Introduction
Creation, once the domain of highly specialized authors and artists, has rapidly become widespread with the advent of the digital content era. Equipped with advanced smart devices and editing software, individuals without professional training can now capture photographs, record videos, and produce original visual content in a matter of hours or even minutes. Although the value of most digital content depreciates rapidly, individuals continue to derive significant psychological benefits from their social networks, which include family, friends, and interested yet unfamiliar viewers. In some cases, this engagement also provides pecuniary rewards through tips or advertising revenue. However, copyright protection, which originated in 1710 with the aim of safeguarding creators' incentives to innovate, has evolved minimally and is increasingly misaligned with the digital landscape. While regulators have sought to introduce new forms of protection for digital content, the theoretical foundation and practical effectiveness of these measures remain subjects of ongoing debate.
There are two key distinctions between digital content and traditional copyrighted works. First, traditional works are primarily rooted in imagination, whereas digital content is initially more derived from capturing reality. As Shakespeare writes in A Midsummer Night's Dream, ‘The lunatic, the lover, and the poet/Are of imagination all compact’, whether in novels, plays, operas, or films, most creative works are produced by authors with ‘seething brains’ and ‘shaping fantasies’, blending their life experiences with deep emotion and insight into human nature. These works consistently evoke empathy and emotion, possessing a lasting vitality. The creation of digital content, however, does not adhere to the same principles. Take the short video industry, discussed later, as an example. Initially, most individuals simply capture treasured moments, express their emotions, or showcase aspects of their daily lives. This process typically involves neither significant imagination nor substantial creative effort, thereby diminishing the perceived necessity for copyright protection. However, as an increasing number of digital content creators seek attention beyond their close social circles, and some begin to rely on it for their livelihood, the creation of short videos increasingly mirrors traditional visual works such as films and plays. The demand for protection resurfaces, driven by the need to safeguard original ideas, plots, designs, and frameworks from misuse and piracy.
Second, digital content does not require the same level of effectiveness or value as traditional works. Traditionally, a creative work must be both ‘original’ and ‘effective/valuable’ to qualify for copyright protection. Originality implies that the work originates from the author and is not copied from others (Parchomovsky & Stein, Reference Parchomovsky and Stein2009), while effectiveness refers to the market's valuation of the creation (Runco & Jaeger, Reference Runco and Jaeger2012). Generally speaking, the costs and market value associated with creating a movie or play are significantly higher than those for a piece of digital content, such as a short video. However, the piracy costs for digital content are considerably lower, often resulting from a mere click. This presents a new challenge for copyright law. While many digital content pieces may lack substantial value, it remains essential to protect originality in order to prevent an oversaturation of similar content. As the volume of digital content increases, traditional regulations on copyright infringement – typically managed by a central authority and relying on a ‘notice and takedown’ mechanism – are beginning to face significant challenges. Fortunately, the introduction of artificial intelligence (AI)-powered identification systems offers a potential solution. While AI may not excel in assessing the artistry and value of a content piece, it is highly proficient at detecting whether a digital work has been copied, pieced together, or adapted from other sources. The subsequent question is whether authority-led copyright protection remains effective in safeguarding the originality of digital content today, and how it will be influenced by advanced approaches and technologies.
With these concerns in mind, we focus on the emerging short video industry and the external regulation of its copyright enforced by the Chinese government. We collected a unique dataset comprising 15,536 short videos from Douyin, a leading Chinese short video platform, produced between January 2019 and February 2020. This dataset was gathered through stratified random sampling to ensure representativeness. The identification of originality for each short video is conducted using the platform's AI-powered identification system, supplemented by manual verification. Using a difference-in-differences (DID) framework, we find that the proportion of original short videos during the period of strengthened regulation is higher, suggesting that authority-led copyright protection – partly facilitated through intra-platform regulation – still plays an important role in the creation environment for digital content. Additionally, professional creators tend to produce more than amateur creators, possibly due to their greater sensitivity to infringements on the pecuniary and psychological benefits associated with short videos.
This article contributes to the existing literature from two perspectives. The first is about the effectiveness of copyright protection. Researchers have pointed out that an optimal regime of intellectual property rights is a balance between securing private returns of creators and avoiding undue restrictions on the diffusion of ideas that benefit society (Arrow, Reference Arrow1962; Gilbert & Shapiro, Reference Gilbert and Shapiro1990; Green & Scotchmer, Reference Green and Scotchmer1995; Kumar & Turnbull, Reference Kumar and Turnbull2008). However, most previous research has focused on other forms of intellectual property rather than copyright due to challenges in collecting data on the quantity and quality of copyright-protected works and the scarcity of significant shocks in copyright regulation that allow for DID tests. Only two influential papers have empirically examined this issue, both leveraging wars as shocks and offering conflicting perspectives. Giorcelli and Moser (Reference Giorcelli and Moser2020) estimated the effect of Italy's varying copyright protection under Napoleon's conquest on the historical success and popularity of operas, while Biasi and Moser (Reference Biasi and Moser2021) examined the adverse effects of copyright on follow-on creation, measured by citations, using the World War II Book Republication Program as an external regulation. These pioneering attempts mark the beginning of empirical literature on the effectiveness of copyright, leaving considerable room for further research. Our article expands their perspectives by focusing on the digital content industry and how platforms are involved in the authority-led copyright regulation.
