Overview

Public-employment digital interventions are more likely to matter when they address the bottleneck between a job seeker and a real opportunity. The Swedish and Dutch experiments support conditional contrasts: reporting-focused messages moved compliance without the intended employment gain, while personalized occupational information improved employment for a targeted structurally poor, non-Covid population. The design rule is to fit message, institutional relationship, recipient, and labor market, then measure downstream outcomes; scale effects and mechanisms remain conditional.

Learning outcomes

A reminder should be judged by reporting, but not credited with employment until downstream outcomes confirm the connection.
Supportive language needs institutional credibility; occupational advice needs feasible alternatives and current market evidence.
Scale decisions require subgroup, job-quality, and spillover monitoring because neither experiment establishes a universal hierarchy.

Keywords

bottleneckmotivationcomplianceinformationpersonalizationspillovers

Designing Messages That Reach the Right Bottleneck

Public-employment digital intervention is the subject that connects the two experiments examined here: a service sends information or motivational communication to an unemployed person, and the result depends on the relationship, the search decision, and the labor market around them. This final synthesis answers the opening question—why can motivational emails increase compliance while weakening job search?—and turns the answer into a conditional way to design and evaluate interventions. The evidence does not support a universal rule that motivation is useless or information is always superior. It supports a stricter rule: a message matters when it changes the bottleneck that stands between a job seeker and a real opportunity.

What the two experiments actually establish

The Swedish and Dutch interventions are not two arms of one experiment. They took place in different countries, under different benefit and employment-service systems, among different populations, and with different definitions of success. Their headline percentages cannot be placed on a common ranking scale. What can be compared is the causal pathway each intervention was designed to change.

The Swedish intervention entered an already controlling relationship. Job seekers had to report activity each month, and insufficient activity could trigger review and possible benefit sanctions. The supportive messages tried to make search more autonomous by emphasizing choice, competence, personal goals, and understanding. The controlling messages emphasized obligations, monitoring, deadlines, and sanctions. Both types increased activity-report submission. The controlling arm had no measurable effect on the primary application or job-finding outcomes; the supportive arm modestly reduced applications and lowered job finding in selected horizons, without producing better observed job quality.

That is how compliance can rise while search weakens. The message can make the administrative requirement more salient without improving the recipient’s search strategy. In the supportive case, the wording may also have changed the perceived pressure of the relationship. A soft message from a strict monitoring institution can be heard as evidence that the institution has relaxed its control. For people who relied on that pressure, or who interpreted the mismatch negatively, the result can be less search rather than more self-directed effort. For people who were already autonomously motivated, added control may instead crowd out the reason they had already endorsed for searching. These explanations are plausible and consistent with subgroup patterns, but they are not direct measurements of what happened inside every recipient.

The Dutch intervention addressed a different bottleneck. It targeted people searching in occupations with poor prospects and gave them personalized comparisons with alternatives that were both more promising and supported by evidence of feasible transitions. The intervention did not primarily ask recipients to supply more effort to the same search. It changed the information available when they decided where to look. In the main population—people with no paid employment who were searching in structurally poor, non-Covid occupations—employment rose by about 2.9 percentage points after 18 months, from a control mean of 55.9%. Monthly hours increased by about four from a control mean of 66.1, and monthly earnings by about €72 from a control mean of €1,118. Those are relative increases of 5.2%, 6.0%, and 6.4%, respectively; the percentage-point employment change should not be confused with the relative percentage increase.

The occupational evidence makes the information mechanism more credible. At a detailed occupational level, 21% of control-group respondents who found work entered their primary search occupation, compared with 15% in treatment.

Published by DeepMox

FAQ

Does the Swedish evidence prove that supportive language harms job seekers?

No. It shows a modest adverse pattern for one Swedish treatment package in a monitoring relationship, not that all supportive language or sustained autonomy-supportive service harms job seekers.

What remains unresolved after the Swedish and Dutch experiments?

They do not settle cross-country transfer, the psychological mechanism of Swedish responses, the best follow-up for Dutch recommendations, or the full equilibrium effect of rollout.

How should a public employment service decide whether to scale a message?

It should scale only after testing the intended downstream outcome in the relevant target group and checking job quality, subgroup effects, and spillovers.

Previous11 minDutch Occupational Information: Employment and Spillovers

All chaptersBack to Topic