{"id":"601d187a6cee070007e6c11b","slug":"calling-bullshit-en","title":"Calling Bullshit","audio_url":"https://hls.blinkist.io/bibs/601d187a6cee070007e6c11b/601d187a6cee070007e6c11d-T1612519636.m4a?Expires=1643412363&Signature=HDuboPcXIJTmo2r03KoMqKzL-wWIfrUiqncq3AtuKFTjS10uVuaK9p8M346Wx~CBo319DEd4XYuWy1rH-f7tVqsmbxzdySqMtoeOEAQfe~ArFXLRVVu9Z4A-7~YHCkxVvFDvERTlBQs7G6WUHcarXoy2CTlU10-r-O1j9TgYyoYE~VagQlNEfAtoq7rl6znU8u-rKFvG~FbhOdsiHtKHlxUQjs7K4YLL4x45cZwb0rE7SxkRKJzH~IIPK~AMHsQTw9G2U4hb1UOYJT71aevozyl~-AeNR1QlgCwsEQum96kOVbRbgy0BAtcldvc1ULEK6CXP3hh7FnQut4ca~dl9GA__&Key-Pair-Id=APKAJXJM6BB7FFZXUB4A","chapters":[{"id":"601d187a6cee070007e6c11d","title":"We all need to be alert to the dangers of bullshit.","text":"

In 1998, the medical journal The Lancet published a study co-written by British physician Andrew Wakefield. It claimed that there may be a link between the widely used MMR vaccine and autism.

But there wasn’t. Multiple later studies found no link at all, and it turned out that Wakefield’s study was deeply flawed. The Lancet even retracted the paper in 2010. It now counts as one of the most comprehensively discredited studies of all time. It was bullshit.

And yet its influence has been huge. The “antivax” movement is still going strong, vaccine rates in the US are lower than they were before, and cases of measles are up.

The sad truth is that it’s far, far easier to make people believe bullshit than it is to make them change their minds.

But we all have a duty to try.

The key message here is: We all need to be alert to the dangers of bullshit.

Bullshit isn’t just a modern phenomenon. Back in ancient Greece, Plato accused the Sophists, a rival school of philosophy, of dealing in bullshit. He said they were only interested in winning arguments, not in what was actually true.

The twenty-first century, however, provides bullshit with an especially fertile breeding ground. These days, bullshit often claims to be rigorously grounded in science, like Wakefield’s vaccination study. Or it uses seemingly irrefutable evidence like photographs.

You might remember a story that came out after the Boston Marathon bombing in 2013. The story claimed that an eight-year-old girl from Sandy Hook Elementary School had been killed – there was even a photo of her running. The story was shared by more than 92,000 people on social media.

You guessed it: it was false. The girl in the photo hadn’t even run the race – it wasn’t open to children. But the story just proved too irresistible not to share.

That’s just one example of how modern technology like social media can spur bullshit on. If bullshit like the vaccination study can catch on as widely as it did back in 1998, imagine the harm that can be done in the age of Twitter.

Factor in our hyperpartisan news networks, factories churning out fake news, and the ease of image manipulation, and we’ve got ourselves a full-on bullshit crisis. We need to take action – now.

"},{"id":"601d187a6cee070007e6c11e","title":"Bullshitters try to persuade people to believe something is true, without really caring about the evidence.","text":"

Before we get into how to disprove bullshit, let’s ask a simple question. What exactly is bullshit?

According to the authors, bullshit arises out of a determination to impress or persuade people. Bullshitters care more about winning the argument than about whether they’re telling the truth.

The key tools that a modern bullshitter has at his disposal are language, statistics, and graphics. And he uses them to barrage the audience with an overwhelming amount of information. A lie can simply be something that isn’t true – but bullshit will be dressed up to look as much like the truth as possible.

Here’s the key message: Bullshitters try to persuade people to believe something is true, without really caring about the evidence.

One classic bullshitting method is to use what the science sociologist Bruno Latour called black boxes. Think of it like this: You collect some data, and then feed that data through a complicated scientific process like an algorithm. That process is the black box, and whatever comes out of it tends to get treated as objective fact.

But here’s the thing. Even without getting into the details of how the black box processes the data, there’s often still plenty to criticize. For starters, always look at what data is actually being used.

Here’s an example. An experiment in 2016 claimed to prove that criminals and non-criminals had differently shaped heads – and that an algorithm proved it.

According to the results, criminals and non-criminals had a slightly different angle between the nose and the mouth, and a different curvature of the lip. But before you get your tape measure out, consider this: in the study, the photos of criminals were all government-issued ID photos, whereas the non-criminals were represented by professional headshots.

