We study access to ride-sharing across geographical regions using both theoretical and empirical analyses. We specifically model and examine the effects of economies of density in ride-sharing. Our model predicts that (i) economies of density skew access to ride-sharing away from less dense regions; (ii) the skew will be more pronounced for smaller platforms (i.e., "thinner markets"); and (iii) ride-sharing platforms do not find this skew efficient and thus, use price and wage levers to mitigate (but not eliminate) it. We show that these insights are robust to whether the source of economies of density is the supply side or the demand side. We then calibrate our model using ride-level Uber data from New York City. We use the model to simulate counterfactual scenarios, offering a quantitative evaluation of our theoretical results and informing platform strategy and policy.

We propose a novel theory-based approach to the reinforcement learning problem of maximizing profits when faced with an unknown demand curve. Our method is based on multi-armed bandits, which are a collection of minimal assumption non-parametric models that balance exploration and exploitation for maximizing rewards while learning across arms. Specifically, we build on Gaussian process bandits, which represent a flexible non-parametric model that unlike other non-parametric alternatives also provide principled estimates of uncertainty. We leverage the informational externality inherent in price experimentation across arms (price levels) in two ways: correlation between demands correspond to closer price levels, and demand curves are weakly monotonically decreasing. Incorporating these informational externalities limits unnecessary exploration of certain prices and performs better (characterized by greater profitability or reduced experimentation) than currently advanced approaches like UCB (with partial identification), or baseline Gaussian process bandits. Additionally, our method can be used in conjunction with methods like partial identification. Across a wide spectrum of true demand distributions our algorithm demonstrated a significant increase in rewards, most notably for right-skewed underlying WTP distributions where current approaches tend to underperform. Our algorithm performed consistently achieving between 96.9% and 99.1% of the optimal rewards depending on the simulation setting.

A critical element of word of mouth (WOM) or buzz marketing is to identify seeds, often central actors with high degree in the social network. Seed identification typically requires data on the relevant network structure, which is often unavailable. We examine the impact of WOM seeding strategies motivated by the friendship paradox, which can obtain more central nodes without knowing network structure. Higher degree nodes may be less effective as seeds if these nodes communicate less with neighbors or are less persuasive when they communicate; therefore, whether friendship paradox–motivated seeding strategies increase or reduce WOM and adoption remains an empirical question. We develop and estimate a model of WOM and adoption using data on microfinance adoption across village social networks in India. Counterfactuals show that the proposed strategies with limited seeds are about 13%–30% more effective in increasing adoption relative to random seeding. These strategies are also on average 5%–11% more effective than the firm’s leader seeding strategy. We also find these strategies are relatively more effective when we have fewer seeds. This paper was accepted by Juanjuan Zhang, marketing. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2024.4991.

We provide the mathematical and empirical foundations of the friendship paradox in networks, often stated as “Your friends have more friends than you.” We prove a set of network properties on friends of friends and characterize the concepts of ego-based and alter-based means. We propose a network property called inversity that quantifies the imbalance in degrees across edges and prove that the sign of inversity determines the ordering between ego-based or alter-based means for any network, with implications for interventions. Network intervention problems like immunization benefit from using highly connected nodes. We characterize two intervention strategies based on the friendship paradox to obtain such nodes, with the alter-based and ego-based strategy. Both strategies provide provably guaranteed improvements for any network structure with variation in node degrees. We demonstrate that the proposed strategies obtain several-fold improvement (100-fold in some networks) in node degree relative to a random benchmark, for both generated and real networks. We evaluate how inversity informs which strategy works better based on network topology and show how network aggregation can alter inversity. We illustrate how the strategies can be used to control contagion of an epidemic spreading across a set of village networks, finding that these strategies require far fewer nodes to be immunized (less than 50%, relative to random). The interventions do not require knowledge of network structure, are privacy-sensitive, are flexible for time-sensitive action, and only require selected nodes to nominate network neighbors.

This article develops a method to automatically discover and quantify human-interpretable visual characteristics directly from product image data. The method is generative and can create new visual designs spanning the space of visual characteristics. It builds on disentanglement methods in deep learning using variational autoencoders, which aim to discover underlying statistically independent and interpretable visual characteristics of an object. The impossibility theorem in the deep learning literature indicates that supervision with ground truth characteristics would be required to obtain unique disentangled representations. However, these are typically unknown in real-world applications, and are in fact exactly the characteristics that need to be discovered. Extant machine learning methods are unsuitable since they require ground truth labels for each visual characteristic. In contrast, this method postulates the use of readily available product characteristics (such as brand and price) as proxy supervisory signals to enable disentanglement. This method discovers and quantifies human-interpretable and statistically independent characteristics without any specific domain knowledge on the product category. It is applied to a dataset of watches to automatically discover interpretable visual product characteristics, obtain consumer preferences over visual designs, and generate new ideal point designs targeted to specific consumer segments.

This paper develops a theory-based, explainable deep learning convolutional neural network classifier to predict the time-varying emotional response to music.

We obtain consumer willingness to pay for subscription product lacking price variation by leveraging the usage and subscription choice variation.

