HookThe pattern was obvious. The mark was not.
On a summer fieldtrip a student walks a shingle beach with a metre rule, measuring the long axis of pebbles at eight points spaced 50 metres apart in the direction of longshore drift. The mean pebble size falls from 62 mm at the drift-source end to 40 mm at the far end — a drop of roughly 35%. The pattern looks obvious. The temptation is to write the pebbles get smaller, which proves attrition and move on. That sentence earns almost nothing.
The marks live in what the student does next: rank both variables, run Spearman's rank correlation, get rs = −0.95, and test it against the critical value before claiming the relationship is real. That is Assessment Objective 3 in miniature — investigate a question, handle the data honestly, and only then draw a conclusion you can defend. Every skills mark in AQA Geography, worth close to a quarter of the whole A-level, is a version of the same demand: not what does the map show? but what can you prove it shows, and how sure are you?
ModelQuantitative, qualitative — and how you sampled
Two families of evidence. Quantitative data are numbers you can test: pebble sizes, pedestrian counts, rainfall totals, house prices. Qualitative data carry meaning rather than magnitude: interview transcripts, field sketches, photographs, newspaper and social-media sources. AQA wants you to engage critically with both, and the sharpest lever it pulls is sampling. Random sampling gives every item an equal chance (a random-number table laid over a grid) and removes selection bias, but it can clump and leave gaps. Systematic sampling takes every nth item — every 50 m along the beach — which is quick and even but can hide or exaggerate a spatial rhythm that shares its interval. Stratified sampling divides the population into sub-groups and samples each in proportion (survey 60% of respondents in the ward holding 60% of residents), which controls for known structure but needs prior data to set it up.
Two words the examiner rewards: reliability (would you get the same result if you repeated it?) and validity (are you measuring what you claim to?). A badly worded questionnaire can be perfectly reliable and completely invalid. Qualitative sources add the problems of representation and ethics: a Victorian photograph of a slum was framed to persuade, an interview is one person's perspective, and any research involving people needs consent and anonymity.
ModelCore skills — annotate, don't just label; count properly
Literacy in geography means annotation, and this single distinction decides a surprising number of marks. A label names a feature — 'groyne'. An annotation explains it — 'timber groyne interrupting longshore drift and trapping sediment on its updrift side, so the beach is around a metre higher to the left'. The AO3 credit is in the explanation; the naming is worth almost nothing on its own.
Numeracy is the quiet half of the paper. You are expected to handle percentage change (from 62 mm to 40 mm is (40 − 62) ÷ 62 = −35%), rates, ratios, proportions and per-capita figures without a second thought, because the analysis questions assume it. Questionnaire and interview technique is examined too: closed questions (tick-boxes, Likert scales) give codeable quantitative data but constrain the answer, while open questions give qualitative depth but resist comparison. Pilot the questionnaire, strip out leading questions, and think hard about where, when and whom you sample — a weekday-morning high-street survey silently excludes commuters, workers and anyone without the time to stop.
MechanismCartographic skills — every map answers a different question
OS maps first: four- and six-figure grid references read eastings before northings ('along the corridor, then up the stairs'); scale is literal (1:50,000 means 2 cm on the map is 1 km on the ground); and gradient comes from spot heights and contour spacing — 60 m of climb over 1.5 km of ground is a gradient of 1 in 25. Synoptic (weather) charts use isobars, fronts and the pressure gradient, where tightly packed isobars signal strong winds.
Then the thematic family, each matched to a data type. A choropleth shades areas by value and suits rates and densities (population per km²) but wrongly implies every point inside a zone is identical. An isoline map joins points of equal value (contours, rainfall isohyets). A dot map places one dot per n units and shows distribution well but is almost impossible to count. Proportional symbols scale a circle's area to a value, and flow or desire lines scale arrow width to movement — migration, trade, commuting. Map overlays stack layers so a spatial correlation becomes visible. Choosing the wrong map for the data is itself a lost AO3 mark.
MechanismGraphical skills — the chart is a choice, and it is marked
Pick the wrong graph and you lose credit before plotting a point. Line graphs show continuous change over time (a storm hydrograph). Bar charts and histograms handle discrete categories and frequency. Scatter graphs with a line of best fit expose the relationship between two variables — you eyeball the correlation here, then test it with statistics.
AQA also expects you to read the specialised forms fluently. Triangular graphs plot three components that sum to 100% — soil as sand, silt and clay, or a workforce split into primary, secondary and tertiary employment — and each axis is read in its own direction, which is where candidates trip. Logarithmic scales compress vast ranges so you can plot earthquake energy or settlement size under the rank-size rule on one page. Dispersion diagrams show the spread of a single variable and reveal the median and quartiles at a glance. Kite diagrams show species abundance along a transect — the classic tool for a sand-dune succession at somewhere like Studland Bay in Dorset. Radial and rose diagrams handle directional or cyclical data such as wind direction or pebble orientation. Proportional and divided circles (pie charts scaled by total) show composition and size together.
DataStatistical skills — the numbers that turn a pattern into a proof
Start with the averages, because each answers a different question. The mean (sum ÷ n) uses every value but is dragged by outliers; the median (the middle value) shrugs off a freak reading; the mode (most common value) is the only average you can use on categorical data such as 'most common land use'. Then measures of spread: the range is crude, the interquartile range captures the middle 50% and ignores extremes, and the standard deviation measures the average distance of values from the mean — a large standard deviation means a dispersed, less reliable dataset.
