Sharks and ice cream

Every organisation has had the meeting. Someone says “we’ve always done it this way”. Someone else remembers it differently. Twenty minutes later nobody has changed their mind and nothing has been decided.

More often than not, the answer is already sitting in a spreadsheet, a booking system or a mailing list report that nobody has opened. So how do we get from a pile of numbers to a better decision? And what do sharks and ice cream have to do with it? This post covers three ideas: what data actually is, how it can fool you, and how to put it to work.

Start with the evidence

“It is a capital mistake to theorize before you have all the evidence. It biases the judgment.” Sherlock Holmes, in Arthur Conan Doyle’s A Study in Scarlet.

Well over a hundred years later, plenty of decisions are still made on a hunch. Sometimes the hunch is right. Often it is just the loudest opinion in the room. Suhail Doshi, who founded the analytics company Mixpanel, put it bluntly: most people decide by guessing or by gut, and they will be either lucky or wrong.

Evidence does not remove judgement. It gives judgement something solid to stand on.

Data, information and knowledge

These three words are often used as if they mean the same thing. They don’t, and the difference matters.

• Data is the raw fact, with no context. 100.

• Information is data with context. 100 miles per hour.

• Knowledge is what you understand from it. 100 miles per hour is quite fast. On a UK motorway, it is also illegal.

A number on its own means very little. Add a unit and some experience, and it starts to tell you something. Photo: Unsplash.

A single number tells you almost nothing. Put it next to something, a date, a unit or last year’s figure, and it starts to mean something. Add experience and you can act on it.

Try it with your own work. Take a number from your last report, say 40 bookings in a week. On its own it is information. It only becomes knowledge when you can say whether 40 is a good week, a bad week or a normal one, and what you will do about it.

Data is everywhere

We create more data now than at any point in history, and most of it comes from things we barely notice.

• Your heating: a smart thermostat records when you are home, how warm you like it and how quickly the house loses heat. It turns that into a schedule that warms the house before you wake up.

• Your bank: the app on your phone sorts every payment into categories and can tell you that you spent more on takeaways this month than last. Your bank uses the same data to spot a card being used somewhere it has never been before.

• Healthcare: data from scans, tests and appointments helps spot patterns across thousands of patients that no single clinician would ever see.

A thermostat turns data (temperatures and times) into knowledge (when to switch the heating on). Photo: Unsplash.

Each example follows the same path: collect the data, give it context, and use what you learn to make a better decision. A school, a charity or a small business can do exactly the same with the data it already has.

Patterns and trends

Once you have data, you start to look for shapes in it. Two words come up a lot.

A pattern is something that happens in a predictable sequence. Room bookings rise every September. Donations peak in December. Fewer volunteers are free in August.

A trend is a general direction over time. Membership has crept up every quarter for two years. Printing costs are higher this year than last.

The warning that comes with both: you need enough data to see a real pattern, and you have to resist forcing the data to fit the story you already believe.

Correlation is not causation

Here is a fact. When ice cream sales go up, so do shark attacks. So does eating an ice cream make you more likely to meet a shark?

More ice cream, more shark attacks. Does one cause the other? Photo: Unsplash.

Of course not. Both go up in the summer. Hot weather sells ice cream, and hot weather also puts far more people in the sea, which is where the sharks are. The two numbers move together, which is a correlation, but one does not cause the other. The summer is causing both.

Ice cream sales and shark attacks rise and fall together, and both peak in the summer. An illustrative chart, not real figures.

Nobody blames the ice cream. Photo: Unsplash.

It sounds obvious with sharks and ice cream. It is much harder to spot in your own data, especially when the result is the one you were hoping for. Did the new website bring in more enquiries, or did enquiries go up because it was the start of term? Did the newsletter fill the event, or was it the sunny weekend?

If you want to see how silly this can get, Tyler Vigen’s Spurious Correlations collects real data sets that line up almost perfectly by coincidence, such as the amount of cheese eaten per person and the number of people who died by becoming tangled in their bedsheets. The lines match. Nobody thinks one causes the other.

Before you act on a link between two things, ask three questions:

• Could something else be causing both?

• Could it be the other way round?

• Could it be coincidence, because there is not enough data yet?

Diagnose, describe, predict

There is an old line, usually credited to the economist Ronald Coase: if you torture the data long enough, it will confess. The way to avoid that is to be clear about the job you are asking the data to do. There are three.

• Diagnose: use the data to find the cause of a problem you already have, and a way to fix it. Why did fewer people renew this year?

• Describe: summarise the key parts of the data so others can understand it quickly. What happened last term, on one page?

• Predict: use what has happened before to estimate what will happen next. If bookings follow the same pattern as the last three years, how busy will March be?

A supermarket is a good place to see all three at work. Your loyalty card tells the shop what you buy. Describing that data shows which products sell together. Diagnosing it explains why one branch sells less bread than another. Predicting from it means the shelves are full of barbecue food before the first warm weekend, not after it.

Every shelf in a supermarket is a decision made from data. Photo: Unsplash.

How data can mislead

Data can be used honestly, or not. Three things to watch for, in other people’s reports and in your own:

• Misleading: presenting data so people quickly see what you want them to see. A chart whose axis does not start at zero can make a small change look huge.

• Manipulating: picking the parts of the data that support your point and quietly leaving out the rest.

• Skewing: collecting too little data, or data from only one group, so the results lean one way. A survey of the people who came to an event will not tell you why everyone else stayed away.

Three things to take away

1. Use evidence to make decisions, not guesses or the loudest voice in the room.

2. Correlation does not mean causation. Look for the summer behind the sharks and the ice cream.

3. Know which job you are asking the data to do: diagnose, describe or predict.

Most organisations already hold the data they need. The skill is asking it good questions, and being honest about the answers. Start small: pick one decision coming up this month, and ask what the data says before anyone gives their opinion.