Second, our study provides empirical evidence to the growing literature on the creation of digital content through both theoretical analysis (Litman, Reference Litman1996; Rong, Xiao, Zhang, & Wang, Reference Rong, Xiao, Zhang and Wang2019) and case studies (Arthurs, Drakopoulou, & Gandini, Reference Arthurs, Drakopoulou and Gandini2018; Chen, Valdovinos Kaye, & Zeng, Reference Chen, Valdovinos Kaye and Zeng2021; Cheng, Liu, & Dale, Reference Cheng, Liu and Dale2013; Lu & Lu, Reference Lu and Lu2019; Shutsko, Reference Shutsko2020). Most digital content is created on platforms that exert looser control over authors and offer less support for creation compared with traditional organizations such as film companies or animation studios. These platforms have stronger incentives for scaling expansion and profit maximization rather than quality control (Jeon & Rochet, Reference Jeon and Rochet2010; Teh, Reference Teh2022). As a result, platforms may deliberately allow less original short videos and weaken regulation of infringing works. Using the short video industry as a representative example, our article illustrates how strengthened external regulation may compensate for insufficient internal regulation and address potential incentive incompatibility.
The remainder of this article is organized as follows. Section ‘Institutional background’ introduces the background and the policy shock on copyright protection. Section ‘Theoretical framework and hypotheses’ presents the theoretical framework and hypotheses. Section ‘Methods’ introduces our methods and data. Section ‘Results’ presents the empirical results. Section ‘Discussion’ discusses the contributions, implications, and limitations. Section ‘Conclusion’ concludes the article.
Institutional Background
Copyright Protection of Digital Content in China
In contrast to many other countries, China has implemented a ‘dual-track’ copyright protection system that encompasses both administrative and judicial safeguards (Hong, Edler, & Massini, Reference Hong, Edler and Massini2022). The administrative protection is overseen by the National Copyright Administration of China (NCAC) and, for digital content, is jointly managed by the Cyberspace Administration of China (CAC). Since the establishment of the NCAC in 1985, the ‘dual-track’ system has functioned effectively and independently. The administrative track, in particular, has been more proactive and timelier, exerting greater influence on the creation of digital content. Aligned with the Chinese central government's push for an originality-driven virtual environment, the administrative track has increasingly emphasized copyright protection (Huang, Reference Huang2017; Peng, Ahlstrom, Carraher, & Shi, Reference Peng, Ahlstrom, Carraher and Shi2017; Zhao, Reference Zhao2020). A notable example of this ‘campaign-style’ regulation is the Campaign against Online Infringement and Piracy (COIP), which, as shown in Figure 1, spanned from May to October 2019 and has had a significant impact on the digital content industry.
Figure 1. Timeline of ‘Campaign against Online Infringement and Piracy’ (COIP)
Campaign Against Online Infringement and Piracy
The COIP, colloquially known as ‘Jianwang’ in Chinese, was officially initiated by the NCAC. The campaign aimed to combat online infringement and emphasize the importance of copyright protection, particularly for short videos. Stricter measures were implemented against the unauthorized streaming of copyrighted content, including news, movie clips, live performances, and original music. Severe violators faced stringent administrative penalties, while short video platform operators benefiting from ‘safe harbor’ protection were required to strengthen their regular self-regulation practices.
These regulations were positively received in practice. For example, Douyin enhanced its AI-powered infringement detection algorithm and imposed intra-platform penalties on violators. These penalties ranged from rating reductions to account restrictions, and in more serious cases, temporary or permanent bans. Numerous copyright violators were penalized during the campaign, and an increasing number of creators joined Douyin, drawn by its increasingly supportive environment for originality protection.
Theoretical Framework and Hypotheses
Creation of Original Content and Copyright Protection
With the rise of smart devices and the establishment of user-generated content platforms, billions of people have started photographing, filming, and capturing their daily lives to attract attention, leading to a boom in digital content creation. Through same-side and cross-side network effects, an individual work spreads, and its social value amplifies much more quickly than traditional forms of content. Contributors receive not only pecuniary rewards from platforms and advertisers but also socio-cognitive benefits such as sensemaking, group recognition, influence, fulfillment, subculture formation, self-expression, and collaboration opportunities (Adarves-Yorno, Postmes, & Alexander Haslam, Reference Adarves-Yorno, Postmes and Alexander Haslam2006; Birnhack, Reference Birnhack2001; Chang & Chen, Reference Chang and Chen2020; Litman, Reference Litman1996), all of which further motivate them to create more. This positive feedback loop rapidly accumulates viewers and creators, and ultimately strengthens the attention economy.
Unfortunately, the creators of digital content suffer greatly from infringement and abuse, often even more so than in the past. The benefits of copying and reposting others' work have grown with the development of networks, while the costs have decreased due to less-detectable plagiarism and the popularization of self-editing applications (Liu, Shi, Teixeira, & Wedel, Reference Liu, Shi, Teixeira and Wedel2018). The short video industry is particularly affected. Unlike traditional media such as books and movies, short videos are highly substitutable and created by decentralized individuals, making it challenging for viewers to identify and subscribe to the original creator (Cheng et al., Reference Cheng, Liu and Dale2013; Wang, Gu, & Wang, Reference Wang, Gu and Wang2019). This lack of recognition significantly reduces the traffic and benefits accruing to original creators, diminishing their bargaining power against platforms (Rao, Legout, Lim, Towsley, Barakat, & Dabbous, Reference Rao, Legout, Lim, Towsley, Barakat and Dabbous2011). Despite some top short video creators signing contracts with platforms to gain market power, most contributors lack effective channels to protect their work. According to interviews with individual creators and the managerial board of short video platforms, the risk of infringement has significantly hindered creators' incentives and has become a stumbling block for the industry.
