Seminars 2020-2021

A widespread belief is that the traditional brick-and-mortar K–12 education system in the US is broken. The private sector, specifically education technology (EdTech) companies, have stepped in to try to help. In this paper, we study DonorsChoose, a nonprofit that works to improve the traditional brick-and-mortar system with a teacher crowdfunding platform. Given the constraints of working with the current struggling system, we ask whether DonorsChoose moves the needle on effectiveness and inequality. Combining DonorsChoose data with data on student test scores in Pennsylvania from 2012–2013 to 2017–2018, we find an increase in the number of DonorsChoose projects funded at a school leads to higher student performance, after controlling for selection biases. In high schools, a 10% increase in the number of funded projects leads to a 0.1 to 0.2 percentage point (pp) increase in students scoring basic and above in all tested subjects. A 10% increase in the number of funded projects at an elementary or middle school leads to a 0.06 pp increase in the percentage of students scoring basic and above in language arts and a 0.15 pp increase in science. We find these effects are driven primarily by teacher projects from the lowest income schools, suggesting the platform helps reduce inequality in educational outcomes. Based on a textual analysis of thousands of statements from all funded teachers describing how resources are used, we find two channels of improvement uniquely effective in the lowest income schools. Our study suggests that those in the education sector can harness the wisdom of front-line workers — teachers — to improve effectiveness, efficiency, and equity.

Link to paper: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3556208

https://michiganross.umich.edu/faculty-research/faculty/samantha-keppler

We use a large and unique corpus of job specifications for C-suite positions to document and explain heterogeneity in the skills demanded for high-level executives across firms. A novel algorithm maps the text for each executive search into six separate skill clusters that reflect cognitive, interpersonal, and operational dimensions. Patterns in the social skills cluster are particularly striking: it features the highest growth in the sample, it is the relatively most common cluster in CEO searches, and is very heterogeneous across firms. We propose a mechanism whereby executive social skills facilitate the exchange of problems between workers and managers; construct proxies for the need for such coordination; and show these correlate with the presence of social skills language. The results suggest that the varied structure of firms induces demand for different executive skills.

https://www.hbs.edu/faculty/Pages/profile.aspx?facId=541712

We provide a theory that unifies the blockbuster and long tail phenomena. Specifically, we analyze a model where a large number of firms compete in making market entry and product quality decisions and then sequentially arriving customers with (random) private preferences make purchase decisions based on product quality and historical sales. We show that a growing network effect always contributes to more sales concentration on a small number of products, supporting the blockbuster phenomenon. However, product variety and investments in quality, as an outcome of firms’ ex ante competitive decisions, may increase or decrease as the network effect grows. If the strength of the network effect is below a threshold, an increasing network effect will shift more sales towards the products with higher quality, preventing more products from entering the market ex ante and inducing firms to adopt the blockbuster strategy by making high-budget products. Otherwise, the network effect will easily cause the market to be concentrated on a few products; even some low-quality products may have a chance to become a “hit” due to luck. In this case, when the network effect is growing, the ex-ante equilibrium product variety will be wider, and firms make lower-budget products, a finding that is consistent with the long tail theory. We test our theory with the movie box office data and find strong supporting evidence.

https://www.rotman.utoronto.ca/FacultyAndResearch/Faculty/FacultyBios/Hu

Although it is well-established that people’s opinions and tastes are deeply divided along political lines, less is known on what can bring liberals and conservatives together. In this research, we examine whether and how counter-stereotypical political endorsement (i.e., individuals who endorse [reject] a policy that most from their own group are perceived to reject [endorse]) can help reduce political polarization. Across three studies conducted within the Brazilian context, we show that relative to controls and stereotypical endorsers, counter-stereotypical endorsers attenuate the association between an individual’s self-reported political orientation and his/her policy preferences. This is true for cannabis legalization (study 1), gun rights (study 2), and abortion (study 3). As important, we reveal that the attenuation in polarization is asymmetric. When a counter-stereotypical endorser supports a given policy (e.g., a conservative politician supporting cannabis legalization), it persuades in-groups (e.g., increasing support for cannabis legalization among conservatives) more than it dissuades out-groups (e.g., reducing support for cannabis legalization among liberals). The role of changes in beliefs about (a) policy effectiveness and (b) in-group social acceptance is discussed.

