Data and methods

Machine learning (ML) improves economic policy analysis by addressing the complexity of modern data. It complements traditional econometric methods by handling numerous control variables, managing interactions and non-linearities flexibly, and uncovering nuanced differential causal effects. However, careful validation and awareness of limitations such as risk of bias, transparency issues, and data requirements are essential for informed policy recommendations.

Randomized control trials are often considered the gold standard to establish causality. However, in many policy-relevant situations, these trials are not possible. Instrumental variables affect the outcome only via a specific treatment; as such, they allow for the estimation of a causal effect. However, finding valid instruments is difficult. Moreover, instrumental variables (IV) estimates recover a causal effect only for a specific part of the population. While those limitations are important, the objective of establishing causality remains; and instrumental variables are an important econometric tool to achieve this objective.

The Mincer equation—arguably the most widely used in empirical work—can be used to explain a host of economic, and even non-economic, phenomena. One such application involves explaining (and estimating) employment earnings as a function of schooling and labor market experience. The Mincer equation provides estimates of the average monetary returns of one additional year of education. This information is important for policymakers who must decide on education spending, prioritization of schooling levels, and education financing programs such as student loans.

Measuring workers’ productivity is important for public policy and private-sector decision-making. Due to the lack of a general measure that captures workers’ productivity, firms often use one- or multi-dimensional performance measures, which can be used, for example, to analyze how different incentive systems affect workers’ behavior. The public sector itself also uses measures to monitor and evaluate personnel, such as teachers. Policymakers and managers need to understand the advantages and disadvantages of the available metrics to select the right performance measures for their purpose.

Gross domestic product (GDP) is the key indicator of the health of an economy and can be easily compared across countries. But it has limitations. GDP tells what is going on today, but does not inform about sustainability of growth. The majority of time is spent in home production, yet the value of this time is not included in GDP. GDP does not measure happiness, so residents can be dissatisfied even when GDP is rising. In addition, GDP does not consider environmental factors, reflect what individuals do outside paid employment, or even measure the current or future potential human capital of a country. Hence, complementary measures may help to show a more comprehensive picture of an economy.

Statistical models can help public employment services to identify factors associated with long-term unemployment and to identify at-risk groups. Such profiling models will likely become more prominent as increasing availability of big data combined with new machine learning techniques improve their predictive power. However, to achieve the best results, a continuous dialogue between data analysts, policymakers, and case workers is key. Indeed, when developing and implementing such tools, normative decisions are required. Profiling practices can misclassify many individuals, and they can reinforce but also prevent existing patterns of discrimination.

Anti-discrimination policies play an important role in public discussions. However, identifying discriminatory practices in the labor market is not an easy task. Correspondence testing provides a credible way to reveal discrimination in hiring and provide hard facts for policies, and it has provided evidence of discrimination in hiring across almost all continents except Africa. The method involves sending matched pairs of identical job applications to employers posting jobs—the only difference being a characteristic that signals membership to a group.

When hiring new workers, employers use a wide variety of different recruiting methods in addition to posting a vacancy announcement, such as adjusting education, experience, or technical requirements, or offering higher wages. The intensity with which employers make use of these alternative methods can vary widely depending on a firm’s performance and with the business cycle. In fact, persistently low recruiting intensity partly helps to explain the sluggish pace of job growth in the US economy following the Great Recession, and the historically subpar wage growth during the subsequent expansion.

In the context of growing worldwide inequality, it is important to know what happens when the demand for low-skilled workers changes. Because natural resource shocks are global in nature, but have highly localized impacts on labor prospects in resource extraction areas, they offer a unique opportunity to evaluate low-skilled men's behavior when faced with extreme variations in local labor market conditions. This situation can be utilized to evaluate a broad range of outcomes, from education and income, to marital and fertility status, to voting behavior.

The open science and research transparency movement aims to make the research process more visible and to strengthen the credibility of results. Examples of open research practices include open data, pre-registration, and replication. Open science proponents argue that making data and codes publicly available enables researchers to evaluate the truth of a claim and improve its credibility. Opponents often counter that replications are costly and that open science efforts are not always rewarded with publication of results.