Thinking tool: Be sceptical

Headshot of Adam Jones

Adam Jones

There are a lot of surprising claims in the world. Some of them are even true!

But people often overstate evidence, or communicate imprecisely. Unless you're careful, you end up with wrong beliefs.

So when evaluating important claims, dig in carefully. (Or at least, only hold beliefs as strongly as the evidence supports!)

Wheeler Hall at the University of California, Berkeley: a large neoclassical stone building seen across an open plaza, framed by trees, with the Campanile bell tower rising behind it against a blue sky (Wheeler Hall at UC Berkeley. Photo: Bob Collowan, CC BY-SA 3.0, via Wikimedia Commons.)

UC Berkeley's 1973 gender bias controversy is a good case study: it shows two ways a claim can mislead.

The first is that a true statistic can still point at the wrong conclusion. Concern began when the university realised 44% of the 8,442 men who applied were admitted, against 35% of the 4,321 women - a gap too big to be chance. The obvious conclusion: bias against women. But department by department, women tended to be favoured - they had applied disproportionately to the competitive departments. (This is the canonical Simpson's paradox example.) Understanding the underlying facts lets you act on how the world actually is - here, prompting the more fruitful follow-up "why do men and women apply to such different departments?".

The second is that shit just gets made up. Many textbooks say Berkeley was sued for discrimination over this. It (probably) wasn't - more likely, administrators worried about bias asked a statistician to check. But I'm not confident there definitely wasn't a lawsuit either. Many things are lost to time, and while it's unsatisfying, "we don't know" beats asserting a false statement either way.

More examples of statistics being misleading:

  • The denominator is missing. Insurers, law firms and driving centres like to report that 52% of crashes happen within five miles of home, so you should concentrate when driving near home. But there are several issues:

    • To know if near-home driving is actually more dangerous, you'd need the share of miles driven within five miles of home - which is probably large. I can't find that stat, and I'm fairly confident almost nobody quoting the 52% knows whether near-home driving is more or less dangerous per mile.
    • Even if it were more dangerous, it's not clear concentrating more helps - perhaps it raises stress or fatigue, and backfires.
    • And even if concentrating did help, why only near home? The other 48% of crashes presumably happen further out, and it would be odd if concentration helped within five miles of home but not beyond.

  • The causation runs the other way. Countries with more doctors per capita record more cancer - are doctors bad for you? More likely: with more doctors, people live long enough to get cancer rather than dying young of infectious diseases, or cancer was happening at the same rate but not being recorded.

    • Similarly, cancer screening programmes often substantially increase 5-year survival rates (the percentage of people alive 5 years after diagnosis). But you'd expect this even if screening changed nothing: earlier detection means longer survival after diagnosis even if people die at the same point, and screening catches milder cases (severe ones were already presenting to doctors).

  • True facts, selected to imply a conclusion. Consider this classic warning about dihydrogen monoxide:

    Dihydrogen monoxide is the major component of acid rain, contributes to the greenhouse effect, may cause severe burns, accelerates corrosion of many metals, and has been found in excised tumours of terminal cancer patients.

    Despite this it is used as an industrial solvent and coolant, in nuclear power plants, in the production of styrofoam, and in the distribution of pesticides - and even after washing, produce remains contaminated by this chemical.

    In 2013, a statement like this caused concerned residents to call their water company after a radio show alerted them to dihydrogen monoxide coming from their taps. The radio show hosts were threatened with a felony (although no charges were pursued) for sharing this information.

    Dihydrogen monoxide is, of course, the chemical name for water. All of the facts there are true, but the framing and implications suggest a false conclusion. The technique is sadly common in reporting and analysis.

More examples of made-up facts:

  • You might believe spinach is an exceptional source of iron. It isn't - it's fine, but comparable to plenty of other green vegetables, and its iron is poorly absorbed.

    A bowl of fresh spinach leaves on a grey surface, with more loose leaves scattered beside it

    A popular explanation for the myth: a 19th-century German chemist misplaced a decimal point, inflating the number tenfold. But that explanation is itself probably not true! Someone suggested it as a possibility in 1972, a doctor repeated it as fact in 1981 and later admitted he couldn't remember where he'd got it from, and everyone since has cited him. So: a made-up fact, explained by a made-up fact about how facts get made up.

  • "Cybercrime will cost the world $10.5 trillion a year by 2025" is everywhere - the World Economic Forum, vendor whitepapers, conference keynotes. It comes from Cybersecurity Ventures, extrapolated from a similarly made-up 2015 figure at an assumed 15% growth rate.

    I contacted them to ask where they got the base 2015 figure or the growth rate and received no reply.

    (I do advocate for made up numbers - but they need to be vaguely grounded in reality, or treated as fully made up! It saddens me to see this figure quoted so widely. And my guess is it's not a one-off - just one of the few figures I've tried to trace.)

Cautions

Being sceptical doesn't mean being cynical. Scepticism means being open to new evidence, and wanting to pursue it properly.1 Default-dismissing evidence is directly at odds with this!

Evidence can take many forms. In most of life you won't find a peer-reviewed, replicated, randomised controlled trial telling you exactly what to do. Someone random making a claim online is some evidence - just hold it with much more uncertainty.

You can still believe and act on things with partial evidence. Almost nothing in the world is fully "proven". Believing the status quo without evidence is just as silly as believing an alternative theory without evidence. (Although a reasonable status quo bias is usually supported by some evidence!)

This is part of my thinking tools series. Also consider:

Footnotes

  1. There are different working definitions of scepticism, and perhaps this is a semantics question. I like the Wiktionary definition:

    A methodology that starts from a neutral standpoint and aims to acquire certainty through scientific or logical observation.

    The takeaway: be rigorous about getting to the root of claims before relying on them for important things, rather than attached to a specific word for all this. If you have a better word, let me know! ↩