Why every answer from data raises the next question

0:00Onions cost thirty-five rupees a kilo. There. Any questions? Now here is a year of onion prices in two towns — twenty-four numbers instead of one. Something just happened. The single number closed the conversation and the table opened one. And the strange part is that thirty-five was not a lie. It is a real price. It really was paid, in one month, in both towns. In the other twenty-two months of that year it is simply wrong.

0:34The cheapest month is seventeen and the dearest is sixty. The spread that one number hides is forty-three rupees, which is wider than the number itself. One of those is something you get told. The other is something you get to think with. So what do you want to know? Not what I want to tell you. What the table made you want to ask. Here are six questions people actually raise when they see it.

1:03Do the seasons move the price? Where are these two towns, and how far apart are they? How much do prices differ between two shops in the same street? Do other things behave this way — rice, tomatoes, petrol? What decides the price of an onion in the first place? And whom do these swings hurt, and whom do they help? Six questions, and they are not all the same kind of question.

1:32Sort them with one test. Could collecting data settle this? Do seasons move the price — yes. Collect several years and look. Where the towns are — yes, that is something you can go and find out. Shop to shop in one street — yes. Walk down it and write the prices down. Other goods — yes. Same table, one more column. That is four you could settle. The last two are a different animal. What decides an onion's price is a question about weather and land and trade.

2:08And whom the swings help is a question about people's lives. Data can inform both of those. It cannot settle either. That is not a flaw in the questions. Knowing which of your questions is which is most of the skill. Take one of the four and make it properly answerable. Do seasons move the price is a wondering, not yet a question you could act on. A question you can settle names three things.

2:39What you would collect: monthly prices. From where: these two towns, and ideally a few more. And over what stretch: several years, because one year cannot separate a season from an accident. And how you would show it, which for prices over time means a line or a column for each month. Now it is not a thought. It is something a person could go and do on Monday. Your turn, with the labels taken off.

3:10Two rows of twelve numbers. One row for each of two places, one number for each month. I am not going to tell you what they measure. Look at them and work out what kind of thing they could be. You are allowed to add them up, compare them, or divide them by anything you like. Take your time. Those numbers are the only evidence you get. Here is what is visible already.

3:39The first row climbs to its highest in June, and falls to its lowest in December. The second row does exactly the opposite. Highest in December, lowest in June. Each place's best month is the other place's worst. And it is not only at the two ends. From one month to the next there are eleven steps. In all eleven, when one row goes up the other goes down. Whatever these are, the two places are opposed all year long.

4:13Now add each row up. Four thousand five hundred and twelve, against four thousand four hundred and twenty-five. Eighty-seven apart, across an entire year. Under two per cent. So the two places get almost exactly the same amount of whatever this is. They just get it at opposite times. That already rules a great deal out. Whatever this is, neither place is simply richer in it than the other. That is the clue worth holding on to. Same total, opposite shape.

4:50One more move before I tell you. Divide each month by its own number of days, because a monthly total flatters the long months. Now the first place runs from eighteen point eight at the top down to six point zero at the bottom. Six point zero is exactly a quarter of a day. The second place runs from fifteen point one down to nine point two. The first place swings twelve point eight. The second swings five point nine.

5:23More than twice as wide. They are hours of daylight. Hours of daylight, month by month, in two places on the same planet. And now every single thing you noticed means something. Opposite all year, because one place is north of the Equator and the other is south. Almost the same yearly total, because over a full year everywhere gets roughly the same share. The wider swing, because the first place sits further from the Equator than the second.

5:57Eighteen hours of daylight in June and six in December is a long way north. Nothing in the table said Equator, or hemisphere, or tilt. All of that came out of the shape. You just read the tilt of a planet off two rows of numbers. And here is the thing worth noticing about what happened just then. You have your answer. Does it feel finished? If a place that far north gets eighteen hours in June, what happens further north still?

6:30Nineteen. Twenty. Twenty-four. There are places where, in midsummer, the Sun goes round the sky and never sets. You did not learn that from the table. The table made you ask. The answer did not close anything. It pointed further north. Once you see that pattern you cannot stop seeing it. Twelve class averages from two schools, and one school is taller grade for grade, every time. Four questions come out of that, and they split two and two.

7:04Two you could answer by going and measuring: how tall the students in your own school are, and what the average across every child of that age would be. The other two are why one school is taller, and where on earth these two schools actually are. Those questions do not get answered. Not because nobody thought of them, but because those twelve averages do not hold the answer. A question left standing is an honest place to stop.

7:36Draw what you have been doing all the way through, and it does not come out as a line. You start with a question. You collect. You display it. You summarise it. And at the end you have a new question, which starts the whole thing again. It is a ring, not a pipeline. The tempting picture is that data goes in one end and answers come out the other.

8:01What actually happens is that every answer hands you the next question. If your analysis ever stops handing you questions, that is worth being suspicious about. Which means you can run the ring yourself, this week, without anyone's permission. Count the letters in every classmate's name, and draw the shape that makes. Tally how often you leave the house, every day for a month. Measure everyone in your family against your own height.

8:33Ask several people to say when they think a minute has passed, and time them. Take one page of two different books and count the words in every sentence on each. Each one is a question, a collection, a picture, and an answer that raises the next question. So here is the only thing I would like you to keep. An analysis does not end with an answer. It ends with a better question.