https://professor.fgv.br/ebape/eduardo-b-andrade

Forthcoming

In many online platforms, customers’ decisions are substantially influenced by product rankings as most customers only examine a few top-ranked products. Concurrently, such platforms also use the same data corresponding to customers’ actions to learn how these products must be ranked or ordered. These interactions in the underlying learning process, however, may incentivize sellers to artificially inflate their position by employing fake users, as exemplified by the emergence of click farms. Motivated by such fraudulent behavior, we study the ranking problem of a platform that faces a mixture of real and fake users who are indistinguishable from one another. We first show that existing learning algorithms—that are optimal in the absence of fake users—may converge to highly sub-optimal rankings under manipulation by fake users. To overcome this deficiency, we develop efficient learning algorithms under two informational environments: in the first setting, the platform is aware of the number of fake users, and in the second setting, it is agnostic to the number of fake users. For both these environments, we prove that our algorithms converge to the optimal ranking, while being robust to the aforementioned fraudulent behavior; we also present worst-case performance guarantees for our methods, and show that they significantly outperform existing algorithms. At a high level, our work employs several novel approaches to guarantee robustness such as: (i) constructing product-ordering graphs that encode the pairwise relationships between products inferred from the customers’ actions; and (ii) implementing multiple levels of learning with a judicious amount of bi-directional cross-learning between levels.

Link to the paper: https://arxiv.org/abs/2009.05138

http://www.mit.edu/~golrezae/

High impact / low probability supply chain disruptions can pose major challenges for a firm. When the firm can backlog some portion of its unmet customer orders, such disruptions represent a form of bottleneck shifting because the firm’s constrained resources change over time. The firm’s disruption exposure is driven by (i) lost orders in the disruption stage due to part constraints and (ii) lost orders in the recovery stage due to production capacity constraints. The firm would like to reduce its disruption exposure in both stages without incurring material incremental costs. To be practically useful, a solution must account for the operational reality that the firm’s part inventories are constantly changing. We show that this complexity can be exploited to solve the firm’s problem. First, we analytically prove that the firm’s disruption exposure is (i) decreasing at a decreasing rate with the inventory quantity of the disrupted part and (ii) decreasing at a decreasing rate with the inventory quantity of the non-disrupted parts. We then develop an optimization model to examine the practical implications of these effects using detailed data from our research partner, a large diversified manufacturing firm (DMF). With targeted changes to its inventory policies across a set of parts, DMF can reduce its aggregate disruption exposure by 52.6% to 55.4% while reducing its inventory holding and ordering costs by 1.7% to 8.1%. Achieving these results, however, requires inventory policy changes to many parts, and trades off disruption exposure decreases to some parts with increases to other parts. We introduce an alternative solution, a “strategic portfolio” of parts, that is simpler to implement, also inexpensive, and allows the firm to materially reduce its disruption exposure across all parts. Holding inventory is known to mitigate a firm’s disruption exposure, but it is perceived to be costly relative to other mitigation strategies, such as supply chain insurance or cultivating alternative source of supply. Our part portfolio approach overcomes these limitations, thereby adding a new strategy to management’s arsenal of disruption risk mitigation options.

