What is the Funnel Experiment?
Deming's Funnel Experiment is a demonstration designed to illustrate the adverse effects of making adjustments to a stable process without understanding the causes of variation. This is a practice especially common in managers who do not understand Systems of Human Performance. In the original experiment, marbles are dropped through a funnel onto a target on a sheet of paper. The goal is to get each marble as close to the target as possible.
The experiment demonstrates four different "rules" or management approaches for adjusting the funnel position based on where previous marbles landed:
- Rule 1: Leave the funnel alone - No adjustments are made regardless of results
- Rule 2: Adjust opposite to the last result - Move the funnel in the opposite direction by the amount of the last deviation
- Rule 3: Move funnel over the target - After each marble, reposition the funnel directly over the target
- Rule 4: Move funnel over the last marble - Position the funnel directly above where the previous marble landed
What the Experiment Teaches About Tampering
The Funnel Experiment reveals that Rule 1 (no adjustment) consistently produces the best results with the least variation around the target. The other three rules represent different forms of "tampering" - well-intentioned adjustments that actually make the process worse by increasing variation and moving results further from the desired target.
Key Lessons:
- Natural Variation is Normal: Every stable process has inherent variation that cannot be eliminated through adjustments
- Tampering Increases Variation: Making adjustments based on individual data points typically doubles, triples, or exponentially increases process variation
- Good Intentions, Poor Results: Logical-seeming adjustments (like Rule 3's "aim for the target") often backfire
- System vs. Special Causes: Adjustments should only be made when there are identifiable special causes, not for common cause variation
Deming's Funnel Experiment
Interactive Simulation of Process Tampering Effects
Statistics
Rule 1: Leave the Funnel Alone
Do not adjust the funnel position regardless of where the marble lands. This represents the natural variation of a stable process. This rule typically produces the best long-term results with the least variation around the target.
Key Insights
Start dropping marbles to see how different adjustment rules affect process variation. Rule 1 (no adjustment) typically performs best, while the other rules demonstrate various forms of "tampering" that increase variation over time.
How the HTML Simulation Works
The interactive HTML simulation recreates the Funnel Experiment digitally, allowing users to experience the effects of different adjustment rules without physical materials.
Learning Experience:
Users can:
- Observe Patterns: Watch how different rules create different scatter patterns
- Compare Outcomes: Switch between rules to see immediate differences in performance
- Understand Statistics: See numerical proof of which approaches work better
- Experience Insights: Discover counterintuitive truths about process management
The simulation makes Deming's abstract concepts tangible and provides immediate feedback on the consequences of different management philosophies. Users typically discover that their intuitive responses (Rules 2-4) perform worse than simply leaving the process alone (Rule 1), providing a powerful lesson about the dangers of tampering with stable processes.
This hands-on experience helps managers and quality professionals understand when to act and when to resist the urge to "fix" processes that are actually performing normally within their natural variation limits.
Real-World Applications:
The experiment directly relates to common management mistakes such as:
- Adjusting production processes after every quality measurement
- Changing budgets based on monthly variances from previous periods
- Modifying organizational policies after each employee survey
- Setting quotas based on recent output rather than system capability
- Making personnel changes in response to normal performance variation