Designing Geographic Inquiry and Sampling Techniques

A geographic inquiry is a structured investigation that starts with a question or hypothesis, collects and analyzes data, and reaches a conclusion, and sampling is the method of choosing a small, representative set of locations or people from a larger population because it is not possible to measure everything. In brief: the main sampling methods are random (every item has an equal chance), systematic (every nth item), stratified (proportional to subgroups) and pragmatic or convenience sampling (what is easiest). A well-chosen sample lets you draw conclusions about the whole area, while a biased sample misleads. For a simple survey of a proportion, a sample of about 385 gives a margin of error of about 5% at 95% confidence.

Good design decides whether an investigation can answer its question. This guide explains the stages of an inquiry, how to write hypotheses and choose data, the main sampling techniques and their strengths, compares them in a table, works through sample size and stratified sampling calculations, outlines the history of methods and corrects common mistakes.

What are the stages of a geographic inquiry?

1. Question and hypothesis: start with a focused question, such as “How does pebble size change downstream?” A hypothesis is a testable statement, such as “Pebble size decreases with distance from the source.” A null hypothesis states that there is no relationship, and statistical tests assess whether the evidence is strong enough to reject it.
2. Planning and risk assessment: decide what to measure, where, when and how, check safety and permissions, and consider ethical issues such as consent and privacy.
3. Data collection: gather primary data in the field and secondary data from existing sources, using a suitable sampling plan.
4. Data presentation: graphs, maps and tables.
5. Analysis: describe patterns and test them statistically.
6. Conclusion and evaluation: answer the question using the evidence, and assess the limits of the methods, data quality and what could be improved.

What types of data are used?

Primary data are collected first-hand by the researcher (measurements, questionnaires, interviews, sketches, photographs), which fit the question exactly but are limited by time and resources. Secondary data are collected by others (census data, weather records, maps, satellite images, published studies), which cover large areas and long periods but may not match the question exactly and should be checked for source and reliability. Data can also be quantitative (numerical, such as river depth or traffic counts) or qualitative (descriptive, such as opinions and perceptions from interviews and photographs). Many inquiries use a mix of methods to cross-check results, called triangulation.

What sampling techniques are there?

The population is everything you want to learn about (all the pebbles in a river, all the residents of a town), and the sample is the part you measure. Sampling saves time and cost, and if done properly, gives results close to those of the entire population.

• Random sampling: every member of the population has an equal chance of being chosen, using random numbers, a calculator or a generator. For a place, you can generate random pairs of coordinates on a grid placed over the study area. It avoids bias but may miss parts of the area, and needs a list or map of the whole population.
• Systematic sampling: choosing items at regular intervals, such as every 10th house along a street or a measurement every 50 m along a transect. It is simple and gives even coverage, but may match a hidden pattern (such as every shop at a corner).
• Stratified sampling: dividing the population into subgroups (strata), such as neighborhoods, age groups or land uses, and sampling each in proportion to its size. It makes sure that all groups are represented and is useful when the population is mixed, but requires knowledge of the strata.
• Cluster sampling: selecting entire groups (such as certain streets or villages) at random and sampling within them, which saves travel time but may be less accurate.
• Pragmatic (convenience) sampling: choosing the easiest or most accessible places or people, such as sites close to the road. It is quick but likely to be biased.
• Transect sampling: measuring along a line (such as from the sea to the back of a sand dune system or from the center to the edge of a city), which is good for showing change over a distance.

How can sampling errors and bias be reduced?

Larger samples reduce random error, but cost more time. Bias arises from non-random choices: for example, interviewing only people who are friendly or stand near the station. Ways to reduce it include random or stratified methods, choosing sites before going to the field, interviewing at several times of the day and week, asking neutral questions and recording how refusals were handled. A pilot study tests the methods and questionnaire before the main survey. The sample size should be large enough for the type of analysis: many statistical tests need at least 10 to 30 observations, and a survey of public opinion needs several hundred responses for a small margin of error. The researcher should record how the sample was chosen, so that others can judge it and repeat it.

Sampling methods compared

Worked example: sample sizes and intervals

1. Hypotheses: a student writes H₁: “pebble length decreases with distance downstream” and H₀: “there is no relationship between pebble length and distance.” The data will be used to test H₀.
2. Systematic interval: a street has 1,200 houses and the student wants a sample of 60. The sampling interval is 1,200 ÷ 60 = 20, so every 20th house is chosen, starting from a random number between 1 and 20 (for example 7, then 27, 47, and so on).
3. Stratified sample: a town of 10,000 people has four wards with 2,000, 3,000, 1,500 and 3,500 residents. For a questionnaire sample of 100, each ward gets its share: 2,000 ÷ 10,000 × 100 = 20, 3,000 ÷ 10,000 × 100 = 30, 1,500 ÷ 10,000 × 100 = 15 and 3,500 ÷ 10,000 × 100 = 35. The total is 20 + 30 + 15 + 35 = 100.
4. Sample size for a proportion: the sample size for estimating a proportion is n = z² × p × (1 − p) ÷ e². With z = 1.96 (95% confidence), p = 0.5 (the most cautious assumption) and a margin of error e = 0.05, n = 3.8416 × 0.25 ÷ 0.0025 = 384.2, so about 385 people.
5. A smaller sample: for a margin of error of 10% (e = 0.10), n = 3.8416 × 0.25 ÷ 0.01 = 96.04, so about 97 people. Halving the margin of error needs a sample about four times larger.
6. Random coordinates: a study area is 400 m × 400 m. Random numbers between 0 and 400 give the points (73, 312), (205, 41) and (388, 150), which are used as quadrat positions.

How did research methods develop?

Early geography was largely descriptive and based on exploration. In the 1950s and 1960s, the “quantitative revolution” brought statistical methods and sampling into geography, with figures such as William Bunge and Peter Haggett. In the same period, Ronald Fisher’s work on experimental design and statistical inference had changed science in general. Later, humanistic and critical geographers in the 1970s and 1980s argued for qualitative methods, such as interviews and ethnography, and today, mixed methods are common. Digital tools, such as GPS, mobile data collection, drones and online surveys, have made data collection easier, and open data have increased the use of secondary sources, but the basic principles of sampling and design remain.

Which mistakes are common?

Mistake 1: Choosing sites for convenience and calling it random

Sampling near the road or at easy spots is pragmatic sampling and may be biased. Random sampling requires a chance procedure.

Mistake 2: Using too small a sample

Very small samples give unreliable results and make statistical tests weak. Decide the sample size in advance, based on the analysis planned.

Mistake 3: Writing a hypothesis that cannot be tested

A hypothesis should be clear and measurable, such as a stated relationship between two variables that can be investigated with data.

Mistake 4: Confusing bias with random error

Random error averages out in larger samples. Bias is a systematic distortion that remains however large the sample is.