https://www.johnson.cornell.edu/faculty-research/faculty/ws366/

Problem definition: This paper describes the real-time participatory modeling work that our team of academics, public health officials, and clinical decision-makers has been undertaking to support the regional efforts to tackle COVID-19 in the East of England. Methodology: Since March 2020, we have been studying four research questions that have allowed us to address the pandemic’s current and near-future rapidly evolving epidemiological state, as well as the bed capacity demand in the short (a few weeks) and medium (several months) term. Frequent data input from and consultations with our public health and clinical partners allow our academic team to apply dynamic data-driven approaches using time series modeling, Bayesian estimation, and system dynamics modeling. We thus obtain a broad view of the evolving situation. Results: The academic team presents the model outcomes and insights during weekly joint meetings among public health services, national health services, and academics to support COVID-19 planning activities in the East of England, contributing to the discussion of the COVID-19 response and issues beyond immediate COVID-19 planning. Academic/practical relevance: As COVID-19 planning efforts necessitate rapid response, our portfolio of scratch models aims to achieve the right balance between rigor and speed in the face of an uncertain and changing situation. Managerial implications: Our regional and local focus enables us to better understand the pandemic’s progression and to help decision-makers make more informed short- and medium-term capacity plans in different localities in the East of England. In addition, the learnings from our collaborative experiences may present guidance on how academics and practitioners can successfully collaborate in rapid response to disasters such as COVID-19.

(In collaboration with Cambridge Judge Business School Covid-19 Planning team and Public Health England)

https://www.jbs.cam.ac.uk/faculty-research/faculty-a-z/feryal-erhun/

Six studies show that majority group members misperceive diversity policies as unbeneficial to their ingroup, even when policies benefit them. Majority members perceived non-zero-sum university admission policies—policies that increase the acceptance of both URM (i.e., underrepresented minority) and non-URM applicants—as harmful to their ingroup when merely framed as “diversity” policies. Even for policies lacking diversity framing (i.e., “leadership” policies), majority members misperceived that their ingroup would not benefit when initiatives provided relatively greater benefit to URMs, but not when they provided relatively greater benefit to non-URMs. No evidence emerged that these effects were driven by ideological factors:  Majority members’ misperceptions occurred even when accounting for beliefs around diversity, groups, hierarchy, race, and politics. Instead, we find that majority group membership itself predicts misperceptions, such that both Black and White participants accurately perceive non-zero-sum policies as also benefiting the majority when participants are represented as a member of the minority group.

https://haas.berkeley.edu/faculty/jacoby-senghor-drew//k/l/g.a.vankleef/g.a.vankleef.html?cb

Billions of tons of solid waste are generated every year globally, estimated at 60M tons of electronic waste, 730M tons of medical waste, 1B tons of hazardous waste, and 2B tons of municipal solid waste (World Bank Group 2018). Not all this waste is properly treated in the country of origin; rather, it is either dumped locally or exported. While some value-added recovery happens in export locations when high-value waste is exported, when this waste is low-quality, unwanted waste, it leads to a myriad of health and environmental problems in the destination country. In this talk, I will draw on two papers addressing these issues. “Truthful Mechanisms for Medical Product Surplus Allocation” (Zhang, Atasu, Ayer, Toktay) is motivated by the estimated 6M tons of medical surplus waste generated in the US annually, some of which is exported by Medical Surplus Recovery Organizations (MSROs) to under-resourced hospitals abroad. Yet the World Health Organization estimates that over seventy percent of donated medical equipment was inappropriate. In this paper, we analyze a resource allocation problem faced by an MSRO under information asymmetry regarding the needs of a heterogeneous set of recipients. We identify implementable strategies to support recipient selection decisions that maximizes value to recipients. “Treat, Dump, or Export? How Domestic and International Waste Management Policies Shape Waste Chain Outcomes” (Wijnsma, Lauga, Toktay) turns to the regulatory environment and studies the role of anti-dumping and anti-export policies in shaping waste outcomes under double-sided information asymmetry between waste producers and waste treatment operators.

