Sampling, data quality and outliers
In a minute
Statistics is only as good as the data behind it. Here you learn the difference between a and a , how to choose a fair sample, and how to spot data you can't trust. You also meet the rule for an , and what to do when you find one.
You'll be able to
- Use the words population, sample and , and tell from .
- Judge whether a data source is reliable, and explain where , missing values and recording errors come from.
- Use the rule to identify outliers, and decide whether each one is a valid value or an error.
- Describe simple random, systematic, stratified, quota and convenience sampling, and judge how well each represents the population.
- Work out how many members a stratified sample takes from each group.
Before you start3 quick questions on what this builds on
Populations, samples and types of data
A school wants to know how long its students spend on homework. Asking all of them is slow. Asking is quicker, and gives a good estimate if the are chosen well.
You use the sample to estimate facts about the population. Collecting data from every member instead is called a census.
Watch for two traps. Shoe sizes are discrete: they jump from to to , with nothing in between. A height recorded as cm is still continuous: it was only rounded.
Try it first, then reveal one step at a time.
- Ask of each one: is it counted or measured?
- (a) Races are counted, so this is discrete. (c) Gears are counted too, so this is discrete.
- (b) Time is measured and can take any value in a range, so this is continuous.
- (d) Mass is measured, so this is continuous, even if it is recorded to the nearest kg.
Try it first, then reveal one step at a time.
- (a) The population is all trees in the orchard. The sample is the trees inspected.
- (b) In the sample, of the trees are diseased. The sample is random, so assume the same fraction holds for the orchard.
- Scale up: trees. Another random sample would give a slightly different estimate.
Reliable data and bias
A large data set can still mislead. Check it for missing or impossible values too.
To judge a data source, ask who collected the data, and why. A firm that surveys people about its own product gains from a good result. Then ask how the data was collected. A source that doesn't say how its members were chosen is hard to trust.
Bias has three common causes. Part of the population may be out of reach: an online survey misses people without internet. People who volunteer choose themselves, and they often hold strong views. Some people don't reply, and they may differ from those who do.
A bigger sample doesn't remove bias. Asking volunteers instead of still leaves out everyone who didn't volunteer.
Missing data: someone skips a question, or a sensor fails. Never record a blank as . Leave it out, and say how many values are missing.
Recording errors: a slipped decimal point, a wrong unit or a typing slip. Look for values that are impossible or far out of line. Correct one only if you know the true value; otherwise remove it and say so.
Try it first, then reveal one step at a time.
- (a) Six masses lie between and g. An egg of under g is not realistic here, so is almost certainly an error: the decimal point has slipped.
- (b) Replace with and add: .
- Divide by : g. This lies among the typical masses, as it should.
Try it first, then reveal one step at a time.
- (a) In the sample, train at least four times a week. Apply this to all members: .
- (b) Say who could be chosen: only people in the gym at that time.
- Members who train often are far more likely to be there on a given evening than members who rarely come. So frequent trainers are over-represented, and is likely to be too high.
Outliers
Look at these test scores: , , , , , , . The sits far above the rest, but is it far enough to count as unusual? You need a rule that everyone applies the same way.
The rule uses the quartiles. With the data in order, the lower quartile has a quarter of the values below it, and the upper quartile has three quarters. You'll find them in 4.3; here they are given, or read from your GDC.
The (IQR) is the gap between the quartiles: the spread of the middle half of the data. It is in the formula booklet.
This rule is not in the booklet, so learn it. The two boundaries are called the lower and upper fences.
The inequalities are strict. A value exactly equal to a fence is not an outlier.
Try it first, then reveal one step at a time.
- Find the IQR: , so .
- Upper fence, measured from , not the median: .
- A score is an outlier only if it is greater than , so a score of exactly is not an outlier.
The rule only flags a value as unusual; it doesn't make it wrong. Check each outlier against the context.
If the value is possible, it is a valid member of the : keep it. If it is impossible, or clearly a slip, it is an error: correct it if you know the true value, or remove it and say so.
Try it first, then reveal one step at a time.
- (a) Enter the times in a list and run one-variable statistics. Write down what you read off: and .
- (b) , so .
- Lower fence: . Upper fence: .
- No time is below , and only . So s is the only outlier.
- (c) A student who got stuck could take nearly two minutes. The value is possible, so it is valid: keep it.
Sampling methods
The five methods below differ in how random they are, and in what you need before you start. Most need a numbered list of the whole .
Effectiveness: no sampling bias, but it needs a full list, and by chance it can miss a small group.
Effectiveness: quick and evenly spread. It goes wrong if the list has a repeating pattern that lines up with .
Effectiveness: usually the most representative method. It needs each stratum's size and a list of its members.
Effectiveness: quick and needs no list, so it suits street surveys. The collector's choices can bias it, for example by approaching only people who look friendly.
Effectiveness: the quickest and cheapest method, but the most likely to be biased.
Try it first, then reveal one step at a time.
- Total workers: .
- Each shift's share: day , evening and night .
- You can only choose whole people. Round every share down: . That is one short of .
- Give the missing place to the shift with the largest decimal part. That is the day shift, with , so it gets .
- Check the total: . Choose the workers at random within each shift.
Rounding each share to the nearest whole number can give the wrong total. A safe method is to round every share down, then hand out the places still missing. Give them one each to the strata with the largest decimal parts. No stratum can go negative, and the total always comes out right.
If two decimal parts tie for the last place, choose between those strata at random. For equal strata and a sample of , each share is . Rounding down gives , so one stratum, chosen at random, gets the tenth place.
Try it first, then reveal one step at a time.
- (a) .
- (b) Choose a random whole number from to as the first loaf, then take every th loaf after it.
- (c) The loaves are , , , . The th is steps of after the first: .
- Check: the next one would be , beyond , so the sample has exactly loaves.
Common mistakes
- Calling a value that equals a fence an outlier. The inequalities are strict, so a value on the fence is not an outlier.
- Adding to the median. Measure from for the upper fence and from for the lower one.
- Saying only that a sample is too small or not representative. Name the group the method favours, and say why its answers would differ.
- Calling the first few people to arrive a systematic sample. That is convenience sampling; systematic takes every th member of an ordered list.
- Dividing by the sample size in stratified sampling. Each stratum's fraction is its size over the population size.
- Deleting an outlier just because it is one. Remove a value only if it is impossible or a known error.
Exam tips
- State or identify a sampling method: the name is enough. Describe one: say how members are picked; for systematic sampling, give and the random start.
- Explain why a sample is biased: the mark needs a reason tied to the context, such as who is left out and why they would answer differently.
- Verify that a value is or is not an outlier: write the fence with numbers substituted, then compare the value with it using or .
- Estimate a population count from a sample: write the fraction you scale by, such as , for the method mark.
Key takeaways
Practise this
2 past-paper questions on 4.1 in the AA HL bank, each with a mark scheme and a worked explanation.