More or Less Behind the Statistics
More or Less Behind the Statistics

WS More or Less: When maths mistakes really matter

Tim Harford talks to Matt Parker on how simple maths mistakes can cause big problems.

Featured Speakers

BBC HostMatt Parker Guest

Topics Discussed

Episode Summary

Executive Summary: The episode explores how small numerical or procedural mistakes can cascade into major consequences, using a famous BA windscreen failure, rounding tricks in policy and commerce, Excel’s floating-point quirks, and the looming 2038 computer time overflow. Matt Parker argues that errors often arise from systems, assumptions, and too much faith in approximation, not just individual carelessness.

Main Topics: Cascading engineering errors: the BA windscreen disaster (Priority: 5/5): A maintenance chain of tiny errors—wrong bolts, missing tools, poor visibility, and storage mistakes—ended with a cockpit window failure and the pilot being partially sucked out of the aircraft. The story illustrates how minor slipups can interact disastrously. Rounding as a tool for manipulation (Priority: 5/5): Tim Harford and Matt Parker discuss how rounding can be useful, but also how it can be exploited. A Trump-era regulatory proposal attempted to reinterpret a 3x insurance limit as 3.49x by claiming 3.49 rounds down to 3. Everyday pricing and misleading numerical framing (Priority: 3/5): The conversation notes that labeling an item as 'a pound' while charging £1.49 shows how imprecise language and consumer assumptions can obscure the real cost, akin to borrowing more than you repay. Excel and floating-point arithmetic errors (Priority: 5/5): Excel can produce results that are mathematically close but not exactly equal to zero because it stores numbers in binary with limited precision. A simple subtraction example reveals how rounding and representation interact. The 2038 problem and limits of 32-bit timekeeping (Priority: 5/5): Computers that count seconds since 1970 using 32-bit signed integers will overflow in 2038, causing time-related failures in affected systems unless they are updated. Tradeoffs in system design and future-proofing (Priority: 4/5): The episode explains that 1970s engineers chose 32-bit timekeeping partly because memory was expensive, but this short-term efficiency created a long-term technical debt that remains unresolved.

Key Arguments: Small errors can accumulate into catastrophic outcomes when systems lack safeguards or redundancy. Rounding is not just a mathematical convenience; it can be used strategically to reinterpret rules and stretch legal limits. Limited-precision binary storage means computers may display numbers that are mathematically approximate rather than exact. Older design choices can create long-lasting vulnerabilities, as seen in the 2038 time overflow problem. Engineering and software systems need checks at the right stage; otherwise an apparently harmless reformulation can bypass validation. Historical memory constraints explain some design decisions, but they do not eliminate the need to plan for future compatibility.

Data Points: Bolt size difference: 0.026 inches (about 0.66 mm) wider - Difference cited between the A2117D and A2118C bolts used in the BA maintenance example Bolt length difference: 0.1 inches (about 2.5 mm) longer - Second size distinction mentioned in the BA bolt comparison Insurance ratio limit: 3 times - Affordable Care Act rule limiting how much more insurers can charge their most expensive customers than their cheapest customers Proposed reinterpretation: 3.49 times - Trump administration proposal to recast the 3x limit by claiming 3.49 rounds down to 3 Example price: £1.49 - Used to illustrate how something advertised as 'a pound' can still be priced above £1 Subtraction example: 0.5 - 0.4 - 0.1 - Excel example intended to show floating-point rounding behavior Scientific-notation result: Approximately e^-27 / 0.00000000 display - Describes Excel returning a near-zero but nonzero value after forcing a recalculation check 32-bit maximum signed integer: 2,147,483,647 - Maximum number of seconds storable in a 32-bit signed time counter 2038 overflow date: 3:14 a.m., Tuesday 19 January 2038 - Moment when 32-bit Unix-like timekeeping systems will overflow Unix epoch: Beginning of 1970 - Reference point used for computer time counting Mac Terminal command: date +%s - Command given to retrieve the current number of seconds since 1970

Pivotal Quotes: "One tiny mistake after another. One tiny mistake after another after another." — Tim Harford: Describing the chain of maintenance errors that led to the BA windscreen failure "3.49 rounds down to three." — Tim Harford quoting the regulatory argument: Illustrating how rounding was used to justify a looser insurance rule "If you put it all in brackets and multiply it by 1, it doesn't do the check at the Right time, and you get an answer which is close to zero, but not zero." — Matt Parker: Explaining Excel's floating-point and evaluation-order quirk

Implications: Listeners are reminded to treat rounded figures, software outputs, and legacy systems with caution. Small numerical shortcuts can mislead, and old design decisions may create future failures unless systems are tested and updated.

🔓 Sign Up for Unlimited Episode Search

About More or Less Behind the Statistics

Tim Harford and the More or Less team try to make sense of the statistics which surround us. From BBC Radio 4

View all episodes from More or Less Behind the Statistics