https://www.scheller.gatech.edu/directory/faculty/toktay/index.html

What does it mean to be human? Seven studies explore this age-old question and show that the attribution of gender is a critical component of seeing human (i.e., anthropomorphizing). Given gender’s primacy in social cognition, we propose gender is linked to seeing human in a way that cannot be said of other social categories (race, age, sexual-orientation, religion, disability). We test this hypothesis in seven studies: six that induce humanization (i.e., anthropomorphism) and measure social-category ascription; and one that includes (versus removes) gender and measures humanization. From recalling personal experience (Study 1), to perceiving “human-like” movement (Study 2a–2b), to anthropomorphizing stimuli (Study 3a–3b), and even creating a “human form” (Study 4), we demonstrate the heightened tendency to see gender (versus other social categories) when anthropomorphizing non-human entities, and further show that gender ascription does not happen when merely describing them (Studies 2–4). In addition, we show the reciprocal process, where assigning an anthropomorphized entity with a gender increases its humanness (Study 5). These results highlight the fundamental role of gender in humanization and have theoretical implications for research on gender, anthropomorphism, and mind perception. Further, these findings have practical relevance for current discussions around “genderlessness” and the rapidly growing movement towards a genderless society.

https://www.gsb.stanford.edu/faculty-research/faculty/ashley-martin

Leadership isn’t about you. It’s about how effective you are at empowering other people—and making sure this impact endures even in your absence. The origins of great leadership are found, paradoxically, not in worrying about your own status and advancement, but in the unrelenting focus on other people’s potential.

The session will show how the boldest, most effective leaders use a special combination of trust, love, and belonging to create an environment in which other people can excel.

https://www.hbs.edu/faculty/Pages/profile.aspx?facId=6587

Existing research illustrates a contingent association between a team’s work assignment and the ideal network for superior performance. For a basic task, the ideal network is organized around a central individual. If the task is complex, the ideal network is decentralized and democratic. Language similarity is one reason why complex work requires a decentralized network. To perform well, team members must apply the same problem-solving framework, and decentralized teams have an advantage in reaching consensus. Recent research suggests that language similarity is more beneficial for performance when a network is centralized. The implications of this potential outcome are underappreciated. Even if centralized teams struggle to agree on what framework to use, if the performance implication of language similarity is larger in a centralized team, centralized teams could be preferable, even if the focal task is complex. We analyze the performance of seventy-seven teams working to identify abstract symbols, which is a complex task and which also requires language similarity. People are randomly assigned to different network conditions and work together for a number of trials. We find that language similarity improves with experience at a slower rate in centralized teams, but we also find that the language similarity effect on team performance is larger for centralized teams, large enough to shift the overall advantage to centralized teams. We also estimate the performance of teams working in networks that combine elements of centralized and decentralized networks. Performance is higher in teams that combine both network features.

https://mitsloan.mit.edu/faculty/directory/ray-reagans

We consider a parallel processing network with removable servers. Beginning with the single server model with power and service rate control, we study the importance of a delayed restart when the server is off. In particular, we show that an optimal policy exists (under the average cost criterion) that delays restarting until a “safety stock” of work is in the system. It then behaves similarly to that of the classic service rate control models. With that as the backdrop, we consider scheduling with the ability to remove servers. We introduce “delay-JSQ” (join the shortest queue) policies, show their stability and asymptotic optimality in the two-server case, and conclude with a detailed numerical study that shows they outperform JSQ by up to 80%. This is joint work with Professor Douglas Down from McMaster University and Dr. Pamela Badian-Pessot (now at Proctor and Gamble).