You probably don’t need a supercomplex algorithm to realize that people are more likely to smile in a professional headshot than in an ID photo.

In other words, the data set used in this study was useless – so the results were, too. It’s possible to refute the study even without opening up the algorithmic black box.

Did the study’s authors set out to deceive people? Probably not. But they were so concerned with proving their hypothesis that they looked past the fact that the data they were using was flawed. The result? You’ve guessed it. Pure bullshit.

"},{"id":"601d187a6cee070007e6c11f","title":"Correlation does not imply causation.","text":"

Not all bullshit scientific studies produce surprising results. Some, in fact, produce results that are incredibly obvious. For instance, a recent study looking at college students found a correlation between positive self-esteem and having had a first kiss before starting college.

But what does that result actually prove? Does it mean that naturally confident people start kissing younger? Or is the first kiss the cause of a boost in self-esteem?

And hold on – why are we even talking about kissing? What is the role of romantic relationships? Isn’t it possible that both kissing and high self-esteem are the outcomes of finding yourself in a relationship?

Sure, the study’s findings are obvious. But even though it shows a correlation, it doesn’t tell us anything about why these two things are linked.

The key message is this: Correlation does not imply causation.

This is a form of bullshit that the media plays a big role in spreading. Sometimes, a statistical study will be very careful not to imply a causal link between two things – but when the study ends up in the newspapers, all the subtlety is often lost.

A 2018 report by Zillow, a real estate website, provides an example. It showed that cities with rising house prices also tended to have lower fertility rates for women in their late twenties. But, as the report noted, the one trend hadn’t necessarily caused the other.

For instance, underlying concerns about money and career development might be affecting people’s decisions about both housing and parenting. Perhaps they were simply planning to have children later – the study didn’t even consider women over 30. Zillow’s report didn’t provide answers – it merely described a correlation.

But when the press got hold of it, they were less careful. Stories used words like “cause” and “effect,” suggesting that it was specifically the high house prices that were causing fewer young women to have children. The original study wasn’t bullshit – but the reporting about it was.

It’s also worth remembering that sometimes a correlation doesn’t tell us anything interesting at all. If you plot the prevalence of autism against the level of organic food sales, you’ll find a very tight correlation indeed: both lines have risen quite sharply in recent years. But it would be ridiculous to suggest a connection. They’re simply two things that have risen over time.

"},{"id":"601d187a6cee070007e6c120","title":"It’s disturbingly easy to make numbers say anything you want them to.","text":"

One evening, one of the authors – Carl – found himself in a hotel lobby and in need of a hot drink. He picked up a packet of cocoa. Not wanting to dose up on caffeine before bed, Carl decided to check the packaging and was pleased to read that it was “99.9 percent caffeine free.” Until he thought about it for a moment.

It turns out that a Starbucks coffee contains 415 milligrams of caffeine per 20 ounces, making a standard cup approximately 0.075 percent caffeine. Which means that coffee itself is 99.9 percent caffeine free.

So on a packet of hot cocoa, it’s not much of a claim.

The key message here is: It’s disturbingly easy to make numbers say anything you want them to.

Not all examples are as harmless as a cup of hot cocoa. In 2017, the website Breitbart wrote that 2,139 people with DACA status – people who entered the US illegally when they were minors and were granted amnesty – had been convicted or accused of crimes.

But that was out of a total of 700,000 people – fewer than one in every 300. There’s twice as much chance of a US citizen currently being incarcerated as there is of a DACA recipient having been accused of a crime. That number – 2,139 – sure sounds scary, though.

By choosing whether to use numbers or percentages, you can make the scale of a problem appear big or small. And you can do the same thing when talking about percentage increases.

According to a press release from The Lancet, drinking one alcoholic drink every day, rather than zero, increases the risk of developing an alcohol-related health problem by 0.5 percent. Which sounds pretty serious.

But how prevalent are these alcohol-related problems in the first place? They occur in just one percent of people who don’t drink. So a 0.5 percent increase takes the risk to 1.005 percent.

It’s vital to remember the difference between percentage differences and percentage point differences. In that example, the percentage difference of 0.5 seemed large, but the percentage point difference was a tiny 0.005.

In other words, people can produce bullshit with numbers even without resorting to lies. Presenting them in the right way – or the wrong way – can have a huge impact. So it’s vital to keep on your toes.

"},{"id":"601d187b6cee070007e6c121","title":"When the data you use for a test isn’t neutral, the results will show selection bias.","text":"

Statistics are all around us. But where do they come from? How do we know, for example, that Dutch men are the world’s tallest? Very often, the data used is only a sample. In the Netherlands, they didn’t measure every single man, but rather just a sample from across the population.