Effective copyright protection can address this issue to some extent. Unlike patents, which emphasize novelty, require costly applications, and primarily protect firms, copyrights focus on originality, are automatically acquired, and predominantly protect individuals. Given the nature of digital content, copyright protection does not require a new work to be completely unprecedented or significantly different from previous works. It only requires that the creation process be original and not directly copied from others. Copyright generally protects the fixation of an idea or thought in an ‘expression’, not the idea or thought itself. For example, filming one's daily life in a humorous way creates an original work protected by copyright, preventing others from using clips or frames directly extracted from this work without permission. However, it does not prevent others from filming in a similar way and creating a new copyrighted work with comparable themes or plots. In this sense, copyright strikes a balance between protecting individuals' incentives and promoting the growth of the content market.
Based on our discussions above, we hypothesize a positive relationship between copyright protection and the creation of original short videos. Under the ‘campaign-style’ governance, Chinese ex officio copyright enforcement is publicly broadcast and communicated through platforms on time, quickly influencing users' behavior. Therefore, we expect that an external policy like the COIP will likely increase the originality rate of short videos. A preliminary analysis of Figure 2 suggests the possibility of this relationship.
Figure 2. Distribution of videos and original rate by months
Formally, we propose the following hypothesis for empirical testing:
Hypothesis 1 (Effect of Copyright Protection on Originality) (H1): After the COIP strengthening the regulation on copyright infringement towards short videos, creators tend to produce a higher fraction of original works.
Influencers and Heterogeneous Effect of Copyright Protection
Building on our hypothesis that strengthened copyright protection stimulates originality, we propose that its influence may vary among individuals. We hypothesize that influencersFootnote 1 – those who are well-known and popular on the platform – are likely to experience a greater change in incentive due to the following reasons.
First, influencers are more vulnerable to copyright protection campaigns than other users. Given their high visibility and large fan bases on the platform, influencers are easier to identify and more likely to attract followers, making their infringing activities more prominent. Consequently, they face greater risks of being targeted by enforcement efforts.
More importantly, influencers derive greater benefits from their original work compared with other contributors. In the competition for attention, influencers invest substantial time and effort into creating original content, which they produce more frequently and at a higher quality than others. Their large fan bases enable them to attract significant attention from viewers, granting them substantial influence over public opinion and behavior. Unlike regular users, influencers reap greater social and pecuniary rewards from their large followings, with some earning a living through viewer tips, advertising fees, and revenue-sharing agreements with platforms. Stronger copyright regulation benefits influencers by driving more traffic to their original content, thereby increasing their incentives to create. Conversely, influencers stand to lose more if they engage in infringing activities, as such behavior can severely damage their reputation and reduce their business value. For influencers who rely on their content for income, the impact of reputational harm can be particularly detrimental.
Another reason influencers may react more strongly than regular users is that they closely monitor regulatory updates from authorities and platforms, making them more likely to be aware of new regulations in a timely and clear manner. With a clear understanding of the benefits of creating original works and the penalties for unoriginal contributions, influencers are likely to adjust their creative activities promptly.
Following this logic, we propose that the impact of the COIP on influencers is greater than on other users. Formally, our second hypothesis is:
Hypothesis 2 (Effect of Copyright Protection for Influencers) (H2): Compared to other users, influencers with a larger fan base are more inclined to produce original works after the implementation of strengthened copyright regulation.
The Relationship Between Intra-Platform Regulation and External Regulation
While Hypothesis 1 emphasizes the potential effect of strengthened copyright protection on creators' incentives, an important question remains: what role do the platforms play? As shown in Figure 2, both the number of aggregated videos and the number of unoriginal videos increase alongside the rise in the originality rate of short videos. This trend eliminates the possibility that the increasing originality rate is solely explained by the platform's strategy of taking down unoriginal works in response to regulation. A more likely mechanism is that the platform has implemented several intra-platform measures to align with external regulation, thereby shifting users' inclination to create more original works.
This hypothesis of the mechanism is supported by both theory and our interviews with the management board of Douyin. From a regulatory theory perspective, the legitimacy of rules, the competencies of entities, and intra-entity interactions are three essential factors for the success of nonbinding governance (Gorwa, Reference Gorwa2019a). Compared with external regulation from administrative or legal institutions, platforms have significant informational and cost advantages in supervising user behavior and maintaining quality control. Thus, it is expected that the social impact of external regulation will be more favorable when the platform's strategy aligns with government supervision.
Additionally, the management board of Douyin confirms that external regulation has indeed reduced internal resistance to strengthening the detection and punishment of unoriginal short videos on the platform. Figure 3 further supports this, showing a significant increase in both the absolute number and proportion of accounts temporarily banned on Douyin following the implementation of the COIP.
Figure 3. The dynamics of intra-platform regulation
Based on the analysis above, we expect that promoting intra-platform regulation serves as a viable channel for external copyright regulation to take effect. Formally, we propose our final hypothesis:
Hypothesis 3 (Effect of Intra-platform Regulation) (H3): The COIP raises the proportion of original short videos partly through promoting intra-platform regulation.
Methods
Dependent Variable
Evaluating the originality level of short videos presents a significant challenge. To address this issue, we have adopted a novel method involving supervised machine learning, facilitated by Douyin. This approach draws on similar research conducted by Wu and Zhu (Reference Wu and Zhu2022). Specifically, the identification process, implemented by Douyin's Copyright Center, combines supervised machine learning techniques with two rounds of manual checks by video auditors.
