Sebastian Smith • 3rd October
How To Lie With Statistics
When taken at face value, what looks like an honest statistic could be inflated, exaggerated, oversimplified, and weaponized; such is the effect that a terror attack could look like child’s play and a cough a pandemic.
“I’m so glad the vaccination program is working, virus deaths have decreased dramatically,” said my mother, not too long ago after she had switched off the television. And so they had – in the neatly cut graph displayed by the British Broadcasting Corporation. It was one that correlated the rollout of the vaccine against the decrease in ‘virus’ deaths in 2021.
My mother’s conclusion was statistically backed in an unusual way. It was a graph, an exceptionally devious one. Like many other sophisticated graphs, it was responsible for selectively cutting the graph to assume a correlation between x and y. Little had the BBC mentioned that the mass downfall of ‘virus’ deaths had taken the exact same trend of downfall last year; rising in the winter, and falling in the summer.
Only this time, the vaccine had been rolled out at the beginning of summer, at the exact same rate that ‘virus’ deaths were going to fall anyway. The BBC had cut out the 2020 deaths, as not to remind viewers that the virus had taken the same pattern last year, and had solely focused on the deaths in 2021, assuming that this time the decrease was the result of the vaccine.
Give me a declining pandemic, and I shall happily give you a vaccine for such the event. Water in a fancy should pass. And so it is all too often that we see, often without knowledge, statistics missing vital context to create an image of solidity in pure wind. When taken at face value, what looks like an honest statistic could be inflated, exaggerated, oversimplified, and weaponized; such is the effect that a terror attack could look like child’s play, and a cough could become a pandemic.
Influenced by and dedicated to Darrel Huff’s 1954 How To Lie With Statistics, the title of this article might sound like a devious endeavour, but should actually be taken as a guard against statistical manipulation. The corporate media may already know these tricks, but the average person must learn them in self-defence. Here is how to lie with statistics.
‘A graduate degree from Oxford can greatly enhance your skills and career prospects’ wrote Oxford University in a recent publication, ‘as well as exclusive resources and opportunities’. According to the university’s research, after just 15 months of graduation, a student makes an average annual salary of £41,600. Well, perhaps I ought to go to Oxford.
But before we come to such a conclusion and pay anywhere between ten to thirty thousand pounds for a degree, what does this figure truly mean? It would appear that if you take the orthodox route down to Oxford University, neither you nor your parents will be struggling for much longer. By assumption, if you can take a forty grand salary within just 15 months of graduation, by ten, twenty years, you could be placed within the top 1% of the country and comfortably retire.
That’s a lovely scheme and a fancy statistic, but beyond its veneer, it lies at two major faults. First of all, the statistic was founded upon a survey sent out to 1,286 Oxford graduates, asking them about their career choices 15 months after graduation. Of those 1,286 surveys sent, 721 responded, but just 320 were willing to disclose their salary. This gave the salary poll a 24% feedback rate, relatively high for a poll, but far too low to find any truth. The misconception here is that the 320 students were a representative sample. They were not. They were the sample who received a poll asking them whether they would disclose their salary and were then confident enough to be motivated to do so.
We cannot be sure about the missing 966 graduates who either completely ignored the poll, or did not disclose their salary – but it can be assumed that they are not the ones living the life as a prominent London lawyer, a marketing guru, or a successful businessman. Most likely, they are just not making the career choice to brag about it. These are the men who, a year after completing their business degree, are at the bottom of some major corporation, waiting for something to happen. Or maybe they are mechanics, amateur writers, barely surviving musicians, and so on.
Even after the fact that over three times as many people ignored the poll than disclosed their salary, the statistic still lies at a second major fault. Namely, the figure shows what Oxford graduates said they earned, and not necessarily what they actually earned. Out of vanity, some people like to exaggerate their incomes, particularly if it is lower than usual or if they are being asked in person. In other polls, if the participant is being asked about something pressuring, such as income, for their income tax, they may very well under-exaggerate their answer.
And so when one of the fanciest universities in the country asks its graduates for their salary, the pressure might be there to at least round it up, or, like most who received it, completely ignore it. Perhaps then it becomes clear that the forty-one thousand average came not from a fair representation, but only those already earning enough to brag.
In March 2019, Forbes magazine named Kylie Jenner the youngest ever billionaire ranking a net worth of $1.2b at just 21. Billionaire at twenty-one, quite an impressive achievement. Two years later, after some new audits, Forbes removed Kylie from their billionaires’ list learning she had a net worth of ‘just under $900m’ – a not so humble exaggeration of $300 million. Although not usually to the tune of nine figures, most people, if not out of fear of judgement, will exaggerate their answers out of pure snobbery. Similarly, in a questionnaire titled ‘how many times per day do you brush your teeth?’ It is almost certain that anybody who does so less than once a day will not admit to it as such. “Twice per day madam – always”.
