Using machine learning to identify incentives in forestry policy: Towards a new paradigm in policy analysis

• DOI: 10.1016/j.forpol.2021.102624
• Publication Date: 2021-11-11T22:16:37Z
• Authors: Daniel Firebanks-...; Jordi Planas; Kathleen Buckingham; Cristina Taylor; David Silva; Galina Naydenova; René Zamora-Crist...
• Keywords: Incentive program

As 2021 saw the launch of United Nations Decade on Ecosystem Restoration, it highlighted need to prepare for success over decade and understand what public economic financial incentives exist support sustainable forest landscape restoration. To date, Initiative 20 × 20, a coalition 18 Latin American countries, has committed place 50 million hectares under restoration conservation by 2030. Understanding policies in these countries that turn those commitments into action, however, is very labor-intensive, requiring decision makers read analyze thousands pages documents span multiple sectors, ministries, scales lie outside their areas expertise. address this, we developed semi-automated policy analysis tool uses state-of-the-art Natural Language Processing (NLP) methods mine documents, assist labeling process carried out experts, automatically identify contain classify them incentive instrument from following categories: direct payments, fines, credit, tax deduction, technical assistance supplies. Our best model achieves an F1 score 93–94% both identifying its instrument, as well accuracy above 90% 5 6 instruments, reducing weeks work matter minutes. In particular, properly identified relative frequency credits, fines primary instruments countries. We also found deductions, supplies, are much less used among most that, oftentimes, describe vague intangible terms. addition, our designed constantly improve performance with more data feedback experts. Furthermore, while experiments were run Spanish framework be widely scalable different languages, limited only number languages supported current multilingual NLP models. Using standardized approach generate could provide evidence-based transparent system find complementarity between help remove barriers implementers policymakers enable informed decision-making process.