Data science has emerged as a critical discipline for deriving insights and guiding decisions. However, as the field matures, a schism is becoming apparent between "Pragmatic Data Science" and "Purist Data Science." The former is grounded in industry whereas the later is traditionally a more academic approach. They offer different philosophies that can significantly impact the quality of decision-making in various contexts. We explore the distinctions between these two perspectives and argue that pure models are only as good as the data and the model choice that define them.
Purist data science is often heralded as staying true to statistical rigor and the scientific method. Here, the emphasis is on developing models that are accurate, reliable, and easily generalised to new use cases. The underlying logic is that relationships in data, a meticulous data collection process, exhaustive exploratory data analysis, and a strict regimen of hypothesis testing will yield superior results to the human decision making process.
It is highly dependent on the quality and completeness of the data. Garbage in, garbage out, as the saying goes. Furthermore, this approach is often constrained by the choice of the model, which might not always capture the complexity or the nuances of the real-world scenario it's supposed to represent.
Contrary to the rigidity of the purist approach, pragmatic data science is much more flexible and relies to a certain extent on human expert judgement.
Art more than science - in this sense, pragmatic data science often resembles more art than science. It allows for human intuition, domain expertise, and even "gut feeling" to play a role in the decision-making process. The models used in this approach might not be pure, but they are useful.
One of the core key advantages of the pragmatic approach is its forward-looking and strategic perspective. Unlike purist models, which are highly optimized and constrained by available data, pragmatic data science can incorporate future scenarios, trends, and unknown variables into the decision-making process.
Consider a retail business deciding on the optimal location for a new store.
Both pragmatic and purist approaches have their merits and drawbacks, but it's essential to recognize that pure models are only as good as the data, the model choice that defines them and the stability of the backdrop from which predictions are determined. Pragmatic data science offers a more adaptable, forward-looking, and ultimately practical approach to decision-making. In a world that's always changing, perhaps it's time we embrace the art of data science as much as the science itself.
Get in touch to discuss how we can help you design the best approach to data science that drives values to your business.