We constructed the variable ‘Originality’, assigning a value of 1 to videos identified as original by the platform and 0 otherwise. If the identification metric for a particular video surpasses the predetermined threshold, it is recognized as an original video. This threshold exceeds legal norms and industry standards. Although some videos require case-by-case assessment, those featuring live content, rich material, unique concepts, and distinct styles generally qualify as original, while their counterparts are categorized as unoriginal. Importantly, it is worth noting that unoriginal videos are not inherently infringing videos that need to be punished, as the platform regularly removes infringed content through the ‘notice-and-takedown’ process. When we refer to unoriginal videos, we specifically denote those with lower levels of originality compared with the videos deemed original within our sample.
Key Independent Variables
Regulation is a binary variable that indicates whether creators produce videos after the launch of strengthened copyright protection regulation. To be specific, Regulation is assigned 1 if a video is produced after the launch of the COIP, and 0 otherwise.
Treat is a dummy variable that indicates whether a short video belongs to the treatment group. According to the proportion of unoriginal videos in each theme in Table 1Footnote 2, the themes of ‘Film & TV’, ‘Plot’, and ‘ACG’ rank in the top three. Supporting evidence from interviews with Douyin platform management and official monitoring reports indicates that popular films, TV shows, comics, and plot-based dramas are primary targets for short video infringement. Therefore, Treat is assigned a value of 1 if a short video belongs to the ‘Film & TV’, ‘Plot’, or ‘ACG’ themes (the treatment group) and 0 if it belongs to any other theme (the control group). Short videos in the treatment group are more likely to infringe copyright compared with those in the control group, and consequently, they are also more likely to be subject to regulation under copyright protection policies. Thus, the variable of interest is the interaction term between Regulation and Treat, representing the joint effect of copyright regulation and the categorization of videos vulnerable to regulation.
Table 1. Distribution of themes and proportion of unoriginal videos
Notes: ACG refers to ‘Animation, Comics, and Games’, which is a subculture of Southeast Asia. The ‘Proportion of unoriginal videos’ metric represents the average rate of nonoriginal videos across various themes from January to April 2019.
Mediating Variable and Moderator Variable
To assess the mediating role of intra-platform regulation in the relationship between copyright protection and originality, we identified a mediating variable: the monthly proportion of temporarily banned accounts among all accounts (proportion of accounts temporarily banned) by the platform. It provides a relative measure by comparing the number of temporarily banned accounts to the total number of accounts. The moderator variable in our study is a binary variable, Influencer, which is coded as 1 for creators who have amassed more than 10,000 followers and 0 for those who have not.Footnote 3
Control Variables
We control for other factors that may influence the main relationships we investigate, specifically focusing on Duration, Music, and Special Effects. The variable Duration measures the time length of short videos, Music denotes whether the creator incorporates background music (BGM) into the short videos, and Special Effects indicates whether special effects are integrated into the content.
Model
We adopt a DID model to estimate the hypotheses using the COIP policy as a natural experiment. The main DID model is specified as follows:
where the dependent variable (Y it) is the originality of videos (termed as Originality). ν i and μ t represent province-fixed effects and month-fixed effects to control the location and time of video posting. e i,t is a randomly distributed error term. The magnitude of the coefficient (β) indicates the impact of copyright protection on originality. Since the dependent variable is a dummy variable, we employ Probit models to estimate the coefficients.
Sample
Our unique dataset comprises 15,536 short videos generated by individual users on the Douyin platform, covering the period from January 2019 to February 2020. This time frame is selected due to the COIP policy, which lasted from May to October 2019. Our dataset includes observations from four months before the campaign's launch to four months after its end, focusing exclusively on videos created by users from mainland China. To ensure representativeness, the observations are randomly sampled through stratification. Initially, we categorized all videos into five groups based on their number of views. Subsequently, we calculated the proportion of videos in each group relative to the total number of videos and then randomly sampled videos from each group in accordance with its corresponding proportion. For example, if Group 1 accounts for 10% of all videos, we would randomly sample 2,000 videos from this group (20,000 × 10%). To mitigate the influence of outliers, continuous variables are winsorized at the 1st and 99th percentiles.
Given the low proportion of the treatment group in the full sample (approximately 5%), we employed the propensity score matching (PSM) method to conduct a more balanced analysis. Following the approach used in previous studies (e.g., Hendricks, Hora, & Singhal, Reference Hendricks, Hora and Singhal2015; Qiu, Tian, & Zeng, Reference Qiu, Tian and Zeng2022), we conducted month-by-month PSM to identify matching videos in the control group for each video in the treatment group. The matching process took into account covariates including Duration, Music, Special Effects, and province-fixed effects. After implementing 1:4 nearest-neighbor matching, the means of the covariates in the treatment and control groups showed no significant differences, confirming that the two groups were balanced and well-matched. The final balanced matching sample consisted of 3,087 short videos.
Results
Statistical Analysis
Tables 2 and 3 present descriptive statistics and correlations of variables. Notably, 57% of the videos in our dataset are original, highlighting their prevalence. The mean value of Treat is 0.239, suggesting that the treatment group accounts for approximately a quarter of the sample. Most videos feature music but lack special effects. Video durations range from 5 s to 1 min, averaging 20.7 s, emphasizing the dominance of very short videos. Regarding intra-platform regulation, the average monthly figure for temporarily banned accounts exceeds 10,000, underscoring the significance that Douyin places on intra-platform regulation. Influencers represent approximately 3.6% of the total, indicating that the vast majority of creators are noninfluencers with a relatively smaller follower base. Table 3 shows a positive correlation between Regulation and Originality, indicating a positive link between copyright protection and originality.
