2 Data Collection and Quality

Choose and document data sources

Business analysis depends on information that fits the decision, is collected with methods that reduce bias, and is documented well enough to interpret. More data are not automatically better: weak coverage, unclear measurements, or unsuitable use can lead to misleading conclusions.

Compare source types

A data source is where information originates. Internal sources include sales transactions, customer-service records, inventory systems, employee records, and website activity. They can be detailed and timely, but may reflect only a business’s existing customers or processes.

External sources include government statistics, industry reports, supplier information, and research studies. They can provide broader context, although their definitions, coverage, and collection dates may not match the business’s needs.

Primary data are collected directly to answer a particular question, such as through a customer survey or product test. Secondary data were collected earlier for another purpose, such as a public report or an existing company database.

Before using any source, check who collected the data, why and how they were collected, which people or events they cover, when collection occurred, and whether the definitions match the decision. Record these details as metadata so the dataset and its methods can be interpreted.

Define the population and sampling method

The population is the full group a business wants to understand, while a sample is the subset from which it collects data. For example, if a retailer wants to estimate satisfaction among everyone who shopped last month, those customers are the population; the customers selected to answer questions are the sample.

A sampling frame is the list or mechanism used to reach potential participants. If it leaves out an important part of the population—for example, customers without email addresses—the sample may have undercoverage. Increasing the sample size does not fix a biased frame or selection method.

Sampling approaches

In probability sampling, each population member has a known, nonzero chance of selection. Simple random sampling selects members at random. Stratified sampling first divides the population into relevant groups and then samples from each group. Probability methods support estimates of sampling uncertainty.

In nonprobability sampling, selection chances are unknown. Examples include a voluntary online poll and a convenience sample. Such results may be useful for exploration, but may not represent the full population. Sampling limitations also include coverage, nonresponse, and measurement problems; probability sampling does not eliminate these nonsampling problems.

Even with a sound design, some selected people may not respond. Nonresponse bias arises when responders differ in relevant ways from nonresponders. Report how participants were selected and recruited, the population studied, and important limitations. Do not imply that a convenience poll represents everyone.

Design surveys and clear measures

A survey gathers answers to standardized questions from people or organizations. Use one when existing data do not adequately answer the question. First define the decision and target population; then write questions that are clear, neutral, and limited to one idea at a time. Avoid leading wording, unexplained terms, and response choices that omit plausible answers. Test questions with a small group before using them more widely.

For example, “How satisfied are you with the price and quality of our service?” asks about two things at once. Separate it into questions about price and service quality. When appropriate, offer “not applicable,” “don’t know,” or “prefer not to answer,” and distinguish these from a substantive response such as “none.” Survey methods and reporting should be transparent.

Measurement is the process of assigning values to a concept. Concepts such as satisfaction or loyalty are not directly observable, so defined questions or indicators are used to represent them. An operational definition states exactly how a measure is obtained. For example, “repeat purchase” could mean a second completed order within 90 days.

Evaluate the measure

Validity concerns whether a measure captures the concept it is intended to measure. Reliability concerns whether the method produces consistent results when the underlying condition has not changed. A measure can be reliable but invalid: a consistently miscalibrated scale gives repeatable readings that are still wrong.

Use an appropriate reference period and clear units. “How often did you order in the past 30 days?” is more precise than “Do you order often?”

Assess data quality

Data quality means fitness for use: whether information is suitable for a particular purpose. It has several dimensions, including relevance, accuracy, timeliness, accessibility, interpretability, and coherence. These dimensions can involve trade-offs; for example, a very fast estimate may leave less time for checking.

Practical checks include:

• Relevance: Does the dataset answer the decision question?

• Accuracy: Do values correctly describe what they are intended to measure?

• Completeness: Are required records or fields missing? Missing data can distort results when their absence is systematic.

• Consistency: Are definitions, units, and formats applied uniformly across records and over time?

• Timeliness: Is the information current enough for the decision?

• Interpretability and coherence: Are methods, definitions, and limitations documented so results can be understood and compared?

For example, monthly sales comparisons across stores may be unreliable if one store records returns in the month of the original sale while another records them when returned. Standard definitions and checks for missing, duplicate, out-of-range, or conflicting records help identify problems. Clean data by correcting documented errors; do not silently alter inconvenient observations.

Collect and protect data responsibly

Responsible data collection begins with a clear purpose. Tell people what information is being collected and how it will be used. Collect only what is needed, protect it against unauthorized access, limit access to people with a legitimate need, and set a retention period. Securely dispose of data when it is no longer needed. These practices reflect privacy principles such as collection limitation, purpose specification, data quality, and security safeguards; businesses are also advised to limit personal-information collection and retention to legitimate needs.

Consider whether sensitive information is necessary, whether participants can decline or skip sensitive questions, and whether results could expose individuals or unfairly affect a group. De-identifying records can reduce risk, but linked details may still make people identifiable.

Legal duties depend on the jurisdiction and type of information. Organizations should check the relevant requirements rather than assume one general rule applies.