Full Caseworker Discretion and System Sabotage

In Simple Words

Imagine giving a veteran ship captain an automated weather navigation computer, but leaving its use completely optional. If the captain believes personal intuition is superior and fears the computer might eventually replace human judgment, the captain will leave the computer turned off and navigate strictly by memory.

Research question

Implementing decision-support software under a policy of full caseworker discretion without binding incentives leads frontline staff to disregard or actively sabotage the system. While restricting discretion risks crowding out intrinsic caseworker motivation and ignoring valuable private information, complete discretion renders statistical targeting tools ineffective.

Participants

N = 142 treatment caseworkers managing jobseekers across Swiss regional employment offices in a randomized pilot experiment (excluding Geneva).

Study Design

Randomized field study utilizing server access logs to track download frequencies, combined with Instrumental Variables (2SLS) estimation using caseworker randomization as an instrument for prediction downloads.

Results

15% of treatment caseworkers (21 out of 142) never downloaded predictions for any client, overall predictions were inspected for only 37% of jobseekers, and IV estimates confirmed that downloading predictions had no statistically significant effect on compliance rates (0.00 to -0.05 across specifications).

Takeaway

Design algorithmic implementation protocols that combine caseworker discretion with explicit compliance incentives or mandatory justification requirements when deviating from statistical predictions.

Read alongside the research

Plain-English glossary

Original explanations, informed by the references below. Examples are illustrations, not data from the studies above. The research text stays unchanged.

Statistically significant

A label used when a result passes a chosen statistical threshold, often p < 0.05. It does not show how large or useful an effect is. A result that is not statistically significant does not prove there is no effect.

ExampleA study reports p = 0.03 against a 0.05 threshold. It passes that test, but the effect could still be too small to matter.

American Statistical Association · Statement on p-values ↗Read more about statistically significant →

Related findings

• Administrative Exclusion of Inactive Work Scheme Beneficiaries →Policy & public sector

  This is like upgrading a school communications portal but sending setup instructions only to students currently sitting in class today. Former students who stepped away temporarily are left completely out of the loop and face steep hurdles attempting to re-enroll.

• Bi-Weekly Adaptive Digital Skills Training in One-Stop Window →Policy & public sector

  Handing someone an instruction manual written in a foreign language does not teach them how to read it. Setting up bi-weekly group practice sessions alongside one-on-one personal tutoring acts like having a bilingual tutor sitting beside the student, guiding their hands as they practice typing and completing online forms step-by-step.

• AUNA Integrated Social and Employment One-Stop Model →Policy & public sector

  When someone has a complex health problem involving multiple organs, visiting separate independent doctors who do not talk to each other leads to conflicting treatments. An integrated one-stop model acts like a lead attending physician who coordinates all medical specialists in bi-weekly joint consultations to ensure a single, harmonious recovery plan.

• Coordinated Needs-Assessment One-Stop Model for Welfare Recipients →Policy & public sector

  Navigating municipal social assistance without guidance is like trying to find specific goods in a giant, unlabeled warehouse without an information desk. A coordinated one-stop shop provides a dedicated guide who evaluates an individual's overall needs and directly connects them to available social benefits and municipal programs.