pages and videos than consumers acquired through social referrals and propose.

However, social referrals might not outperform every other online channel. While consumers acquired through search engines are drawn to the website by clicking on the best possible result according to their search query, consumers acquired through social referrals access a website as a response to a referral from a social contact (Shi et al. 2013). The referrer chooses the content and to a certain extent, also the context of the referral, whereas the recipient only acts on the referral (Burtch et al. 2014). Consumers accessing a website via search engines are actively searching for the content themselves and have relatively more control over the situation. It is therefore reasonable to expect that consumers seeking information might be more inclined to engage with a website than consumers acting on a social referral (Matook et al. 2015). In the context of content websites, this means that consumers who land on a website because of organic search engine results are likely to view more pages and videos to satisfy their information or browsing needs, while social referrals commonly comprise the advocacy of a particular page or video on the website. As a result, this search behavior might also be associated with a comparably longer stay on a website than is the case with consumers acquired through social referrals. Therefore, we hypothesize that consumers accessing a content website via search engines might be more sticky than consumers acquired through social referrals.

Consumers’ referral propensity reflects their tendency to share information with acquaintances, colleagues, family members, and friends. Individuals are more likely to be influenced by those with whom they share ties, common understandings, and interests (Borgatti and Everett 1992). Prior research has also shown that consumers’ fear of being negatively perceived by their social contacts for forwarding messages deters them from doing so (Phelps et al. 2004; Zhang et al. 2014). The fact that a consumer has been referred to the website by another consumer may be interpreted as a signal of social appreciation, which, in turn, positively influences the likelihood of sharing (Pihlström and Brush 2008). Receiving content from others provides social proof or a seal of approval that the content is likely to be good, which should increase the referral propensity (Chen and Berger 2016; van Herpen et al. 2009). Moreover, receiving referrals from social contacts rather than from a company is believed to enhance recipients’ trust in these messages (Godes and Mayzlin 2004; Villanueva et al. 2008). Therefore, we propose that consumers acquired through social referrals may be associated with an increased likelihood to forward this information to others as social acceptance may be more likely and, owing to higher value perception, the perceived value of the information increases over that of both other channels (Koch and Benlian 2015). Moreover, consumers arriving at the website through search engines or advertisements are primarily driven by self-interest and might be associated with a decreased likelihood to start communicating with their social contacts by sharing a referral. Following this logic, we expect that consumers acquired through social referrals will be linked to a higher referral propensity than consumers acquired through organic search engine results or social media advertisements.

We obtained a clickstream data set from a well-known European media company that provides professional VOD content, such as short video clips, on their website. This VOD website is ideal for our research objective for three reasons: First, the website displays only proprietary, producer-generated videos, which are typical experience goods and one of the most-consumed content forms on the Internet (Godes and Mayzlin 2004; Nelson 1970). Second, the website contains only one form of experience good, since previous research showed differences in consumer behavior (e.g., time per page and number of page views) across product types (Huang et al. 2009). Third, the primary goal of the video content is entertainment, which is likely to be forwarded to others (Phelps et al. 2004).

The studied website has roughly 650,000 visits per day, of which approximately 530,000 are unique consumers. Social sharing buttons are displayed on each page next to the video to make it easier for consumers to share the website with others. The company did not explicitly incentivize social referrals with rewards, which enabled us to examine unsolicited and unrewarded referral decisions (Schmitt et al. 2011). The company tracks outbound social referrals from existing consumers to identify consumers arriving at the website through these social referrals. This recording allowed us to investigate consumer-to-consumer communication, which is typically not made publicly available and is therefore difficult to study (Trusov et al. 2009; Bucklin and Sismeiro 2009).

The goal of this study is to analyze consumers’ stickiness to a website and their referral behavior. Therefore, our data set includes only visitors who started to view a video on the website, as we do not seek to understand the factors that drive consumers to respond, for example, to social media advertising or social referrals (Mao and Zhang 2015). We limit the data to consumers from one country, since there might be differences in the presentation of the website’s content based on a consumer’s country. Furthermore, we analyze unique consumers and do not consider multiple visits in our analysis (Chatterjee et al. 2003). Any time a consumer accesses a URL, website providers record the details of the session with “cookies” on consumers’ devices. Among other session-related details, these data points include the online channel from which the consumer arrives, how long the consumer remains on the website (i.e., visit length), how many pages were viewed (i.e., page views), the number of video starts, whether the consumer clicked on the website’s social sharing buttons (i.e., referral propensity), the device category, and an identification number. It is important to note that social referral receivers do not know whether the referral was conducted through social sharing buttons or other means, such as copying a URL link.