Marketing models typically focus on how structured product characteristics impact consumer preferences. However, visual characteristics of products present in unstructured image data play an important role in impacting preferences for many categories. We seek to automatically discover and quantify visual characteristics (attributes) from image data using a disentanglement-based approach. While the deep learning literature has shown that supervision is required to obtain unique disentangled representations, ground truth visual characteristics are typically unknown. We develop a method that does not require such supervision, and instead uses readily available structured product characteristics as supervisory signals to enable disentanglement. Our method does not need prior knowledge of characteristics, yet we are able to discover semantically interpretable and statistically independent characteristics. Moreover, the method quantifies the levels of each discovered product characteristic, necessary for managerial tasks such as demand modeling and conjoint analysis. We apply this method to automatically discover visual product characteristics of watches, and discover 6 semantically interpretable visual characteristics providing a disentangled representation. Our results find the supervisory signal `brand' best promotes disentanglement relative to an unsupervised approach. We lastly demonstrate how consumers preferences may be assessed over these discovered visual characteristics using a choice-based conjoint analysis.

Network intervention problems often benefit from selecting a highly-connected node to perform interventions using these nodes, e.g. immunization. However, in many network contexts, the structure of network connections is unknown, leading to a challenge. We develop and examine the mathematical properties of two distinct informationally light strategies, a novel global strategy and local strategy, that yield higher degree nodes in virtually any network structure. We further identify a novel network property called Inversity, whose sign determines which of the two strategies, local or global, will be most effective for a network. We demonstrate that local and global strategies obtain a several-fold improvement in node degree relative to a random selection benchmark for generated and real networks (including contact, affiliation and online networks). In some networks, they achieve a 100-fold improvement. We show how these new strategies can be used to control contagion of an epidemic spreading across a set of village networks, finding that the strategies developed here require far fewer ($<50\%$) nodes to be immunized, relative to the random strategy baseline. Prior research has typically used the complete network structure to choose nodes for optimal seeding. The relevant network is often costly to collect, and is privacy-invasive, requiring knowing each person's network neighbors, and might not be possible to obtain for time-sensitive interventions. Our interventions are less invasive of individual privacy, since each selected node only needs to nominate some network neighbors for intervention, while mathematically guaranteed to provide better connected nodes.

Music evokes emotion in listeners and emotions impact our state of mind. Being able to identify emotion in music provides information about the experienced emotion of the listener. This information is useful in a wide array of settings, ranging from music therapy to advertising. In this paper, we develop a deep neural network emotion classifier that uses different audio transformations (spectrograms) designed to capture specific music concepts and find that different emotions are best captured by different transformations. We also compare time and frequency filters with traditional black box square filters in a convolutional neural network to understand what the square filters may be capturing. It is challenging to identify how music conveys emotion, which has generated vast amounts of psychology research on music and emotion (Johnson-Laird and Oatley 2016; Juslin and Laukka 2003; Juslin and Zentner 2001). The research suggests that different emotions are associated with different settings of music concepts, such as pitch and tempo (Johnson-Laird and Oatley 2016). For example, happy music typically has a wide range of pitches and a medium tempo while sad music typically has a small range of pitches and a slow tempo. Based on research about how humans perceive sound, engineers have developed a number of sound wave transformations to capture different music concepts. For example, the Mel spectrogram highlights the frequencies of perceptual relevance to a human listener while the shorttime Fourier transform (STFT) spectrogram reflects linear frequencies. Following the trend in many fields, deep neural networks (DNNs) have demonstrated significant performance gains in Music Emotion Recognition (MER) (Liu et al. 2017; Malik et al. 2017). While DNNs flexibly incorporate features from the data, with high dimensional data such as audio and video, conceptually developed input transformations using domain knowledge may improve performance. We examine which input transformations are most useful for emotion classification in music. DNNs tradeoff interpretability for performance, making it difficult to understand what features are captured (Lakkaraju, Bach, and Leskovec 2016). We therefore seek to answer the following two research

Disruptive technologies like Artificial Intelligence (AI), Machine Learning (ML), Virtual Reality (VR), Augmented Reality (AR), Mixed Reality (MR), Internet of Things (IoT) etc. affected healthcare, tourism, fashion, entertainment, gaming, education industry etc. in a big way. These technologies have also positively influenced the day today functioning of business management. These technologies are being used for product development, promotional activities, market research, merchandising, inventory management, experience management, and in retail business etc. Customers tend to welcome these technological usages in marketing in a big way, as these technologies are not only providing them functional benefits but also aspirational benefits. All stakeholders are gaining due to the increasing usage of these technologies as the marketers are able to automate lots of processes, customers are getting better purchase experience at relatively lesser cost and convenience of place and time. Their use and importance are increasing significantly in the last decade. In this paper, these technologies will not be covered from a technical point of view. The scope of this paper will only confine to their application in the marketing function of the business management. This research paper is expected to explore the impact of these emerging technologies on marketing of goods and services with appropriate examples/ case studies from the industry.

