Correlation in G2 Geography describes an association between variables. A scatter graph plots paired observations so you can examine whether the variables tend to change together. A positive pattern rises from left to right; a negative pattern falls. Neither direction, by itself, proves that one variable causes the other.
The quickest useful routine is to read the axes and units, identify what each point represents, describe the overall pattern and then examine its strength and exceptions. Only afterward should you suggest a geographical explanation. This keeps the observed relationship separate from the mechanism you propose.
This guide supports scatter-graph and fieldwork learning in the 2027 standalone G2 Geography syllabus K226. The examples are invented to make the reasoning visible. For the wider investigation process, read fieldwork and sampling; for source-based claims, use the unfamiliar-evidence guide. Advanced statistical calculations are not assumed here.
eduKateSG · Geography learning guide
Find your next learning step
Start with the question closest to your task. Every teaching chapter remains open below; the grouped index helps you return to a particular explanation.
Choose a chapter
Chapters 1–5 · Read paired observations
Chapters 6–10 · Describe direction and strength
Chapters 11–15 · Examine lines and unusual points
Chapters 16–20 · Connect patterns to fieldwork
Chapters 21–25 · Practise a justified conclusion
A scatter graph uses paired data. Each point combines a value of X with its corresponding value of Y for the same site, person, time or other observation unit. The pairing is the relationship you are examining.
Imagine measuring distance from a gate and pedestrian count at several sites. A point must use the distance and count for the same site. Sorting the distances and counts separately before matching them would create artificial pairs and destroy the original evidence.
Begin with a table containing an observation identifier and both values. Check the row before plotting. The identifier helps trace an unusual point back to its source.
This is more than a technical plotting detail. If the pairs are wrong, a beautifully drawn graph can show a convincing pattern that was never observed. Accurate pairing establishes the foundation for every later description. When reviewing a graph, ask what one dot represents before deciding what the overall shape means.
The horizontal and vertical axes identify the quantities being compared. Their labels, units and scales establish the meaning of the point positions. A rising cloud of points is not yet a useful geographical statement without those variables.
If X is distance in metres and Y is a count over five minutes, the graph compares those measures. It does not automatically show travel time, daily footfall or population.
Write the units beside the variable when reading. Then choose a point and translate its position into a sentence. This quick check often catches a mistaken scale or swapped axis before it affects the whole answer.
Do not assume that the horizontal variable is proven to be the cause because it appears on the X axis. The plotting arrangement helps organise a proposed relationship; it does not establish causation. A strong interpretation uses the actual labels and the collection design. The axes tell you what was compared, while the investigation and geographical mechanism inform what can reasonably be explained.
Counts collected over different durations are not directly comparable as though they measured the same exposure time. A pedestrian count over five minutes and one over twenty minutes need either a matched design or an appropriate conversion.
Use invented values: 40 people in five minutes corresponds to an average of eight per minute during that interval. Sixty in twenty minutes corresponds to three per minute. The larger raw count has the lower average rate.
The conversion does not show that movement was uniform throughout either interval. It simply places the totals on a common rate basis. If the task concerns peak flow, the averages may conceal the relevant variation.
Before plotting, check what the question actually asks you to compare. Standardising units can improve the graph, but it cannot repair every difference in timing, weather or site conditions. A fair method begins with comparable collection and records the remaining limits. The graph is a presentation of that method, not a replacement for it.
A positive association means higher values of one variable tend to occur with higher values of the other. On the usual axes, the points show an upward tendency from left to right.
The word tend allows variation. A positive pattern does not require every successive point to be higher than the previous one. Examine the overall arrangement rather than rejecting the relationship because one point differs.
Use the variable names in your description. ‘As distance increases, the recorded value tends to increase’ is more informative than ‘the graph is positive’. Add appropriate evidence if the task requires it.
Positive is not a judgement that the outcome is good. A positive association between two undesirable measures would still be positive in the statistical sense. Keep the direction label separate from evaluation. This small distinction helps students write clearly and avoids importing everyday meanings into a technical description of how the values are arranged.