VOD websites offering free content often only monitor observable, session-related behavior and do not record financial or demographic data, because they do not require consumers to register on their websites (Chatterjee et al. 2003; Burtch et al. 2014). While this prevents consumers’ resistance to providing personal data, as well as the effort of having to log in prior to viewing the content, it also restricts website providers’ knowledge of their customers. For our study, this poses additional challenges to account for consumer heterogeneity. Following prior literature (Burtch et al. 2014; Bhatnagar et al. 2016), we focus on several critical session-related variables (i.e., location and device) in our estimations to address consumer heterogeneity. Because we analyze consumer data supplied by the media company that is associated with all daily visits, our data does not suffer from the recall problems associated with self-reported data (Huang et al. 2009) and has high external validity.

The data used in this study spans 8 weeks from February 20 to April 20, 2016, and includes all daily visits from the following three online channels: social referral, search engine, and social media advertisements. As bot activity is common on websites and not every bot can be detected by the website provider beforehand, we anticipated some bots in our data set as well (Schwarzer et al. 2016). Therefore, we limit the number of page views to 100 and the time spent on the website to a maximum of 5 h (18,000 s). 882 observations were deleted due to a higher number of pages viewed, and 242 observations were removed due to higher visit length. However, to better assess the impact of such suspicious activity, we reran the analyses and obtained consistent result patterns. Our final data set comprises 956,920 visits from consumers with a unique identification number and 21 different videos. In general, each video was accessed by consumers referred from all three channels during that time period, and each video in our data set was started more than 1000 times. The average visit contained 1.7 page views lasting a total of 2.28 min.

Based on the hypotheses presented above, we developed three general categories of variables: online channel measures (independent variables), session measures (dependent variables), and control measures.

We present the regression results for Visit Length and Page Views in Table 6. The results for Video Starts are displayed in Table 7. Model 1 presents the results of the control-only model and Model 2 presents the full model. For page views and video starts, we report Nagelkerke’s pseudo R² value (Nagelkerke 1991), acknowledging that low pseudo R² statistics in logistic regression are the norm (Hemmert et al. 2018; Hosmer et al. 2013).

The first two columns provide the results for visit length. In accordance with Hypotheses 1 and 2, we find that consumers acquired through social referrals spend on average approximately Exp(0.453) = 1.57 times more time on the website than consumers acquired through social media advertisements and approximately Exp(− 1.990) = 0.14 times less time than consumers acquired through organic search engine results. The third and fourth column present the results with regard to the breadth of the visit. Model 1 presents the control-only model. The expected number of page views is approximately Exp(0.114) = 1.12 times higher for consumers acquired through social referrals than for those acquired through social media advertisements. As hypothesized, we observe that consumers referred from organic search engine results are associated with viewing more pages. Search results compared with social referrals are expected to have a rate Exp(1.621) = 5.06 times greater for page views.

We also measured time per page (i.e., the average number of seconds spent per page) with similar results. Consumers acquired through search engines have a significantly higher time per page than consumers acquired through social referrals, and these consumers have a significantly higher time per page than consumers acquired through social media advertisements.

Column one and two show the results for video starts for the total sample and column three and four the results for a smaller sample size (see Sect. 6.3).

Confirming Hypothesis 1, consumers acquired through social referrals are associated with starting significantly more videos than consumers arriving at the website via social media advertisements. The odds of viewing more than one video on the website are approximately Exp(3.006) = 20.21 times greater for social referrals than for social media advertisements. In accordance with Hypothesis 2, consumers acquired through search engines are associated with an increased likelihood to start more than one video, compared with consumers arriving at the website via social referrals. The odds of viewing more than one video on the website are approximately Exp(1.437) = 4.21 times greater for search results than for social referrals.

