Can a continuous variable have a limited range?

Yes; a uniform variable can be confined to a finite interval.

Can an exact observed value have probability zero?

Yes in the idealized model; probability is distributed over infinitely many values.

Does converting seconds to minutes preserve continuity?

A nonzero linear change of units preserves this distributional property, although the density's numerical scale changes.

AI Glossary / Statistics and probability / Intermediate

Continuous random variable

Definition

A continuous random variable represents an uncertain numerical quantity through a distribution that assigns zero probability to each individual value. In the common density-based setting, probabilities are calculated for intervals by integrating a probability density function. Examples include modeled durations, lengths, and measurement errors. The variable is the uncertain quantity; its distribution specifies how probability is allocated across possible outcomes.

What it means

Before observing a task, its completion time is uncertain. A random variable T can represent that time in minutes. After the task finishes, the recorded number is one observed realization of T. The observation and the variable's probability model serve different roles.
Continuous modeling allows values across a continuum rather than restricting the quantity to a countable list of outcomes with probability masses. It does not require every real number to be possible, nor does it mean every value is equally likely. A duration model may allow only nonnegative values and favor some intervals over others.

How it works

Define the measurement carefully: when timing begins, when it ends, which units are used, and which population or process is represented. Then specify or estimate a distribution. For a density-based model, P(a < T <= b) equals the area under its density from a to b. The cumulative distribution function summarizes the probability of falling at or below each threshold.
A density can vary across the range and need not be bounded by one. Probabilities concern areas, so they also depend on interval width. Exact endpoints do not change interval probabilities when each endpoint has zero probability.
Real instruments round and sometimes censor measurements. A displayed three minutes may represent an interval around three, while a timeout may indicate only that the duration exceeded a limit. These are different observation mechanisms and should be modeled accordingly. Repeated displayed values can occur even when the underlying quantity is modeled continuously.
A mathematical random variable is a function from outcomes to numbers; the continuous label describes its probability distribution. Some advanced continuous distributions lack an ordinary density, although applied introductions generally focus on the density-based case. An arbitrary real-valued variable is not necessarily continuous: it may be discrete or combine discrete and continuous components.
Check the model against the process. A continuous approximation can be useful without claiming that an instrument measures infinitely many decimal places or that every physical process is literally continuous.

Three practical examples

A support team models the time until an agent responds, with a separate rule for unanswered requests.
An engineering team models deviations of manufactured lengths from a target dimension.
An AI forecasting system treats a future measured temperature as an uncertain numerical quantity and estimates probabilities of exceeding operating thresholds.

Think of it this way

Think of a random variable as a labeled measurement waiting to be revealed. Its distribution is the map of uncertainty around that measurement. Reading a thermometer supplies one value; the map describes the chances of ranges of values before that reading is known.

Worked example

Let T represent a hypothetical waiting time in minutes, modeled by an exponential distribution with mean three. Its density is f(t) = (1/3) exp(-t/3) for t >= 0, and its cumulative distribution is F(t) = 1 - exp(-t/3).
The probability of waiting at most six minutes is F(6) = 1 - exp(-2), approximately 0.864665. The probability of waiting between three and six minutes is F(6) - F(3) = exp(-1) - exp(-2), approximately 0.232544.
The 90th percentile solves 1 - exp(-t/3) = 0.90, giving t = -3 ln(0.10), approximately 6.9078 minutes. This threshold describes the modeled distribution, not a promise for every individual wait.
P(T = 3) is zero. If a device rounds to the nearest minute, however, a displayed three corresponds to 2.5 <= T < 3.5. That probability is exp(-2.5/3) - exp(-3.5/3), approximately 0.123195.
The example assumes this exponential model for illustration. A three-minute mean by itself does not imply exponential waiting times; other distributions with the same mean can give different answers to all these probability questions.

Illustrative case study

Fictional case: Cedar Analytics studies service latency recorded to the nearest second. A manager assumes that repeated values mean the underlying latency cannot be modeled continuously. Analysts explain the distinction between a duration and its rounded recording, then examine the measurement process.
They also find that timeouts are stored as a fixed maximum value. Treating these as completed durations would distort the upper tail, so the team preserves their censored status. It compares candidate models and checks whether latency differs across workloads. The team gains a more defensible description of uncertainty without claiming that a continuous model automatically fits the system.

Common misconceptions and mistakes

A decimal number does not automatically identify a continuous random variable: prices restricted to fixed increments remain discrete on that scale. Conversely, integer recordings can represent rounded continuous measurements. Zero probability for an exact value does not mean no observation can occur. Finally, continuity does not imply independence between observations, a normal distribution, or a finite mean and variance.

Business use

This concept helps teams define prediction targets, choose statistical models, and interpret uncertainty statements. Specify the underlying quantity separately from how it is recorded. Include units and observation limits in data documentation. When translating model output into a service threshold or operating decision, use interval or tail probabilities that correspond to the actual business question.

Check your understanding

In the worked example, is a displayed three-minute wait the same event as T = 3 exactly?

Answer: No; rounding represents an interval.

Does a mean of three minutes identify the distribution?

Answer: No.

Would recording all waits above ten minutes as ten preserve their exact durations?

Answer: No; it introduces an observation limit that must be documented.

Related terms

Continuous probability distribution
Random variable
Realization
Cumulative distribution function
Censoring
probability density
measurement precision
rounding
exponential distribution

Learn next

Practice defining an underlying variable and its observation mechanism for a sensor dataset. Then compare interval probabilities, quantiles, and rounded-value frequencies under candidate continuous models, paying particular attention to long tails and observation limits.

Sources

Hossein Pishro-Nik, Introduction to Probability, Statistics, and Random Processes, Probability Density Function: https://www.probabilitycourse.com/chapter4/4_1_1_pdf.php KNIME, Data Science Glossary: https://www.knime.com/blog/data-science-glossary

Previous: Continuous probability distribution Next: Continuous-time normalizing flow