A negative association shows a downward tendency from left to right. Higher values of one variable tend to occur with lower values of the other.
An invented investigation might find lower recorded pedestrian counts at sites farther from a particular entrance. The pattern could be described as negative if the points support it. The graph alone does not show why it occurs.
Negative does not mean the data are wrong or the relationship is weak. A negative pattern can be strong and clear. Direction and strength are different aspects of the description.
Practise stating them separately: a strong negative association, a weak positive association or no clear linear association. Then connect the terms to the observed points. Precise language reduces the need for vague phrases such as ‘a bad correlation’. The graph is describing an arrangement of measurements, not expressing approval or disapproval of the places or people involved.
For a linear pattern, strength concerns how closely the points follow the overall straight-line tendency. A tight descending arrangement can show a strong negative relationship, while a widely scattered rising arrangement can show a weaker positive one.
The slope is not the same as strength. A steep-looking line can have widely dispersed points, and a shallow-looking pattern can be very consistent. Axis scaling also affects visual steepness.
Describe strength using the actual scatter rather than the angle alone. If the evidence is ambiguous, avoid false precision. A qualitative school task may require a supported description rather than an invented correlation coefficient.
One useful check is to imagine moving along X. How consistently does Y follow the proposed pattern? That question keeps your attention on the points. A line drawn through them should summarise the relationship, not hide the amount of variation that remains. The unexplained scatter is part of the evidence.
A graph may show no obvious straight-line tendency. The appropriate description is often no clear linear association in the supplied observations. This is narrower than declaring that the variables can never be related.
A curved pattern may exist, or different groups may have different relationships. The range may also be too limited or the measurements too variable to reveal a clear pattern.
Read the points before selecting the label. If values rise and then fall, a single positive or negative description may miss the structure. A school question may use simple examples, but unfamiliar evidence can require a more careful statement.
The conclusion should reflect what the graph shows. You can say that these observations do not support a clear linear trend. To claim absence of every relationship across all places and times would need much more evidence. This is a useful distinction because it allows a definite result without turning a limited dataset into a universal statement.
The figure below uses the same horizontal and vertical scales in three panels. The left panel has a positive tendency, the middle a negative tendency and the right no clear linear tendency. The values are invented and have no real geographical location.
Using identical scales makes the visual comparison easier. The purpose is to focus on direction and scatter without differences in axis design distracting from the relationship.
Notice the repeated Y values in the first panel. They do not prevent an overall positive pattern. Also notice that the negative pattern is just as meaningful a relationship as the positive one.
Use the data table if you want to plot the points yourself. Explain one panel aloud using the labels and then compare it with another. The task is not to memorise the arrangement. It is to recognise the relationship when the variables, units or drawing change. Open the full-size figure through the image link if the labels appear small on your device.
| X | Positive panel Y | Negative panel Y | No clear linear panel Y |
|---|---|---|---|
| 1 | 2 | 9 | 5 |
| 2 | 3 | 8 | 2 |
| 3 | 3 | 7 | 8 |
| 4 | 5 | 7 | 4 |
| 5 | 6 | 5 | 7 |
| 6 | 6 | 4 | 3 |
| 7 | 8 | 3 | 6 |
| 8 | 9 | 2 | 5 |
Read X first and Y second. Locate the horizontal value, then the vertical value, and mark their intersection. Continue using the same procedure for every pair.
Choose a scale that covers the data and uses the available space clearly. Label units and avoid irregular intervals unless the axis explicitly indicates a justified break.
After plotting, check several points against the original table, including the largest, smallest and an unusual value. This catches errors that can alter the apparent relationship.
A tidy graph is helpful because it makes the evidence readable, but neatness does not prove accuracy. The important checks concern paired values, scale and placement. If a point is difficult to locate, use the grid and unit intervals rather than estimating from a nearby symbol. A consistent routine reduces avoidable errors and leaves more attention for interpreting the geographical pattern.
Scatter points are usually not joined one after another because the order of rows in the table may have no substantive meaning. A zigzag connecting them can suggest a sequence that the investigation did not collect.