Table 2. Descriptive statistics
Table 3. Correlation coefficient matrix
Note: ***p< 0.01.
Figure 2 illustrates the monthly distribution of videos and the corresponding original ratios (proportion of original videos) from January 2019 to February 2020. It is evident that both the volume of videos and the original rate exhibit a consistent upward trend during this timeframe. Additionally, there is a significant acceleration in the growth of the original rate post-May 2019, suggesting that the implementation of the COIP likely contributed to an increase in the creation of original videos. A more in-depth analysis is provided in the following sections.
Baseline Results
Table 4 presents the baseline findings from Probit models employed to test our hypotheses, alongside an OLS (ordinary least squares) model for comparison. Originality is the dependent variable in the baseline regressions. Robust standard errors are consistently applied throughout our analysis. Model 1 does not incorporate control variables. Model 2 introduces control variables, leading to a significant improvement in explanatory power, as evidenced by the Pseudo R 2 metric. Model 3 introduces month-fixed effects and province-fixed effects. After controlling for month-fixed effects, the variable Regulation is not necessary to include in the regression model due to multicollinearity. Consequently, we exclude it from Model 4, whose results align with those in Model 3. Model 5 acts as a comparison using the OLS method.
Table 4. Baseline results
Notes: Robust standard errors are reported in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.
Model 4's outcome reveals a statistically significant positive effect of the interaction term between Regulation and Treat on Originality (β = 0.337, P < 0.05). This validates our H1, indicating that bolstered copyright protection (the COIP) stimulates creators' impetus to produce original short videos. Particularly noteworthy, the marginal effect of the interaction term in Model 4 is 0.102, signifying that enhanced copyright protection can amplify the probability of originality in short videos by 10.2%. For robustness, Model 5 replaces the Probit model with an OLS model, yielding consistent results aligned with Model 4.
Furthermore, we have observed intriguing patterns regarding the control variables. First, the coefficient of Music demonstrates a significantly positive influence, indicating that incorporating music aids creators in crafting original short videos. This aligns with Douyin's standout characteristic – the seamless integration of music and video content, which sets it apart from competitors. Second, there seems to be a negative relationship between Duration and Originality. This trend could be attributed to the strong correlation between duration and the production cost of original videos. While long videos may not entail the same crew and equipment expenses as movies, they demand thorough scripting, well-crafted plots, and adept editing. For instance, the market cost for a 60-s video stands at around ¥500 in China and $500 in the US. Generally, the cost of audiovisual creations rises exponentially with their duration. In order to reduce production costs, longer videos are more likely to imitate or infringe upon original works. Consequently, the considerable cost associated with producing lengthy videos could explain the negative effect of Duration on Originality.
Parallel Trend Test
Two methods are utilized to verify the parallel trend assumption. First, we conduct a time trend analysis. The results are presented in Figure 4, which details the time trends in the proportions of original videos within the treatment and control groups. Prior to the launch of the COIP in May 2019, the proportions of original videos in both groups displayed similar trends, indicating that they were following parallel paths. Notably, after the launch of the COIP, the proportion of original videos in the treatment group increased rapidly, whereas it declined in the control group. Consequently, the disparity in the proportions of original videos between the two groups significantly narrowed following the COIP's initiation. This shift suggests that the COIP had a differential impact on the treatment group, supporting the validity of our DID approach.
Figure 4. Time trend of originality
Second, the placebo test method provides additional verification of the parallel trend assumption. We hypothesized an earlier launch of the COIP, specifically in February, March, or April of 2019. Subsequently, we reconstructed the Regulation variable and its interaction term with Treat. The results, as depicted in Table 5, indicate that the estimated coefficients for the interaction term between Regulation and Treat are not significant in any scenario, confirming that the observed changes in originality can be attributed to the actual implementation of the COIP and not to other underlying trends.
Table 5. Placebo tests
Notes: Robust standard errors are reported in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.
To conclude, both the time trend analysis and the placebo test provide evidence supporting the parallel trend assumption, validating the reliability of our model and its conclusions.
Heterogeneity from Different Creators
The preceding analysis emphasizes the consistently positive impact of copyright protection on short videos. In the next phase, we delve into the creator heterogeneity of this impact across different creators. According to our theoretical analysis, influencers may respond differently from other contributors to copyright protection measures. To explore this creator heterogeneity, we introduce a creator variable, Influencer, which is derived from a new dataset obtained from the Douyin website.Footnote 4 This variable, along with its interaction with Regulation × Treat, is incorporated into our baseline regression. To estimate the differential effect, we adopt a triple-differences model, building on the DID model used in the baseline regressions. The triple-differences model is expressed as follows:
where the variables are the same as those in the baseline model specification except for Influencer.
Table 6 presents our findings. The statistically significant coefficient of Regulation × Treat × Influencer in Model 1 indicates that influencers put more creative effort into producing original videos after the implementation of the COIP compared with others. Figure 5 visually illustrates the moderating effect of influencers. To provide more proxies for influencers, we introduce two related variables: Number of fans and Number of likes per video. Number of fans, measured in tens of thousands, represents the current fan count of a creator, while Number of likes per video is calculated by dividing the total number of likes by the number of videos that a creator possesses. The descriptive statistics for these two new variables are displayed in Appendix Table A1. Overall, it is observed that creators generally have a substantial number of fans, yet the average number of likes per video is relatively low.
