What Is Data? Recorded Representations That Need Context

Computers are described as machines that process data, companies call themselves data-driven, and everyday devices collect data continuously. Yet the word can feel so broad that it explains almost nothing.

Data is a recorded representation of facts, observations, events, or ideas.

The representation is not the thing itself. A temperature reading is not heat, a photograph is not the scene, and a customer record is not the person. Data is a form that allows some aspect of reality or thought to be stored, transmitted, compared, and processed.

A concrete example: the number 42

By itself, 42 is ambiguous. It might represent:

• A person's age in years
• A temperature in degrees Celsius
• A product quantity
• A database identifier
• An answer in a survey
• The text printed on a shirt

The stored symbols are identical, but their meanings differ. To interpret the value, we need context: a field name, unit, definition, source, and purpose.

temperature_celsius = 42 carries more information than 42, but questions remain. Where was it measured? At what time? With what instrument? How accurate was the sensor?

Data is created through representation

Something becomes usable as computer data when it is encoded in an agreed form.

A sound wave can be sampled as a sequence of numbers. A color image can be represented as pixel values. A delivery event can be represented by a timestamp, location, package identifier, and status code.

Encoding always makes choices:

• Which properties to include
• Which details to omit
• How precise values should be
• Which categories are allowed
• How missing information is marked

No representation captures everything. A map of a city is useful precisely because it leaves out most details and preserves those needed for navigation.

Data, information, and knowledge

These terms are often used interchangeably, but a useful distinction is:

• Data: Recorded values or symbols
• Information: Data interpreted within context
• Knowledge: Understanding that supports explanation or action

Suppose a server reports response times of 120, 135, and 910.

Those numbers become information when we know they are milliseconds for three requests. They contribute to knowledge when analysis shows that image uploads consistently cause the slow responses and an engineer understands why.

The boundaries are not rigid. The important point is that storing values does not automatically produce understanding.

Measurements are constructed

Many data values come from measurement. Measurement connects an observable phenomenon to a scale or category.

A thermometer converts a physical condition into a number. A survey converts a person's response into selected categories. Website analytics convert events such as clicks and page loads into records.

Each process can introduce error:

• Instruments have limited precision.
• People misunderstand questions.
• Sensors drift.
• Events fail to send.
• Definitions change over time.
• Sampling misses part of the population.

Data does not become neutral merely because it is numeric. The collection method shapes what the numbers can support.

Types constrain interpretation

Software commonly distinguishes types such as:

• Integer
• Decimal
• Boolean
• Text
• Date and time
• Geographic coordinate
• Binary file
• Category

A date stored as plain text may be difficult to sort or validate. A price stored as a floating-point value may introduce rounding surprises. An identifier that looks numeric may belong as text because arithmetic on it has no meaning.

Choosing a type declares which operations make sense. Data type is part of context, not just a storage detail.

Data can be qualitative or quantitative

Quantitative data represents amounts or counts, such as height, duration, revenue, or number of visits.

Qualitative data represents qualities or categories, such as color, product type, interview themes, or written feedback.

Qualitative does not mean unstructured or unimportant. A category can be encoded consistently, and a paragraph can contain evidence that a single score misses.

Converting rich observations into numbers can make comparison easier, but it may discard nuance. Good analysis remembers what was compressed.

Missing data has meaning

A blank value can mean several different things:

• The property does not apply.
• The value is unknown.
• Collection failed.
• The user declined to answer.
• The value has not arrived yet.
• The system lost it.

Treating all blanks as zero can create false conclusions. Zero sales is different from no sales report.

Systems often use null, special codes, or separate status fields to represent missingness. The chosen convention must be documented and handled consistently.

Data quality is fit for purpose

High-quality data is not simply data without spelling mistakes. Useful dimensions include:

• Accuracy: Does it reflect reality closely enough?
• Completeness: Are required values present?
• Consistency: Do sources agree and use the same definitions?
• Timeliness: Is it current enough for the decision?
• Validity: Does it follow expected rules and formats?
• Uniqueness: Are duplicate records controlled?

Quality depends on purpose. A location rounded to the nearest kilometer may be enough for weather forecasting but unsafe for directing an ambulance.

Raw data is never entirely raw

People sometimes call original collected values raw data. The term is useful, but even early data has already passed through choices:

• What was observed
• When collection began and ended
• Which sensor or form was used
• How values were encoded
• Which events were filtered out

Calling data raw should not hide its history. Provenance, or the record of where data came from and how it changed, helps people judge whether it can answer a question.

Data has a lifecycle

Data usually moves through stages:

1. Creation or collection
2. Validation
3. Storage
4. Transformation
5. Use and sharing
6. Archiving
7. Deletion

Responsibilities exist at every stage. Collection needs consent and purpose. Storage needs protection. Transformation needs traceability. Sharing needs access control. Deletion needs retention rules and reliable execution.

Keeping data forever is not automatically valuable. Old, undocumented, sensitive data can become cost and risk.

More data is not always better

Larger datasets can improve analysis, but only when the added data is relevant and trustworthy.

Millions of biased observations can produce a precisely wrong conclusion. Collecting extra personal data "just in case" increases security and privacy exposure. Repeated copies can make it unclear which version is authoritative.

A stronger question than "How much data can we collect?" is "What data is necessary to support this purpose, and how will we maintain its meaning?"

Knowledge check

1. Why is data a representation rather than the thing itself?
2. What context would you need to interpret the number 42?
3. How can the collection method influence data?
4. Why should missing data not automatically be treated as zero?
5. Give one example where data quality depends on its intended use.

The one idea to remember

Data is a recorded representation. Context, definitions, collection methods, and quality determine what that representation means and whether it can responsibly support a decision.