If the data are a time series, a line graph may answer a different question about change through time. Choose the graph type according to the relationship being examined rather than because one style looks more complete.
A scatter graph can include a suitable trend line when required. That line summarises a relationship; it is different from joining each point in row order.
When reading a supplied graph, identify what the line represents. A best-fit line, a theoretical relationship and a connected sequence have different purposes. The labels or question should establish the intended interpretation. The central evidence remains the observations and their context. A line is a tool for explaining them, not permission to invent a path through every point.
A best-fit line represents the broad linear tendency of the plotted data under the method used. It does not pass through every point unless the data happen to lie exactly on a line.
For an introductory visual line, follow the teacher’s instructions and aim to represent the overall pattern rather than connect the two most convenient points. A statistically calculated line uses a defined method and should not be confused with an arbitrary sketch.
The distance of points from the line remains important. It shows variation around the summary relationship. A line without the points can conceal that variation.
Do not describe the line as proof that every observation must behave exactly according to it. It can support estimates within a suitable scope, but individual values may differ. The usefulness of the summary depends on the pattern, sample and purpose. When a question supplies the line, read it as the stated model and keep its limitations attached to the interpretation.
An outlier is an observation that lies unusually far from the general pattern or other values. It can reflect a recording error, an unusual site or a genuine feature of the process. Its appearance alone does not tell you which explanation is correct.
Trace the point back to its identifier and field notes. Was the count taken at a different time? Is the site beside another entrance? Was the value entered incorrectly? These are questions to investigate rather than conclusions to assume.
Do not delete a point simply because it weakens the expected pattern. Removing inconvenient observations can make the graph look cleaner while making the investigation less honest.
If a justified correction is made, record it. If the point is valid, discuss what it suggests about the proposed explanation. A useful outlier can reveal an additional influence or a limit to the simple model. The investigation becomes stronger when unusual evidence is examined rather than hidden.
Points can form clusters rather than one continuous cloud. Different site types, times or participant groups may contribute different patterns. A single overall description can conceal that structure.
Imagine invented observations from both a shopping street and a quiet residential route. Their counts and distances may occupy different parts of the graph. Combining them without identifying the groups may encourage an overly simple explanation.
If the source gives group labels, compare the patterns within and between groups. Do not invent group identities from point positions alone.
A useful conclusion can state that the combined pattern may partly reflect differences between the sampled settings. The next investigation step would examine those settings and collect relevant contextual information. This is a constructive result: the scatter graph has revealed a better question. Good analysis does not require every dataset to support one smooth line. The arrangement of groups can itself be geographically important.
Correlation describes an association in the observations. Causation claims that changing one factor produces a change in another through a process. The first does not by itself establish the second.
A scatter graph can suggest a relationship worth explaining. Geographical knowledge helps propose a mechanism, while study design and additional evidence help assess whether that mechanism accounts for the pattern.
Use two sentences to keep the distinction clear. First describe what the graph shows. Then suggest why the relationship may occur, identifying the relevant process and uncertainty.
This is more useful than adding ‘correlation is not causation’ as an isolated slogan after an otherwise overconfident paragraph. The distinction should shape the wording throughout. You can make a clear observational finding and offer a plausible explanation without claiming that the graph has proved every link. The reader should be able to see where the evidence ends and the interpretation begins.
Two variables can be associated because another factor influences them both. This possibility is often called confounding. The useful task is to identify a plausible third factor in the actual geographical context.
Imagine an invented sample where ice-cream sales and park attendance rise together. Warm weather could contribute to both. The graph would not prove that purchasing ice cream causes people to visit the park.
The example is a reasoning illustration, not a real study. Its purpose is to show why a visible association can support more than one causal account.
In a fieldwork answer, suggest a relevant factor and explain its pathway. A generic list of every possible influence is less helpful. If the proposed third factor was measured, examine it. If it was not, state how additional data could test the explanation. This turns a limitation into a practical next step and keeps evaluation connected to the question rather than repeating a stock phrase about unreliable data.