Table 6. Heterogeneity from different creators
Notes: Robust standard errors are reported in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.
Figure 5. Moderating effect of Influencer creators
The results of the user heterogeneity tests, conducted using these two new variables, are presented in Models 2 and 3 of Table 6. These findings align with those observed in Model 1. In summary, H2 is verified.
Mediation Analysis: Intra-Platform Regulation
Next, we consider the role of intra-platform regulation, augmented by AI, in protecting intellectual property rights and creating a supportive environment for content creators. Recent enhancements to the platform's AI-based infringement detection algorithm and the imposition of intra-platform penalties – such as reducing a video's search priority, limiting the visibility of new releases, and temporarily or permanently banning infringing accounts – illustrate the platform's commitment to copyright enforcement.
To test the mediating role of intra-platform regulation between copyright protection and originality, we adopt the causal steps approach (Baron & Kenny, Reference Baron and Kenny1986), introducing the mediating variable: the proportion of accounts temporarily banned.Footnote 5 The results of the mediation tests are presented in Table 7. Based on Model 1 in Table 7, we find that enhanced copyright protection (the COIP) amplifies the probability of originality. The results in Model 2 indicate a significant positive impact of the COIP on intra-platform regulation. Figure 3 further illustrates the strengthening of intra-platform regulation following the launch of the COIP. Additionally, intra-platform regulation has a notable impact on user behavior (Gorwa, Reference Gorwa2019a, Reference Gorwa2019b). In summary, these findings support the notion that the COIP promotes originality through strengthened intra-platform regulation, validating H3.
Table 7. The mediating role of intra-platform regulation
Notes: Robust standard errors are reported in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.
Robustness Analysis
To guarantee the robustness of our key findings, we conduct various robustness tests. First, we modified the sample period to address concerns about potential confusion caused by the COIP implementation period (May 2019–October 2019). We replaced the original sample period (January 2019–February 2020) with a shorter period (January 2019–October 2019). Model 1 of Table 8 displays the results of the adjustment, yielding similar findings to our initial analysis.
Table 8. Robustness check 1: Change the sample period and the independent variables
Notes: Robust standard errors are reported in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.
Second, we further investigate whether the COIP has a lasting impact beyond the conclusion of the policy. In this analysis, we introduce two new binary variables: Campaign Period and After Campaign. Campaign Period denotes whether a short video was crafted between May and October 2019, while After Campaign signifies videos produced from November 2019 to February 2020. Next, we replaced the previous independent variable with the interactions of Treat and Campaign Period, as well as Treat and After Campaign, respectively. Model 2 of Table 8 presents the results. The interaction of Treat and Campaign Period yields a significant positive effect on Originality (β = 0.410, p < 0.05) while the coefficient of the interaction of Treat and After Campaign is positive, yet not statistically significant. This outcome provides evidence that the impact of strengthened copyright protection on originality is primarily observed during the COIP implementation period.
Third, to explore differences among the three themes within the treatment group, we reconstructed the independent variable. Specifically, Treat was replaced with three thematic dummies (Film, Plot, and ACG), each serving as an indicator variable. Additionally, we introduced interaction terms between these dummies and Regulation. The results for these interaction terms are displayed in Model 3 of Table 8. Notably, the coefficients for all three interaction terms are positive, with the interaction between Plot and Regulation showing a significantly positive effect. This suggests that the COIP has a more pronounced impact on plot videos compared with other types.
Fourth, to address the potential issue of omitted variables, we incorporated a title indicator variable about video characteristics (Title) into the baseline model. The results from Model 1 of Table 9 confirm that our main findings remain valid. Furthermore, we refined our control for geographic variation by substituting province-fixed effects with city-fixed effects. The findings from Model 2 of Table 9 indicate that the primary results are still robust, with the inclusion of city-fixed effects.
Table 9. Robustness check 2: Consider more control variables and city-fixed effects
Notes: Robust standard errors are reported in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.
In addition, we incorporated some characteristics of creators into our analysis. Specifically, we introduce Organization in Model 3 of Table 9. Organization represents whether a creator is an organization (e.g., multi-channel network or commercial association) certified by the short video platform. The descriptive statistics pertaining to the creators are detailed in Appendix Table A1. In our sample, organizations make up less than 1.2%. The results with the characteristic variable of creators in Model 3 of Table 9 demonstrate that the main findings related to copyright protection remain consistent with the baseline results.
Fifth, we closely examine the cross-sectional nature of the dataset. To address this, we incorporate province–month-fixed effects into the model. The results in Model 1 of Table 10 demonstrate that our main findings remain robust after accounting for these effects. Additionally, by integrating theme-fixed effects, we further refine our control over the influence of video themes, which are closely related to the COIP. The results in Model 2 of Table 10 also highlight a significant impact of copyright protection on originality.
Table 10. Robustness check 3: Add province–month-fixed effects and theme-fixed effects
Notes: Robust standard errors are reported in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.
Sixth, we modify the dependent variable and conducted a DID estimation using theme-level data instead of video-level data to validate our main findings. In the baseline regression, we utilize video-level data, and now we construct theme-level panel data to reassess the primary results. Specifically, we calculate the number of videos and original videos per month per theme, allowing us to determine the percentage of original videos in each theme, referred to as the Original rate. Subsequently, we perform DID regression using the theme-level data with two-way fixed effects. The results, presented in Table 11, showcase that Regulation × Treat still has a significantly positive impact on the Original rate. Therefore, the findings using theme-level data align with the baseline results, reinforcing the robustness of our conclusions.