https://www.engineering.cornell.edu/faculty-directory/mark-e-lewis

Norms uphold the social order by guiding behavior without the force of laws. Behaviors that violate norms therefore pose a potential threat to organizations and societies. Accordingly, norm violations often trigger negative reactions in observers, such as unfavorable impressions, moral outrage, and gossip. Despite these reputational detriments, norm violations are omnipresent. I propose that one reason why norm violations persist is that the readiness to violate norms – despite potential costs in the form of sanctions – serves as a social signal that can afford benefits for actors’ social rank. Against this theoretical background, I will first present evidence that individuals who violate social norms are perceived as more powerful than those who abide by norms. Second, I will elucidate when observers are willing to actively grant power to norm violators by supporting them as leaders. A number of interrelated research projects indicate that the effects of norm violations on leadership granting depend on (1) the prosociality of the norm violation (i.e., whether it benefits others), (2) the hierarchical position of the observer (i.e., high or low power or social-economic status), (3) the type of norm that is violated (i.e., societal level or group level), and (4) the cultural context of the norm violation (i.e., variations in individualism-collectivism and tightness-looseness). I will discuss implications of these findings for understanding the perpetuation of norm-violating behavior in organizations and society at large.

https://www.uva.nl/profiel/k/l/g.a.vankleef/g.a.vankleef.html?cb

Using data from 38,916 businesses in war-torn Colombia and from 5,138 attacks by the two rebel groups, FARC and ELN, we study how firms manage inventory during civil war. We obtain exogenous variation in the conflict intensity via a difference-in-differences model, which hinges on the peace process between Colombia’s government and FARC. Relying on this identification strategy, we hypothesize and show that war causes two effects on firm-level inventories. First, it leads firms to replace physical assets (inventory) with fungible assets (cash), causing them to operate with an over-secured financial buffer but a fragile operational buffer. Second, this inventory reduction occurs mostly in unprocessed inventories (finished-goods inventories are insensitive to violence), meaning that whereas war-torn businesses are equipped to fulfill planned orders, they become inflexible at handling uncertain future demand. We then show that the magnitude of these effects is highly contingent on the firm’s position in the supply chain, its proximity to distribution markets, and the type of attacks it is subject to. We propose policies to address war-related risk in supply chains.

https://www.mcgill.ca/desautels/juan-camilo-serpa

In the first, modeling, part of this work I will discuss the tradeoff between accuracy and availability of tests and show how the accuracy of a test in detecting the underlying state affects the demand for the information product differentially across heterogeneous agents. Correspondingly, the test accuracy can serve as a rationing device to ensure that the limited supply of information products is appropriately allocated to the heterogeneous agents. When test availability is low and the social planner is unable to allocate tests in a targeted manner to the agents, we find that moderately good tests can outperform perfect tests in terms of social outcome.

In the second part of the talk, I will discuss the work that we recently completed with screening travelers at the Greek Border. From July 1st to November 1st we designed, implemented and deployed an online learning system to allocate the country’s limited testing resources on the incoming tourist population. Specifically, for each of the 40 points of entry and given the daily number of tests available, we use travelers’ characteristics to decide who to test. Using this approach, we essentially double the effectiveness of testing resources and provide early warnings for outbreaks around the world.

https://www.marshall.usc.edu/personnel/kimon-drakopoulos

Matching demand with supply has been a long-standing challenge in operating residential electricity markets. The utility firms often face stochastic demand functions that are affected by the unpredictable exogenous random shocks (e.g., outdoor weather condition). Although various Demand Response programs are in place to regulate electricity consumption, the effectiveness of these programs has been undermined, largely because the consumers have demonstrated limited capability in adjusting their household appliances’ settings. In this paper, we construct a demand model to describe how consumers make consumption decisions in response to random external factors representing their ambient environment at a given price. To that end, we adopt the notion of “rational inattention” to capture the consumers’ inertia in readjusting their decisions over time. Subsequently, we investigate an electricity firm’s pricing decision as well as the important role of smart appliances in driving the overall consumption patterns. Our findings highlight the nuanced implications of rationally inattentive consumers, and lead to guidelines for better regulating retail electricity markets.

