Search results

The growth of social media and the sharing economy is generating abundant unstructured image and video data. Computer vision techniques can derive rich insights from unstructured data and can inform recommendations for increasing profits and consumer utility – if only the model outputs are interpretable enough to earn the trust of consumers and buy-in from companies. To build a foundation for understanding the importance of model interpretation in image analytics, the first section of this article reviews the existing work along three dimensions: the data type (image data vs. video data), model structure (feature-level vs. pixel-level), and primary application (to increase company profits vs. to maximize consumer utility). The second section discusses how the “black box” of pixel-level models leads to legal and ethical problems, but interpretability can be improved with eXplainable Artificial Intelligence (XAI) methods. We classify and review XAI methods based on transparency, the scope of interpretability (global vs. local), and model specificity (model-specific vs. model-agnostic); in marketing research, transparent, local, and model-agnostic methods are most common. The third section proposes three promising future research directions related to model interpretability: the economic value of augmented reality in 3D product tracking and visualization, field experiments to compare human judgments with the outputs of machine vision systems, and XAI methods to test strategies for mitigating algorithmic bias.

This article is a response to the two articles about our chapter (Cohen & Doveh, this volume). The first article was written by Viechtbauer and Budescu and the second written by Hanges and Lyon (both in this volume). The main contribution in the first article relates to the statistical methodology, while in the second article the authors introduce further applications to our method and discuss the interpretability of intra-class correlation coefficients (ICC). We concur with most of the ideas expressed in these articles and elaborate on some of the points raised in them.

Nonperforming assets in any banking system have stressed the economic health of nations. Resultantly, literature has given considerable impetus to predict failures and bankruptcy. Past studies have focused on the outcome of failures, while, there is a dearth of studies focusing on ongoing firms in bad shape. We plug this gap and attempt to identify underlying communication patterns for firms witnessing prolonged underperformance. Using text mining, we extract and analyze semantic, linguistic, emotional, and sentiment-based features in non-numeric communication channels of these poor-performing firms and their peers. These uncovered patterns highlight the use of vocabulary and tone of communication, in correspondence to their financial well-being. Furthermore, using such patterns, we deploy various Machine Learning algorithms to identify loser firm(s) way ahead in time. We observe promising accuracy over a time window of five years. Such early warning signals can be of critical importance to various stakeholders of a firm. Exploration of writing style-related features for any firm would help its investors, lending agencies to assess the likelihood of future underperformance. Firm management can use them to take suitable precautionary measures and preempt the future possibility of distress. While investors and lenders can be benefitted from this incremental information to identify the likelihood of future failures.

In this paper, we consider the analysis of models for univariate and multivariate ordinal outcomes in the context of the latent variable inferential framework of Albert and Chib (1993). We review several alternative modeling and identification schemes and evaluate how each aids or hampers estimation by Markov chain Monte Carlo simulation methods. For each identification scheme we also discuss the question of model comparison by marginal likelihoods and Bayes factors. In addition, we develop a simulation-based framework for analyzing covariate effects that can provide interpretability of the results despite the nonlinearities in the model and the different identification restrictions that can be implemented. The methods are employed to analyze problems in labor economics (educational attainment), political economy (voter opinions), and health economics (consumers’ reliance on alternative sources of medical information).

We argue that resource-based view (RBV) researchers must take into account three interdependencies, (i) intrafirm resource complementarity, (ii) interfirm resource complementarity or rivalry, and (iii) compatibility or incompatibility of firm resources to broader socio-economic institutions, when attempting to empirically verify the RBV. However, these interdependencies lead to three potential causes of statistical bias, which can reduce the interpretability of such empirical examinations. First, omitted variable bias results from a researcher's inability to find and include in empirical analyses appropriate operationalizations of constructs. Second, selection bias can arise when a researcher samples only from one subset of the population, and not others. Bias in estimates can occur if a correlation between unobserved determinants of the outcome and factors affecting the selection process exist. Finally, joint dependence, where two explanatory variables are themselves mutual determinants, can lead to biased estimation.

Increasing availability of the financial data has opened new opportunities for quantitative modeling. It has also exposed limitations of the existing frameworks, such as low accuracy of the simplified analytical models and insufficient interpretability and stability of the adaptive data-driven algorithms. I make the case that boosting (a novel, ensemble learning technique) can serve as a simple and robust framework for combining the best features of the analytical and data-driven models. Boosting-based frameworks for typical financial and econometric applications are outlined. The implementation of a standard boosting procedure is illustrated in the context of the problem of symbolic volatility forecasting for IBM stock time series. It is shown that the boosted collection of the generalized autoregressive conditional heteroskedastic (GARCH)-type models is systematically more accurate than both the best single model in the collection and the widely used GARCH(1,1) model.

Across disciplines, researchers and practitioners employ decision tree ensembles such as random forests and XGBoost with great success. What explains their popularity? This chapter showcases how marketing scholars and decision-makers can harness the power of decision tree ensembles for academic and practical applications. The author discusses the origin of decision tree ensembles, explains their theoretical underpinnings, and illustrates them empirically using a real-world telemarketing case, with the objective of predicting customer conversions. Readers unfamiliar with decision tree ensembles will learn to appreciate them for their versatility, competitive accuracy, ease of application, and computational efficiency and will gain a comprehensive understanding why decision tree ensembles contribute to every data scientist's methodological toolbox.