The headline techniques are relational. Spearman's rank correlation coefficient tests the strength and direction of a relationship between two ranked variables on a scale from +1 to −1. Chi-squared tests whether an observed set of frequencies differs significantly from what chance would produce — is pebble roundedness independent of beach zone, or not? The rule students forget, every year, is the same: a coefficient means nothing until you compare it against the critical value for your sample size and chosen confidence level. Calculating rs is the easy half; the significance test is the half that scores.
Eight sampling points along a shingle beach, taken every 50 m in the direction of longshore drift. Mean pebble long-axis (mm): 62, 58, 60, 51, 49, 45, 47, 40. Hypothesis: pebble size decreases with distance (attrition). Step 1 — rank each variable. Distance ranks run 1 to 8. Ranking size with smallest = 1 gives, in distance order: 8, 6, 7, 5, 4, 2, 3, 1. Step 2 — find d (distance rank − size rank) for each point: −7, −4, −4, −1, +1, +4, +4, +7, and square each: d² = 49, 16, 16, 1, 1, 16, 16, 49. Step 3 — sum them: Σd² = 164. Step 4 — apply the formula rs = 1 − (6 × Σd²) ÷ (n × (n² − 1)) = 1 − (6 × 164) ÷ (8 × 63) = 1 − 984 ÷ 504 = −0.95. That is a strong negative correlation. Step 5 — the mark-winning step almost everyone rushes: compare it to the critical value. At n = 8 the critical value is about 0.64 at the 95% confidence level and 0.83 at 99%. Because 0.95 exceeds both, we reject the null hypothesis of 'no relationship' at 99% confidence — the size decrease is very unlikely to be down to chance. The trap: had n been only 5, the 95% critical value is 0.90, so even rs = −0.88 would not have been significant. Strength of correlation and statistical significance are not the same thing.
MechanismICT and GIS — layering data to see what a table hides
A Geographical Information System (GIS — QGIS, ArcGIS) stores spatial data in layers you can switch on, query and overlay. Put a choropleth of deprivation beneath a point layer of fast-food outlets and a spatial correlation appears that no spreadsheet would ever have shown you. That is the everyday power the examiner wants you to understand: GIS makes a pattern visible, but it does not explain it — the explanation is still your job.
Remote sensing uses satellite imagery from programmes such as the US Landsat series and the EU's Sentinel-2 to measure change over time: deforestation on the Amazon frontier, the outward creep of a city's built-up area, or an urban heat island read off the thermal band across Birmingham on a summer night. Electronic databases — the ONS and the census, the Environment Agency's flood-risk maps, Met Office climate records — supply free secondary data at national scale. Two judgements matter for AO3: every satellite image has a resolution and a date, and every database has a source and a collection method. Know all four before you trust the data, and say so.
CaseThe enquiry route — from question to a conclusion you can defend
Fieldwork skills are examined on both Paper 1 and Paper 2 (the independent investigation, Component 3, is marked separately and is not what these written-paper questions test). The route to enquiry is a six-stage loop the examiner can interrogate at any point. First, pose a specific, measurable question or hypothesis — 'pebble size decreases with distance in the direction of drift' — tied to a named process or theory. Second, plan data collection: primary data (your own measurements, counts, questionnaires) and secondary data (census figures, tide tables), with a justified sampling strategy, a written risk assessment (tides, unstable cliffs, traffic) and ethical clearance where people are involved. Third, present the data with the graph or map that actually fits it. Fourth, analyse — describe the pattern, quantify it with a statistical test, and identify anomalies so you can explain them rather than quietly delete them. Fifth, conclude by answering the original question with direct reference to the evidence. Sixth, evaluate — where did sampling, instruments or weather threaten reliability and validity, and what specific change would improve the method next time?
In the exam you may be asked about your own fieldwork or handed an unfamiliar dataset and location. The same six-stage route structures your answer to both, which is why examiners can set a fieldwork question on a place you have never visited and still expect full marks.
VocabularyKey terms the mark scheme pays for
TrapsMisconceptions that cost marks
ExamWhat examiners want
These marks sit under AO3 — roughly a quarter of the whole A-level — and they appear on both written papers, from a one-mark 'give the six-figure grid reference', through a four-mark 'suggest one strength and one limitation of this sampling method', up to an extended fieldwork question worth as many as nine marks. They are the most reliably scoreable marks in Geography because the technique is fixed; the students who drop them are the ones who never practised the method under time.
For any statistical calculation, do three things in order: state the formula, substitute the numbers, and then — the step candidates skip — interpret the result against the critical value. 'rs = −0.95' is half an answer. 'rs = −0.95, which exceeds the 99% critical value of 0.83 at n = 8, so we reject the null hypothesis of no relationship' is the full one. The significance sentence is where the method mark is won or lost, and writing 'null hypothesis' explicitly signals you understand what the test is for.
For extended fieldwork and evaluation questions the mark schemes are levelled, and they reward specific, self-aware method detail over generic caution. 'The weather affected our results' scores nothing. 'A rising tide forced us to abandon the last two sampling points, which biased the mean pebble size upward because the finest sediment sits lowest on the beach' scores, because it ties a named limitation to its effect on a named variable. Always route the answer back to reliability and validity, and finish with one concrete improvement.
Finally, show that you can choose. Naming the right graph, sampling method or statistical test for a given dataset — and saying in one line why the alternatives are worse — is itself an AO3 skill the examiner is looking for. Never present a technique without a short justification of why it fits the data in front of you.