Can a discrete variable have decimal values?

Yes, when its permitted values are distinct specified possibilities, such as scores in half-point steps.

Can a discrete variable have infinitely many possible values?

Yes. A mathematical count model may allow any nonnegative integer, a countably infinite set.

Does an integer-valued database field prove the underlying variable is discrete?

No. It may store a rounded continuous measurement or a categorical code.

AI Glossary / Data science and analytics / Beginner

Discrete variable

Definition

A discrete variable takes values from a finite or countably infinite set of distinct possibilities. In introductory data analysis, the term often refers to numerical counts or scores with specified steps, such as the number of requests received or a rating recorded in half-point increments. Discrete values need not be equally likely, and the variable's possible values need not be whole numbers.

What it means

Counting completed orders produces values such as 0, 1, 2, and 3; there cannot be 2.4 completed orders in one observation under that definition. A scoring system allowing 0, 0.5, 1, and 1.5 is also discrete despite using decimals. The distinction concerns the set of allowable values. Numeric-looking identifiers, such as account numbers, are usually category labels rather than measurements for which ordinary arithmetic has a useful interpretation.

How it works

Define the observation unit and list or describe the allowed values. Record their frequencies and, when using a probability model, assign probability mass to individual values. Different discrete models suit different mechanisms: a yes-or-no result, a fixed number of trials, and a count of arrivals are not interchangeable. Preserve meaningful zero values and distinguish them from missing records. If a continuous quantity is rounded or grouped during recording, document that process rather than assuming the stored representation captures the underlying quantity without loss.

Three practical examples

Service operations: the number of incoming tickets per hour is a discrete count, with the hour and duplicate-handling rule clearly defined.
Product testing: the number of failed checks out of a fixed set is a discrete outcome with a known upper bound.
Learning assessment: a quiz score awarded in half-point steps is discrete even though some permitted scores contain a decimal fraction.

Think of it this way

Think of a staircase whose steps mark the values you are allowed to occupy. A count has one-unit steps, while a scoring scale might have half-unit steps. The analogy conveys separate possibilities. It does not mean all discrete variables have equally spaced steps, a fixed maximum, or the same chance of landing on each value.

Worked example

A fictional help desk records incoming ticket counts for five hours: 0, 1, 1, 3, and 5. The total is 10 and the mean is 10/5 = 2 tickets per hour. Zero appears once, one appears twice, three appears once, and five appears once. No observed hour has exactly two tickets, but two is still a valid average.
Now suppose the fifth hour's record is missing rather than equal to five. The known total is 5 over four observed hours, giving an observed-hour mean of 1.25. Replacing the missing value with zero would instead produce a five-hour mean of 1, but that calculation would assume there were no tickets in an hour whose count is unknown. The original variable remains discrete even when its average is fractional. The example shows why the type of individual observations, the type of a summary, and the handling of missing data are separate questions.

Illustrative case study

In a fictional subscription service, an analyst prepares a model of monthly support contacts. Some rows contain zero contacts, while others lack a complete month's observation because an account was newly opened. The team separates genuine zero counts from incomplete exposure and records how much time each account was observable. It also checks whether multiple messages belong to one contact or several, because the counting rule changes the outcome. The project illustrates careful count definition before modeling; it does not assume a particular probability distribution or claim a performance improvement.

Common misconceptions and mistakes

Discrete does not mean uniformly distributed, categorical codes do not automatically become quantities, and a decimal value is not automatically continuous. A mean of 2.6 requests is compatible with integer-valued observations. Conversely, recording duration to whole seconds does not make the underlying passage of time a count of events. Discrete counts may be zero-inflated, bounded, dependent, or highly variable; choosing a model solely because a column contains integers can ignore those properties. A missing value is not an additional numerical outcome unless explicitly defined as such.

Business use

For business reporting and AI preparation, specify the counting or scoring rules before selecting a chart or model. Include the exposure period when comparing counts across customers, locations, or time windows. A branch with more visitors may naturally have more complaints even if its complaint rate is lower. Validate impossible values, but investigate unusual valid values rather than automatically deleting them. When a model produces a fractional prediction for a count, identify whether it represents an expected count or an intended actionable integer; rounding choices belong to the decision process.

Check your understanding

The average number of tickets across several hours is 1.25. Does that make ticket count continuous?

Answer: No. Each hourly count remains an integer. Averaging discrete observations can produce a fractional summary without changing the nature of the original variable.

Related terms

Learn next

Practice classifying counts, fixed-step scores, category codes, and rounded measurements. Build a frequency table from a small count dataset, including zeros and missing observations separately. Then study discrete probability models and rates so you can compare observations with different exposure periods appropriately.

Sources

Statistics Canada, types of variables: https://www150.statcan.gc.ca/n1/edu/power-pouvoir/ch8/5214817-eng.htm . Statistics Canada, frequency distributions: https://www150.statcan.gc.ca/n1/edu/power-pouvoir/ch8/5214814-eng.htm . Primary educational excerpts were consulted for numerical variable types, stepped scores, and frequency summaries. The definition also includes countably infinite discrete supports used in probability theory. Examples and calculations are original; editorial review remains pending.

Previous: Discrete Uniform Distribution Next: Discrete variational autoencoder