Data science for better productivity

When it comes to performance evaluation, productivity analytics, and benchmarking, Data Envelopment Analysis (DEA) is the most widely used tool. In particular, DEA is data-enabled analytics (Zhu, 2020). It embraces four types of analytics, namely descriptive, diagnostic, predictive, and prescriptive, depending on application. The research question(s) posed on the data will determine the type of analytic lens adopted by DEA. The issue includes twelve research articles by authors from Australia, China, Germany, Spain, Taiwan, the United Kingdom, and the United States; and it spans a spectrum of research areas, encompassing business administration and resource allocation, corporate diversification, pension funds and financial loans, banking, traffic congestion, healthcare, entertainment and hospitality, among others.

Our Special Issue opens with a comprehensive review by Shi et al. (2020): “Data Science and Productivity: A Bibliometric Review of Data Science Applications and Approaches in Productivity Evaluations”. The manuscript performs a bibliometric analysis in combination with an empirical review of 533 research articles and offers a comprehensive review of data science techniques and methodologies used in productivity analysis. The authors identify current predominant trends and patterns in data science and productivity and provide an outlook on future research directions. Among others and unsurprisingly, the authors find that DEA is one of the most frequently applied data science approaches in the dataset studied. Moreover, an ample number of studies used other data science techniques in combination with DEA.

The following seven papers focus on the use of DEA. The range of domains is varied, encompassing business, pension funds, financial loans, traffic congestion, and healthcare. In “A new hybrid method for the fair assignment of productivity targets to indirect corporate processes”, Ihrig et al. (2019) develop a method allowing the allocation of an ex-ante defined resource level across various processes of an organisation. The authors propose a mixed-integer/linear programme that incorporates a social welfare function, allowing decision-makers to consider fairness aspects. The core idea of the proposed method is to define processes and their corresponding costs as inputs, and their value as output within a DEA framework. The paper makes both a theoretical and practical contribution by providing an insight into how to allocate resources and productivity targets when productivity targets are to be allocated among processes of a single organisation and in the absence of (external) benchmarking figures.

In “Corporate diversification, firm productivity and resource allocation decisions: The data envelopment analysis approach”, Jiang et al. (2019) develop new DEA models as analytical tools that help diversified firms to more accurately calculate productivity at both the firm and its business segment levels, and then reallocate resources among business segments in order to achieve better firm overall performance. Based on a sample of U.S. public firms from 2012 to 2016, the empirical results illustrate that, compared to the conventional DEA models, the proposed models yield more useful productivity expression for diversified firms, and that the model-suggested resource reallocation is linked to better firm performance.

A critical issue for ensuring long-term sustainability of pension funds is how the Public Service Pension Fund Management Board (PSPFMB) makes entrusting decisions to choose investment trust corporations (ITCs). In “Entrusting decisions to the public service pension fund: An integrated predictive model with additive network DEA approach”, Lin et al. (2020) use the additive network DEA approach to measure the efficiency scores of 34 ITCs, including operating performance, equity fund performance, and bond fund performance. The authors further propose a method to effectively forecast the future efficiency scores of each ITC by using a trend analysis technique. This forecasting model could help support the PSPFMB when estimating and prioritising ITCs to improve long-term sustainability of retirement pension financial revenue. Among others, the paper advocates a mechanism for ITC evaluation that helps improve the deficiency of old decision-making processes which only look at ITCs’ past performances.

Finding a model that can predict effective customers of Internet financial loan products and then providing a strategy for precision marketing is the focus of research by Zhu et al. (2020) in their paper “A DEALG methodology for prediction of effective customers of internet financial loan products”. The authors propose a decision-making framework for combining data mining algorithms to achieve precision marketing of Internet financial loan products. To this aim, the paper considers pre-processed data via DEA and adds the DEA efficiency value into the logistic regression model, which can improve the accuracy of the basic logistic regression model. The proposed novel data analytics approach is termed DEALG and works to significantly enhance the customer response rate of Internet loan products according to its results.

In “Should drivers cooperate? Performance evaluation of cooperative navigation on simulated road networks using network DEA”, Summerfield et al. (2020) propose an approach using agent-based simulation and network DEA to answer whether having higher proportions of cooperative drivers on a road network is always better. To this aim, the authors use a heuristic cooperative routing algorithm that minimises network congestion, simulates a road network, and measures its performance under various proportions of cooperative drivers. They measure both the congestion (city manager’s perspective) and the average drivers’ perceived road network performance. Among others, the paper demonstrates that the network DEA model is an effective way of determining road traffic performance.