For the good of example, let’s say you were in the property market looking for a property to purchase. Intending to get my commission, I offer you a house on a road near Oxford. For argument’s sake, you are a snob, so I take it upon myself to tell you that the average income in the neighbourhood is £85,000 per year. Such a statistic might be at fault, but like most, you didn’t question it and are instead bought into the lifestyle. Like most in Britain, sliding in that statistic will reassure your self-esteem, and so you make the sale and mention that statistic to your friends a thew times on the blue moon.
For the latter half of the year, that statistic goes unnoticed. This is until you get approached by me, a neighbour, complaining about the increase in council tax. Understandably so, you are disappointed, but it comes as much more of a surprise when I mention that the average income in the neighbourhood is “just £24,000”. Quite the decrease, yet the neighbourhood hasn’t changed. After such a perversion, you wonder, am I lying now, or was I lying earlier? The charm with manipulative statistics is that you don’t need to lie to create a false impression. In fact, both statistics were based upon the same data and were mathematically arrived at. I never once lied.
The misconception here is that most people, like yourself, will assume that there is just one type of average. There is actually three, the mean, median and mode. The most common type is the mean, which is found by adding up all numbers and dividing them by how many there are. The median is the number value in the middle, and the mode is the most commonly found number. Although they all have the same goal, different types of averages can have massively different conclusions.
The latter figure, I arrived at by using a median, putting all the incomes in order and finding the middle value, equally, I could have found the most common salary and used a mode. When I wanted to inflate that number – incentivize my sale, with £85,000, I used the most common type of average – a mean. I added all the salaries and divided them by how many there were. In this particular neighbourhood, the two averages are so far apart because it happened that most of my neighbours were simple farmers, but two of these inhabitants were millionaire country housers, and therefore the total average was massively skewed.
The tendency to lie – legally, and without falsifying data thus becomes a beautiful trick in the book for any corporation which wants tell sell something that isn’t. One may come across a company publishment that says, “at the Alternative Column, we pride ourselves that our employees are paid on average £65,000 per annum”. Employment at a company that pays on average £65,000 per annum, one might feel their life is on an incline.
After a humble wait, the company called back, offering a salary of… £22,000 per year. “That’s the average salary at our company. We simply can’t afford much more”. The prospecting employee, presumably, might be crying, but the company never once lied. In fact, using the same data collected at the same time, they can both say that the average salary at the company is £65,000 and the average salary at the company is £22,000. Both figures may be wildly different, but both are correct and factually true.
It isn’t, but let’s say that the Alternative Column was run by a crude figure on a £135,000 salary, employing eight writers earning £22,000 per year, and a marketing manager on a £35,000 salary. If this CEO was honest with his averages, he would use a mode to calculate his average salary, which would be £22,000. On the contrary, if he wanted to glaze his figures, he could go for a mean, giving him an average salary of £33,600, and if he had a thew more different salaries at his company, he could go for a median, giving him a gamble at a totally different number. Without a doubt, most executives wouldn’t want to disclose such a disparity in wages between them and their employees. He could, then, bring his wage down and pay himself the rest in bonuses.
The British Broadcasting Corporation (BBC), which I have unfortunately had the burden of noticing is Britain’s most trusted media outlet, wrote in a headline recently, “Fully vaccinated people are much less likely to die with Covid-19 than those who aren’t, or have had only one dose”. For a country that has given its power to the government for over a year, this might sound like a smug headline for the government to brag about. The headline continues… “figures from the Office for National Statistics (ONS).” It is almost universally felt that when data comes from a government agency, it must be more credible. Such is the effect that many sceptics may have failed to look through it.
The headline doesn’t stand on it’s own either: “Out of more than 51,000 Covid deaths in England between January and July 2021, only 256 occurred after two doses.” Impressive information – a statistic with such large numbers, a clear correlation in data and no manipulative use of averages could not possibly be wrong. If I haven’t clarified yet – statistics are genius. And this one, like all the many before, gives a false impression. For most readers, the keywords in this article may be “much less likely to die”, but should actually have been “between January and July”.
The falsity in this figure is the assumption that x causes y; that because such a small number of ‘covid’ deaths were people who were vaccinated, this must be the result of the vaccine. What the BBC did not mention (as they often fail to), is that most vaccines were only administered in early summer, at which point nobody dies from ‘Covid 19’ anyway. In fact, of all 51,000 deaths, almost all occurred in January and early February, at which point most people could not get vaccinated. If the BBC had any respect for Journalism and data, they would have used the one-month data, such as deaths during October. This would have given them the information that 71% of virus deaths were vaccinated – but that would serve neither the BBC nor the government-funded statistics agency very well.