https://som.yale.edu/faculty/saed-alizamir

How can we improve information markets for actors in smallholder supply chains, increase incentives to not deforest, or assign students more efficiently and equitably to public schools? Many socially important problems involve markets where scarce societal resources are allocated, or information or incentives are provided to individuals with differing needs and preferences. In this talk, I will present ongoing work with Joann de Zegher on crowdsourcing market information from competitors in an Indonesia-based smallholder supply chain. I will then discuss how we can build on theoretical tools from mechanism design to design and implement effective markets that give the right resources to those who need them most. (paper and abstract attached)

https://msande.stanford.edu/people/irene-lo

We present a near-optimal solution to the problem of causal inference on panel data. Specifically, we present a convex estimator, which for a certain `tangent space’ condition on the design matrix recovers the treatment effect at an optimal rate. A negligible violation of this tangent-space condition renders recovery of the treatment effect impossible. Our work alleviates the need for strong structural assumptions on the design matrix (synthetic control) or strong distributional assumptions (common to propensity based methods for factor models). Our results are made possible through a simple insight on a deficiency of existing estimation approaches (they fail to leverage all the information in ‘treated’ observations), an exploitation of the connection between certain convex and non-convex estimators, and an adaptation of the leave-one-out approach to analyzing entry-wise guarantees for matrix completion problems.

https://mitsloan.mit.edu/faculty/directory/vivek-f-farias

Food and agriculture supply chains are essential to any society and economy, but at the same time pose significant risks to human health. In this talk, we will discuss how supply chain analytics and sensing, machine learning and modeling can inform regulatory policies and resource allocation to address and manage food adulteration and safety risks as well as zoonotic disease risks. Much of the talk will focus on a major multidisciplinary research effort to study risks related to economically motivated adulteration (EMA) of food in China, particularly ones originated from the upstream parts of the corresponding supply chains.

The talk is based on work done under a new collaborative project funded by the Walmart Foundation as well as work under a contract with the US FDA. It is joint work with multiple faculty and students at MIT and in Chinese universities.

https://mitsloan.mit.edu/faculty/directory/retsef-levi

Abstract forthcoming

Motivated by online labor markets, we consider the online assortment optimization problem faced by a two-sided matching platform that hosts a set of suppliers waiting to match with a customer. Arriving customers are shown an assortment of suppliers, and may choose to issue a match request to one of them. Before leaving the platform, each supplier reviews all the match requests he has received, and based on his preferences, he chooses whether to match with a customer or to leave unmatched. We study how platforms should design online assortment algorithms to maximize the expected number of matches in such two-sided settings.
We show that, when suppliers do not immediately accept/reject match requests, our problem is fundamentally different from standard (one-sided) assortment problems, where customers choose over a set of commodities. We establish that a greedy algorithm, that offers to each arriving customer the assortment that maximizes the expected increase in matches, is 1/2 competitive when compared against the clairvoyant algorithm that knows in advance the full sequence of customers’ arrivals.  In contrast with related online assortment problems, we show that there is no randomized algorithm that can achieve a better competitive ratio, even in asymptotic regimes. Next, we introduce a class of algorithms, termed preference-aware balancing algorithms, that achieve significantly better competitive ratios when suppliers’ preferences follow the Multinomial Logit and the Nested Logit choice models. Using prior knowledge about the “shape” of suppliers’ preferences, these algorithms are calibrated to “balance” optimally the match requests received by suppliers. Overall, our results suggest that the timing of suppliers’ decisions and the structure of suppliers’ preferences play a fundamental role in designing online two-sided assortment algorithms. (joint work with Ali Aouad)

Link: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3712553

https://www.gsb.stanford.edu/faculty-research/faculty/daniela-saban

We present evidence that differences in profiling accuracy are rarely influenced by which data broker is doing the profiling, and instead depend on who is being profiled. Consumers who are better off – those with high income and home ownership, employment, college education – are profiled accurately more often. Occupation – white collar vs. blue collar jobs as well as age and household arrangements also affect profiling accuracy.
Our analyses suggest profiling errors are not driven by how lucrative an individual is, but by the nature of people’s digital footprint and regularity of online activities, which in turn reflect socio-economic and demographic status. The results that better-off people are more likely to be accurately profiled has consequences for both policy and marketing practice.

