home. cv. research. teaching. thoughts. personal. failed projects.

research

My dissertation is on learning to design, deliver, and diffuse interventions.

working papers:

First Law of Motion:
Influencer Video Advertising on TikTok
Jeremy Yang^*, Juanjuan Zhang, Yuhan Zhang

[paper] [slides] [tweet]

Under review at Marketing Science

- MSI Alden G. Clayton Doctoral Dissertation Proposal Award (2021/5)
- MIT Sloan Doctoral Research Forum Thesis Prize (2021/4)

Abstract: This paper develops an algorithm to predict the effect of influencer video advertising on product sales. We propose the concept of motion-score, or m-score, a summary statistic that captures the extent to which a product is advertised in the most engaging parts of a video. We locate pixel-level product placement with an object detection algorithm and estimate pixel-level engagement as a saliency map by fine-tuning a deep 3D convolutional neural network on video-level engagement data. M-score is then defined as pixel-level engagement-weighted advertising intensity of a video. We construct and evaluate the algorithm with around 40,000 influencer video ads on TikTok, the largest short video platform of the world. We leverage variation in video posting time to identify the causal effect of video ads on product sales. Videos of higher m-score indeed lift more sales. This effect is sizable, robust, and more pronounced among impulsive, hedonic, or inexpensive products. We trace the mechanism to influencers’ incentives to promote themselves rather than the product. We discuss how various stakeholders in entertainment commerce can use m-score in a scalable way to optimize content, align incentives, and improve efficiency.

presentations:
- ISMS Marketing Science Conference, University of Rochester (2021/6)
- Theory + Practice in Marketing, University of Pennsylvania (2021/6)
- Initiative on the Digital Economy Annual Conference, MIT (2021/5)
- Artificial Intelligence in Management Conference, USC (2021/5)
- Sloan Doctoral Research Forum, MIT (2021/4)
- Paris Conference on Digital Economics Workshop (2021/4)
- Management Workshop, Peking University (2021/3)
- Rotterdam School of Management, Erasmus University (2021/3)
- Columbia Business School, Columbia University (2021/2)
- ICCIT & Rotman School of Management, University of Toronto (2021/2)
- Darden School of Business, University of Virginia (2021/2)
- School of Business, University of Washington (2021/1)
- Cheung Kong Graduate School of Business (2021/1)
- Harvard Business School, Harvard University (2021/1)
- HEC Paris & Hi! Paris Center on AI and Data Analytics (2021/1)
- Krannert School of Management, Purdue University (2021/1)
- College of Business, City University of Hong Kong (2021/1)
- Center For Big Data in Mobile Analytics, Temple University (2020/12)
- Conference on AI/ML, NYU, CMU & Temple University (2020/12)
- HKU Business School, University of Hong Kong (2020/11)
- Conference on Digital Experimentation, MIT (2020/11)
- Marketing Seminar, MIT (2020/11)
- HBS Digital Doctoral Workshop, Harvard University (2020/11)

Targeting for Long-Term Outcomes
Jeremy Yang, Dean Eckles, Paramveer Dhillon, Sinan Aral

[paper] [slides] [talk] [lecture] [tweet] [media]

Minor revision at Management Science

- American Statistical Association Dissertation Proposal Award (2021/1)
- INFORMS Annual Meeting Best Paper Award (2020/11)
- Finalist of WISE Best Student Paper Award (2020/11)
- Accepted for Presentation at the QME Conference (2020/10)
- Accepted for Presentation at NeurIPS (CausalML Workshop) (2019/12)

Abstract: Decision-makers often want to target interventions (e.g., marketing campaigns) so as to maximize an outcome that is observed only in the long-term. This typically requires delaying decisions until the outcome is observed or relying on simple short-term proxies for the long-term outcome. Here we build on the statistical surrogacy and off-policy learning literature to impute the missing long-term outcomes and then approximate the optimal targeting policy on the imputed outcomes via a doubly-robust approach. We apply our approach in large-scale proactive churn management experiments at The Boston Globe by targeting optimal discounts to its digital subscribers to maximize their long-term revenue. We first show that conditions for validity of average treatment effect estimation with imputed outcomes are also sufficient for valid policy evaluation and optimization; furthermore, these conditions can be somewhat relaxed for policy optimization. We then validate this approach empirically by comparing it with a policy learned on the ground truth long-term outcomes and show that they are statistically indistinguishable. Our approach also outperforms a policy learned on short-term proxies for the long-term outcome. In a second field experiment, we implement the optimal targeting policy with additional randomized exploration, which allows us to update the optimal policy for each new cohort of customers to account for potential non-stationarity. Over three years, our approach had a net-positive revenue impact in the range of $4-5 million compared to The Boston Globe’s current policies.

presentations:
- Center for Causal Inference Symposium, RAND Corporation (2021/6)
- American Economic Association Annual Meeting (2021/1)
- School of Business, Hebrew University of Jerusalem (2020/12)
- Workshop on Information Systems and Economics (2020/12)
- Inference and Statistics Reading Group, Lyft (2020/12)
- INFORMS Annual Meeting (2020/11)
- International Conference on Computational Social Science, MIT (2020/7)
- Initiative on the Digital Economy Annual Conference, MIT (2020/5)
- Marketing Seminar, MIT (2020/5)
- NeurIPS CausalML Workshop (2019/12)
- HBS Digital Doctoral Workshop, Harvard University (2019/12)
- Conference on Digital Experimentation, MIT (2019/11)
- Advances on Field Experiments Conference, University of Chicago (2019/7)