But imagine if that sample had included a basketball team.

Or imagine you go to an organic market, and ask people their political views. Chances are, you won’t get a representative sample of the population – you’ll probably find disproportionately more liberals there.

This is known as selection bias, and it’s a serious problem in statistics.

This is the key message: When the data you use for a test isn’t neutral, the results will show selection bias.

Here’s an example that shows the strange effects selection bias can have. True or false: Hot guys are jerks.

Let’s assume there’s no actual correlation: if you plot attractiveness against niceness on a graph, the dots are randomly distributed. However, some dots on that graph are men you’d never consider dating. So by filtering out all the complete jerks, and everyone you find extremely unattractive, we can basically remove all the dots on one side of a straight line.

The dots that now remain will show a correlation, because of the way we’ve just adjusted the sample. So while hot guys aren’t jerks in general, of the ones you’d consider dating, there is a trend in that direction.

That’s an unfortunate fact of life, but it isn’t bullshit. The next example, however, is. Pretty much every car insurance agency claims that switching to them saves people an average of some high figure like $500 a year. How can that possibly be true?

Well, the only people who actually bother to switch are the ones who’ll save serious money by doing so – otherwise, they’ll just stay put. Inevitably, that figure of $500 only includes an unrepresentative sample.

In clinical trials, this problem is known as data censoring. Sometimes people drop out of trials for a new drug if they experience side effects, so their experiences aren’t recorded in the data. If a nonrandom selection of people drops out of an experiment, that will inevitably skew the results.

Truly random samples are harder to obtain than you might imagine, because selection bias is everywhere. It’s one more reason why you need to think carefully about the statistics you hear.

"},{"id":"601d187b6cee070007e6c122","title":"Don’t be dazzled by big data and machine learning – the underlying data still has to be sound.","text":"

Using modern technology, graphic designers can create elaborate and sometimes ridiculous diagrams – like a chart about goats that’s made to fit into the shape of a ram’s horn, or a “subway map” of Bible books or music genres.

Those things may simply offer a fun way of listing a few facts or points about something. But that doesn’t mean that conventional-looking graphs are automatically more serious or accurate. If you’re looking at a bar chart, always check that the y-axis goes all the way down to zero. If it doesn’t, you won’t see things in the correct proportion.

The other problem that tech presents us with is the rise of big data, in which the research isn’t always built on sturdy foundations.

Here’s the key message: Don’t be dazzled by big data and machine learning – the underlying data still has to be sound.

The huge datasets that get called “big data” tend to be fed into mysterious algorithms that teach themselves to do anything from recognizing faces to trading stocks. This is known as machine learning, and it’s a great example of the sort of mysterious, inscrutable “black box” we talked about earlier.

We’ve already seen one example of machine learning creating bullshit: the algorithm that allegedly identified criminals based on their faces. Other examples are even more ridiculous.

One machine was fed chest X-rays, in the hope that it could pick out images in which there were heart or lung issues. It did so correctly – but only because it noticed that the scans of unhealthy chests tended to have some text printed in the corner. All of the unhealthy scans it had been fed had been made by one particular machine. Tested on other scans, it didn’t work.

Another example went wrong on a larger scale. In 2008, Google started developing Google Flu Trends, aiming to predict how the flu would move around the US based on searches for terms like “flu symptoms” or “pharmacies.”

Google’s algorithm ended up looking for terms clearly unrelated to flu – like “high school basketball” – simply because searches for them also peaked in winter. And over time, its flu predictions got worse and worse.

Looking at the past, the algorithm was able to find correlations between flu trends and other factors. But, as we know, correlation does not imply causation. So when it looked to the future, it fell flat on its face.

The potential of machine learning is considerable. But sometimes it takes a human to tell it’s just producing bullshit.

"},{"id":"601d187b6cee070007e6c123","title":"The imperfections of modern science mean that bullshit creeps in everywhere.","text":"

One of the great things about science is that it corrects itself. If one study finds an interesting result, others will try to replicate it, ironing out any problems and gradually making progress.

That means that there’s really no such thing as absolute scientific truth. Rather, science is simply the accumulation of all the experiments that have been done up until now.

In fact, the scientific system we have today is filled with problems. For instance, think about which studies tend to get published in journals. Usually, it’s the ones that show positive results. But what if ten other studies have already tried to prove the same thing, and all failed? Those findings are often not reported at all.

In other words, scientific publication suffers from selection bias. And that’s just one way in which bullshit is baked right into the modern scientific system

The key message here is: The imperfections of modern science mean that bullshit creeps in everywhere.

