Lesson 7 of 7 · 20 min
Solving problems with bivariate data
Put the whole topic together to solve a practical problem — display and describe an association, fit and check a linear model, predict, and draw a conclusion that respects association versus causation.
What you will be able to do
- Follow a complete process to analyse an association between two variables.
- Combine the scatterplot, r, R², the least-squares line and the residual plot into one argument.
- Evaluate a claim based on bivariate data, including claims of cause.
A grain grower on the Darling Downs wants to know how much the growing-season rainfall matters for her wheat yield, and what yield to expect in a dry year. She has rainfall and yield records for eight seasons. Every tool from this topic and the last one has a job in answering her.
Analysing an association between two numerical variables
Identify the variables
Decide which is explanatory and which is response.
Display and describe
Draw a scatterplot. Describe the direction, form and strength, and calculate .
Fit the model
Use technology to find the least-squares line, and write it with the variable names.
Check the model
Look at the residual plot. Random scatter means a linear model is appropriate.
Interpret and use
Interpret the slope, the intercept (if meaningful) and . Make the predictions asked for.
Conclude and evaluate
Comment on reliability (strength, residuals, interpolation or extrapolation), and on whether any cause can be claimed.
Eight seasons on one Darling Downs farm. The residual plot has no clear pattern, so a linear model is appropriate.
Worked example
Analyse rainfall and wheat yield from start to finish
A Darling Downs grower recorded growing-season rainfall and wheat yield for eight seasons.
| Rainfall (mm) | 210 | 245 | 260 | 290 | 315 | 340 | 360 | 395 |
|---|---|---|---|---|---|---|---|---|
| Yield (kg/ha) | 1600 | 1900 | 2300 | 2200 | 2600 | 2900 | 2800 | 3300 |
The residual plot for a linear model shows the residuals scattered randomly about 0. The grower expects a dry season of 230 mm. Analyse the association and advise her on the yield to expect, commenting on the reliability of your advice.
Step 1: Identify the variables
Rainfall is the explanatory variable and yield is the response variable.
Step 2: Describe the association with r
Technology gives : a strong positive linear association. Seasons with more rain tend to have higher yields.
Step 3: Fit the least-squares line
Technology gives and , so
Step 4: Check the model
The residual plot shows random scatter, so a linear model is appropriate.
Step 5: Interpret the slope and R²
Each extra millimetre of rain is predicted to increase the yield by 8.71 kg/ha. : 94.8% of the variation in yield can be explained by the variation in rainfall.
Step 6: Predict for 230 mm
Step 7: Comment on reliability
230 mm is inside the data (210 to 395 mm), so this is interpolation. With a strong linear association, a random residual plot and interpolation, the prediction is reliable — though with only eight seasons, and other factors (temperature, frost, soil moisture at planting) affecting yield, treat it as an estimate.
Step 8: Advise
In a 230 mm season the grower should expect a yield of about 1800 kg/ha, well below the eight-season mean of 2450 kg/ha.
Try it yourself first, then check the working one step at a time.
Answer
Strong positive linear association (, ); ; the predicted yield for 230 mm is about 1825 kg/ha, which is reliable because it is an interpolation from a strong linear model with random residuals.
Method
Identify the variables, then describe the association with r.
Fit the line with technology and check the residual plot.
Interpret the slope and R², then predict.
Judge reliability with strength, residuals and interpolation, and answer the question in context.
Watch out for
Skipping the residual plot and assuming the association is linear.
Giving a prediction without the reliability comment the question asks for.
Using , predict the wheat yield in a season with 300 mm of rain. Give your answer to the nearest kilogram per hectare.
The grower uses the same line to predict the yield in a season with 600 mm of rain: kg/ha. Which comment on this prediction is best?
Exam tip
Complex unfamiliar questions
CU questions in this topic usually give you a context and a decision to make — which variable is the better predictor, whether a prediction can be trusted, whether a claim is justified. Use the steps above as a checklist, quote the statistics you use, and finish with a clear conclusion in context.
The last step — the conclusion — is where cause sneaks in. Rainfall plausibly does cause higher yields (wheat needs water), but even so the data only shows an association. On other farms, soil type or fertiliser could be confounding variables. When someone makes a claim from bivariate data, test it against everything you know.
Worked example
Evaluate a claim of cause
For 12 regional Queensland towns, a council report plots the number of fast-food outlets per 10 000 people against the hospital admissions for heart disease per 10 000 people. It finds and concludes: "Limiting the number of fast-food outlets will reduce heart disease in our town." Evaluate this conclusion.
Step 1: Describe the association
shows a strong positive linear association: towns with more fast-food outlets per person tend to have more heart-disease admissions per person.
Step 2: Quantify it
, so about 67% of the variation in admission rates can be explained by the variation in fast-food outlet numbers. About a third is due to other factors.
Step 3: Association is not causation
The data is observational — the council didn't change the number of outlets and watch what happened — so it shows association only.
Step 4: Suggest confounding variables
Towns with an older population, lower average income or higher smoking rates may have both more fast-food outlets and more heart disease. Any of these could produce the association.
Step 5: Consider the data
Twelve towns is a small sample, so the association may also partly be chance.
Step 6: Conclude
The data supports an association but not the claim that limiting outlets will reduce heart disease. To test the claim you would need to control for the other variables or run a controlled study.
Try it yourself first, then check the working one step at a time.
Answer
The conclusion is not justified. There is a strong positive association (, ), but the data is observational; age, income and smoking rates could cause both variables, and 12 towns is a small sample. Limiting outlets may not reduce heart disease.
Method
Describe and quantify the association.
Say what kind of data it is: observational data shows association, not cause.
Name specific confounding variables and how they link to both.
Reach a clear conclusion about the claim.
Watch out for
Agreeing with the claim because is large.
Rejecting the claim without giving a specific alternative explanation.
What bivariate data can show
- Whether two variables are associated, and how strongly
- The form of the association
- A model for predicting one variable from the other within the data
- How much of the variation is explained ()
What it can't show on its own
- That one variable causes the other
- That the trend continues outside the data
- That the model suits individuals rather than the overall trend
- That an association is not coincidence
Summary
Analyse an association in order: variables, display and describe, fit, check the residuals, interpret and predict, conclude.
Quote the statistics you rely on (r, R², the line) and interpret each in context.
Reliability depends on strength, the residual plot and whether you are interpolating.
Evaluate any claim of cause by considering confounding variables and coincidence; only an experiment can establish cause.
Finished the lesson?
Try the 2 quick checks above and the lesson completes itself, or mark it complete now.