Nordic Economic Policy Review 2026: Advancing Policy Through Randomised Experiments

Nordic Economic Policy Review 2026

Comments on Elias Einiö and Annika Nivala: Large-Scale Randomised Experiments with Population Data

Antti Kauhanen

1 Introduction

This article offers a timely, practice-oriented guide to designing and executing national-scale randomised controlled trials (RCTs) in settings where population registers cover the relevant units of observation. Anchored in the Finnish Recruitment Subsidy Experiment (FRSE) conducted along with the Ministry of Economic Affairs and Employment, the study synthesises econometric and organisational lessons highly relevant to public policies characterised by costly treatments and imperfect take-up. It constitutes a valuable contribution to the growing literature and practice of experimentation at scale.

2 A Nordic advantage – if we remove bottlenecks

The Nordic countries’ population registers provide a genuine comparative advantage for policy experimentation. The ability to measure outcomes for all target units in administrative data reduces the usual cost constraints on data collection, enabling large samples, high statistical power, and credible estimates of the policy's effect, which are crucial in contexts where participation is voluntary. Registers also facilitate planning (through ex ante power calculations using population data) and implementation (e.g., precise identification of the target population).

However, realising this advantage requires attention to three practical bottlenecks:

Access to data: Streamlined and predictable access arrangements are necessary to conduct experiments on policy-relevant timelines.
Timeliness of data: Outcome registers must be updated quickly enough to support interim analyses and timely reporting to policymakers.
Cost of access: Excessive data access fees can deter experimentation and make learning slow.

3 What low take-up reveals – and how to learn more

The recruitment subsidy experiment highlights a central puzzle: take-up was only 1.5% in the target population, but 32% among companies that became new employers during the experiment. This pattern sheds light on where the main frictions arise and suggests that policy design and communication may be as important as financial incentives.

The article rightly emphasises that public policy experiments often estimate the impact of offering access to a treatment rather than forcing uptake. To translate offers into behaviour, the evaluation should budget for complementary process evidence to help explain non-acceptance by eligible companies that would plausibly benefit. Concrete steps include:

Verifying whether letters were received and read (e.g., address accuracy checks and open-rate statistics for digital channels).
Diagnosing perceived complexity or risk around eligibility, compliance, and audits.
Solving timing and cashflow problems that might reduce benefits for small businesses.
A/B testing message framing (e.g., simplification, deadlines, testimonials) to optimise applications without changing the underlying generosity.

These inexpensive additions can substantially improve the interpretation of experimental impact and guide rapid policy iterations under the same programme.

4 From quasi-experiments to RCTs: it is not just identification

One intriguing insight from the Finnish case is that the transition from observational evidence to randomised experiment was closely linked with a significant policy redesign. The revamped programme included clearer eligibility criteria, quicker, automatic payments, and direct outreach to eligible companies, elements that probably boosted both participation and impact. Consequently, the experiment revealed significant positive effects on hiring, contrasting with a previous observational study on a similar instrument that registered no effects. This highlights a crucial point: the differences between evaluations involve not only identification but also the policy framework. In Duflo’s “practice of policy,” experiments generate evidence that informs improved programme design, which is then tested in subsequent experiments. The Finnish case exemplifies this iterative approach.

5 Design–implementation–evaluation as a single contract

Another important lesson pertains to the institutional setup. The team collaborated with the ministry from the early stages of design through to implementation and evaluation. This included conducting power calculations using administrative data and providing feedback on the bill in Parliament and information letters before the legislation was enacted in January 2022. Researchers advocated for randomisation at the level of the entire target population, identified from the tax register, rather than from an applicant pool. This approach allowed for the estimation of population-wide impact, which is the most informative data for policymaking.

The “comprehensive contracting” model, which integrates design, implementation support, and evaluation, effectively aligns incentives, ensures that randomisation is feasible within legal and operational frameworks, and increases the likelihood that the results will be both policy-relevant and timely.

6 Beware general equilibrium effects

When experiments are extensive and national in scope, it is crucial to consider the possibility that the treatment may influence outcomes in the control group, potentially through labour-market displacement or wage pressure. In the context of recruitment subsidies, spillovers might be envisioned if treated companies increase wages or attract workers away from control companies. Although population registers enable the monitoring of market-level outcomes, credibly identifying spillovers is more complex and often necessitates specific design choices, such as cluster-level randomisation and varying saturation across regions, along with pre-analysis plans for measuring spillovers. This issue deserves explicit attention on the agenda for future national experiments.

7 Concluding remarks

The study compellingly demonstrates that conducting national-scale RCTs using population data is both feasible and highly informative for public policy, especially when treatments are expensive, and uptake is suboptimal. The FRSE experiment exemplifies how thoughtful design choices, such as clarity, speed, and proactive outreach, can enhance programme effectiveness and how experimentation can be seamlessly integrated into policymaking as a standard practice for improvement. To expedite learning, the Nordic policy ecosystem should intensify its focus on enabling infrastructure (data access, timeliness, and cost) and on institutional models that integrate design, implementation, and evaluation into a cohesive whole. This approach will produce more credible evidence, faster iterations, and improved public programmes.