As a rule of thumb, scientists tend to accept that a correlation is statistically significant if it has a p-value of 0.05 or less. That means there’s less than a five percent likelihood that the correlation is the result of chance.

Sounds reasonable, right? But it’s worth thinking about Goodhart’s Law. This states that as soon as a measure becomes a target, it becomes ineffective, because people try to game the system to get there.

Sadly, that includes scientists. As we’ve already seen, it’s easy to manipulate numbers to make them say what you want. By being selective about an experiment’s results, it’s often quite easy to engineer a p-value of 0.05 or less. It’s called p-hacking, and it’s widespread.

Another major issue is how stuff gets reported in the media – or whether it gets reported at all. Only a small fraction of scientific research gets reported on, and inevitably the studies that make the best headlines are the ones that get covered. So of course there’s selection bias galore.

There are problems within scientific journals, too. Less-esteemed journals tend to be quite unselective about what they publish, if the author is willing to pay. Thankfully, that gives rise to one relatively easy way to spot scientific bullshit: if you see a big claim made in a tiny journal, be skeptical. If there was anything credible about the result, rest assured that the researchers would have got it published somewhere better.

"},{"id":"601d187b6cee070007e6c124","title":"Through a few simple techniques, you can equip yourself in the fight against bullshit.","text":"

Journalists don’t get everything right, especially when they report on science. But one habit of theirs is well worth bearing in mind. When you’re trying to work out whether something is true, ask three questions, like a journalist would: Who is behind this information? How did they get it? And what are they trying to sell?

It’s not the only way we can identify bullshit – but it’s a great way to start.

This is the key message: Through a few simple techniques, you can equip yourself in the fight against bullshit.

You know the expression “Too good to be true”? Most of the time, it’s pretty much on the money. If you hear something that just doesn’t seem plausible, there’s a very healthy possibility that it’s bullshit.

One great way to work out how plausible something is is to make Fermi estimations. These are very rough approximations you can do in your head that’ll tell you if something’s more-or-less right in terms of scale.

Say someone told you there were 121,000 John Smiths in the UK. Let’s fact-check quickly. How many people are in the UK? More than ten million, but fewer than a billion. So let’s say 100 million. How many of those people are called John? Fewer than one in ten, but surely more than one in a thousand – so let’s say one in a hundred. It’s probably about the same for people called Smith.

That gives us the number of John Smiths as 100 million divided by a hundred, divided by a hundred – making it ten thousand. That isn’t very accurate, but it’s enough to make it clear that 121,000 is bullshit.

Another tip against bullshit is to be wary of confirmation bias. This is the tendency to believe information that confirms what we already think. We’re all guilty of this from time to time – so exercise extra caution with information you like the sound of.

Remember correlation and causation, too, and be especially skeptical if you read that one trend has “caused” another – there could be a completely separate reason for each.

And finally: be extra cautious with information you’ve found via unreliable online sources like Twitter.

What do you do when you find bullshit? Call it out. But the authors ask you to be polite when you do so. After all, we all make mistakes. And what’s more, you’re far more likely to change someone’s mind if you’re nice about it.

"},{"id":"601d187b6cee070007e6c125","title":"Final summary","text":"

The key message in these blinks:

Bullshit is the all-too-common art of persuading people of something, while not really caring about the truth. In the modern world of social media and big data, you need to be extra wary that you don’t get taken in. By understanding that correlation does not imply causation, considering what the numbers we’re told really mean in context, and being skeptical about the merits of datasets, you can better equip yourself to tell fact from bullshit.

Actionable advice:

Call bullshit by getting the facts right.

Simply identifying bullshit isn’t enough. It’s up to all of us to call out bullshit when we see it, so that more and more people can see how often we’re taken in by bogus statistics. But when you do this, it’s vital to get the facts right. So make sure you have the correct figures in hand before you start taking someone else to task. And if you make a mistake, admit it. Otherwise, you’re just another bulshitter.

Got feedback?

We’d love to hear what you think about our content! Just drop an email to remember@blinkist.com with Calling Bullshit as the subject line, and share your thoughts!

What to read next: Not Born Yesterday, by Hugo Mercier

These blinks have explained some of the many forms that bullshit takes in the modern world, and helps you call them out. But maybe humans aren’t so bad at that after all.

In the blinks to Not Born Yesterday, you’ll hear how people make decisions about trust and belief, and why most historical attempts at mass persuasion have failed. To find out more about the hardwired bullshit detectors inside us all, have a listen.

"}]}

Log in to Blinkist

Sign up for Blinkist

Learn more, live more