https://mitsloan.mit.edu/faculty/directory/catherine-tucker

Abstract forthcoming

Physicians, judges, teachers, and agents in many other settings differ systematically in the decisions they make when faced with similar cases. Standard approaches to interpreting and exploiting such differences assume they arise solely from variation in preferences. We develop an alternative framework that allows variation in both preferences and diagnostic skill, and show that both dimensions are identified in standard settings under quasi-random assignment. We apply this framework to study pneumonia diagnoses by radiologists. Diagnosis rates vary widely among radiologists, and descriptive evidence suggests that a large component of this variation is due to differences in diagnostic skill. Our estimated model suggests that radiologists view failing to diagnose a patient with pneumonia as more costly than incorrectly diagnosing one without, and that this leads less-skilled radiologists to optimally choose lower diagnosis thresholds. Variation in skill can explain 44 percent of the variation in diagnostic decisions, and policies that improve skill perform better than uniform decision guidelines. Failing to account for skill variation can lead to highly misleading results in research designs that use agent assignments as instruments.

https://healthpolicy.fsi.stanford.edu/people/david_chan

We analyze a resource allocation problem faced by Medical Surplus Recovery Organizations (MSROs) that recover medical surplus products to fulfill the needs of under-served healthcare facilities in developing countries. Due to the uncertain, uncontrollable supply and limited information about recipient needs, delivering the right product to the right recipient in MSRO supply chains is particularly challenging. The objective of this study is to identify strategies to improve MSROs’ value provision capability. In particular, we propose a mechanism design approach, and determine which recipient to serve at each shipping opportunity based on recipients’ reported preference rankings of different products. We find that when MSRO inventory information is shared with recipients, the only truthful mechanism is random selection among recipients, which defeats the purpose of eliciting information. Consequently, we propose two operational strategies to improve MSROs’ value provision: i) not sharing MSRO inventory information with recipients; and ii) withholding information regarding other recipients. We characterize the set of truthful mechanisms under each setting, and show that eliminating inventory and competitor information provision both improve MSROs’ value provision. Further, we investigate the value of cardinal mechanisms where recipients report their valuations. We show that in our setting, eliciting valuations has no value added beyond eliciting rankings under a wide class of implementable mechanisms. Finally, we present a calibrated numerical study based on historical data from a partner MSRO, and show that a strategy consisting of a ranking-based mechanism in conjunction with eliminating inventory and competitor information can significantly improve MSROs’ value provision.

https://www.scheller.gatech.edu/directory/faculty/toktay/index.html

The Centers for Medicare and Medicaid Services (CMS) star rating methodology for publicly evaluating hospitals uses a latent variable model that is based on the presumption of a single, but unobservable, hospital-specific quality factor shared across a group of performance measures. Performance measures are given higher weight if they statistically appear to be more strongly correlated with this hidden factor. We show how this approach, when applied to measures that are weakly or not correlated with each other, can effectively ignore measures and can exhibit “knife-edge” instability, so that even if hospitals improve relative to all other hospitals, they may nonetheless score lower overall because of weight shifting onto different measures than before. In contrast, we provide an approach to scoring and ranking hospitals that, under reasonable conditions, ensures that hospitals that improve relative to all other hospitals obtain higher scores, while also having the capability to autonomously adjust weights as measures are added or subtracted over time. Rather than exploit statistical correlation, we propose a conic optimization framework that offers a new integrated approach in data envelopment analysis for simultaneous efficiency analysis and performance evaluation. We develop theory that explains the behaviour of our approach, including various properties satisfied by hospital scores at optimality. Using data, we apply our approach to score and rank nearly every hospital in the United States and demonstrate the extent to which it agrees or disagrees with the existing approach to the CMS star ratings.
https://pubsonline.informs.org/doi/abs/10.1287/opre.2019.1972

https://www.chicagobooth.edu/faculty/directory/a/dan-adelman