5(f)–(h), 5(k); 3i Section 4 - Analysing and explaining investigation findings

Choosing analysis to answer the question

An analytical method should follow the question and data, rather than be selected because it appears sophisticated. A comparison of neighbourhood journey times may require measures of centre and spread, while a question about distance and sediment size requires paired spatial analysis. An enquiry into belonging may require thematic interpretation of accounts. A numerical significance test cannot answer every geographical question.

Begin by checking and organising raw records. Confirm units, locations, dates, category meanings and missing values. Distinguish a true zero from a failed measurement. Check unusual observations against original notes; do not remove them merely to strengthen the expected trend. Record corrections and exclusions so their effect can be assessed.

Describe the pattern before explaining it. Establish direction, magnitude, variation and exceptions, with specific evidence. Then ask which geographical processes could account for it. An illustrative sediment enquiry might show smaller particles with distance from a source but a local increase near an eroding cliff. That exception could reveal a second input, modifying a simple transport explanation rather than disproving the entire systems approach.

Select inferential methods only where the data and assumptions fit. Spearman tests monotonic association, chi-squared can test categorical independence, Mann–Whitney compares independent distributions through ranks, and an independent t-test compares means under its assumptions. Paired data need appropriate paired treatment. Explain why the chosen method advances the enquiry, and interpret its result rather than leaving a statistic unexplained.

Analysis can also compare sources, locations, groups or periods without a significance test. A well-supported qualitative explanation or a transparent budget may be more appropriate than a misapplied test. OCR's analysis criterion rewards appropriate, coherent reasoning and explanation within the learner's own investigation, not a compulsory checklist of statistical techniques.

Quantitative evidence and uncertainty

Quantitative analysis should communicate what a result means in the studied setting. A mean describes a centre, while dispersion shows variation around or within the distribution. An illustrative pair of neighbourhood samples might have equal mean journey times but different spreads; the conclusion about inequality changes when the distribution is considered. Report units, sample sizes and relevant conventions, not just calculator outputs.

A significance decision is conditional on a null model and assumptions. State the hypothesis, test, significance level, relevant degrees of freedom or sample size, statistic, critical comparison and decision. Failing to reject a null means insufficient evidence against it, not proof of no effect. A statistically significant relationship does not identify its cause or guarantee practical importance.

Assess independence and representativeness. Ten repeated readings at one site may improve understanding of short-term variability but do not automatically represent ten independent places. An unbalanced sample can overrepresent accessible sites. More precise arithmetic does not repair those design limitations. Explain how they might affect the observed pattern and the population to which it is generalised.

Measurement error and natural variation are also different. If a survey instrument has a positive offset, repeated readings may consistently overestimate a quantity. If real sites differ, a large standard deviation may be a genuine finding. A residual sediment budget may reflect an unmeasured boundary flux or accumulated measurement uncertainty, so it should not automatically be named a precise offshore loss.

Use comparisons to investigate alternatives. Matching tidal stage can improve beach-profile comparisons; contrasting geology can help explain differences in erosion; repeated periods can reveal whether a pattern persists. These steps do not guarantee causal proof, but they strengthen an explanation by addressing plausible confounding influences. The analysis should make the strength of inference visible rather than bury uncertainty in a generic final paragraph.

Worked centre and spread comparison

Soil-pH readings at five sampled plots are 4, 5, 5, 6 and 10. Calculate the centre and full spread:

The high observation pulls the mean above the median. Check it against the record and local conditions before deciding whether it is a genuine patch or an error.

Worked second distribution

Four stream temperatures are 8, 10, 10 and 12°C. Use the same sequence:

Here mean and median agree, but the observations still vary. Centre and spread describe different properties, and values on different measurement scales should not be compared directly.