An association can leave uncertainty about which variable influences which, or whether influence runs both ways. The choice of X and Y does not settle that question.
An invented relationship between tourist facilities and visitor numbers could involve facilities attracting visitors, visitor demand encouraging facilities, or both changing with other influences. A scatter graph alone does not reveal the full sequence.
Time-based evidence and a clear process explanation can help. Ask what happened first, what mechanism is plausible and what alternative explanations remain.
Keep the conclusion proportionate. You can describe a positive association between facilities and visitors in the sample while recognising that the causal relationship needs more investigation. This is a stronger answer than choosing a direction merely because one variable appears on the horizontal axis. The graph organises paired observations; the geographical investigation explains the processes that may connect them.
A scatter graph represents the observations collected. If the sample includes only a narrow range of sites or conditions, the resulting pattern may not represent a wider area.
Suppose all pedestrian observations are taken near one busy gate. The graph cannot establish how counts relate to distance across the whole neighbourhood if distant and differently connected sites are absent.
Sample size matters, but location and range matter too. Many repeated observations at one type of site do not automatically provide broad spatial coverage.
To improve the investigation, identify the missing variation relevant to the question. A more suitable site selection or repeated collection at comparable times may help. Explain how the change strengthens the evidence rather than simply asking for more data. A graph can be accurately plotted and still have limited geographical reach because its collection design did not capture the places or conditions needed for the wider claim.
Interpolation estimates within the observed range; extrapolation extends beyond it. Both depend on a model, but extrapolation usually adds uncertainty because the relationship may change outside the sample.
If observations cover distances from 20 to 100 metres, using a fitted line at 60 metres is different from extending it to two kilometres. The latter assumes the same relationship persists across unobserved conditions.
A geographical process can encounter thresholds, barriers or different land uses. A straight line is not guaranteed to continue indefinitely. For example, a count cannot become physically meaningful as a negative number simply because an extended line crosses below zero.
When a question asks for an estimate, use the supplied line and scale carefully, then recognise the scope. If asked to evaluate the estimate, relate the limitation to distance from the observed range and possible changes in conditions. The distinction is useful because it explains why a neat numerical answer may still have limited reliability.
A scatter graph comparing different sites at one time describes spatial variation. It does not necessarily show what happens as one site changes through time. Those are different comparisons.
Imagine sites farther from a centre having lower counts in one survey. The graph does not prove that a particular site’s count will fall over the next year as though it were moving along the horizontal axis.
Likewise, data from different years need a clear account of what each point represents. Check whether the units are places, dates or people before describing a temporal trend.
Use language that matches the design. ‘Across the sampled sites’ identifies a spatial comparison. ‘Over the recorded years’ identifies a time comparison. These small phrases help prevent a large inference error. Geography often connects space and time, but the connection needs evidence rather than an assumption that one arrangement can stand in for the other.
Use invented pairs of distance and pedestrian count: 10 metres and 90 people; 20 and 80; 30 and 70; 40 and 65; 50 and 50. Assume every count covers the same five-minute period.
A suitable description states that recorded counts tend to decrease as distance increases across these sites. The values show a clear negative association in this small invented set.
Evidence can be given by comparing the endpoints, but do not rely only on them if the intervening points contradict the pattern. Here, the intermediate values support the stated direction.
A possible explanation is reduced accessibility or attraction farther from the specified source, depending on the context. The data do not establish that mechanism by themselves. Other entrances, land uses and collection conditions may matter. A complete answer separates the observed association from the proposed explanation and keeps the scope within the five sampled sites.
An invented table has Site A with X equal to 1 and Y equal to 9, Site B with 2 and 7, and Site C with 3 and 5. The original pairs form a negative pattern.
A student sorts the Y values from smallest to largest while leaving X unchanged, creating pairs 1 and 5, 2 and 7, and 3 and 9. The new graph appears positive, but it no longer represents the observations.
The repair is to retain complete rows when sorting. Each observation’s two values and identifier must stay together. Replot from the original paired records.