A critical element of word of mouth (WOM) or buzz marketing is to identify seeds, often central actors with high degree in the social network. Seed identification typically requires data on the full network structure, which is often unavailable. We therefore examine the impact of WOM seeding strategies motivated by the friendship paradox to obtain more central nodes without knowing network structure. But higher-degree nodes may communicate less with neighbors; therefore whether friendship paradox motivated seeding strategies increase or reduce WOM and adoption remains an empirical question. We develop and estimate a model of WOM and adoption using data on microfinance adoption across 43 villages in India for which we have data on social networks. Counterfactuals show that the proposed seeding strategies are about 15-20% more effective than random seeding in increasing adoption. Remarkably, they are also about 5-11% more effective than opinion leader seeding, and are relative more effective when we have fewer seeds.

We study fairness in the context of classification where the performance is measured by the area under the curve (AUC) of the receiver operating characteristic. AUC is commonly used to measure the performance of prediction models. The same classifier can have significantly varying AUCs for different protected groups and, in real-world applications, it is often desirable to reduce such cross-group differences. We address the problem of how to acquire additional features to most greatly improve AUC for the disadvantaged group. We develop a novel approach, fairAUC, based on feature augmentation (adding features) to mitigate bias between identifiable groups. The approach requires only a few summary statistics to offer provable guarantees on AUC improvement, and allows managers flexibility in determining where in the fairness-accuracy tradeoff they would like to be. We evaluate fairAUC on synthetic and real-world datasets and find that it significantly improves AUC for the disadvantaged group relative to benchmarks maximizing overall AUC and minimizing bias between groups.

This paper studies the effects of economies of density in transportation markets, focusing on ridesharing. Our theoretical model predicts that (i) economies of density skew the supply of drivers away from less dense regions, (ii) the skew will be more pronounced for smaller platforms, and (iii) rideshare platforms do not find this skew efficient and thus use prices and wages to mitigate (but not eliminate) it. We then develop a general empirical strategy with simple implementation and limited data requirements to test for spatial skew of supply from demand. Applying our method to ride-level, multi-platform data from New York City (NYC), we indeed find evidence for a skew of supply toward busier areas, especially for smaller platforms. We discuss the implications of our analysis for business strategy (e.g., spatial pricing) and public policy (e.g., consequences of breaking up or downsizing a rideshare platform)

We examine how operational or technological transformation impacts consumer value, as well as the effectiveness of a firm’s pricing strategies. We develop a model of multidimensional screening featuring forward-looking consumers who make short-run consumption and long-run purchase decisions. Using a detailed panel of consumer data from a rental-by-mail firm, we estimate consumer utility for current consumption, obtaining heterogeneous preferences for bunching and smoothing consumption. Using counterfactual analysis, we evaluate the impact of improving service time. We find that the firm with improved service time might create more value for all consumers, but its profits and even revenues could diminish because value extraction becomes more difficult. We find a novel mechanism that causes this effect, which is driven by increased consumer heterogeneity in the valuation for each product and reduced differentiation across products. This result persists even when the firm can reoptimize its price levels based on the service time. We find that a change in the pricing strategy might be required for the firm to obtain higher revenue with improved service time. This paper was accepted by Matthew Shum, marketing.

We investigate the use of the inclusive value based approach for estimating dynamic discrete choice models of demand with aggregate data. The inclusive value sufficiency (IVS) approach approximates a multi-dimensional state space with a single “sufficient statistic” in order to mitigate the curse of dimensionality and tractability estimate model primitives. Although in widespread use, the conditions under which IVS is appropriate have not been examined. Theoretically, we show that the estimator is biased and inconsistent. We then use Monte Carlo simulations (of a simple model of dynamic durable goods adoption) to demonstrate the degree of bias associated with the inclusive value approximation estimator under an array of parameterizations and data generating processes. In our examination, we show that the estimator performs better when the discount factor is smaller and/or when the price sensitivity of the consumer is larger. Examining how the bias impacts economic quantities of interest, we find that the IVS method under estimates the true long-run own-price elasticities and over estimates the change in profits as prices change. Theses findings highlight the importance of correctly specifying how consumers form expectations. As a result, researchers should consider how to empirically support their assumption for the underlying consumer belief structure.

We study how to obtain the distribution of willingness to pay (WTP) for subscription products, where consumers pay a fixed price each period. In the absence of price variation, we demonstrate how variation in usage and subscription choice together can identify the WTP distribution and elasticities, which are primitives of interest in demand estimation. We then propose a novel estimation strategy to recover the WTP distribution when price does not vary in data. Our approach does not assume specific distributions for the unobservables. In addition, we show how price variation, if exists, can help identify the functional form by which usage impacts WTP. We illustrate our method with an application to a music streaming service.

We develop a new approach using market-level data to model, identify, and estimate a dynamic discrete choice demand model for durable goods with continuous unobserved product-specific state variables. They are specified as serially correlated and correlated with the observed product characteristics, particularly price. We provide a method to estimate all model primitives, including the consumer's discount factor and the state transition distributions of unobserved product characteristics without the need to reduce the dimension of the state space or by other approximation techniques, such as discretizing state variables. We prove the identification of model primitives and provide an estimation algorithm in which the most computationally demanding step is a linear regression. Finally, we show how it can be implemented in an application in which we estimate the demand for smartphones.