3.4.1, 3.4.2.1 - Measurement, sampling and data quality

Measurement, units and protocols

Measurement assigns a value to a clearly defined quantity using a stated method and unit.

Choose a quantity that can answer the question. To compare the slope of two safe paths, measure vertical rise and horizontal distance in the same units; to compare street activity, count the same type of event for equal durations. Record location, time, conditions, instrument and method beside the value. A number without these details can conceal a different measure.

A complete slope calculation

In an original training path, vertical rise is 4.0 m over a horizontal distance of 50.0 m. Percentage gradient is rise divided by horizontal distance, multiplied by 100:

The gradient is also 1 in 12.5: 1 m of rise for 12.5 m horizontally. An 8% gradient is not an 8° angle. A sloping tape distance is longer than the horizontal distance, so using it as the denominator would slightly underestimate this percentage gradient. Choose a suitable levelling or angle/distance method and record which distance was measured; do not silently mix them.

For counts, define whether a return crossing counts again and how simultaneous crossings are recorded. For length, place the instrument at the intended endpoints; for a temperature comparison, use comparable heights, exposure and equilibration time. Instrument resolution is the smallest increment it displays or distinguishes. A digital display with many digits does not guarantee equivalent accuracy. Report precision justified by the instrument and protocol, retaining unrounded values during calculations.

To practise around your centre, select a safe public path or count point, define the quantity, pilot the protocol and record a sample row with units and conditions. Repeat the same protocol at other sites. This is a method you adapt locally, not a ready-made record of your own fieldwork.

Error, accuracy and repeatability

Different sources of error require different remedies; taking more readings does not fix every defect.

Accuracy concerns closeness to a valid reference or accepted value. Precision concerns the fineness or consistency of measurements; a set of tightly clustered readings can still all be wrong. Repeatability asks whether the same observer and method under the same conditions give similar results. Distinguish these from validity: whether the chosen measure actually represents the quantity in the enquiry.

A systematic error shifts measurements consistently, such as a tape with the wrong zero point or a temperature sensor reading 2°C high. Repeating it many times leaves the shift. Check/calibrate against a reference and correct or replace the instrument, documenting any correction. A random error varies unpredictably, such as inconsistent endpoint alignment. A clear protocol and repeated independent readings help reveal its variability; repeats do not guarantee a valid measurement.

A gross error is a mistake such as copying 1.8 m as 18 m or entering centimetres as metres. Range checks can identify suspicious values, but an unusual result may be real. Verify the field log and instrument rather than deleting every extreme value. A change in temperature during a long transect is genuine temporal variation that may confound a spatial comparison, rather than necessarily instrument error.

Check against a reference

Suppose a training instrument reads 52 m on a valid 50 m reference. The signed error is +2 m; its absolute magnitude is 2 m. Absolute percentage error is $2\div50\times100=4\%$. This is a known discrepancy against a reference, not the uncertainty of every future measurement. Three repeat readings of 52 m would be highly repeatable but inaccurate against that reference. Without a valid reference, disagreement among readings reveals variation but does not establish the true value.

An observed difference of 0.1°C between sites should not be interpreted as a robust spatial pattern if instrument uncertainty or inconsistent exposure could be larger. Repeat, standardise and report limitations; do not assert that a very small displayed difference proves a geographical process.

Sampling frames and approaches

Sampling selects part of a defined population so that evidence can be collected feasibly and its representation judged.

For a question about all publicly accessible street segments in a town centre, list or map those segments before choosing them. Sampling only those near a convenient entrance changes what the evidence represents. A larger convenience sample may still be biased if it systematically excludes quiet streets or private space.

A random sample uses a chance procedure so eligible units have known selection probabilities. Number frame units and use random numbers, specifying whether selection is without replacement. Random selection reduces deliberate location choice; a small random sample may still be uneven by chance. A systematic sample selects at regular intervals after a suitable starting rule, such as every 20 m along a transect from a randomly selected start. It gives spatial coverage, but a repeating feature at the same interval can bias the pattern. A stratified sample divides the population into relevant groups and samples within each. It ensures those groups are represented, but depends on accurate group definitions and frame information.

Proportional allocation

Original training frame: 100 public street segments, of which 50 are retail, 30 residential and 20 mixed use. For a sample of 20, allocate $20\times50/100=10$ retail, $20\times30/100=6$ residential and $20\times20/100=4$ mixed-use segments. Then randomly select the required number within each group. The allocations preserve frame proportions; simply taking the nearest 10 retail segments would lose random selection within that stratum. If exact allocations are fractional, use a stated rounding rule that keeps the sample total correct.

Proportional strata are not always the best choice. A small but important category may need a larger allocation for useful comparison, with weighting and an honest account of representation. In a physical transect, regular points can measure a spatial gradient; stratifying by land-cover type can instead compare environments. Neither design automatically represents changes through time. Choose both spatial and temporal coverage to fit the actual question, rather than treating one sampling name as universally superior.

Method fit and defensible conclusions

A suitable design connects the question, eligible population, measurements and feasible collection conditions before the results are interpreted.

Consider the training question “Does pedestrian activity differ between retail and residential street segments on weekdays?” Define activity as crossings of a marked line per five minutes, establish the segment frame and stratify it by land use. Select within strata, pilot the counting rules and visit sites in comparable time windows. Repeat on more than one appropriate weekday if the question concerns a general weekday pattern. If one observer always counts retail at lunchtime and residential early in the morning, land use and time are confounded: either could explain the difference.

For a physical question about ground-surface temperature across a safe land-cover transect, record sites at suitable intervals, standardise sensor exposure/height and measurement time, and record shade or changing cloud. Distances should be defined from a fixed origin and locations retained. A transect that contains only pavement cannot compare pavement with vegetation. A design may combine stratification by surface with systematic locations, provided the selection rules are stated and match the population claim.

Assess data quality using concrete evidence: completeness, consistent units/definitions, calibrated measurements, spatial and temporal coverage, sample selection and missingness. Where access is unsafe or permission unavailable, exclude the site openly and describe the narrower frame. Do not replace it silently with an easy site and continue claiming random representation. Record non-response or equipment failure rather than entering zero. Repeated observations at one site increase temporal evidence there; they do not equal the same number of independent sites.

A strong conclusion states its scope. “In the sampled segments during the matched windows, retail counts were higher” may be supported. “Every retail street is always busier because shoppers prefer it” introduces a universal claim and an unmeasured motive. Corroborate the explanation with relevant evidence and recognise other influences, such as transport stops. Use the same logic in your own local context, modifying categories, safe access and schedule to its geography. A well-selected sample improves inference, but cannot eliminate every source of bias or establish causation alone.

Keep collection proportionate: avoid unnecessary personal identifiers, respect permission and do not expose precise private locations. These are generic method skills. They neither certify fieldwork days nor assess a student's independent investigation.