Dividing things into categories is a human habit, helping us to learn and understand the external world, as well as to share and communicate knowledge. In biology, for example, taxonomy — the classification of species — offered a pathway to understanding the process of evolution. To be useful, a classification scheme cannot be rigid, but has to be flexible enough to change in response to developments in the system to which it applies.
In a recent paper, LML External Fellow Hyejin Youn and colleagues consider systems of economic classification useful in the analysis of urban systems. The economies and societies of cities continually create novel services and products which increase diversity and productivity. Such novelties often escape classification in existing terms. In particular, the authors note, while current industrial classification systems use physical firms or tradable goods as the fundamental descriptive unit of classification, the global economy is increasingly driven by human capital linked to knowledge and skill. As this happens, the natural types of economic products or services rest on differences in the nature of human capital, aspects not captured in earlier classification schemes.
In an effort to identify a useful classification system for such activities, the researchers apply a machine learning algorithm to identify topics from the distributions of occupations across United States urban areas. Their idea is to follow a bottom-up approach to identifying specialized topics from occupation records, and seeing how these vary by geographical location. As they show, this model has several advantages over conventional analysis, being more flexible and useful with new data, and thereby capturing structural changes of an economy. The researchers hope this computational approach can be used to address important questions at the intersection of labour economics and economic structure, and go beyond existing survey methods to provide more timely information about the underlying structures of regional economies.
The paper is available as a pre-print at https://arxiv.org/pdf/2009.09799.pdf