The Use of Quantitative Methods with Two Different Perspectives: Data-Centric versus Problem-Centric

Çağlar Çağlayan

by Çağlar Çağlayan
Ph.D. in Operations Research, Georgia Institute of Technology

Despite the diversity of the data analytics methods and the variety of business problems, the quantitative decision science methods can be grouped into two main categories: data-centric and problem-centric approaches (Wegryn, 2014; Rose, 2016). The ultimate purpose of both data- and problem-centric approaches is the same: to help decision-makers make better (informed, effective and efficient) decisions. Yet, usually, the way they approach a problem is fundamentally different (Figure 1). The goal of this article is to briefly introduce these two “quantitative decision science” approaches, highlight their differences, and discuss their roles in better decision-making.  

Figure 1. Data- and Problem-Centric Approaches

Figure 1. Data- and Problem-Centric Approaches

Data-centric approaches, as the name implies, prioritize the use of data and aims to gain insights from the data about the problem of interest. Accordingly, the efforts of the practitioners of data-centric approaches are primarily concentrated on (1) analyzing (including cleaning, summarizing, and manipulating) data; (2) investigating the extend of its use and limitations; and (3) generating new insights from the dataset by applying (and tailoring) analytical methods, ideally based on justifiable assumptions that can be validated. Some means of translating data into new insights, by applying data-centric approaches, are as follows: Identifying and quantifying relations between a key outcome and variables in the dataset (statistical association), detection of systematic changes that variables exhibit over time (pattern recognition), and estimating the future course of a key outcome by modeling the behavior of certain variables (prediction/forecast).  A few examples of data-centric techniques are time-series analysis, regression models, machine learning methods, and deep neural networks.

The use of data is also critical for applying problem-centric approaches to business problems as the qualities of a dataset (e.g., its size, the uncertainty around its variables, etc.) significantly affect the choice of the analytical model to be used. Yet, instead of data, the primary focus of the practitioners of problem-centric approaches is on the business problem itself. The primary goal is to convert the business problem into a well-defined analytical problem that can be modeled and solved. Identifying the decisions to be made, the key outcome to be improved, and understanding the mechanisms governing the key dynamics of the business problem are usually the initial steps of problem-centric approaches. These steps are followed by the development of an analytical model that captures the key dynamics of the business problem and links these dynamics to an objective function to be optimized through the decision variables. The final step is to use an algorithm that solves the analytical problem and identifies the optimal decisions generating the best result(s). A few examples of problem-centric methods are as follows: deterministic optimization (e.g., linear and integer programing), discrete-event simulation, queuing theory, and Markov decision processes.

Understanding the critical features of the business problem and the content and limits of the dataset are required steps, up to a certain extent, both for data- and problem-centric approaches. Whether the primary efforts are concentrated on extracting information from the data or mathematical modeling of the problem is where data- and problem-centric methods begin to differentiate. One way to see the difference between these two approaches is to look at what kind of questions they address through their validation (and debugging) efforts. A typical question for a data-centric approach is the following: Does the employed quantitative method describe the patterns and relationships that the dataset manifests with a high level of accuracy/precision [and hence, can be trusted to make future predictions]? On the other hand, a standard question for a problem-centric technique is as follows: Does the utilized analytical model correctly capture the key dynamics of the underlying problem without over-simplification [and hence, can be used to identify the best course of action]? As it can be seen by these two sample questions, a primary concern regarding the validity of one approach is mainly on the correct use of the data whereas the validity of the other is challenged via its capability to capture of the key problem features. Accordingly, while debugging, a data-centric method might be dealing with over- and under-fitting issues while a problem-centric approach might need to address problems such as a missing constraint or a wrong objective function.

The differences between data- and problem-centric approaches are also related with their roles and objectives. To explain these differences better, we can get assistance from a few terms: descriptive, diagnostic, predictive and prescriptive analysis (Maydon, 2017).

  • Descriptive Analysis: The quantitative description of important information contained in the dataset
  • Diagnostic Analysis: Examination of the historical course of the process of interest and identification the relations of system behavior and process outcomes with the variables in the dataset
  • Predictive Analysis: Projection of the future behavior of the process as a function of certain variables
  • Prescriptive analysis: Identification of the best course of action to be taken to improve the system/process of interest.

Generally, descriptive and diagnostic analyses are conducted via data-centric methods; prescriptive analysis is performed by problem-centric techniques; and both data-centric and problem-centric methods are utilized for predictive analysis.

To conclude, there is a wide range of variety in analytical methods employed by researchers and practitioners to solve their business problems. Despite the variety, the “quantitative decision science” techniques can still be placed into one of the following categories, based on how they approach a problem: data-centric and problem-centric.  Generally, these two perspectives not only have different ways to help decision-makers, but also conduct analyses at different dimensions and hence, have different roles at generating better data-driven decisions. Although it might be unrealistic to expect from any practitioner to be an expert of both approaches, it is very beneficial (if not required) to be familiar with some analytical methods in both domains to correctly approach a business problem and to have a better (full-picture) understanding of the utility of quantitative methods for decision-making.


 [1] Wegryn, G. (2014). Top 5 Analytics Trends for 2015. INFORMS Podcast.
 [2] Rose, R. (2016). Defining Analytics: a Conceptual Framework. OR/MS Today, 43(3), 36-41.
 [3] Maydon, T. (2017). 4 Types of Data Analytics. Principa, Url: https://insights., accessed June 2018.