Data literacy for journalists
Just as literacy refers to 'the ability to read for knowledge, write coherently and think critically about printed material' data literacy is the ability to consume for knowledge, produce coherently and think critically about data.
Data literacy includes statistical literacy but also understanding how to work with large data sets, how they were produced, how to connect various data sets and how to interpret them.
Poynter’s News University offers classes of Math for journalists, in which reporters get help with concepts such as percentage changes and averages. Interestingly enough, these concepts are being taught simultaneously near Poynter’s offices, in Floridian schools, to fifth grade pupils (age 10-11), as the curriculum attests.
That journalists need help in math topics normally covered before high school shows how far newsrooms are from being data literate. This does not go without problems. How can a data-journalist make use of a bunch of numbers on climate change if she doesn’t know what a confidence interval means? How can a data-reporter write a story on income distribution if he cannot tell the mean from the median?
A reporter certainly does not need a degree in statistics to become more efficient when dealing with data. When faced with numbers, a few simple tricks can help her get a much better story. As Max Planck Institute professor Gerd Gigerenzer says says, better tools will not lead to better journalism if they are not used with insight.
Even if you lack any knowledge of math or stats, you can easily become a seasoned data-journalist by asking 3 very simple questions.
1. How was the data collected?
Amazing GDP growth
The easiest way to show off with spectacular data is to fabricate it. It sounds obvious, but data as commonly commented upon as GDP figures can very well be phony. Former British ambassador Craig Murray reports in his book, Murder in Samarkand, that growth rates in Uzbekistan are subject to intense negotiations between the local government and international bodies. In other words, it has nothing to do with the local economy.
GDP is used as the number one indicator because governments need it to watch over their main source of income - VAT. When a government is not funded by VAT, or when it does not make its budget public, it has no reason to collect GDP data and will be better-off fabricating them.
Crime is always on the rise
'Crime in Spain grew by 3%', writes El Pais. Brussels is prey to increased crime from illegal aliens and drug addicts, says RTL. This type of reporting based on police-collected statistics is common, but it doesn’t tell us much about violence.
We can trust that within the European Union, the data isn’t tampered with. But police personnel respond to incentives. When performance is linked to clearance rate, for instance, policemen have an incentive to reports as much as possible on incidents that don’t require an investigation. One such crime is smoking pot. This explains why drug-related crimes in France increased fourfold in the last 15 years while consumption remained constant.
What you can do
When in doubt about a number’s credibility, always double check, just as you’d have if it had been a quote from a politician. In the Uzbek case, a phone call to someone who’s lived there for a while suffices (‘Does it feel like the country is 3 times as rich as it was in 1995, as official figures show?’).
For police data, sociologists often carry out victimisation studies, in which they ask people if they are subject to crime. These studies are much less volatile than police data. Maybe that’s the reason why they don’t make headlines.
Other tests let you assess precisely the credibility of the data, such as Benford’s law, but none will replace your own critical thinking.
2. What’s in there to learn?
Risk of Multiple Sclerosis doubles when working at night
Surely any German in her right mind would stop working night shifts after reading this headline. But the article doesn’t tell us what the risk really is in the end.
Take 1,000 Germans. A single one will develop MS over his lifetime. Now, if every one of these 1,000 Germans worked night shifts, the number of MS sufferers would jump to 2. The additional risk of developing MS when working in shifts is 1 in 1,000, not 100%. Surely this information is more useful when pondering whether to take the job.
On average, 1 in every 15 Europeans totally illiterate
The above headline looks frightening. It is also absolutely true. Among the 500 million Europeans, 36 million probably don’t know how to read. As an aside, 36 million are also under 7 (data from Eurostat).
When writing about an average, always think 'an average of what?' Is the reference population homogeneous? Uneven distribution patterns explain why most people drive better than average, for instance. Many people have zero or just one accident over their lifetime. A few reckless drivers have a great many, pushing the average number of accidents way higher than what most people experience. The same is true of the income distribution: most people earn less than average.
What you can do
Always take the distribution and base rate into account. Checking for the mean and median, as well as mode (the most frequent value in the distribution) helps you gain insights in the data. Knowing the order of magnitude makes contextualization easier, as in the MS example. Finally, reporting in natural frequencies (1 in 100) is way easier for readers to understand that using percentage (1%).
3. How reliable is the information?
The sample size problem
'80% dissatisfied with the judicial system', says a survey reported in Zaragoza-based Diaro de Navarra. How can one extrapolate from 800 respondents to 46 million Spaniards? Surely this is full of hot air.
When researching a large population (over a few thousands people), you rarely need more than a thousand respondents to achieve a margin of error under 3%. It means that if you were to retake the survey with a totally different sample, 19 times out of 20, the answers you’ll get will be within a three percentage points interval of the value you would have found, had you asked every single person.
Drinking tea lowers the risk of stroke
Articles about the benefits of tea-drinking are commonplace. This short item in Die Welt saying that tea lowers the risk of myocardial infarction is no exception. Although the effects of tea are seriously studied by some, many pieces of research fail to take into account lifestyle factors, such as diet, occupation or sports.
In most countries, tea is a beverage for the health-conscious upper classes. If researchers don’t control for lifestyle factors in tea studies, they tell us nothing more than ‘rich people are healthier - and they probably drink tea’.
What you can do
The math behind correlations and error margins in the tea studies are certainly correct, at least most of the time. But if researchers don’t look for co-correlations (e.g. drinking tea correlates with doing sports), their results are of little value.
As a journalist, it makes little sense to challenge the numerical results of a study, such as the sample size, unless there are serious doubts about it. However, it is easy to see if researchers failed to take into account relevant pieces of information.
This piece is part of the Data Journalism Handbook which is released under Creative Commons CC-BY-SA. Selected chapters of the handbook are being republished in the Media Helping Media Data Journalism section. In each case the author or authors of the piece are mentioned.
Image by JD Hancock and released under Creative Commons