Scientists collect data by making systematic observations or measurements using instruments, experiments, field sampling, surveys, biological specimens, historical records, sensors, simulations and other evidence gathering methods. Data can be quantitative, such as temperature, mass, concentration or population size, or qualitative, such as descriptions of behaviour, geological structures or observed symptoms. The collection method is determined by the scientific question, the variables being studied, the population or environment of interest and the precision required.
Scientists first define what they need to measure and how the measurement will represent the underlying phenomenon. This process is important because data are not simply raw facts. They are observations produced through particular instruments, sampling designs, operational definitions and procedures. The meaning and quality of data therefore depend strongly on how they were collected.
Experimental data are collected by deliberately manipulating conditions and measuring resulting outcomes. Researchers identify independent and dependent variables and determine how these will be measured. Experiments may use control groups, treatment groups, random assignment and repeated trials. Replication helps estimate variability and determine whether an observed effect is consistent. Laboratory experiments can provide substantial control over environmental conditions, while field experiments provide greater realism but often involve more uncontrolled variation.
The experimental protocol should specify procedures before data collection begins where practical, including measurement intervals, sample handling and exclusion criteria. Automated instruments can collect large numbers of observations, but automation does not eliminate the need for calibration and quality control. Scientists must still verify that instruments are measuring the intended quantity and that recorded values are stored correctly.
Observational data are collected without deliberately assigning an experimental treatment. Researchers may observe wildlife, monitor atmospheric conditions, record medical outcomes, survey populations, measure geological formations or analyse astronomical objects. Observational studies are essential when manipulation is impossible, unethical or inappropriate. Scientists carefully design sampling strategies to ensure that observations provide useful information about the population or process being studied. Random sampling can reduce selection bias, while stratified sampling ensures that important subgroups are represented. Systematic sampling can provide regular spatial or temporal coverage. Convenience samples are easier to obtain but can be less representative. Longitudinal studies repeatedly measure the same subjects or locations over time, allowing researchers to examine change and temporal relationships.
Field data collection introduces challenges associated with spatial and environmental variability. Researchers must select sampling locations, determine sampling frequency and account for accessibility, seasonality and local conditions. Environmental monitoring may require sensors operating continuously, while ecological surveys may involve repeated visits at specified times of year. Geological sampling may target particular rock layers or structures, and oceanographic research may collect measurements along transects or at different depths.
Standardised field protocols help ensure that measurements collected by different researchers or at different locations are comparable. Field notes, photographs, coordinates, timestamps and environmental conditions provide essential context. Without adequate metadata, a measurement may be impossible to interpret correctly because its location, timing, method or surrounding conditions are unknown.
Instruments are central to many forms of data collection. Thermometers, balances, spectrometers, microscopes, cameras, satellites, radar systems, sequencing platforms, mass spectrometers and electronic sensors convert physical or biological properties into measurable signals. Calibration establishes how an instrument's output corresponds to known standards. Instruments can experience drift, noise, sensitivity limits and environmental interference. Researchers therefore use reference standards, control measurements and maintenance procedures. Some instruments measure variables directly, while others produce signals that require mathematical conversion or modelling. Data acquisition systems can also introduce errors through incorrect sampling rates, electronic interference, software problems or storage failures. Instrument data should therefore be accompanied by information about calibration, resolution, detection limits and operating conditions.
Biological and medical data collection often requires additional ethical and methodological considerations. Human research may involve informed consent, privacy protection, secure data storage and procedures approved by appropriate ethics bodies. Biological specimens may require controlled collection, preservation and chain of custody. Researchers must define inclusion and exclusion criteria carefully to avoid inadvertently producing a biased sample. Genetic, physiological and behavioural measurements can be influenced by many factors, requiring standardised collection conditions. Animal research also involves ethical requirements and careful experimental design. The goal is to obtain scientifically informative data while protecting participants, animals and ecosystems and maintaining the integrity of samples and records.
Scientists also obtain data from existing sources rather than collecting every observation themselves. Existing datasets may include government monitoring records, historical archives, museum specimens, satellite observations, previous experiments, clinical databases and publicly available scientific repositories. Reusing existing data can increase sample size and enable long term or large geographical studies. However, researchers must understand how the original data were collected, including definitions, sampling design, missing values, measurement changes and quality controls. Combining datasets can introduce inconsistencies if instruments, units, spatial resolutions or definitions differ. Harmonisation procedures may be required before analysis. Data provenance, meaning information about where data came from and how they were processed, is therefore an important component of scientific data management.
After collection, scientists inspect, clean, validate, document and analyse data. Cleaning may identify impossible values, duplicate records, instrument failures and transcription errors, but researchers must avoid removing observations merely because they contradict expectations. Missing data require explicit treatment because different missingness mechanisms can affect conclusions. Quality control may involve comparing measurements with reference values, checking distributions, examining temporal continuity and investigating anomalies. Data are often transformed into standard units or formats before analysis.
Statistical models then estimate relationships, differences, trends or probabilities while accounting for uncertainty. The quality of scientific conclusions depends on the entire process from question formulation and sampling design through measurement, recording, validation and analysis. Large quantities of data cannot compensate for a poorly designed collection strategy.
