Three Essays on Learning and Collaboration in Operations Management


In knowledge-intensive work such as data analysis or emergency medicine, learning and collaboration is integral to the effective “production” process. Uncovering the mechanisms, benefits and costs of learning and collaboration is crucial for the design and operation of a successful process. My Ph.D. research centers on empirically investigating the ways in which learning and collaboration among people affect the performance of the processes they execute. In this collection of essays, I study some of the effects of three modes of learning and collaboration (learning by doing, learning from peers and supervision) on operational performance, together with my advisors at Kellogg and collaborators at eBay and NorthShore Health System. This dissertation incorporates two published papers: Chapter 1 uses material from Yin et al. (2018) while Chapter 3 uses material from Wang et al. (2019). In Chapter 1, we investigate how data-analyst productivity benefits from collaborative platforms that facilitate learning-by-doing (i.e. analysts learning by writing queries on their own) and learning-by-viewing (i.e. analysts learning by viewing queries written by peers). Learning is measured using a behavioral (productivity-improvement) approach. Productivity is measured using the time from creating an empty query to first executing it. Using a sample of 2,001 data analysts at eBay Inc. who have written 79,797 queries from 2014 to 2018, we find that: 1) learning-by-doing is associated with significant productivity improvement when the analyst's prior experience focuses on the focally queried database; 2) only viewing queries that are authored by analysts with high output rate (average number of queries written per month) is associated with significant improvement in the viewer's productivity; 3) learning-by-viewing also depends on the ``social influence'' of the author of the viewed query, which we measure `locally' based on the number of the author's direct viewers per month or `globally' based on the how the author's queries propagate to her peers in the overall collaboration network. Combining results 2 and 3, when segmenting analysts based on output rate and `local' social influence, the viewing of queries authored by analysts with high output but low local influence is associated with the largest improvement in the viewer's productivity; whereas when segmenting based on output rate and `global' social influence, the viewing of queries authored analysts with high output and high global influence is associated with the largest improvement in the viewer's productivity. Overall, regardless of the segmentation, learning-by-viewing is associated with greater productivity improvement than learning-by-doing in our study. In Chapter 2, we investigate whether an individual's productivity and learning ability vary over time. A Hidden Markov Model (HMM) is proposed to capture such dynamics and is applied to the same empirical setting as in chapter 1. This model enables us to segment data analysts into several latent states with respect to their productivity and learning ability. These analysts are allowed to transit between states, the direction and probability of which depend on their participation in two modes of learning activities (writing own queries and viewing peers' queries). The effect of an analyst's participation in either kind of learning activity also varies by her state. We find that the three-state HMM model fits better than the standard learning curve model in relevant measures. This implies that the dynamic model is more appropriate to capture the evolvement of analysts' productivity and learning. We have identified three latent states (novice, intermediate and advanced) in ascending order of the intrinsic productivity of analysts. Our findings reveal different patterns of learning in different states. Only analysts in the novice state benefit from both writing own queries and viewing peers' queries. The learning effects from writing own queries decreases with higher state. We also find that analysts in the intermediate state or above risk transiting to lower states by viewing peers' queries. In Chapter 3, we perform an observational, time-motion study on 25 EPs who worked in a community-academic ED and a non-academic community ED. The objective is to compare attending emergency physician (EP) time spent on direct and indirect patient care activities in emergency departments (ED) with, and without, emergency medicine (EM) residents. Two observations of each EP were performed at each site. Average time spent per 240-minute observation on main-category activities are summarized in percentages. We report descriptive statistics (median and interquartile ranges) for the number of minutes EPs spent per sub-category activity, in total and per patient. We performed a Wilcoxon two-sample test to assess differences between time spent across two EDs. The results show that the 25 observed EPs executed 34,358 tasks in the two EDs. At the community-academic ED, EPs spent 14.2% of their time (8.5 minutes/hour) supervising EM residents. Supervision activities included data presentation, medical decision making, and treatment. The time spent on supervision is offset by a decrease in time spent by EPs on indirect patient care at the community-academic ED compared to the non-academic community ED, specifically from communication and EHR work. There was no statistical difference with respect to direct patient-care time expenditure across two EDs. There was a non-statistically significant difference in attending patient load between sites.

Date created
Resource type
Rights statement