Generalization of scientific knowledge

This page introduces how scientific knowledge is generalized in the Upright net impact model.

Scientific and other relevant knowledge about products and their impacts may pertain to specific or generic product and service categories. Therefore products in the Upright net impact model are modelled as a hierarchy (i.e., parent-child) relationships. These relationships are used to understand what research is relevant to which specific products.

In the above illustration, these relationships and information specific to each parent and child product would translate into the following observations:

  • Facts about green apples are relevant when considering the impact of apples in general. They do not, however, apply to red or yellow apples.

  • Apples are a specific type of pome fruits, and pome fruits are a specific type of fruits. Facts about pome fruits also apply to apples.

The main underlying assumption behind the algorithm itself is that the volume of research associated with a product is indicative of the certainty of the level of impact: a single article, regardless of how conclusive, constitutes a less reliable finding than a 100 articles.

If a single article identified green apples as a cause of cancer in the context of the above illustration, but apples, pome fruits and fruits had no such indications in spite of much more relevant research, the model would weight the lack of evidence in the parent products of green apples as more conclusive, and virtually ignore the single outlier article.

Conversely, if a large number of articles identified red apples to be much more healthy than apples in general, the model would start to accept such evidence as conclusive, and give less weight to the knowledge about the health impacts of apples in general.

Technically the algorithm is implemented using a Bayesian inference framework. The details and an illustrative example of the algorithm are found in this section of the illustrative example of quantifying net impact.

It is worth noting that the hierarchical representation of products and services is also useful in modelling other impact datasets. For example, estimation of EU taxonomy alignment can be similarly associated with specific or generic products and services, and generalized into their respective parent and child products.

Last updated