Perceived control and agency
Believing you can influence what happens reduces stress, even when you never act.
Perceived control is the belief that one's actions affect outcomes. It reduces the stress response to an aversive event independently of whether the control is exercised. The belief has to be credible: control that turns out to be nominal produces a sharper negative reaction than none was offered.
How it shows up in software
Cancel buttons, undo, granular settings and manual overrides carry value beyond their use rate. A long process feels different when a stop control is visible. Automation that cannot be inspected or interrupted generates anxiety in proportion to how consequential it is, which is why background sync and AI actions need visible handles.
Using it well
- Give every long-running or automated action a visible stop, and honour it promptly.
- Make automated changes inspectable after the fact: what changed, when, and how to reverse it.
- Offer a small number of real choices rather than a large number of cosmetic ones. Nominal control is worse than none.
- Let people opt out of an automation without leaving the product, and remember the choice.
Where it turns manipulative
- Placebo controls, such as a close button on an ad that reloads it, or granular privacy toggles that do not change what is collected.
- Consent interfaces that offer many switches and one accept-all button, where the design makes real control expensive.
- Framing a forced change as a user choice by presenting two options that lead to the same outcome.
Where you have seen it
GitHub Copilot
Suggestions appear as ghost text that requires an explicit accept keystroke, so nothing enters the file without an action.
macOS
Time Machine lets a user browse and restore individual past versions of a file rather than only the whole backup.
Gmail
Filters are user-authored rules with visible conditions, so inbox sorting is inspectable rather than opaque.
What the research says
- Glass and Singer, 1972Mixed evidence
Participants exposed to unpredictable noise showed fewer negative aftereffects when given a button that could stop it, even though almost none pressed it.
Laboratory work from the early 1970s with small samples. The core result is widely cited and broadly consistent with later stress research.
- Langer and Rodin, 1976Contested
Nursing home residents given responsibility for small decisions and a plant to care for showed higher rated alertness and wellbeing than a comparison group told staff would handle things.
Small, single-site, and from an era with weaker standards. It is quoted far more confidently than its design supports.
- Bandura, 1977Well evidenced
Self-efficacy theory: expectations about one's own capability predict whether people attempt a task and how long they persist when it goes badly.
Grades are a judgement about the evidence, not about how useful the idea is. Plenty of contested effects are still worth knowing, as long as you do not cite them as settled.