Table 11. Robustness check 4: Conduct DID estimation on the theme-level data
Notes: Robust standard errors are reported in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.
Discussion
Theoretical Contributions
Our study makes several contributions to the theoretical literature. First, we examine whether historical copyright protection mechanisms remain effective in sustaining a healthy environment for digital content, a relatively underexplored yet increasingly significant form of creative work. Although short videos are easier to produce, have less value, and have shorter lifecycles compared with traditional forms of creative works like movies and plays, the logic of copyright – granting original creators exclusive rights to realize private returns and mitigating challenges related to inappropriability and uncertainty in creative activities (Arrow, Reference Arrow1962; Mowery & Rosenberg, Reference Mowery and Rosenberg1989; Nordhaus, Reference Nordhaus1969) – remains unchanged, albeit with additional dimensions of benefit-cost trade-offs.
Beyond pecuniary rewards, short video contributors benefit from socio-cognitive sensemaking, fulfillment, group recognition, subcultural influence, self-expression, and relationship development. The network effect further amplifies and accelerates these social benefits, emphasizing the need for timely regulatory measures. Simultaneously, the cost of infringements – such as imitation and misuse – has significantly decreased due to technological advancements and increasing global interconnectedness, necessitating the implementation of new forms of regulation. Our article provides empirical evidence on the effectiveness of external copyright regulation in promoting originality in short video creation, and demonstrates that authority-led regulation is further reinforced through platform involvement.
Our article also aligns with existing literature on innovation management in the digital era (Adams, Bessant, & Phelps, Reference Adams, Bessant and Phelps2006; Candelin-Palmqvist, Sandberg, & Mylly, Reference Candelin-Palmqvist, Sandberg and Mylly2012; Nambisan, Lyytinen, Majchrzak, & Song, Reference Nambisan, Lyytinen, Majchrzak and Song2017; Tidd, Reference Tidd2001). Nambisan et al. (Reference Nambisan, Lyytinen, Majchrzak and Song2017) argue that digital innovation is less constrained, less predefined, and characterized by fewer boundaries between the innovation process and its outcomes – attributes exemplified by short video creation.
We propose that although open-source and ‘private-collective’ models are well-suited for industries such as software (von Hippel & von Krogh, Reference von Hippel and von Krogh2003), they may be suboptimal for digital content, given the low cost of imitation, rapid iteration cycles, and limited timeframes for capturing commercial value. Using a ‘structure-conduct-performance’ perspective (Spulber, Reference Spulber2013), we advocate for fostering a healthy environment that encourages individuals to invest greater effort in original creation. Stricter limitations on ‘fair use and ‘transformative use’, enhanced penalties for infringements, and intensified platform self-regulation under external supervision have proven to be effective approaches. These discussions contribute to the broader framework of innovation management within the attention economy.
We also contribute to the theory of platform regulation. The implementation of ex-ante regulation and the establishment of government-led supervision frameworks can be highly costly. This aligns with contemporary discussions on achieving efficient and effective intra-platform governance and user self-governance (Gorwa, Reference Gorwa2019a, Reference Gorwa2019b; Teh, Reference Teh2022). Our article supports the notion that a stronger governmental impetus can encourage platforms to proactively engage in intra-platform governance, which is less costly and, as evidenced in our case, effective.
Practical Implications
Our results have rich implications for policymakers. In the era of the Internet, the diffusion of ideas and the development of creations accelerate. Some primitive understandings of intellectual property protection have gradually become outmoded, and the efficacy of strong protection on innovation is still under debate. Our article extends the discussion to the short video industry and provides robust evidence on the effectiveness of strengthened copyright protection. Leveraging a regulatory perturbation, we posit a relationship between enhanced regulation and creators' inclination toward the production of original content (H1). As our theoretical implication indicates, it is necessary to consider the comprehensive effects of the value-realization period, difficulty in imitation, explosive propagation along networks, and a series of other factors when determining the regulatory approach. At least in the short video industry, we recommend stronger copyright protection for originality protection, and suggest that this conclusion may also suit the entire user-generated content industry.
To decompose the aggregated impact of strengthened regulation, we examined the diverse responses of influencers and other creators. Our results align with some patent researchers' suggestions that copyright protection policies can significantly favor one group over others (Acemoglu & Akcigit, Reference Acemoglu and Akcigit2012), underscoring that a successful policy may not be universal and should be contingent on the specific characteristics of targeted creators.
This study also offers practical insights for digital platforms, grounded in theoretical frameworks that emphasize the crucial role of regulation in maintaining service quality (Jeon & Rochet, Reference Jeon and Rochet2010; Teh, Reference Teh2022), particularly within the digital content industry. To strengthen intra-platform regulation, we advocate for a comprehensive strategy that integrates AI-based detection with human review, alongside stringent penalties for creators who violate regulations. This combined approach is likely to be more effective and sustainable for platforms in addressing infringement.
Limitations and Future Research Directions
Conversely, our study is not devoid of limitations. First, the quantification of originality relies on a dummy variable. While adequate for general discussions and prevalent in industry practice, a more nuanced originality index encompassing diverse dimensions – such as editing, music, dubbing, and textual elements – could offer a more refined assessment. With richer dimensions, it is also possible for further research to examine the robustness of our conclusion with supervised or unsupervised clustering.
