The big idea
Statistics answers questions about large groups. Usually you cannot ask everyone, so you ask a smaller group and use its answers to estimate. How you choose that group, and what you ask, decide whether you can trust the answer.
Reminder from Year 7: asking only your friends gives biased (unfair) results.
The data handling cycle
An investigation has four stages, called the data handling cycle. Here they are for the hypothesis (a statement you can test with data) "Pupils who play a musical instrument score higher in a memory test".
| Stage | What you do | For this hypothesis |
|---|
| Plan | decide what data you need, and from whom | who plays an instrument, and memory test scores, for 40 pupils chosen at random |
| Collect | collect the data | a questionnaire and a memory test |
| Process and represent | work out averages, make tables and diagrams | each group's mean score, and a bar chart |
| Interpret and discuss | say what the results show | a higher mean for players would support it (but not prove that playing causes it) |
New questions from the results start the cycle again.
Primary and secondary data
Primary data is data you collect yourself. It fits your question, but it takes time and effort. Secondary data was collected by someone else, such as figures on a website. It is quick to get, but it may be out of date or come from a source you cannot trust.
Populations, samples and a census
The population is everyone or everything you want to find out about. A census collects data from every member of the population; a sample collects data from only some of them. A sample is quicker and cheaper, but its result may differ from the whole population's. A bigger sample usually gives more reliable results.
A bakery makes 4,000 jam doughnuts a day. To check the jam, a doughnut has to be cut open. Why does the bakery use a sample, not a census? In a sample of 40, 3 have too little jam. Estimate how many of the 4,000 have too little jam.
A census would mean cutting open all 4,000 doughnuts, leaving none to sell.
Estimate: \(\dfrac{3}{40} \times 4000 = 300\) doughnuts.
This is only an estimate: a different sample could give a different answer.
Bias and random sampling
Results are biased when the way the data is collected makes them likely to be too high or too low. Think about who is asked, where and when they are asked, and how (people may not answer a personal question honestly out loud).
In a random sample, every member of the population has the same chance of being chosen. Give every member a number, then pick numbers with a random number generator (or draw names from a hat), ignoring repeats. Asking whoever you meet is not random.
To find out whether people in a town would like a skate park, Mo asks the first 50 people leaving a gym at 7 am on a Monday. Why are his results likely to be biased? How could he do better?
At that time, people leaving a gym are mostly adults on their way to work. Young people, the most likely to use a skate park, are hardly asked. So the results will probably show fewer people wanting one than in the whole town.
Better: choose a random sample from a list of everyone in the town.
Writing good questions
A good questionnaire question asks about one thing, with a time period if it needs one. It is not leading (pushing people towards an answer). Its response boxes use amounts, not vague words such as "often", and cover every answer with no gaps or overlaps.
A council is planning a cycle path. What is wrong with this survey question? Write a better one.
- Question: "Cycling is good for you and the planet. How much do you cycle?"
- Boxes: 1 to 5 km / 5 to 10 km / more than 10 km
The first sentence is leading: it says cycling is good, so some people may claim to cycle more. "How much" has no time period. The boxes overlap (5 km is in two of them), and there is no box for 0 km.
- Better question: "How far did you cycle in the last 7 days?"
- Boxes: 0 km / more than 0 and up to 5 km / more than 5 and up to 10 km / more than 10 km
Remember
- A census uses every member of the population; a sample uses some.
- In a random sample, every member has the same chance of being chosen.
- Bias can come from who, where, when and how people are asked.
Common mistakes
- Saying a sample is more accurate than a census: only a census gives the full picture.
- Thinking a bigger sample fixes bias: 500 people leaving the same gym are still biased.
- Saying data is primary because you are using it: it is primary only if you collected it.
Top tip
To spot bias, ask yourself: who has been left out, and would they have answered differently?