Context
The elections are US data from 1961–2025. The estimate is local to close elections and state–industry cells.
01 THE SHORT ANSWER
A win has two places it can travel.
First, through the election record: close wins are followed by more wins in the same state–industry combination. Then, through workers’ minds: information about wins changes stated support and expectations about coworkers.
These are different outcomes. The page keeps them together so a change in opinion does not become a promise about union formation.
Start with the election threshold- 01The result
A close win opens a path to later wins.
- 02The expectation
Win information moves support and beliefs about coworkers.
- 03The boundary
Interest in contacting an organizer does not rise significantly.
02 THE HISTORICAL STEP
First, isolate a win that barely crossed the line.
The authors compare NLRB elections clustered around the 50% threshold. When a few votes change the result, the comparison can isolate what follows a win that almost did not happen.
The reported estimate is 0.284–0.689 additional wins in the same state–industry cell over the next three years. It is not a multiplier for every workplace: it is a range across specifications in a local setting.
SWITCH WHAT CHANGES AT THE THRESHOLD?
A close win and a close loss are not the same thing.
The interaction changes the scenario; it does not invent an estimate for a loss.
CLOSE WIN
+0.284–0.689
additional wins in the same state–industry cell over the next three years.
03 THE EXPERIMENTAL STEP
Then ask what information changes in a worker’s mind.
In an experiment with 9,089 hourly workers in US retail and service jobs, participants received information about union wins, how easy it was to find a job, both, or neither.
The switch below puts the same measures on one stage. Support values are standard deviations; the contact-interest value is in percentage points.
9,089hourly workers
retail + service
US
retail + service
US
6,000 YouGov + 3,089 Prolific. The main support analysis starts with participants who underestimated union activity.
SWITCH WHAT INFORMATION DO YOU GET?
Separate the labour market from the union signal.
Results are effects relative to control, not absolute support scores.
UNION-WIN INFORMATION
Support and expectations about coworkers move after information about wins.
Union support+0.091 SD
Perceived coworker support+0.076 SD
Interest in an organizer+0.3 pp
No statistically significant effect for contact interest. The control baseline for this measure is 23.9%.
04 THE STOPPING POINT
The chain stops at costly local action.
For interest in contacting an organizer, the union-win information effect is +0.3 percentage points and is not statistically significant. The control group mean is 23.9%.
That does not mean information can never change action. It means this experiment measures a precise boundary: support and beliefs about coworkers move more clearly than reported interest in contacting an organizer.
MEASURED / UNMEASURED
✓
Support and expectationspositive reported effects
?
Contact with an organizer+0.3 pp · not statistically significant
?
Union formationnot measured by this design
05 EVIDENCE LEDGER
Read the estimate inside its fences.
Each row answers a different question. Mix them together and the story becomes larger than the source.
EVIDENCENUMBERWHAT IT SUPPORTS / DOES NOT
HISTORICAL+0.284–0.689
Additional wins over 3 years, same state–industry cell. Not a universal multiplier.
EXPERIMENT+0.091 SD
+0.076 SD
+0.076 SD
Union support and perceived coworker support after win information. Not union formation.
BOUNDARY+0.3 pp
n.s.
n.s.
Interest in contacting an organizer. “Not significant” does not mean “zero,” and this is not observed action.
06 WHAT IT DOES NOT SAY
Three fences keep the conclusion honest.
Population
The experiment uses online-recruited hourly workers in retail and service jobs. It is not an estimate for all workers or for Romania.
Status
The source is an August 2026 NBER working paper, not a peer-reviewed article. The tables and PDF may change.
This is not a prediction. It is a map of evidence: a result can change what feels possible, but the page does not claim that signal alone organizes a workplace.
07 SOURCE
A working paper, visible down to the unit.
The page’s figures come from one primary source, checked and archived locally for audit. The unit conversions are explained in the evidence ledger.
Winning is Contagious: Social Learning and the Dynamics of Union Support
Ellora Derenoncourt, Arindrajit Dube, Suresh Naidu, Heather Sarsons and Niharika Singh · NBER Working Paper 35571 · DOI 10.3386/w35571 · August 2026.
This project synthesizes reported results and does not republish individual data, identities, or source figures. For definitions, formulas and limits, open the Markdown version or the NBER source.