Appendix 2 5, 6; 9GE0/04 - Selecting representations and analysing field evidence
Learn how to turn field evidence into defensible geographical analysis by selecting representations, applying quantitative and qualitative methods, and checking what the results establish. The examples and checks are formative practice using original illustrative material; they are not observations from your investigation.
From field records to usable evidence
Analysis begins by checking what each observation measures and which question it can help answer.
Keep the original records unchanged and work on a clearly labelled copy. Record the variable, units, site, date and time, sampling method, equipment resolution, missing values and any changes made during cleaning. A blank reading is not zero: zero is an observed absence or quantity, whereas a blank means no usable observation. Combining these changes the result.
Outlier
An observation unusually far from the rest of a dataset. It may be a recording error, an unusual but real event, or evidence of a different process.
Check an outlier against the field notes before deciding what to do. Correct a documented transcription error and retain the audit trail. Keep a genuine unusual value; if it strongly affects the result, compare analyses with and without it and explain why. Deleting an inconvenient observation to strengthen a hypothesis undermines the evidence.
Before combining primary and secondary data, check that their categories, geographical boundaries, units and dates match. A weekday footfall count at one entrance cannot be compared directly with an annual town-wide visitor total. A population percentage needs its denominator: 30 respondents expressing a view has a different meaning in a sample of 40 and a sample of 300. Keep the original geographical scale visible when aggregating data.
For an independent investigation, you select and interpret the evidence yourself. A group may have collected primary data where permitted, but a shared spreadsheet does not make a group analysis permissible.
Representations that answer the question
A representation is useful when it makes the relationship under investigation visible without implying more than the observations support.
The following is an original illustrative dataset, invented for method practice. It describes five hypothetical sites along a walkway, counted for ten minutes each on one dry weekday between 14:00 and 15:00. It is not a learner's fieldwork.
| Site | Distance from an interchange (m) | Pedestrians counted in ten minutes |
|---|---|---|
| A | 0 | 46 |
| B | 100 | 40 |
| C | 200 | 35 |
| D | 300 | 27 |
| E | 400 | 32 |
A scatter plot of distance on the horizontal axis and pedestrian count on the vertical axis tests whether two numerical variables are associated. A bar chart with one bar per site compares their counts, but equal spacing between bars alone does not establish geographical distance. A line graph is appropriate for change through ordered time or a meaningful continuous transect; joining isolated sites can imply values between them that were never measured. Choose the method for the purpose rather than because it looks impressive.
State the variable and units on each axis; give the observation period and sample size in the caption. Show the complete range of values and make any axis truncation explicit. A narrow vertical range can exaggerate a small difference. Separate observed points from any fitted trend, and do not extend a fitted trend beyond the observed distances without justification.
GIS can retain the spatial context. A site layer needs correctly recorded coordinates and a consistent coordinate reference system. Join attributes using reliable identifiers, then check the points fall where expected. A symbol map can show counts; combining it with a land-use or transport layer helps investigate possible explanations. Overlaying two layers reveals co-location, not proof of causation. For area maps, compare rates or percentages when population sizes differ, check the same boundary/date definitions, and explain classification breaks. A result for an area does not describe every person in it. Avoid mapping identifiable homes or respondents when that breaches confidentiality.
Use the display that exposes the relevant comparison, relationship or location. Labels, scale, provenance and omissions are part of the evidence.
Quantitative patterns and their limits
Numerical analysis should describe the evidence and test a defined relationship, while preserving the geographical meaning of the measurements.
For the illustrative counts 46, 40, 35, 27 and 32, the mean is pedestrians per ten-minute site count. In order, the values are 27, 32, 35, 40 and 46, so the median is 35 and the range is . The range shows that the typical value hides considerable variation between sites. These are counts per fixed observation period, not a measured hourly rate; multiplying by six assumes the ten-minute flow continues unchanged.
Spearman's rank correlation measures the direction and strength of a monotonic association: whether one variable tends consistently to increase or decrease as the other increases. It does not require a straight-line relationship. With no tied ranks, use:
Here is the number of paired observations and is the difference between the two ranks within each pair. Rank both variables using the same direction. The coefficient has no units and lies between −1 and +1. Ties require average ranks and an appropriate tie-aware calculation; do not silently use the simple formula as if there were no ties.
Rank the smallest distance and count as 1:
| Site | Distance rank | Count rank | ||
|---|---|---|---|---|
| A | 1 | 5 | −4 | 16 |
| B | 2 | 4 | −2 | 4 |
| C | 3 | 3 | 0 | 0 |
| D | 4 | 1 | 3 | 9 |
| E | 5 | 2 | 3 | 9 |
Continue this lesson.
You’re previewing NovaLearn without an account. Create a free account to read the rest.
Create free account