This example shows why interpretation cannot be separated from data handling. A graph can be drawn perfectly from a corrupted table. Checking source rows is therefore an essential step, especially when the result looks unexpectedly neat or conflicts with the field notes. The strongest first question is what each point represents and whether its two values actually belong together.
A conclusion can be limited by measurement, sample design, pairing, range or causal interpretation. Identify which issue affects the actual claim rather than listing every possible weakness.
If the claim concerns the whole neighbourhood but the sample covers one street, spatial coverage is relevant. If the claim says X causes Y, confounding and process evidence are relevant. If units differ, comparability is the immediate problem.
An improvement should address that issue. More accurate instruments do not repair a sample missing half the study area. More sites do not fix values paired with the wrong observations.
Write the limitation, explain its consequence and propose a targeted improvement. This makes evaluation purposeful. A study can remain useful within a narrower scope even when a wider claim is unsupported. Recognising that distinction helps you give a balanced judgement and a practical next step rather than treating every imperfection as proof that the whole investigation failed.
Start with a small paired table, plot it and describe the pattern using variable names. Then change the labels and ask whether the reasoning still works. The direction term should follow the points, while the explanation should follow the geographical context.
Next, add an unusual point or divide the observations into groups. Describe how the interpretation changes. Do not simply preserve the first label because it was the expected answer.
Finally, practise a short paragraph with observation, evidence, possible mechanism and limit. These components can be brief. Their value lies in keeping different kinds of claim clear.
A parent or tutor can ask ‘what does one dot mean?’ and ‘what else might explain this pattern?’ These questions reveal understanding without requiring a rehearsed definition. Return later to a new example after a gap. Successful application provides better evidence of learning than recognition of a graph whose answer has just been read.
A graph stretched vertically can make a trend look steeper; one stretched horizontally can make it look flatter. The underlying paired values have not changed. This is another reason slope appearance should not be used as a substitute for strength.
Check the numerical intervals and aspect of the graph before comparing two pictures. Identical-looking angles can represent different rates when the units or scales differ. Conversely, the same relationship can look different under different layouts.
An axis need not always begin at zero for a scatter graph, but its range must be clear. A restricted range can be useful for examining detail and can also influence the impression made on the reader. Describe the actual values rather than relying on visual drama.
For a fair comparison of the three invented panels in this guide, the axes are held identical. You can therefore focus on the point patterns. When the source uses different scales, translate points back into their labelled quantities before making a claim. A clear numerical comparison is more dependable than judging only which line looks most impressive.
Does positive mean strong? No. Positive describes direction; strength concerns how consistently the points follow the relationship.
Does negative mean bad data? No. A negative association can be strong and meaningful.
Must every point lie on the line? No. Variation is common, and a best-fit line summarises rather than replaces the observations.
Should an outlier be removed? Not merely because it is inconvenient. Investigate its source and record any justified correction.
Does the X axis identify the cause? No. Axis placement does not prove a causal direction.
Can no clear linear pattern rule out every relationship? No. Curves, groups or sampling limits may matter.
What is the most useful first check? Read the variables, units and observation unit, then confirm that each point contains values from the same observation.
Continue with the relevant eduKate guide
Fieldwork and sampling: Match collection design to the relationship being investigated.
Unfamiliar evidence: Separate observation, explanation and judgement.
Mental maps and sense of place: Compare numerical association with interpretation of qualitative representations.
Humidity and rainfall: See why related weather variables still need a process explanation.
Compare Geography learning routes. Match the guide to your current subject and task; G2 and G3 references do not imply identical assessment requirements.
Sources and next steps
SEAB: G2 Geography K226 for 2027 · NIST: scatter plots and limits of causal inference. The figure, paired values and worked geographical examples are original learning illustrations.
Worked examples and invented figures are learning illustrations. This guide supports understanding and does not replace your school’s instructions or the official syllabus for your examination year.
Bring one current task and its feedback when asking for help. Ask eduKate about appropriate support and current availability. The Clementi Mathematics guide illustrates the value of finding the learning gap before adding practice. Check Geography support separately.