Next, we present the results for Referral Propensity in Table 8, along with model fit statistics. For each variable, the logistic regression estimates the effect of the variable on the referral propensity, given that all other covariates remain constant. The likelihood ratio test assesses the overall fit of the model. The analysis of each model indicates a good model fit with a highly significant likelihood ratio (p = 0.000). In column three and four, we present the results with a smaller sample size for robustness (see Sect. 6.3).

Using the results in the second column of Table 8, the odds ratio of the variable Social Media Advertisement equals 0.321 (Exp(− 1.137)). The odds of a referral are approximately Exp(1.137) = 3.12 times greater for consumers accessing the website through social referrals than for consumers landing on the website via social media advertisements, supporting Hypothesis 3. The odds ratio of the variable Search Engine equals 0.229 (Exp(− 1.472)). The odds of a referral are approximately Exp(1.472) = 4.36 times greater for consumers accessing the website through social referrals than for consumers landing on the website via organic search engine results, which also supports Hypothesis 3. Video starts do not have a significant influence on consumers’ referral propensity, whereas the influence of the device and number of page views is significant.

Since our goal is to estimate the impact of the online channels on consumers’ stickiness to the website and their referral propensity, we performed a series of robustness tests. First, we repeated the analyses using a model that controls for the different videos in our data set. By adding a dummy for each of the 21 videos (Video) in our model, we absorbed the effects particular to each video. In addition, we included Posting Duration as another control variable to indicate the number of days between the date on which the video had been posted on the website and the date of the visit (Burtch et al. 2014). Since the Posting Duration variable is measured in days and includes zero, we increased the posting duration by one, which allowed us to use the logarithm. To measure visit length and page views while controlling for video effects and posting duration, we limited the data set and excluded all consumers viewing more than one video during their visit. Therefore, the models do not include the number of video starts. The results continue to support the hypotheses tests. The variable Social Media Advertisement is still significantly negatively related to visit length and page views, and Search Engine is positively related to both outcome variables compared with the Social Referral variable. Posting Duration has a negative and significant influence. To estimate referral propensity, we also limited the data to the first video view in order to control for posting duration and the video. Therefore, referral propensity refers to the first video that was viewed on the website. Accordingly, referral propensity captures whether a consumer clicked on a sharing button before or after the first video was viewed. Additionally, we considered the Video Type of the first video viewed, to account for the difference of specific interest videos versus general interest videos. We find that general interest videos have a significantly higher referral propensity than special interest videos. Posting Duration has a positive and significant influence on referral propensity. The variables Search Engine and Social Media Advertisement are still significantly negatively related to consumers’ referral propensity compared with the Social Referral variable. The results of the effects of the channels on referral propensity remain stable, supporting the theoretical argument that the channel, i.e., social influence, is associated with the propensity to share the videos.

Second, the occurrence of 1’s in our dependent variables Video Starts and Referral Propensity can be considered a rare event. The percentage of consumers viewing more than one video and the percentage of consumers using social sharing buttons is low in our data set. The majority of consumers only view one video (N = 955,311) or do not use the share button (N = 955,213). Therefore, our dependent variables Video Starts and Referral Propensity are not balanced. The biases that rare events create in estimating logit models have been discussed in the literature (King and Zeng 2001; Ben-Akiva and Lerman 1985). To overcome the problem of misclassification, we re-estimate the model while deliberately under-sampling the non-sharing consumers so that a more balanced sample of ones and zeros in the dependent variable is obtained. This sampling technique is called choice-based sampling (King and Zeng 2001). Because this method does not yield consistent estimates of the intercept when traditional maximum likelihood methods are used, we adjust the estimated intercepts for each alternative. Manski and Lerman (1977) developed a weighted endogenous sampling maximum likelihood estimator, which accounts for the different weights in the zeros and ones from the population. However, this estimator has the undesirable property of increasing the standard errors of the estimates (Greene 2003). Therefore, we adjust the estimated intercepts for each alternative by subtracting ln(S_i/P_i) from the exogenous maximum likelihood estimates of the intercept, where S_i is the percentage of observations for alternative i in the sample, and P_i is the percentage of observations for alternative i in the population (Ben-Akiva and Lerman 1985).