When seasonal deaths are compared with the strangely perfectly timed vaccine rollout, “much less likely to die” suddenly disappears in the ether. If this inaccuracy had been mentioned by the BBC just once, a completely different impression would have emerged. Thus, one must ask, if not to manipulate data, why did the BBC pick January to July data… in mid-September? The use of a number as large as 51,000 suggests a fair distribution in deaths, when in fact almost all of them had happened in January. The form of manipulation here is the false assumption that x causes y, when indeed the statistic given may have nothing to do with the assumption it is placed next to.
For the salesman without a gleaming product, there is a certain dread in numbers. Not every number can be put into an average, or compared to something ridiculous without the reader picking up its fault. Whatever the case, there is a real problem facing the businessman trying to sell the efficacy of his service, the publisher who wants his books to be popular, or the pharmacist trying to sell the effectiveness of his drug. When figures do not work in your best interest and thus create the wrong impression for your aim, there is often one answer left.
Choose a graph, particularly the most effective graph – of the line variety. This is great for showing trends, something that most people are interested in for forecasting their trends. As avid capitalists, we shall draw a graph representing the increase in revenue at the Alternative Column. To begin with, rule your paper into squares, indicating thousands of pounds up the side and the months along the bottom. Plot your data into your graph, and connect them through a line. Your graph may look like this:
In this instance, and in the same instance for many businesses, the data is clear – but is not at all impressive. All is very well if all you wish to do is present honest data. In this case, this would show that throughout the whole year, revenue at this business increased by a measly £2,000. On the contrary, suppose you have other motivations in mind. For example, you want to attract some wealthy investors. In this case, all this graph would show to your investors is incompetence and a lousy increase in profits. To solve this, it would be a step in the right direction to cut your graph as such to focus on the key data.
Now what the reader sees is a revenue line that has climbed halfway up the graph in one year. The figures still remain the same. You have probably saved some paper, but now the empty space indicating the potential profits to be gained is gone from the reader’s mind. No longer are they looking at a graph in perspective to thousands of lost revenue. Unfortunately, this graph is far too small for any reader to not notice its clear cropping. At this point, to complete your graph, it would be beneficial to change the value on the y axis to a tenth of what it was previously.
Now that the revenue on the side is measured by 0.2 thousand increases, instead of 2 thousand increases, your graph takes a much better shape. More importantly, your line is now animated to show a massive increase in revenue. Here once again we have the charm in manipulative statistics, namely that the same data can be represented in a different form to show a completely different trend. Perhaps if you translate “revenue climbed £2000 over the year” into a percentage, “revenues increased by a shocking 10% in just one year”, now the picture becomes very different. The same revenue, only now it is a successful business with avid investors looking for a share. This form of manipulation is vastly more effective because you have not once lied – you have only misled. But no one can pin it on you.
Lying with statistics is like an art. The trick behind the process is not so much falsifying the image as it is picking the right kind of statistic to create the right impression. Even after the exploration of graphs, there is always one last, final option that can make even the poorest of claims stand on its own feet. Take a lousy number – and turn it into a percentage.
“Coronavirus deaths increase by 515% in one month” an Australian newspaper could have honestly written in August 2021. Of course, the impression behind this, at least to a foreigner, is that Australia must be going through some form of medical hell. Ambulances rushing through the streets, overcrowded hospitals, and all the usual reader attracting nonsense. The data behind this statistic is also factual. In July 2021, Australia faced a total of 13 deaths, and in August it rose to 80 deaths. This may be an increase of just 67 deaths but expressed as a percentage is a 515% increase. For a country that faces 170,000 thousand deaths every year, “sixty-seven more deaths” is hardly as headline-worthy as “a rapid 515% increase”.
If the risk of getting a skin disorder – “centrisis” is usually 1 in 5,000, and for those who visit Venice increase their risk of catching it to a 2 in 5000 chance, the increase in risk is just 0.02 but expressed as an almighty percentage, this is an increase of 100%. Just as equally, turning a small figure into an “x times as likely”, will also be efficient – especially if you use a small group. In this case, going through Venice will double your chances of getting “centris”.
Whatever statistic finds itself on the desk of the virtuous politician, the pharmaceutical researcher, the professional salesman, or the innocent journalist simply trying to make his point, it may be hard to find who the guilty party may be. Yet the beauty in manipulative statistics is that you no longer need to lie, or falsify data, to create the same impact as a lie – a false impression.
At any one point, a simple choice of words, percentages, or averages may be used to present data in any way its abuser may like. Why falsify data with lies when the guilty salesman can simply creatively re-word it to his liking? As Britain’s former Prime Minister Benjamin Disraeli once said: “there are three types of lies: lies, damned lies, and statistics.”
My favourite is this one: Bill Gates walks into a working-class bar. Immediately, everyone in the bar is a billionaire – on average.