Seeing relationships in data

A scatter plot turns two numerical variables into a picture. Each dot represents one observation, with its horizontal position showing one measurement and its vertical position showing the other. This makes clusters, trends, gaps and unusual points visible in a way that a table of numbers can hide.

Scatter plots are useful in algebra, statistics and everyday problem-solving. A student might compare weekly study hours with test scores, while a commuter might compare distance travelled with time spent travelling. In either case, the graph reveals the form and strength of the association.

Correlation measures how closely paired observations follow a linear pattern. A line of best fit summarises that pattern with an equation. Both are models of evidence, not automatic explanations, because neither reveals why two quantities vary together.

Australian examples use kilometres, litres and degrees Celsius. Every graph should be checked against its context, the observed range and the purpose of the prediction; a neat line can still mislead.

What a scatter plot reveals

The horizontal axis records one variable, commonly called the explanatory variable, while the vertical axis records the other, often called the response. Label both axes and include units such as hours, kilometres or dollars. Scales should show the data without distorting differences.

Once the points are placed, inspect their direction, spread and shape. A compact cloud suggests a consistent relationship, while a loose cloud suggests greater variability. Gaps, separate clusters and curved patterns indicate that one straight line may not summarise the data adequately.

Plotting is only one part of the skill. Interpreting a fitted model also requires understanding the equations behind it; this linear equations guide can support revision on equation notation and algebraic form.

Correlation: direction and strength

The Pearson correlation coefficient, written (r), lies between (-1) and (1). Its sign gives the direction of a linear association: positive means one variable tends to increase as the other increases, and negative means the opposite. A value of zero indicates no linear association, not necessarily no relationship.

The magnitude shows how closely the observations follow a straight-line pattern. Values near (1) or (-1) indicate tight clustering; values near zero indicate weak linear association. A coefficient depends on the sample, the chosen variables and any unusual observations.

Consider eight students’ weekly revision hours and maths scores: (2, 48), (3, 55), (4, 52), (5, 64), (6, 69), (7, 75), (8, 79) and (9, 84). Their correlation is approximately (0.98), describing a very strong positive linear relationship between the recorded variables.

Outliers and influential points

An outlier differs notably from the other observations, while an influential point changes the fitted pattern substantially. These ideas overlap but are not identical: a distant point may have little effect if many other observations anchor the trend.

In the revision example, changing the final pair from (9, 84) to (9, 50) lowers (r) from about (0.98) to about (0.27). The upward arrangement remains, but it becomes much less convincing. This sensitivity is a reason to inspect individual records rather than trust a coefficient blindly.

Flagging an unusual point does not mean deleting it. Check the source for a recording error. If the value is genuine, retain it, explain the circumstances and report how the correlation and fitted line change when that difference matters.

Building a line of best fit

A fitted line has the form (\hat{y}=a+bx), where (b) is the slope and (a) is the vertical intercept. The slope gives the predicted change in the response for each one-unit increase in the explanatory variable. The intercept is the model’s predicted response when the explanatory variable equals zero.

For the original eight revision pairs, least squares gives (b\approx5.31) and (a\approx36.55), so the fitted equation is (\hat{y}=36.55+5.31x). The model associates each extra hour of weekly revision with about 5.3 additional maths marks in this sample.

By hand, draw a straight line that balances the points reasonably above and below it rather than connecting every dot. Least squares makes this more precise by minimising the squared vertical residuals—the differences between observed and predicted values. A residual plot can reveal systematic errors that the original graph hides.

Reading a line responsibly

A positive slope does not prove that one variable causes the other. For instance, students who ask for extra help may also spend more time revising. Help-seeking could contribute to higher marks, or stronger preparation could lead them to seek help. The scatter plot cannot separate those explanations.

Predictions are also limited by the observed range. The revision line uses only two to nine hours, so its intercept at zero hours lies outside the data and should not be presented as a reliable prediction. A wider or deliberately restricted sample can also change the correlation.

Check the pattern before committing to a straight line. If the points curve upward, a quadratic model may describe them better; if groups occupy separate bands, one coefficient can conceal important subgroups. Report the variables, units, sample and any exclusions so the reasoning can be reproduced.

Scatter plots in Australian contexts

Public transport creates a useful mixed-data problem. In Melbourne, a Myki card and the tram, bus and rail network shape commuting choices; in Sydney, Opal is used for many public transport journeys. Plotting kilometres travelled against journey time can produce different trends for trains, cars and bicycles because speeds and stopping patterns differ.

Shopping data brings measurement decisions into the analysis. A plot of litres of milk purchased against spending in A$ may show a negative relationship because larger containers often cost less per litre. Around Brisbane, a graph of outdoor temperature in Celsius against chilled-drink sales may rise during hot weather, while promotions and stock levels complicate the pattern.

Regional travel offers another lesson. For journeys between Perth and Bunbury, distance and travel time may show a tight positive pattern because motorways impose fairly consistent speeds. A mixed sample of short urban trips and regional drives may instead produce a curved cloud, revealing two travel regimes rather than one universal relationship.

From data to a defensible analysis

A defensible analysis begins with clearly defined variables, consistent units and a stated question. Record where each observation came from and whether the sample covers the population being discussed. Missing values, repeated trials and differently measured groups should be handled before any calculation.

The working sequence is to plot the data, inspect direction, form, spread and outliers, calculate (r), and then fit a line. Compare the fitted values with the observations, write the equation with units, and explain what an increase in the horizontal variable predicts.

The final check asks whether the line serves the purpose and stays within a sensible range. Numerical precision does not repair a poorly chosen variable or a biased sample. The concrete next step is to collect ten paired observations, label both axes with units, plot the points and calculate their correlation before drawing the line of best fit.