Second, our data lacks details in short video creation, limiting the depth of our discussion on variations among creators before and after the policy shock. The absence of individual-level data confines us to tracing stylistic trends only at the platform level and necessitates stringent analytical approaches to ensure the consistency of estimations. An optimal dataset would comprise panel data capturing each creator's output over a defined period, enabling the assessment of changes in creators' inclinations toward originality, styles, and efforts. Moreover, we encourage further research to gather data from a broader array of video platforms, particularly those individuals engaged in multihoming. A comprehensive examination across various video platforms would yield profound and nuanced results, allowing us to observe differential reactions of the same individual to policy changes on heterogeneous platforms.
Third, future research should further investigate the mechanisms underlying the impact of strengthened copyright protection on video creation, utilizing more detailed data and advanced methodologies. Regarding intra-platform regulation, it remains challenging to accurately estimate the extent of mediation effects, despite our findings suggesting that the COIP likely fosters originality through enhanced platform regulation. These challenges primarily stem from data limitations and the difficulties in measuring intra-platform regulatory mechanisms. Additionally, other potential mechanisms, such as penicury rewards and psychological benefits derived from creating original content, may play significant roles in the effect of strengthened copyright protection. Future studies should focus on evaluating these mediating variables and exploring their contributions to the overall impact of copyright enforcement on video creation.
Conclusion
By exploring the effect of an external policy shock, the COIP, on the creation of short videos, our article supports the notion that strengthened regulation is still effective in stimulating creators' incentives in the era of digital content. The analysis on influencers reveals the possible user-level heterogeneous impact of the regulation, and the mechanism analysis roughly depicts how external regulation influences creators through enhanced intra-platform regulation. The insights derived from our study carry significant implications for the formulation of intellectual property-related policies, particularly within the context of digital content. We anticipate that our research will fuel further interest in exploring the optimal framework for intellectual property protection from both theoretical and empirical perspectives.
Data availability statement
The data that support the findings of this study are available on request from the corresponding author Fei Hao. The data are not publicly available due to privacy restrictions.
Funding statement
This research was supported by the Beijing Social Science Fund (21DTR051), the ‘Fundamental Research Funds for the Central Universities’ in UIBE (XJ2024001202), the National Natural Science Foundation of China (Nos. 71872098, 72373079, and 72342008), Tsinghua University Initiative Scientific Research Program (No. 2024THZWYY06), a research grant from the Institute for Global Industry at Tsinghua University, and CIBER University of South Carolina.
Appendix
The Descriptive Statistics About Creators
Table A1. Descriptive statistics about creators
Extension on Robustness Check
To further check the robustness of the measurement of originality, we introduced a new dataset along with a novel outcome variable, ‘Originality2,’ which utilizes a refined method to distinguish original content from sourced material. Like the previous dataset, the new one was created using a stratified random sampling process. The formulation of the new dependent variable was adjusted according to the submission channel. Specifically, the new variable is assigned a value of 1 if a video is created entirely through filming and editing within the app, and 0 if it represents a finished work uploaded from external sources. Our interviews with platform management confirmed that videos produced using App-based filming and editing are more likely to be original, whereas uploaded videos often involve unauthorized use of others' content. Thus, the former typically aligns with original content, while the latter aligns with sourced content. Consequently, Originality2 serves as a reliable proxy for assessing video originality, aiding in the differentiation between original and sourced content.
The results utilizing this more precise outcome variable are presented in Table A2. Regulation continues to be denoted as 1 for videos created following the launch of the COIP. The control variables remain consistent with those employed in the baseline regressions. Notably, the coefficient of Regulation is significantly positive, indicating that enhanced copyright protection may contribute to an increase in the originality of the videos, which further reinforces the robustness of our conclusions.
Table A2. Robustness check: Adjust the independent variable
Ke Rong ([email protected]) currently serves as the Director of the Institute of Economics at Tsinghua University, China. He earned his PhD from the University of Cambridge after completing his bachelor's degree at Tsinghua University. His primary research interests include business and innovation ecosystems, the digital economy, and AI and data ecosystems. He has authored over 70 peer-reviewed articles in esteemed journals such as the Journal of International Business Studies, Production and Operations Management, and others.
Jinglei Huang ([email protected]) is a PhD candidate in Economics at the School of Social Sciences, Tsinghua University, after completing his bachelor's degree at the School of Economics and Management, Tsinghua University. His research interests include the platform economy, economic growth, the digital economy, and innovation. His work has been published in Management and Organization Review, Humanities and Social Sciences Communications, Applied Economics Letters, and the Journal of Digital Economy.
Fei Hao ([email protected]) is an assistant professor at the Business School, University of International Business and Economics, China. He earned his PhD degree in Economics from the School of Social Sciences, Tsinghua University. His primary research interests lie in the digital economy, and innovation and entrepreneurship. He currently serves on the Editorial Board of the Journal of Digital Economy.
Danxia Xie ([email protected]) is currently a tenured associate professor in Economics at Tsinghua University. He holds a PhD in Economics from the University of Chicago, a master's degree in Public Policy from Harvard University, and a master's degree in Computer Science from Duke University. His teaching and research focus on digital economics, law & economics, macroeconomics, and finance. He has worked at Peterson Institute for International Economics, a top think tank in Washington, DC.
Sali Li ([email protected]) is a Moore Professor at the Sonoco International Business Department at the Darla Moore School of Business. His research on the internationalization of digital innovation was awarded the Rugman Prize by the AIB. His research has won the Lazardis Award for JBV's Best Paper, Temple/AIB Best Paper Award, etc. He also serves as a consulting editor for the Journal of International Business Studies, and an associate editor for the Journal of World Business.
