Scatter graphs, correlation and cautious predictions · Higher
Introduzido| English | Significado |
|---|---|
| correlation/ˌkɒrɪˈleɪʃn/ | correlação |
| interpolation/ɪnˌtɜːpəˈleɪʃn/ | Estimating a value inside the range of the data you have, for example from a line of best fit. |
| extrapolation/ekˈstræpəleɪʃn/ | Estimating a value outside the range of observed data by extending a pattern or model. |
Chocolate and Nobel prizes
- In 2012 a doctor, Franz Messerli, published a scatter graph in the New England Journal of Medicine: countries that eat more chocolate win more Nobel prizes per person. The correlation was strong. He wrote it partly as a joke — richer countries can afford both.
- Higher questions ask how far a correlation and a prediction can be trusted.
Describing correlation
- Give the direction (positive or negative) and the strength (strong, weak). A strong correlation 相关性 has points close to a line.
- Then interpret in context: "students who studied longer tended to score higher".
The line and its gradient
- For $y = 40 + 4.5x$, the gradient 4.5 means each extra hour of study goes with about 4.5 more marks.
- The intercept 40 is the predicted score for 0 hours — only meaningful if 0 is near the data.
A line of best fit for car age $x$ (years) and value $y$ (thousand yuan) is $y = 120 - 9x$. What does $-9$ mean?
Dificuldade 1/5 · Moedas 100 + pequeno bônus na primeira conclusão hoje
Gradient = change per year.
The intercept of a line of best fit is always a sensible prediction.
Dificuldade 1/5 · Moedas 100 + pequeno bônus na primeira conclusão hoje
x = 0 may be far outside the data.
Worked example: how far to trust it
Using $y = 40 + 4.5x$ (data for 1 to 5 hours), predict scores at 3.6 hours and 10 hours, and comment.
- 3.6 hours: $40 + 16.2 = 56.2$ — interpolation, close to real points, so reliable.
- 10 hours: $40 + 45 = 85$ — extrapolation, twice the largest value.
- Comment: at 5 hours the real score (61) is already below the line (62.5), a sign the gains may flatten.
Scatter graphs, correlation and cautious predictions
Plot paired observations as points, with one variable on each axis
Classify the worked-case statements, then explain the units or invariant that justifies each decision.
Using $y = 40 + 4.5x$, predict the score at 3.6 hours.
Dificuldade 2/5 · Moedas 140 + pequeno bônus na primeira conclusão hoje
40 + 16.2.
Third factors
- A shared cause can create a correlation: wealth drives both chocolate eating and research spending.
- To test cause, scientists change one thing and keep everything else the same.
Chocolate eaten and Nobel prizes per person correlate across countries. What is the most likely explanation?
Dificuldade 1/5 · Moedas 100 + pequeno bônus na primeira conclusão hoje
A third factor.
When to doubt a line
- A weak correlation, few points, an outlier pulling the line, or predicting far outside the data.
- A curved pattern should not be fitted with a straight line.
Which make a prediction less reliable?
Dificuldade 2/5 · Moedas 140 + pequeno bônus na primeira conclusão hoje
Strength and range matter.
A strong correlation can still be meaningless. Chocolate and Nobel prizes correlate strongly, but feeding scientists chocolate would not win prizes. Strength measures how close the points are to a line, not whether one causes the other.
A strong correlation shows that one variable causes the other.
Dificuldade 1/5 · Moedas 100 + pequeno bônus na primeira conclusão hoje
Only an experiment can show cause.
Describe direction, strength and context. Interpret the gradient as a rate per unit. Trust interpolation more than extrapolation, look for third factors, and remember that only a controlled experiment can show cause.
A line of best fit for temperature $x$ (°C, data 10–30) and cold drinks sold $y$ é $y = 12x - 40$. Predict sales at 25 °C and at −5 °C, and comment on each.
Trusting predictions.