Ming D. Leung

A field experiment utilizing computer vision and machine learning on an online platform for temporary staffing of low wage jobs (with PhD students Richard Lu and Sibo Lu) Data Analysis in ProcessLabor market discrimination against African American and Hispanic job applicants continues to persist. A job applicant's race may be correlated with how responsible they present themselves, affecting their likelihood of being hired. Unfortunately, causal identification of this mechanism is difficult to isolate and interventions are difficult to implement. We address these challenges with a field-experiment on a mobile platform for employers seeking to hire local unskilled and low-skilled workers on a temporary basis. Employers choose which applicants to hire after viewing their photos online.

Using computer vision and machine learning methodologies, we identified the features of applicant photos that accounted for them being considered more versus less responsible by survey respondents. Results of the machine learning yielded specific changes, such as to wear a tie or to smile, to improve perceptions of responsibility.

We randomly divided African American and Hispanic applicants scoring low on perceived responsibility into two groups. One received a ‘nudge’ from the platform to improve their profile photos. The other served as a control. Nudged subjects improved the responsibility score of their photos. Both within person and between control and treatment group analyses reveal an improvement in the hiring outcomes of those who had been nudged to make improve their profile photos. Our study informs debates as to the mechanisms underlying biased hiring outcomes as well as potential interventions.

An investigation into how online labor market platforms could increase startup activity in the United States heartland and decentralize entrepreneurship ecosystems ** Supported with a Research Grant from the Kauffman Foundation **(with Weiyi Ng, Olenka Aleksandra Kacperczyk, Sampsa Samila and Sibo Lu)

Data Analysis in Preparation

A fundamental challenge to any entrepreneurial venture is the hiring of skilled employees. In order for startups to scale up beyond merely an idea requires founders to identify, hire, and retain individuals who possess very specific and often valuable skills. This is particularly difficult for startups which may have little in the way of location advantages such as propinquity to a large and talented labor pool, relationships with a breadth of skilled workers, or deep pockets of funding. This observation may be one reason why regional dynamics play such a large role in explaining entrepreneurial founding. Simply put, propinquity matters - you need to be near resources to use them.

This dynamic, however, may be changing. Specifically, the rise of virtual contract labor markets, such as UpWork, has eased the hiring of skilled workers in domains as diverse as programming, design, legal, and business services. Any would be entrepreneur, regardless of where they are located or who they know, is now able to quickly and inexpensively tap a global talent pool of skilled freelancers on demand to fill any talent or labor gap

We seek to link geographical founding datasets to activity and adoption data of the online platforms Elance and oDesk (now UpWork). Doing so will create a dataset that allows us to examine both temporal and geospatial dynamics of remote hiring and entrepreneurship. Our current plan involves two datasets: the first is historical, online behavioral data collected through UpWork. The second involves a sample of 33 million online resumes retrieved from the public front of LinkedIn in 2013. The online resumes allow us to observe founding events of all sorts: ranging from contract self-employment to small-medium enterprises to high potential ventures. Importantly, these resumes also allow us to define the base-rate and risk set of entrepreneurs at each location.

A quasi-experimental investigation of how contention in ratings increases contributions to online movie opinion platforms (with USC PhD Student Jue Wang) Manuscript in PreparationWhat drives people to contribute to online opinion forums? We propose that the level of contention that is visible in the discourse leads to increased contribution by others. This theory lies in contrast to the alternative explanation that more contentious topics of discussion necessarily lead to both more contentious opinions being expressed as well as increased contributions. We utilized a quasi-experimental paradigm to causally identify the effects of contention. We matched identical movies on two different movie review platforms (Metacritic and Rotten Tomatoes) to account for difference in the underlying heterogeneity of the contentiousness of the underlying movie. The difference in the level of polarization in posted ratings in one day was used to predict the number of ratings posted the next day. Results reveal that the greater polarization of ratings in one day leads to a greater number of ratings posted the next day. Furthermore, results of text analyses measuring how much subsequent reviews reference previous ones demonstrate that contention leads reviewers to be more likely to engage in discourse with both those that agree and disagree with their ratings. Implications for online reviews and collective production are discussed.