Identification and Bias-Amplification:
Latent Space Approach to Social Contagion on Observational Networks
Jeremy Yang

Abstract: In social networks, ego behavior is usually a function of the behaviors of her alters’. However, such social contagion or peer effect is hard to identify empirically using observational networks due to endogeneity in tie formation. The latent space models have been proposed under the assumption of assortative mixing as a method of modeling the underlying tie generating process, and it also has been used to adjust for unobserved homophily. At the meantime, Pearl (2009, 2010) suggested that adjusting for one common cause of treatment and outcome might actually increase the net bias by amplifying the bias introduced by an unadjusted confounder. The first part of the paper is on identification: I fit a latent space model to a classic dataset on the diffusion of medical innovation (Coleman et al., 1957,1966) to identify social contagion. The second part is on bias-amplification: I extend Pearl’s framework to a nonlinear system and then use simulation to demonstrate that latent space adjustment can sometimes amplify the net bias, thus it should be used with caution.

Uncertainty and Information Diffusion
T. Tony Ke, Jeremy Yang^*

Abstract: We investigate how uncertainty affects information diffusion. We analyze a dataset that contains over 2400 rumors about 5 breaking news on Twitter from 2014-2015 where rumors are verified to be true or false at different time. We exploit the difference in the timing of verification to estimate the causal effect of uncertainty reduction on diffusion via difference-in-differences. We find that in some events the verification decreases diffusion even for rumors that are true. This motivates a microfounded social learning model on the network that highlights the mechanism through which uncertainty shapes individual’s incentive to share and thereby changing the pattern of diffusion. The model also offers a potential explanation to why rumors tend to diffuse wider and faster than verified news, even for information with similar content.

presentations:
- Marketing Seminar, MIT (2017/11)

Award No Longer Motivates Once You Are Awarded:
A Field Experiment in Online Learning
Fan Bi, Qiang Feng, Jeremy Yang^*

Abstract: This paper studies the effect of social recognition and symbolic award on students’ effort and performance in an online English course in China. Students are randomly assigned to classes, study the materials and take quizzes on a daily basis. We conduct a two stage randomized experiment in which we first randomly assign classes to two treatment groups (pre-announced private or public award given out every week) and a control group, then within each treated class we randomly assign some students to actually receive the award conditional on their performance in the past week. We find that students who received the award exert less effort in the future compared to students with similar past performance but did not receive the award due to randomization. They also do not exert more effort in the future than students with similar past performance in the control group. Students who did not receive the award due to randomization exert more effort and score higher in the future compared to students with similar past performance in the control group. There’s no difference between public and private recognition. Taken together, our results suggest that it is the possibility of being awarded (ex ante) that increases future effort and performance, receiving the award (ex post) actually lowers future effort. In other words, symbolic awards don’t have a continuing motivating effect after you are awarded.

presentations:
- Organizational Economics Lunch, MIT (2020/10)

publications:

Interdependence and the Cost of Uncoordinated Responses
to COVID-19
David Holtz, Michael Zhao, Seth G. Benzell, Cathy Y. Cao, M. Amin Rahimiana, Jeremy Yang, Jennifer Allen, Avinash Collis, Alex Moehring, Tara Sowrirajan, Dipayan Ghosha, Yunhao Zhang, Paramveer S. Dhillon, Christos Nicolaides, Dean Eckles, Sinan Aral

Proceedings of the National Academy of Sciences (2020) [paper]

Abstract: Social distancing is the core policy response to COVID-19. But as federal, state and local governments begin opening businesses and relaxing shelter-in-place orders worldwide, we lack quantitative evidence on how policies in one region affect mobility and social distancing in other regions and the consequences of uncoordinated regional policies adopted in the presence of such spillovers. We therefore combined daily, county-level data on shelter-in-place and business closure policies with movement data from over 27 million mobile devices, social network connections among over 220 million of Facebook users, daily temperature and precipitation data from 62,000 weather stations and county-level census data on population demographics to estimate the geographic and social network spillovers created by regional policies across the United States. Our analysis showed the contact patterns of people in a given region are significantly influenced by the policies and behaviors of people in other, sometimes distant, regions. When just one third of a state’s social and geographic peer states adopt shelter in place policies, it creates a reduction in mobility equal to the state’s own policy decisions. These spillovers are mediated by peer travel and distancing behaviors in those states. A simple analytical model calibrated with our empirical estimates demonstrated that the “loss from anarchy” in uncoordinated state policies is increasing in the number of non-cooperating states and the size of social and geographic spillovers. These results suggest a substantial cost of uncoordinated government responses to COVID-19 when people, ideas, and media move across borders.

How Do Successful Scholars Get their Best Research Ideas?
An Exploration
Cathy Cao, Xinyu Cao, Matthew Cashman, Madhav Kumar, Artem Timoshenko, Jeremy Yang^*, Shuyi Yu, Jerry Zhang, Yuting Zhu, Birger Wernerfelt

Marketing Letters (2019) [paper]

Abstract: We interview 24 marketing professors to ask how they got the ideas for 64 of their papers. More than three quarters of the papers were inspired by holes in the literature, by a “stylized fact” that the current literature cannot explain, or by an interaction with a manager. The rest fall into several smaller categories that to a large extent can be seen as special cases of the three big ones. We describe how papers from each of the three big categories help move the literature forward. We also illustrate the range of situations contained in each category by way of several examples. Among the authors we interview, most do not use a single source. As these authors become more senior, managerial contacts play an increasing role, while the balance between literature and stylized facts appears to be unchanged.

^*Authors are listed in an alphabetic order

home