The following paper by Afsharian (2019), “A frontier-based facility location problem with a centralised view of measuring the performance of the network”, studies a variant of the p-median problem which incorporates DEA into the facility location analysis problem. To this aim, a novel way of using DEA to maximise the overall efficiency of the entire network of the p chosen facilities is presented by an intermediate bilinear mixed-integer programme. The author then formulates its equivalent mixed-integer linear programme, which significantly simplifies the implementation of the approach. The practical relevance of the proposed approach is illustrated by an application to real-world data from hospitals in the district of Braunschweig in Germany.

Motivated by an empirical study of the US Government’s Centers for Medicare & Medicaid Services (CMS) data, in “Improving productivity using government data: The case of US Centers for Medicare & Medicaid's ‘Nursing Home Compare’”, Bougnol and Dulà (2020) first report on accessing, collecting, analysing, and validating the CMS’s data. They then provide an alternative analysis on CMS’s data and attributes using frontier analysis (FA), a generalisation of DEA, and compare and contrast the results with CMS’s star ratings. In the authors’ words, this paper makes the case to governments to use quantitative methods such as FA to replace current highly specialised, complex, and arcane practices while still attaining the objectives of effectively summarising large amounts of data and information, simplifying the consumer’s decision-making process, and spotlighting excellence.

The remaining four papers focus on the use of other approaches as data science tools to analyse productivity. Adopted approaches are fuzzy-set qualitative comparative analysis (fsQCA), logistic regression analysis (LRA) in tandem with an artificial neural network (ANN) via the entropy method, natural language processing (NLP) combined with econometric analysis, and a competing risk methodology within a random survival random forest framework. The range of domains is once more diverse, including business, banking, entertainment, and hospitality industries. In “Exploring data conditions to improve business performance”, Grimaldi et al. (2019) analyse the relationship between different combinations of data conditions and the company performance which they measure through the customer management and provider operations efficiency. The proposed methodology is novel compared to previous research studies and is based on the use of a fuzzy-set qualitative comparative analysis (fsQCA) which allows multiple paths to achieve the possible outcomes to be revealed. The conclusion of the study is different combinations of casual conditions (mainly related to 3 items of the data maturity along with the data-driven company profile) drive better customer management and provider operations efficiency (outcomes).

Li and Chen (2019), in their paper “Entropy method of constructing a combined model for improving loan default prediction: A case study in China”, propose a novel credit scoring model that combines the logistic regression algorithm (LRA) and an ANN via the entropy method, in addition to data filtering and a feature selection algorithm based on random forest. The experimental results show that the proposed combined model outperforms the two base models (that is, the individual benchmark classifiers and Stacking) on four evaluation metrics. The application of the novel model supplements the research on credit scoring and provides useful insights for future research. Findings can assist banks and relevant decision-makers in their efforts to develop strategies aimed at improving the good-loan rate and promoting stable development of the loan market.

In “Improving productivity in Hollywood with data science: Using emotional arcs of movies to drive product and service innovation in entertainment industries”, Del Vecchio et al. (2020) propose a new framework which allows data science to be used to optimise content-generation in entertainment and to test this framework for the motion picture industry. The authors use natural language processing (NLP) combined with econometric analysis to explore whether and to what extent emotions shape consumer preferences for media and entertainment content, which, in turn, affect revenue streams. Findings demonstrate that data science can enhance revenue streams (and, thereby increase productivity) through future preference mapping. The paper makes several theoretical, methodological, empirical, and practical contributions, with the proposed approach representing a good example of a potential decision support system.

Lastly, in “Estimating customer churn under competing risks”, Routh et al. (2020) employ a competing risk methodology within a random survival forest framework that accurately computes the risks of customer churn and identifies relationships between the risks and customer behaviour. In order to assess its performance, the authors then apply the method to data from a membership-based firm in the hospitality industry, where customers face two competing churning events. The method outperforms conventional methods by providing higher accuracy in identifying potential churners. Findings can support marketers to identify and understand churners, as well as develop strategies on how to design and implement incentives.

As special issue editors, we are extremely grateful for the support received from the Editors-in-Chief and the editorial team of The Journal of the Operational Research Society, Taylor & Francis, The OR Society. The many academics and researchers who contributed articles and the experts within the field who reviewed the articles have made this Special Issue on Data Science for Better Productivity of the Journal of the Operational Research Society possible – we thank you! We hope that the research presented in this collection of papers helps to demonstrate the usefulness of data-oriented analytics and data science tools in performance evaluation and benchmarking, for better productivity. We wish you, our readers, informative reading!