Budgeting for SDGs: Quantitative methods to assess the potential impacts of public expenditure

After working on various applications of our policy-priority framework, we took a step back to reflect on where it stood against alternative impact-evaluation approaches in the context of multi-dimensional development. This work presents a systematic analysis comparing our framework to other two popular approaches: regression analysis and machine learning.

Using a novel large-scale dataset that links thousands of expenditure programs to the Sustainable Development Goals for over a decade, we analyze the impact of public expenditure on more than 100 different development indicators. Contrary to the single-dimensional view of evaluating expenditure in terms of overall economic growth, we take a multi-dimensional approach. Then, we assess the effectiveness of three quantitative methods for capturing expenditure effects on development: (1) regression analysis, (2) machine learning techniques, and (3) agent computing. We find that, under the existing data and for this particular task, approaches (1) and (2) have difficulties disentangling sector-specific effects (i.e., target effects in the SDG semantics), which is consistent with results in previous empirical research. In contrast, by applying a micro-founded agent-computing model of policy prioritization, we can provide empirical evidence about potential impacts and bottlenecks across a high-dimensional policy space. Our findings suggest that, in the discussion of budgeting for SDGs, one should carefully evaluate the data available, the suitability of data-driven approaches, and consider alternative methods that are richer in terms of incorporating explicit causal mechanisms and scalable to a large set of indicators.