For Video Starts, we estimated the model in Table 7 again with a choice-based sample that balances the number of consumers viewing one video and consumers viewing several videos on the website. Our smaller data set comprises 1609 consumers who decided to view more than one video and 19,107 randomly sampled consumers (2% of the sample if Video Starts = 0) who did not view more than one video during their visit. The estimates are reported in the third and fourth column of Table 7. We find that the logistic regression with rare event correction produces estimates very similar to those generated by the full sample, which further improves our confidence in the finding that social referrals are positively associated with video starts compared to social media advertisements, but not with organic search.

For Referral Propensity, we estimated the models again with a choice-based sample that balances the number of sharing and non-sharing consumers. Our smaller data set to test Hypothesis 3 comprises 1707 consumers who decided to click on social sharing buttons and 19,090 randomly sampled consumers (2% of the sample if Referral Propensity = 0) who did not click on the sharing buttons during their visit. The estimates are reported in the third and fourth column of Table 8. To summarize, having examined the relationship between social referrals, website stickiness and referral propensity via different model specifications and data sizes, we find consistent evidence for our results.

Our study contributes to eWOM literature at the cross-section of IS and marketing by expanding our understanding of the antecedents of website stickiness in general and online social referral behavior in particular. First, prior literature on eWOM has focused on examining the impact of online reviews on consumer behavior or firm performance (e.g., Chen and Xie 2007; Chevalier and Mayzlin 2006; Cheung and Thadani 2012), and there has been little emphasis on exploring the effectiveness of social referrals via personal communication tools on consumer behavior (Shi et al. 2013; Köster et al. 2015). We extend previous research by examining how social referrals are related to consumers’ behavior on a VOD website. Second, we provide results for a website only offering intangible goods (i.e., videos), whereas most academic literature has investigated the effect of interpersonal persuasion on consumer decision making of tangible goods, such as book sales (e.g., Chevalier and Mayzlin 2006; Li and Hitt 2008; Chen et al. 2004). Therefore, our research differs from extant approaches since it measures profitability in terms of consumers’ website usage and not in terms of product or service sales (Cheung and Thadani 2012). Third, this study contributes to the literature on information processing. While prior research has identified website design variables, such as personalization cues (Benlian 2015), perceived social presence (Cyr et al. 2009), and the emotional states of the consumer (Mahnke et al. 2015), as factors that influence consumers’ behavior, our results suggest that how consumers are drawn to the website may also be linked to how they consume and share content. Fourth, we compare online channels at the individual level, unlike prior research, which has focused on aggregated data (e.g., Lu et al. 2013; Villanueva et al. 2008). Essentially, this research responds to the calls from Lu et al. (2013), Trusov et al. (2009), and Bhatnagar et al. (2016), who have stressed the need to compare online channels at the individual level. Lastly, our data set incorporates consumers already acquired through social referrals, which allows us to compare the referral propensity across different stages of dissemination, whereas extant research (e.g., Burtch et al. 2014; De Bruyn and Lilien 2008; Koch and Benlian 2015) has focused on either first-stage (i.e., consumers acquired through firm-initiated channels) or second-stage (i.e., consumers acquired through costumer-initiated channels) actors.

Our research has direct implications for the management of different online channels. Owing to a plurality of content offers on the Internet, one of the primary objectives and challenges for VOD website providers is to attract more consumers and to keep them on their websites as long as possible. While there are multiple online channels that can draw consumers to a website, providers seek to understand whether specific online channels perform better than others in achieving their goals. We have placed a particular focus on social referrals as a rather new and inexpensive means for VOD website providers to utilize their consumers to attract their friends by using a social sharing button. In particular, consumers acquired through organic search results have a higher stickiness to a website than consumers arriving at the website through social referrals. Therefore, VOD website providers should continue investing in SEO to make consumers aware of their content. In addition, they should facilitate social referrals, since consumers acquired via social referrals exhibit higher visit lengths and breadths than consumers acquired through social media advertisements. This is especially worth noting considering every consumer arriving at the website via social media advertisements causes acquisition costs, whereas sharing buttons provide a rather cost-effective mechanism to attract potential consumers to the website (Shi et al. 2013). Furthermore, VOD website providers could make inferences from the observed behavior of the referrers about the video content preferences of referral receivers. As a result, the website provider is able to serve the referred customer in a tailored way, something that takes time to learn for other newly acquired customers. Because of this informational advantage, the provider might be able to generate higher revenues from referred customers at the beginning of the relationship.

A consumer’s decision to share the video on the website is an important measure for companies to investigate because such an act of sharing attracts potential consumers to the website. These consumers, owing to the way they have been drawn to the website, exhibit higher referral propensity. This study shows that consumers acquired through social referrals are more likely to start influencing their friends than consumers who arrived at the website through other channels (i.e., social media advertisements or search engines). Therefore, VOD website providers should not underestimate social referrals and can utilize our findings to design social referral incentive schemes to facilitate these types of referrals (Shi et al. 2013). Existing studies provided guidance about the design and utilization of rewards in other contexts (Biyalogorsky et al. 2001; Köster et al. 2015) and have quantified the influence of reward allocation schemes and tie strength on referral propensity (Ryu and Feick 2007; Shi et al. 2013). Referral rewards do not have to be monetary; they can also be symbolic (Verlegh et al. 2013). VOD website providers could facilitate the referral engagement of consumers by investing in such a social referral incentive system or by promoting the usage of their social sharing buttons (Shi et al. 2013). Consumers can hardly distinguish between advertisements and friends’ recommendations, as social referrals might be incentivized by website providers (e.g., affiliate links, referral reward programs). Marketing managers could foster social referrals, for instance, by integrating multiple sharing buttons, by placing them at prominent places on their websites, and by using sharing buttons that move with the content. Finally, they might integrate statements such as “send this to a friend” at the end of a video clip. Thus, besides financially pushing social referrals, managers also have other options for promoting social referrals since they have full control over their websites in contrast to search engines or social media platforms.

Our research has several limitations that might be addressed by further research. First, we mainly focused on the effects of the online channel on website visitors’ behavior. Additional studies could concentrate on the specific effects induced by different videos for different users and tie those effects to distinct website outcomes. Second, we derived our data from a leading VOD website providing short video clips. Although this is a common type of VOD video platform in the European media market, other types of video platforms might possess different characteristics and our results might therefore not be directly transferable. Future studies could sample different content types (e.g., user-generated videos, podcasts) to test whether the results remain consistent in other domains (Oestreicher-Singer and Zalmanson 2013; Bar-Gill et al. 2018). Third, our study is based on a large set of observational data that allows for a comparably objective measurement of user behavior. Although clickstream data represents an important source of behavioral insights, the data we analyzed in this study limited our modeling effort in a number of ways. Owing to missing detailed sociodemographic information on the users, the results have to be interpreted cautiously. While we found evidence that social referrals are associated with website stickiness and referral propensity, future studies could test whether there exists a selection effect of users being more prone to social referrals, spending more time on websites or having higher sharing dispositions. Although Danaher et al. (2006) found that variability in visit duration is driven mostly by the situation and only to a small degree by individual traits, further research could analyze data collected on websites with mandatory registration, in order to address consumer heterogeneity (Özpolat et al. 2013). Future research could also apply longitudinal or experimental methods that allow to detect direct causal inferences of the reported relationships. Fourth, our data was based on last-click metrics which ignore prior channel touches and, therefore, might give biased estimates on the credits of each channel (Kannan et al. 2016). However, data sets that observe consumers over time and across multiple channels often provide new challenges concerning data reliability (Chatterjee et al. 2003; Anderl et al. 2016). Furthermore, consumers of hedonic, low-involvement products, such as the free online videos in our study, usually do not engage in elaborate information processing and browsing behavior prior to the focal website visit, which confirms the suitability of last-click metrics (Huang et al. 2009; Wiesel et al. 2011; Vakratsas and Ambler 1999). Fifth, although cookies recognize subsequent visits, we have excluded repeated sessions since cookies do not necessarily allow for the possibility of exploiting similarities in click behavior across sessions for each consumer, owing to the usage of one device by multiple consumers or usage of multiple devices by one consumer (Chatterjee et al. 2003). Finally, while our data set provided insights across three online channels, future research might generate further insights by analyzing additional online channels, such as third-party websites. Moreover, analyzing the resulting profit effects across different channels could provide rich insights for companies and would be an interesting extension to our study.